arch/powerpc/kvm/powerpc.c

   1 /*
   2  * This program is free software; you can redistribute it and/or modify
   3  * it under the terms of the GNU General Public License, version 2, as
   4  * published by the Free Software Foundation.
   5  *
   6  * This program is distributed in the hope that it will be useful,
   7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
   8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   9  * GNU General Public License for more details.
  10  *
  11  * You should have received a copy of the GNU General Public License
  12  * along with this program; if not, write to the Free Software
  13  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  14  *
  15  * Copyright IBM Corp. 2007
  16  *
  17  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
  18  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
  19  */
  20
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/hrtimer.h>
  26 #include <linux/sched/signal.h>
  27 #include <linux/fs.h>
  28 #include <linux/slab.h>
  29 #include <linux/file.h>
  30 #include <linux/module.h>
  31 #include <linux/irqbypass.h>
  32 #include <linux/kvm_irqfd.h>
  33 #include <asm/cputable.h>
  34 #include <linux/uaccess.h>
  35 #include <asm/kvm_ppc.h>
  36 #include <asm/tlbflush.h>
  37 #include <asm/cputhreads.h>
  38 #include <asm/irqflags.h>
  39 #include <asm/iommu.h>
  40 #include <asm/switch_to.h>
  41 #include <asm/xive.h>
  42 #ifdef CONFIG_PPC_PSERIES
  43 #include <asm/hvcall.h>
  44 #include <asm/plpar_wrappers.h>
  45 #endif
  46
  47 #include "timing.h"
  48 #include "irq.h"
  49 #include "../mm/mmu_decl.h"
  50
  51 #define CREATE_TRACE_POINTS
  52 #include "trace.h"
  53
  54 struct kvmppc_ops *kvmppc_hv_ops;
  55 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
  56 struct kvmppc_ops *kvmppc_pr_ops;
  57 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
  58
  59
  60 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  61 {
  62         return !!(v->arch.pending_exceptions) || kvm_request_pending(v);
  63 }
  64
  65 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
  66 {
  67         return false;
  68 }
  69
  70 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  71 {
  72         return 1;
  73 }
  74
  75 /*
  76  * Common checks before entering the guest world.  Call with interrupts
  77  * disabled.
  78  *
  79  * returns:
  80  *
  81  * == 1 if we're ready to go into guest state
  82  * <= 0 if we need to go back to the host with return value
  83  */
  84 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  85 {
  86         int r;
  87
  88         WARN_ON(irqs_disabled());
  89         hard_irq_disable();
  90
  91         while (true) {
  92                 if (need_resched()) {
  93                         local_irq_enable();
  94                         cond_resched();
  95                         hard_irq_disable();
  96                         continue;
  97                 }
  98
  99                 if (signal_pending(current)) {
 100                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
 101                         vcpu->run->exit_reason = KVM_EXIT_INTR;
 102                         r = -EINTR;
 103                         break;
 104                 }
 105
 106                 vcpu->mode = IN_GUEST_MODE;
 107
 108                 /*
 109                  * Reading vcpu->requests must happen after setting vcpu->mode,
 110                  * so we don't miss a request because the requester sees
 111                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
 112                  * before next entering the guest (and thus doesn't IPI).
 113                  * This also orders the write to mode from any reads
 114                  * to the page tables done while the VCPU is running.
 115                  * Please see the comment in kvm_flush_remote_tlbs.
 116                  */
 117                 smp_mb();
 118
 119                 if (kvm_request_pending(vcpu)) {
 120                         /* Make sure we process requests preemptable */
 121                         local_irq_enable();
 122                         trace_kvm_check_requests(vcpu);
 123                         r = kvmppc_core_check_requests(vcpu);
 124                         hard_irq_disable();
 125                         if (r > 0)
 126                                 continue;
 127                         break;
 128                 }
 129
 130                 if (kvmppc_core_prepare_to_enter(vcpu)) {
 131                         /* interrupts got enabled in between, so we
 132                            are back at square 1 */
 133                         continue;
 134                 }
 135
 136                 guest_enter_irqoff();
 137                 return 1;
 138         }
 139
 140         /* return to host */
 141         local_irq_enable();
 142         return r;
 143 }
 144 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
 145
 146 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 147 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
 148 {
 149         struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
 150         int i;
 151
 152         shared->sprg0 = swab64(shared->sprg0);
 153         shared->sprg1 = swab64(shared->sprg1);
 154         shared->sprg2 = swab64(shared->sprg2);
 155         shared->sprg3 = swab64(shared->sprg3);
 156         shared->srr0 = swab64(shared->srr0);
 157         shared->srr1 = swab64(shared->srr1);
 158         shared->dar = swab64(shared->dar);
 159         shared->msr = swab64(shared->msr);
 160         shared->dsisr = swab32(shared->dsisr);
 161         shared->int_pending = swab32(shared->int_pending);
 162         for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
 163                 shared->sr[i] = swab32(shared->sr[i]);
 164 }
 165 #endif
 166
 167 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
 168 {
 169         int nr = kvmppc_get_gpr(vcpu, 11);
 170         int r;
 171         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
 172         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
 173         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
 174         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
 175         unsigned long r2 = 0;
 176
 177         if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
 178                 /* 32 bit mode */
 179                 param1 &= 0xffffffff;
 180                 param2 &= 0xffffffff;
 181                 param3 &= 0xffffffff;
 182                 param4 &= 0xffffffff;
 183         }
 184
 185         switch (nr) {
 186         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
 187         {
 188 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 189                 /* Book3S can be little endian, find it out here */
 190                 int shared_big_endian = true;
 191                 if (vcpu->arch.intr_msr & MSR_LE)
 192                         shared_big_endian = false;
 193                 if (shared_big_endian != vcpu->arch.shared_big_endian)
 194                         kvmppc_swab_shared(vcpu);
 195                 vcpu->arch.shared_big_endian = shared_big_endian;
 196 #endif
 197
 198                 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
 199                         /*
 200                          * Older versions of the Linux magic page code had
 201                          * a bug where they would map their trampoline code
 202                          * NX. If that's the case, remove !PR NX capability.
 203                          */
 204                         vcpu->arch.disable_kernel_nx = true;
 205                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 206                 }
 207
 208                 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
 209                 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
 210
 211 #ifdef CONFIG_PPC_64K_PAGES
 212                 /*
 213                  * Make sure our 4k magic page is in the same window of a 64k
 214                  * page within the guest and within the host's page.
 215                  */
 216                 if ((vcpu->arch.magic_page_pa & 0xf000) !=
 217                     ((ulong)vcpu->arch.shared & 0xf000)) {
 218                         void *old_shared = vcpu->arch.shared;
 219                         ulong shared = (ulong)vcpu->arch.shared;
 220                         void *new_shared;
 221
 222                         shared &= PAGE_MASK;
 223                         shared |= vcpu->arch.magic_page_pa & 0xf000;
 224                         new_shared = (void*)shared;
 225                         memcpy(new_shared, old_shared, 0x1000);
 226                         vcpu->arch.shared = new_shared;
 227                 }
 228 #endif
 229
 230                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 231
 232                 r = EV_SUCCESS;
 233                 break;
 234         }
 235         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 236                 r = EV_SUCCESS;
 237 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 238                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 239 #endif
 240
 241                 /* Second return value is in r4 */
 242                 break;
 243         case EV_HCALL_TOKEN(EV_IDLE):
 244                 r = EV_SUCCESS;
 245                 kvm_vcpu_block(vcpu);
 246                 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
 247                 break;
 248         default:
 249                 r = EV_UNIMPLEMENTED;
 250                 break;
 251         }
 252
 253         kvmppc_set_gpr(vcpu, 4, r2);
 254
 255         return r;
 256 }
 257 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
 258
 259 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
 260 {
 261         int r = false;
 262
 263         /* We have to know what CPU to virtualize */
 264         if (!vcpu->arch.pvr)
 265                 goto out;
 266
 267         /* PAPR only works with book3s_64 */
 268         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
 269                 goto out;
 270
 271         /* HV KVM can only do PAPR mode for now */
 272         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
 273                 goto out;
 274
 275 #ifdef CONFIG_KVM_BOOKE_HV
 276         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 277                 goto out;
 278 #endif
 279
 280         r = true;
 281
 282 out:
 283         vcpu->arch.sane = r;
 284         return r ? 0 : -EINVAL;
 285 }
 286 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
 287
 288 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 289 {
 290         enum emulation_result er;
 291         int r;
 292
 293         er = kvmppc_emulate_loadstore(vcpu);
 294         switch (er) {
 295         case EMULATE_DONE:
 296                 /* Future optimization: only reload non-volatiles if they were
 297                  * actually modified. */
 298                 r = RESUME_GUEST_NV;
 299                 break;
 300         case EMULATE_AGAIN:
 301                 r = RESUME_GUEST;
 302                 break;
 303         case EMULATE_DO_MMIO:
 304                 run->exit_reason = KVM_EXIT_MMIO;
 305                 /* We must reload nonvolatiles because "update" load/store
 306                  * instructions modify register state. */
 307                 /* Future optimization: only reload non-volatiles if they were
 308                  * actually modified. */
 309                 r = RESUME_HOST_NV;
 310                 break;
 311         case EMULATE_FAIL:
 312         {
 313                 u32 last_inst;
 314
 315                 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
 316                 /* XXX Deliver Program interrupt to guest. */
 317                 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
 318                 r = RESUME_HOST;
 319                 break;
 320         }
 321         default:
 322                 WARN_ON(1);
 323                 r = RESUME_GUEST;
 324         }
 325
 326         return r;
 327 }
 328 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
 329
 330 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 331               bool data)
 332 {
 333         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 334         struct kvmppc_pte pte;
 335         int r;
 336
 337         vcpu->stat.st++;
 338
 339         r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 340                          XLATE_WRITE, &pte);
 341         if (r < 0)
 342                 return r;
 343
 344         *eaddr = pte.raddr;
 345
 346         if (!pte.may_write)
 347                 return -EPERM;
 348
 349         /* Magic page override */
 350         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 351             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 352             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 353                 void *magic = vcpu->arch.shared;
 354                 magic += pte.eaddr & 0xfff;
 355                 memcpy(magic, ptr, size);
 356                 return EMULATE_DONE;
 357         }
 358
 359         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 360                 return EMULATE_DO_MMIO;
 361
 362         return EMULATE_DONE;
 363 }
 364 EXPORT_SYMBOL_GPL(kvmppc_st);
 365
 366 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 367                       bool data)
 368 {
 369         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 370         struct kvmppc_pte pte;
 371         int rc;
 372
 373         vcpu->stat.ld++;
 374
 375         rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 376                           XLATE_READ, &pte);
 377         if (rc)
 378                 return rc;
 379
 380         *eaddr = pte.raddr;
 381
 382         if (!pte.may_read)
 383                 return -EPERM;
 384
 385         if (!data && !pte.may_execute)
 386                 return -ENOEXEC;
 387
 388         /* Magic page override */
 389         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 390             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 391             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 392                 void *magic = vcpu->arch.shared;
 393                 magic += pte.eaddr & 0xfff;
 394                 memcpy(ptr, magic, size);
 395                 return EMULATE_DONE;
 396         }
 397
 398         if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
 399                 return EMULATE_DO_MMIO;
 400
 401         return EMULATE_DONE;
 402 }
 403 EXPORT_SYMBOL_GPL(kvmppc_ld);
 404
 405 int kvm_arch_hardware_enable(void)
 406 {
 407         return 0;
 408 }
 409
 410 int kvm_arch_hardware_setup(void)
 411 {
 412         return 0;
 413 }
 414
 415 void kvm_arch_check_processor_compat(void *rtn)
 416 {
 417         *(int *)rtn = kvmppc_core_check_processor_compat();
 418 }
 419
 420 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 421 {
 422         struct kvmppc_ops *kvm_ops = NULL;
 423         /*
 424          * if we have both HV and PR enabled, default is HV
 425          */
 426         if (type == 0) {
 427                 if (kvmppc_hv_ops)
 428                         kvm_ops = kvmppc_hv_ops;
 429                 else
 430                         kvm_ops = kvmppc_pr_ops;
 431                 if (!kvm_ops)
 432                         goto err_out;
 433         } else  if (type == KVM_VM_PPC_HV) {
 434                 if (!kvmppc_hv_ops)
 435                         goto err_out;
 436                 kvm_ops = kvmppc_hv_ops;
 437         } else if (type == KVM_VM_PPC_PR) {
 438                 if (!kvmppc_pr_ops)
 439                         goto err_out;
 440                 kvm_ops = kvmppc_pr_ops;
 441         } else
 442                 goto err_out;
 443
 444         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
 445                 return -ENOENT;
 446
 447         kvm->arch.kvm_ops = kvm_ops;
 448         return kvmppc_core_init_vm(kvm);
 449 err_out:
 450         return -EINVAL;
 451 }
 452
 453 bool kvm_arch_has_vcpu_debugfs(void)
 454 {
 455         return false;
 456 }
 457
 458 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 459 {
 460         return 0;
 461 }
 462
 463 void kvm_arch_destroy_vm(struct kvm *kvm)
 464 {
 465         unsigned int i;
 466         struct kvm_vcpu *vcpu;
 467
 468 #ifdef CONFIG_KVM_XICS
 469         /*
 470          * We call kick_all_cpus_sync() to ensure that all
 471          * CPUs have executed any pending IPIs before we
 472          * continue and free VCPUs structures below.
 473          */
 474         if (is_kvmppc_hv_enabled(kvm))
 475                 kick_all_cpus_sync();
 476 #endif
 477
 478         kvm_for_each_vcpu(i, vcpu, kvm)
 479                 kvm_arch_vcpu_free(vcpu);
 480
 481         mutex_lock(&kvm->lock);
 482         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 483                 kvm->vcpus[i] = NULL;
 484
 485         atomic_set(&kvm->online_vcpus, 0);
 486
 487         kvmppc_core_destroy_vm(kvm);
 488
 489         mutex_unlock(&kvm->lock);
 490
 491         /* drop the module reference */
 492         module_put(kvm->arch.kvm_ops->owner);
 493 }
 494
 495 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 496 {
 497         int r;
 498         /* Assume we're using HV mode when the HV module is loaded */
 499         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
 500
 501         if (kvm) {
 502                 /*
 503                  * Hooray - we know which VM type we're running on. Depend on
 504                  * that rather than the guess above.
 505                  */
 506                 hv_enabled = is_kvmppc_hv_enabled(kvm);
 507         }
 508
 509         switch (ext) {
 510 #ifdef CONFIG_BOOKE
 511         case KVM_CAP_PPC_BOOKE_SREGS:
 512         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 513         case KVM_CAP_PPC_EPR:
 514 #else
 515         case KVM_CAP_PPC_SEGSTATE:
 516         case KVM_CAP_PPC_HIOR:
 517         case KVM_CAP_PPC_PAPR:
 518 #endif
 519         case KVM_CAP_PPC_UNSET_IRQ:
 520         case KVM_CAP_PPC_IRQ_LEVEL:
 521         case KVM_CAP_ENABLE_CAP:
 522         case KVM_CAP_ENABLE_CAP_VM:
 523         case KVM_CAP_ONE_REG:
 524         case KVM_CAP_IOEVENTFD:
 525         case KVM_CAP_DEVICE_CTRL:
 526         case KVM_CAP_IMMEDIATE_EXIT:
 527                 r = 1;
 528                 break;
 529         case KVM_CAP_PPC_PAIRED_SINGLES:
 530         case KVM_CAP_PPC_OSI:
 531         case KVM_CAP_PPC_GET_PVINFO:
 532 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 533         case KVM_CAP_SW_TLB:
 534 #endif
 535                 /* We support this only for PR */
 536                 r = !hv_enabled;
 537                 break;
 538 #ifdef CONFIG_KVM_MPIC
 539         case KVM_CAP_IRQ_MPIC:
 540                 r = 1;
 541                 break;
 542 #endif
 543
 544 #ifdef CONFIG_PPC_BOOK3S_64
 545         case KVM_CAP_SPAPR_TCE:
 546         case KVM_CAP_SPAPR_TCE_64:
 547                 /* fallthrough */
 548         case KVM_CAP_SPAPR_TCE_VFIO:
 549         case KVM_CAP_PPC_RTAS:
 550         case KVM_CAP_PPC_FIXUP_HCALL:
 551         case KVM_CAP_PPC_ENABLE_HCALL:
 552 #ifdef CONFIG_KVM_XICS
 553         case KVM_CAP_IRQ_XICS:
 554 #endif
 555         case KVM_CAP_PPC_GET_CPU_CHAR:
 556                 r = 1;
 557                 break;
 558
 559         case KVM_CAP_PPC_ALLOC_HTAB:
 560                 r = hv_enabled;
 561                 break;
 562 #endif /* CONFIG_PPC_BOOK3S_64 */
 563 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 564         case KVM_CAP_PPC_SMT:
 565                 r = 0;
 566                 if (kvm) {
 567                         if (kvm->arch.emul_smt_mode > 1)
 568                                 r = kvm->arch.emul_smt_mode;
 569                         else
 570                                 r = kvm->arch.smt_mode;
 571                 } else if (hv_enabled) {
 572                         if (cpu_has_feature(CPU_FTR_ARCH_300))
 573                                 r = 1;
 574                         else
 575                                 r = threads_per_subcore;
 576                 }
 577                 break;
 578         case KVM_CAP_PPC_SMT_POSSIBLE:
 579                 r = 1;
 580                 if (hv_enabled) {
 581                         if (!cpu_has_feature(CPU_FTR_ARCH_300))
 582                                 r = ((threads_per_subcore << 1) - 1);
 583                         else
 584                                 /* P9 can emulate dbells, so allow any mode */
 585                                 r = 8 | 4 | 2 | 1;
 586                 }
 587                 break;
 588         case KVM_CAP_PPC_RMA:
 589                 r = 0;
 590                 break;
 591         case KVM_CAP_PPC_HWRNG:
 592                 r = kvmppc_hwrng_present();
 593                 break;
 594         case KVM_CAP_PPC_MMU_RADIX:
 595                 r = !!(hv_enabled && radix_enabled());
 596                 break;
 597         case KVM_CAP_PPC_MMU_HASH_V3:
 598                 r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300));
 599                 break;
 600 #endif
 601         case KVM_CAP_SYNC_MMU:
 602 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 603                 r = hv_enabled;
 604 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 605                 r = 1;
 606 #else
 607                 r = 0;
 608 #endif
 609                 break;
 610 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 611         case KVM_CAP_PPC_HTAB_FD:
 612                 r = hv_enabled;
 613                 break;
 614 #endif
 615         case KVM_CAP_NR_VCPUS:
 616                 /*
 617                  * Recommending a number of CPUs is somewhat arbitrary; we
 618                  * return the number of present CPUs for -HV (since a host
 619                  * will have secondary threads "offline"), and for other KVM
 620                  * implementations just count online CPUs.
 621                  */
 622                 if (hv_enabled)
 623                         r = num_present_cpus();
 624                 else
 625                         r = num_online_cpus();
 626                 break;
 627         case KVM_CAP_NR_MEMSLOTS:
 628                 r = KVM_USER_MEM_SLOTS;
 629                 break;
 630         case KVM_CAP_MAX_VCPUS:
 631                 r = KVM_MAX_VCPUS;
 632                 break;
 633 #ifdef CONFIG_PPC_BOOK3S_64
 634         case KVM_CAP_PPC_GET_SMMU_INFO:
 635                 r = 1;
 636                 break;
 637         case KVM_CAP_SPAPR_MULTITCE:
 638                 r = 1;
 639                 break;
 640         case KVM_CAP_SPAPR_RESIZE_HPT:
 641                 r = !!hv_enabled;
 642                 break;
 643 #endif
 644 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 645         case KVM_CAP_PPC_FWNMI:
 646                 r = hv_enabled;
 647                 break;
 648 #endif
 649 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 650         case KVM_CAP_PPC_HTM:
 651                 r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) ||
 652                      (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST));
 653                 break;
 654 #endif
 655         default:
 656                 r = 0;
 657                 break;
 658         }
 659         return r;
 660
 661 }
 662
 663 long kvm_arch_dev_ioctl(struct file *filp,
 664                         unsigned int ioctl, unsigned long arg)
 665 {
 666         return -EINVAL;
 667 }
 668
 669 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 670                            struct kvm_memory_slot *dont)
 671 {
 672         kvmppc_core_free_memslot(kvm, free, dont);
 673 }
 674
 675 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 676                             unsigned long npages)
 677 {
 678         return kvmppc_core_create_memslot(kvm, slot, npages);
 679 }
 680
 681 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 682                                    struct kvm_memory_slot *memslot,
 683                                    const struct kvm_userspace_memory_region *mem,
 684                                    enum kvm_mr_change change)
 685 {
 686         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 687 }
 688
 689 void kvm_arch_commit_memory_region(struct kvm *kvm,
 690                                    const struct kvm_userspace_memory_region *mem,
 691                                    const struct kvm_memory_slot *old,
 692                                    const struct kvm_memory_slot *new,
 693                                    enum kvm_mr_change change)
 694 {
 695         kvmppc_core_commit_memory_region(kvm, mem, old, new);
 696 }
 697
 698 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 699                                    struct kvm_memory_slot *slot)
 700 {
 701         kvmppc_core_flush_memslot(kvm, slot);
 702 }
 703
 704 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 705 {
 706         struct kvm_vcpu *vcpu;
 707         vcpu = kvmppc_core_vcpu_create(kvm, id);
 708         if (!IS_ERR(vcpu)) {
 709                 vcpu->arch.wqp = &vcpu->wq;
 710                 kvmppc_create_vcpu_debugfs(vcpu, id);
 711         }
 712         return vcpu;
 713 }
 714
 715 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 716 {
 717 }
 718
 719 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 720 {
 721         /* Make sure we're not using the vcpu anymore */
 722         hrtimer_cancel(&vcpu->arch.dec_timer);
 723
 724         kvmppc_remove_vcpu_debugfs(vcpu);
 725
 726         switch (vcpu->arch.irq_type) {
 727         case KVMPPC_IRQ_MPIC:
 728                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 729                 break;
 730         case KVMPPC_IRQ_XICS:
 731                 if (xive_enabled())
 732                         kvmppc_xive_cleanup_vcpu(vcpu);
 733                 else
 734                         kvmppc_xics_free_icp(vcpu);
 735                 break;
 736         }
 737
 738         kvmppc_core_vcpu_free(vcpu);
 739 }
 740
 741 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 742 {
 743         kvm_arch_vcpu_free(vcpu);
 744 }
 745
 746 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 747 {
 748         return kvmppc_core_pending_dec(vcpu);
 749 }
 750
 751 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 752 {
 753         struct kvm_vcpu *vcpu;
 754
 755         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 756         kvmppc_decrementer_func(vcpu);
 757
 758         return HRTIMER_NORESTART;
 759 }
 760
 761 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 762 {
 763         int ret;
 764
 765         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 766         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 767         vcpu->arch.dec_expires = get_tb();
 768
 769 #ifdef CONFIG_KVM_EXIT_TIMING
 770         mutex_init(&vcpu->arch.exit_timing_lock);
 771 #endif
 772         ret = kvmppc_subarch_vcpu_init(vcpu);
 773         return ret;
 774 }
 775
 776 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 777 {
 778         kvmppc_mmu_destroy(vcpu);
 779         kvmppc_subarch_vcpu_uninit(vcpu);
 780 }
 781
 782 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 783 {
 784 #ifdef CONFIG_BOOKE
 785         /*
 786          * vrsave (formerly usprg0) isn't used by Linux, but may
 787          * be used by the guest.
 788          *
 789          * On non-booke this is associated with Altivec and
 790          * is handled by code in book3s.c.
 791          */
 792         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 793 #endif
 794         kvmppc_core_vcpu_load(vcpu, cpu);
 795 }
 796
 797 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 798 {
 799         kvmppc_core_vcpu_put(vcpu);
 800 #ifdef CONFIG_BOOKE
 801         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 802 #endif
 803 }
 804
 805 /*
 806  * irq_bypass_add_producer and irq_bypass_del_producer are only
 807  * useful if the architecture supports PCI passthrough.
 808  * irq_bypass_stop and irq_bypass_start are not needed and so
 809  * kvm_ops are not defined for them.
 810  */
 811 bool kvm_arch_has_irq_bypass(void)
 812 {
 813         return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
 814                 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
 815 }
 816
 817 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
 818                                      struct irq_bypass_producer *prod)
 819 {
 820         struct kvm_kernel_irqfd *irqfd =
 821                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 822         struct kvm *kvm = irqfd->kvm;
 823
 824         if (kvm->arch.kvm_ops->irq_bypass_add_producer)
 825                 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
 826
 827         return 0;
 828 }
 829
 830 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
 831                                       struct irq_bypass_producer *prod)
 832 {
 833         struct kvm_kernel_irqfd *irqfd =
 834                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 835         struct kvm *kvm = irqfd->kvm;
 836
 837         if (kvm->arch.kvm_ops->irq_bypass_del_producer)
 838                 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
 839 }
 840
 841 #ifdef CONFIG_VSX
 842 static inline int kvmppc_get_vsr_dword_offset(int index)
 843 {
 844         int offset;
 845
 846         if ((index != 0) && (index != 1))
 847                 return -1;
 848
 849 #ifdef __BIG_ENDIAN
 850         offset =  index;
 851 #else
 852         offset = 1 - index;
 853 #endif
 854
 855         return offset;
 856 }
 857
 858 static inline int kvmppc_get_vsr_word_offset(int index)
 859 {
 860         int offset;
 861
 862         if ((index > 3) || (index < 0))
 863                 return -1;
 864
 865 #ifdef __BIG_ENDIAN
 866         offset = index;
 867 #else
 868         offset = 3 - index;
 869 #endif
 870         return offset;
 871 }
 872
 873 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu,
 874         u64 gpr)
 875 {
 876         union kvmppc_one_reg val;
 877         int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
 878         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 879
 880         if (offset == -1)
 881                 return;
 882
 883         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 884                 val.vval = VCPU_VSX_VR(vcpu, index);
 885                 val.vsxval[offset] = gpr;
 886                 VCPU_VSX_VR(vcpu, index) = val.vval;
 887         } else {
 888                 VCPU_VSX_FPR(vcpu, index, offset) = gpr;
 889         }
 890 }
 891
 892 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu,
 893         u64 gpr)
 894 {
 895         union kvmppc_one_reg val;
 896         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 897
 898         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 899                 val.vval = VCPU_VSX_VR(vcpu, index);
 900                 val.vsxval[0] = gpr;
 901                 val.vsxval[1] = gpr;
 902                 VCPU_VSX_VR(vcpu, index) = val.vval;
 903         } else {
 904                 VCPU_VSX_FPR(vcpu, index, 0) = gpr;
 905                 VCPU_VSX_FPR(vcpu, index, 1) = gpr;
 906         }
 907 }
 908
 909 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu,
 910         u32 gpr)
 911 {
 912         union kvmppc_one_reg val;
 913         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 914
 915         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 916                 val.vsx32val[0] = gpr;
 917                 val.vsx32val[1] = gpr;
 918                 val.vsx32val[2] = gpr;
 919                 val.vsx32val[3] = gpr;
 920                 VCPU_VSX_VR(vcpu, index) = val.vval;
 921         } else {
 922                 val.vsx32val[0] = gpr;
 923                 val.vsx32val[1] = gpr;
 924                 VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0];
 925                 VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0];
 926         }
 927 }
 928
 929 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu,
 930         u32 gpr32)
 931 {
 932         union kvmppc_one_reg val;
 933         int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
 934         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
 935         int dword_offset, word_offset;
 936
 937         if (offset == -1)
 938                 return;
 939
 940         if (vcpu->arch.mmio_vsx_tx_sx_enabled) {
 941                 val.vval = VCPU_VSX_VR(vcpu, index);
 942                 val.vsx32val[offset] = gpr32;
 943                 VCPU_VSX_VR(vcpu, index) = val.vval;
 944         } else {
 945                 dword_offset = offset / 2;
 946                 word_offset = offset % 2;
 947                 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset);
 948                 val.vsx32val[word_offset] = gpr32;
 949                 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0];
 950         }
 951 }
 952 #endif /* CONFIG_VSX */
 953
 954 #ifdef CONFIG_ALTIVEC
 955 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu,
 956                 int index, int element_size)
 957 {
 958         int offset;
 959         int elts = sizeof(vector128)/element_size;
 960
 961         if ((index < 0) || (index >= elts))
 962                 return -1;
 963
 964         if (kvmppc_need_byteswap(vcpu))
 965                 offset = elts - index - 1;
 966         else
 967                 offset = index;
 968
 969         return offset;
 970 }
 971
 972 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu,
 973                 int index)
 974 {
 975         return kvmppc_get_vmx_offset_generic(vcpu, index, 8);
 976 }
 977
 978 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu,
 979                 int index)
 980 {
 981         return kvmppc_get_vmx_offset_generic(vcpu, index, 4);
 982 }
 983
 984 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu,
 985                 int index)
 986 {
 987         return kvmppc_get_vmx_offset_generic(vcpu, index, 2);
 988 }
 989
 990 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu,
 991                 int index)
 992 {
 993         return kvmppc_get_vmx_offset_generic(vcpu, index, 1);
 994 }
 995
 996
 997 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu,
 998         u64 gpr)
 999 {
1000         union kvmppc_one_reg val;
1001         int offset = kvmppc_get_vmx_dword_offset(vcpu,
1002                         vcpu->arch.mmio_vmx_offset);
1003         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1004
1005         if (offset == -1)
1006                 return;
1007
1008         val.vval = VCPU_VSX_VR(vcpu, index);
1009         val.vsxval[offset] = gpr;
1010         VCPU_VSX_VR(vcpu, index) = val.vval;
1011 }
1012
1013 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu,
1014         u32 gpr32)
1015 {
1016         union kvmppc_one_reg val;
1017         int offset = kvmppc_get_vmx_word_offset(vcpu,
1018                         vcpu->arch.mmio_vmx_offset);
1019         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1020
1021         if (offset == -1)
1022                 return;
1023
1024         val.vval = VCPU_VSX_VR(vcpu, index);
1025         val.vsx32val[offset] = gpr32;
1026         VCPU_VSX_VR(vcpu, index) = val.vval;
1027 }
1028
1029 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu,
1030         u16 gpr16)
1031 {
1032         union kvmppc_one_reg val;
1033         int offset = kvmppc_get_vmx_hword_offset(vcpu,
1034                         vcpu->arch.mmio_vmx_offset);
1035         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1036
1037         if (offset == -1)
1038                 return;
1039
1040         val.vval = VCPU_VSX_VR(vcpu, index);
1041         val.vsx16val[offset] = gpr16;
1042         VCPU_VSX_VR(vcpu, index) = val.vval;
1043 }
1044
1045 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu,
1046         u8 gpr8)
1047 {
1048         union kvmppc_one_reg val;
1049         int offset = kvmppc_get_vmx_byte_offset(vcpu,
1050                         vcpu->arch.mmio_vmx_offset);
1051         int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK;
1052
1053         if (offset == -1)
1054                 return;
1055
1056         val.vval = VCPU_VSX_VR(vcpu, index);
1057         val.vsx8val[offset] = gpr8;
1058         VCPU_VSX_VR(vcpu, index) = val.vval;
1059 }
1060 #endif /* CONFIG_ALTIVEC */
1061
1062 #ifdef CONFIG_PPC_FPU
1063 static inline u64 sp_to_dp(u32 fprs)
1064 {
1065         u64 fprd;
1066
1067         preempt_disable();
1068         enable_kernel_fp();
1069         asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs)
1070              : "fr0");
1071         preempt_enable();
1072         return fprd;
1073 }
1074
1075 static inline u32 dp_to_sp(u64 fprd)
1076 {
1077         u32 fprs;
1078
1079         preempt_disable();
1080         enable_kernel_fp();
1081         asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd)
1082              : "fr0");
1083         preempt_enable();
1084         return fprs;
1085 }
1086
1087 #else
1088 #define sp_to_dp(x)     (x)
1089 #define dp_to_sp(x)     (x)
1090 #endif /* CONFIG_PPC_FPU */
1091
1092 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
1093                                       struct kvm_run *run)
1094 {
1095         u64 uninitialized_var(gpr);
1096
1097         if (run->mmio.len > sizeof(gpr)) {
1098                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
1099                 return;
1100         }
1101
1102         if (!vcpu->arch.mmio_host_swabbed) {
1103                 switch (run->mmio.len) {
1104                 case 8: gpr = *(u64 *)run->mmio.data; break;
1105                 case 4: gpr = *(u32 *)run->mmio.data; break;
1106                 case 2: gpr = *(u16 *)run->mmio.data; break;
1107                 case 1: gpr = *(u8 *)run->mmio.data; break;
1108                 }
1109         } else {
1110                 switch (run->mmio.len) {
1111                 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
1112                 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
1113                 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
1114                 case 1: gpr = *(u8 *)run->mmio.data; break;
1115                 }
1116         }
1117
1118         /* conversion between single and double precision */
1119         if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4))
1120                 gpr = sp_to_dp(gpr);
1121
1122         if (vcpu->arch.mmio_sign_extend) {
1123                 switch (run->mmio.len) {
1124 #ifdef CONFIG_PPC64
1125                 case 4:
1126                         gpr = (s64)(s32)gpr;
1127                         break;
1128 #endif
1129                 case 2:
1130                         gpr = (s64)(s16)gpr;
1131                         break;
1132                 case 1:
1133                         gpr = (s64)(s8)gpr;
1134                         break;
1135                 }
1136         }
1137
1138         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
1139         case KVM_MMIO_REG_GPR:
1140                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
1141                 break;
1142         case KVM_MMIO_REG_FPR:
1143                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1144                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP);
1145
1146                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1147                 break;
1148 #ifdef CONFIG_PPC_BOOK3S
1149         case KVM_MMIO_REG_QPR:
1150                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1151                 break;
1152         case KVM_MMIO_REG_FQPR:
1153                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
1154                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
1155                 break;
1156 #endif
1157 #ifdef CONFIG_VSX
1158         case KVM_MMIO_REG_VSX:
1159                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1160                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX);
1161
1162                 if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD)
1163                         kvmppc_set_vsr_dword(vcpu, gpr);
1164                 else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD)
1165                         kvmppc_set_vsr_word(vcpu, gpr);
1166                 else if (vcpu->arch.mmio_copy_type ==
1167                                 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP)
1168                         kvmppc_set_vsr_dword_dump(vcpu, gpr);
1169                 else if (vcpu->arch.mmio_copy_type ==
1170                                 KVMPPC_VSX_COPY_WORD_LOAD_DUMP)
1171                         kvmppc_set_vsr_word_dump(vcpu, gpr);
1172                 break;
1173 #endif
1174 #ifdef CONFIG_ALTIVEC
1175         case KVM_MMIO_REG_VMX:
1176                 if (vcpu->kvm->arch.kvm_ops->giveup_ext)
1177                         vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC);
1178
1179                 if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD)
1180                         kvmppc_set_vmx_dword(vcpu, gpr);
1181                 else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD)
1182                         kvmppc_set_vmx_word(vcpu, gpr);
1183                 else if (vcpu->arch.mmio_copy_type ==
1184                                 KVMPPC_VMX_COPY_HWORD)
1185                         kvmppc_set_vmx_hword(vcpu, gpr);
1186                 else if (vcpu->arch.mmio_copy_type ==
1187                                 KVMPPC_VMX_COPY_BYTE)
1188                         kvmppc_set_vmx_byte(vcpu, gpr);
1189                 break;
1190 #endif
1191         default:
1192                 BUG();
1193         }
1194 }
1195
1196 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1197                                 unsigned int rt, unsigned int bytes,
1198                                 int is_default_endian, int sign_extend)
1199 {
1200         int idx, ret;
1201         bool host_swabbed;
1202
1203         /* Pity C doesn't have a logical XOR operator */
1204         if (kvmppc_need_byteswap(vcpu)) {
1205                 host_swabbed = is_default_endian;
1206         } else {
1207                 host_swabbed = !is_default_endian;
1208         }
1209
1210         if (bytes > sizeof(run->mmio.data)) {
1211                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1212                        run->mmio.len);
1213         }
1214
1215         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1216         run->mmio.len = bytes;
1217         run->mmio.is_write = 0;
1218
1219         vcpu->arch.io_gpr = rt;
1220         vcpu->arch.mmio_host_swabbed = host_swabbed;
1221         vcpu->mmio_needed = 1;
1222         vcpu->mmio_is_write = 0;
1223         vcpu->arch.mmio_sign_extend = sign_extend;
1224
1225         idx = srcu_read_lock(&vcpu->kvm->srcu);
1226
1227         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1228                               bytes, &run->mmio.data);
1229
1230         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1231
1232         if (!ret) {
1233                 kvmppc_complete_mmio_load(vcpu, run);
1234                 vcpu->mmio_needed = 0;
1235                 return EMULATE_DONE;
1236         }
1237
1238         return EMULATE_DO_MMIO;
1239 }
1240
1241 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1242                        unsigned int rt, unsigned int bytes,
1243                        int is_default_endian)
1244 {
1245         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
1246 }
1247 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
1248
1249 /* Same as above, but sign extends */
1250 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
1251                         unsigned int rt, unsigned int bytes,
1252                         int is_default_endian)
1253 {
1254         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
1255 }
1256
1257 #ifdef CONFIG_VSX
1258 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1259                         unsigned int rt, unsigned int bytes,
1260                         int is_default_endian, int mmio_sign_extend)
1261 {
1262         enum emulation_result emulated = EMULATE_DONE;
1263
1264         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1265         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1266                 return EMULATE_FAIL;
1267
1268         while (vcpu->arch.mmio_vsx_copy_nums) {
1269                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1270                         is_default_endian, mmio_sign_extend);
1271
1272                 if (emulated != EMULATE_DONE)
1273                         break;
1274
1275                 vcpu->arch.paddr_accessed += run->mmio.len;
1276
1277                 vcpu->arch.mmio_vsx_copy_nums--;
1278                 vcpu->arch.mmio_vsx_offset++;
1279         }
1280         return emulated;
1281 }
1282 #endif /* CONFIG_VSX */
1283
1284 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1285                         u64 val, unsigned int bytes, int is_default_endian)
1286 {
1287         void *data = run->mmio.data;
1288         int idx, ret;
1289         bool host_swabbed;
1290
1291         /* Pity C doesn't have a logical XOR operator */
1292         if (kvmppc_need_byteswap(vcpu)) {
1293                 host_swabbed = is_default_endian;
1294         } else {
1295                 host_swabbed = !is_default_endian;
1296         }
1297
1298         if (bytes > sizeof(run->mmio.data)) {
1299                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
1300                        run->mmio.len);
1301         }
1302
1303         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
1304         run->mmio.len = bytes;
1305         run->mmio.is_write = 1;
1306         vcpu->mmio_needed = 1;
1307         vcpu->mmio_is_write = 1;
1308
1309         if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4))
1310                 val = dp_to_sp(val);
1311
1312         /* Store the value at the lowest bytes in 'data'. */
1313         if (!host_swabbed) {
1314                 switch (bytes) {
1315                 case 8: *(u64 *)data = val; break;
1316                 case 4: *(u32 *)data = val; break;
1317                 case 2: *(u16 *)data = val; break;
1318                 case 1: *(u8  *)data = val; break;
1319                 }
1320         } else {
1321                 switch (bytes) {
1322                 case 8: *(u64 *)data = swab64(val); break;
1323                 case 4: *(u32 *)data = swab32(val); break;
1324                 case 2: *(u16 *)data = swab16(val); break;
1325                 case 1: *(u8  *)data = val; break;
1326                 }
1327         }
1328
1329         idx = srcu_read_lock(&vcpu->kvm->srcu);
1330
1331         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
1332                                bytes, &run->mmio.data);
1333
1334         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1335
1336         if (!ret) {
1337                 vcpu->mmio_needed = 0;
1338                 return EMULATE_DONE;
1339         }
1340
1341         return EMULATE_DO_MMIO;
1342 }
1343 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
1344
1345 #ifdef CONFIG_VSX
1346 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val)
1347 {
1348         u32 dword_offset, word_offset;
1349         union kvmppc_one_reg reg;
1350         int vsx_offset = 0;
1351         int copy_type = vcpu->arch.mmio_copy_type;
1352         int result = 0;
1353
1354         switch (copy_type) {
1355         case KVMPPC_VSX_COPY_DWORD:
1356                 vsx_offset =
1357                         kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset);
1358
1359                 if (vsx_offset == -1) {
1360                         result = -1;
1361                         break;
1362                 }
1363
1364                 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1365                         *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset);
1366                 } else {
1367                         reg.vval = VCPU_VSX_VR(vcpu, rs);
1368                         *val = reg.vsxval[vsx_offset];
1369                 }
1370                 break;
1371
1372         case KVMPPC_VSX_COPY_WORD:
1373                 vsx_offset =
1374                         kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset);
1375
1376                 if (vsx_offset == -1) {
1377                         result = -1;
1378                         break;
1379                 }
1380
1381                 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) {
1382                         dword_offset = vsx_offset / 2;
1383                         word_offset = vsx_offset % 2;
1384                         reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset);
1385                         *val = reg.vsx32val[word_offset];
1386                 } else {
1387                         reg.vval = VCPU_VSX_VR(vcpu, rs);
1388                         *val = reg.vsx32val[vsx_offset];
1389                 }
1390                 break;
1391
1392         default:
1393                 result = -1;
1394                 break;
1395         }
1396
1397         return result;
1398 }
1399
1400 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1401                         int rs, unsigned int bytes, int is_default_endian)
1402 {
1403         u64 val;
1404         enum emulation_result emulated = EMULATE_DONE;
1405
1406         vcpu->arch.io_gpr = rs;
1407
1408         /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */
1409         if (vcpu->arch.mmio_vsx_copy_nums > 4)
1410                 return EMULATE_FAIL;
1411
1412         while (vcpu->arch.mmio_vsx_copy_nums) {
1413                 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1)
1414                         return EMULATE_FAIL;
1415
1416                 emulated = kvmppc_handle_store(run, vcpu,
1417                          val, bytes, is_default_endian);
1418
1419                 if (emulated != EMULATE_DONE)
1420                         break;
1421
1422                 vcpu->arch.paddr_accessed += run->mmio.len;
1423
1424                 vcpu->arch.mmio_vsx_copy_nums--;
1425                 vcpu->arch.mmio_vsx_offset++;
1426         }
1427
1428         return emulated;
1429 }
1430
1431 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu,
1432                         struct kvm_run *run)
1433 {
1434         enum emulation_result emulated = EMULATE_FAIL;
1435         int r;
1436
1437         vcpu->arch.paddr_accessed += run->mmio.len;
1438
1439         if (!vcpu->mmio_is_write) {
1440                 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr,
1441                          run->mmio.len, 1, vcpu->arch.mmio_sign_extend);
1442         } else {
1443                 emulated = kvmppc_handle_vsx_store(run, vcpu,
1444                          vcpu->arch.io_gpr, run->mmio.len, 1);
1445         }
1446
1447         switch (emulated) {
1448         case EMULATE_DO_MMIO:
1449                 run->exit_reason = KVM_EXIT_MMIO;
1450                 r = RESUME_HOST;
1451                 break;
1452         case EMULATE_FAIL:
1453                 pr_info("KVM: MMIO emulation failed (VSX repeat)\n");
1454                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1455                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1456                 r = RESUME_HOST;
1457                 break;
1458         default:
1459                 r = RESUME_GUEST;
1460                 break;
1461         }
1462         return r;
1463 }
1464 #endif /* CONFIG_VSX */
1465
1466 #ifdef CONFIG_ALTIVEC
1467 int kvmppc_handle_vmx_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
1468                 unsigned int rt, unsigned int bytes, int is_default_endian)
1469 {
1470         enum emulation_result emulated = EMULATE_DONE;
1471
1472         if (vcpu->arch.mmio_vsx_copy_nums > 2)
1473                 return EMULATE_FAIL;
1474
1475         while (vcpu->arch.mmio_vmx_copy_nums) {
1476                 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes,
1477                                 is_default_endian, 0);
1478
1479                 if (emulated != EMULATE_DONE)
1480                         break;
1481
1482                 vcpu->arch.paddr_accessed += run->mmio.len;
1483                 vcpu->arch.mmio_vmx_copy_nums--;
1484                 vcpu->arch.mmio_vmx_offset++;
1485         }
1486
1487         return emulated;
1488 }
1489
1490 int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val)
1491 {
1492         union kvmppc_one_reg reg;
1493         int vmx_offset = 0;
1494         int result = 0;
1495
1496         vmx_offset =
1497                 kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1498
1499         if (vmx_offset == -1)
1500                 return -1;
1501
1502         reg.vval = VCPU_VSX_VR(vcpu, index);
1503         *val = reg.vsxval[vmx_offset];
1504
1505         return result;
1506 }
1507
1508 int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val)
1509 {
1510         union kvmppc_one_reg reg;
1511         int vmx_offset = 0;
1512         int result = 0;
1513
1514         vmx_offset =
1515                 kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1516
1517         if (vmx_offset == -1)
1518                 return -1;
1519
1520         reg.vval = VCPU_VSX_VR(vcpu, index);
1521         *val = reg.vsx32val[vmx_offset];
1522
1523         return result;
1524 }
1525
1526 int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val)
1527 {
1528         union kvmppc_one_reg reg;
1529         int vmx_offset = 0;
1530         int result = 0;
1531
1532         vmx_offset =
1533                 kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1534
1535         if (vmx_offset == -1)
1536                 return -1;
1537
1538         reg.vval = VCPU_VSX_VR(vcpu, index);
1539         *val = reg.vsx16val[vmx_offset];
1540
1541         return result;
1542 }
1543
1544 int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val)
1545 {
1546         union kvmppc_one_reg reg;
1547         int vmx_offset = 0;
1548         int result = 0;
1549
1550         vmx_offset =
1551                 kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset);
1552
1553         if (vmx_offset == -1)
1554                 return -1;
1555
1556         reg.vval = VCPU_VSX_VR(vcpu, index);
1557         *val = reg.vsx8val[vmx_offset];
1558
1559         return result;
1560 }
1561
1562 int kvmppc_handle_vmx_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
1563                 unsigned int rs, unsigned int bytes, int is_default_endian)
1564 {
1565         u64 val = 0;
1566         unsigned int index = rs & KVM_MMIO_REG_MASK;
1567         enum emulation_result emulated = EMULATE_DONE;
1568
1569         if (vcpu->arch.mmio_vsx_copy_nums > 2)
1570                 return EMULATE_FAIL;
1571
1572         vcpu->arch.io_gpr = rs;
1573
1574         while (vcpu->arch.mmio_vmx_copy_nums) {
1575                 switch (vcpu->arch.mmio_copy_type) {
1576                 case KVMPPC_VMX_COPY_DWORD:
1577                         if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1)
1578                                 return EMULATE_FAIL;
1579
1580                         break;
1581                 case KVMPPC_VMX_COPY_WORD:
1582                         if (kvmppc_get_vmx_word(vcpu, index, &val) == -1)
1583                                 return EMULATE_FAIL;
1584                         break;
1585                 case KVMPPC_VMX_COPY_HWORD:
1586                         if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1)
1587                                 return EMULATE_FAIL;
1588                         break;
1589                 case KVMPPC_VMX_COPY_BYTE:
1590                         if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1)
1591                                 return EMULATE_FAIL;
1592                         break;
1593                 default:
1594                         return EMULATE_FAIL;
1595                 }
1596
1597                 emulated = kvmppc_handle_store(run, vcpu, val, bytes,
1598                                 is_default_endian);
1599                 if (emulated != EMULATE_DONE)
1600                         break;
1601
1602                 vcpu->arch.paddr_accessed += run->mmio.len;
1603                 vcpu->arch.mmio_vmx_copy_nums--;
1604                 vcpu->arch.mmio_vmx_offset++;
1605         }
1606
1607         return emulated;
1608 }
1609
1610 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu,
1611                 struct kvm_run *run)
1612 {
1613         enum emulation_result emulated = EMULATE_FAIL;
1614         int r;
1615
1616         vcpu->arch.paddr_accessed += run->mmio.len;
1617
1618         if (!vcpu->mmio_is_write) {
1619                 emulated = kvmppc_handle_vmx_load(run, vcpu,
1620                                 vcpu->arch.io_gpr, run->mmio.len, 1);
1621         } else {
1622                 emulated = kvmppc_handle_vmx_store(run, vcpu,
1623                                 vcpu->arch.io_gpr, run->mmio.len, 1);
1624         }
1625
1626         switch (emulated) {
1627         case EMULATE_DO_MMIO:
1628                 run->exit_reason = KVM_EXIT_MMIO;
1629                 r = RESUME_HOST;
1630                 break;
1631         case EMULATE_FAIL:
1632                 pr_info("KVM: MMIO emulation failed (VMX repeat)\n");
1633                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1634                 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1635                 r = RESUME_HOST;
1636                 break;
1637         default:
1638                 r = RESUME_GUEST;
1639                 break;
1640         }
1641         return r;
1642 }
1643 #endif /* CONFIG_ALTIVEC */
1644
1645 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1646 {
1647         int r = 0;
1648         union kvmppc_one_reg val;
1649         int size;
1650
1651         size = one_reg_size(reg->id);
1652         if (size > sizeof(val))
1653                 return -EINVAL;
1654
1655         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1656         if (r == -EINVAL) {
1657                 r = 0;
1658                 switch (reg->id) {
1659 #ifdef CONFIG_ALTIVEC
1660                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1661                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1662                                 r = -ENXIO;
1663                                 break;
1664                         }
1665                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1666                         break;
1667                 case KVM_REG_PPC_VSCR:
1668                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1669                                 r = -ENXIO;
1670                                 break;
1671                         }
1672                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1673                         break;
1674                 case KVM_REG_PPC_VRSAVE:
1675                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
1676                         break;
1677 #endif /* CONFIG_ALTIVEC */
1678                 default:
1679                         r = -EINVAL;
1680                         break;
1681                 }
1682         }
1683
1684         if (r)
1685                 return r;
1686
1687         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1688                 r = -EFAULT;
1689
1690         return r;
1691 }
1692
1693 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1694 {
1695         int r;
1696         union kvmppc_one_reg val;
1697         int size;
1698
1699         size = one_reg_size(reg->id);
1700         if (size > sizeof(val))
1701                 return -EINVAL;
1702
1703         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1704                 return -EFAULT;
1705
1706         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1707         if (r == -EINVAL) {
1708                 r = 0;
1709                 switch (reg->id) {
1710 #ifdef CONFIG_ALTIVEC
1711                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1712                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1713                                 r = -ENXIO;
1714                                 break;
1715                         }
1716                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1717                         break;
1718                 case KVM_REG_PPC_VSCR:
1719                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1720                                 r = -ENXIO;
1721                                 break;
1722                         }
1723                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1724                         break;
1725                 case KVM_REG_PPC_VRSAVE:
1726                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1727                                 r = -ENXIO;
1728                                 break;
1729                         }
1730                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
1731                         break;
1732 #endif /* CONFIG_ALTIVEC */
1733                 default:
1734                         r = -EINVAL;
1735                         break;
1736                 }
1737         }
1738
1739         return r;
1740 }
1741
1742 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1743 {
1744         int r;
1745
1746         vcpu_load(vcpu);
1747
1748         if (vcpu->mmio_needed) {
1749                 vcpu->mmio_needed = 0;
1750                 if (!vcpu->mmio_is_write)
1751                         kvmppc_complete_mmio_load(vcpu, run);
1752 #ifdef CONFIG_VSX
1753                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1754                         vcpu->arch.mmio_vsx_copy_nums--;
1755                         vcpu->arch.mmio_vsx_offset++;
1756                 }
1757
1758                 if (vcpu->arch.mmio_vsx_copy_nums > 0) {
1759                         r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run);
1760                         if (r == RESUME_HOST) {
1761                                 vcpu->mmio_needed = 1;
1762                                 goto out;
1763                         }
1764                 }
1765 #endif
1766 #ifdef CONFIG_ALTIVEC
1767                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1768                         vcpu->arch.mmio_vmx_copy_nums--;
1769                         vcpu->arch.mmio_vmx_offset++;
1770                 }
1771
1772                 if (vcpu->arch.mmio_vmx_copy_nums > 0) {
1773                         r = kvmppc_emulate_mmio_vmx_loadstore(vcpu, run);
1774                         if (r == RESUME_HOST) {
1775                                 vcpu->mmio_needed = 1;
1776                                 goto out;
1777                         }
1778                 }
1779 #endif
1780         } else if (vcpu->arch.osi_needed) {
1781                 u64 *gprs = run->osi.gprs;
1782                 int i;
1783
1784                 for (i = 0; i < 32; i++)
1785                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1786                 vcpu->arch.osi_needed = 0;
1787         } else if (vcpu->arch.hcall_needed) {
1788                 int i;
1789
1790                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1791                 for (i = 0; i < 9; ++i)
1792                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1793                 vcpu->arch.hcall_needed = 0;
1794 #ifdef CONFIG_BOOKE
1795         } else if (vcpu->arch.epr_needed) {
1796                 kvmppc_set_epr(vcpu, run->epr.epr);
1797                 vcpu->arch.epr_needed = 0;
1798 #endif
1799         }
1800
1801         kvm_sigset_activate(vcpu);
1802
1803         if (run->immediate_exit)
1804                 r = -EINTR;
1805         else
1806                 r = kvmppc_vcpu_run(run, vcpu);
1807
1808         kvm_sigset_deactivate(vcpu);
1809
1810 #ifdef CONFIG_ALTIVEC
1811 out:
1812 #endif
1813         vcpu_put(vcpu);
1814         return r;
1815 }
1816
1817 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1818 {
1819         if (irq->irq == KVM_INTERRUPT_UNSET) {
1820                 kvmppc_core_dequeue_external(vcpu);
1821                 return 0;
1822         }
1823
1824         kvmppc_core_queue_external(vcpu, irq);
1825
1826         kvm_vcpu_kick(vcpu);
1827
1828         return 0;
1829 }
1830
1831 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1832                                      struct kvm_enable_cap *cap)
1833 {
1834         int r;
1835
1836         if (cap->flags)
1837                 return -EINVAL;
1838
1839         switch (cap->cap) {
1840         case KVM_CAP_PPC_OSI:
1841                 r = 0;
1842                 vcpu->arch.osi_enabled = true;
1843                 break;
1844         case KVM_CAP_PPC_PAPR:
1845                 r = 0;
1846                 vcpu->arch.papr_enabled = true;
1847                 break;
1848         case KVM_CAP_PPC_EPR:
1849                 r = 0;
1850                 if (cap->args[0])
1851                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1852                 else
1853                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1854                 break;
1855 #ifdef CONFIG_BOOKE
1856         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1857                 r = 0;
1858                 vcpu->arch.watchdog_enabled = true;
1859                 break;
1860 #endif
1861 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1862         case KVM_CAP_SW_TLB: {
1863                 struct kvm_config_tlb cfg;
1864                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1865
1866                 r = -EFAULT;
1867                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1868                         break;
1869
1870                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1871                 break;
1872         }
1873 #endif
1874 #ifdef CONFIG_KVM_MPIC
1875         case KVM_CAP_IRQ_MPIC: {
1876                 struct fd f;
1877                 struct kvm_device *dev;
1878
1879                 r = -EBADF;
1880                 f = fdget(cap->args[0]);
1881                 if (!f.file)
1882                         break;
1883
1884                 r = -EPERM;
1885                 dev = kvm_device_from_filp(f.file);
1886                 if (dev)
1887                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1888
1889                 fdput(f);
1890                 break;
1891         }
1892 #endif
1893 #ifdef CONFIG_KVM_XICS
1894         case KVM_CAP_IRQ_XICS: {
1895                 struct fd f;
1896                 struct kvm_device *dev;
1897
1898                 r = -EBADF;
1899                 f = fdget(cap->args[0]);
1900                 if (!f.file)
1901                         break;
1902
1903                 r = -EPERM;
1904                 dev = kvm_device_from_filp(f.file);
1905                 if (dev) {
1906                         if (xive_enabled())
1907                                 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]);
1908                         else
1909                                 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1910                 }
1911
1912                 fdput(f);
1913                 break;
1914         }
1915 #endif /* CONFIG_KVM_XICS */
1916 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1917         case KVM_CAP_PPC_FWNMI:
1918                 r = -EINVAL;
1919                 if (!is_kvmppc_hv_enabled(vcpu->kvm))
1920                         break;
1921                 r = 0;
1922                 vcpu->kvm->arch.fwnmi_enabled = true;
1923                 break;
1924 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1925         default:
1926                 r = -EINVAL;
1927                 break;
1928         }
1929
1930         if (!r)
1931                 r = kvmppc_sanity_check(vcpu);
1932
1933         return r;
1934 }
1935
1936 bool kvm_arch_intc_initialized(struct kvm *kvm)
1937 {
1938 #ifdef CONFIG_KVM_MPIC
1939         if (kvm->arch.mpic)
1940                 return true;
1941 #endif
1942 #ifdef CONFIG_KVM_XICS
1943         if (kvm->arch.xics || kvm->arch.xive)
1944                 return true;
1945 #endif
1946         return false;
1947 }
1948
1949 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1950                                     struct kvm_mp_state *mp_state)
1951 {
1952         return -EINVAL;
1953 }
1954
1955 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1956                                     struct kvm_mp_state *mp_state)
1957 {
1958         return -EINVAL;
1959 }
1960
1961 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1962                                unsigned int ioctl, unsigned long arg)
1963 {
1964         struct kvm_vcpu *vcpu = filp->private_data;
1965         void __user *argp = (void __user *)arg;
1966
1967         if (ioctl == KVM_INTERRUPT) {
1968                 struct kvm_interrupt irq;
1969                 if (copy_from_user(&irq, argp, sizeof(irq)))
1970                         return -EFAULT;
1971                 return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1972         }
1973         return -ENOIOCTLCMD;
1974 }
1975
1976 long kvm_arch_vcpu_ioctl(struct file *filp,
1977                          unsigned int ioctl, unsigned long arg)
1978 {
1979         struct kvm_vcpu *vcpu = filp->private_data;
1980         void __user *argp = (void __user *)arg;
1981         long r;
1982
1983         switch (ioctl) {
1984         case KVM_ENABLE_CAP:
1985         {
1986                 struct kvm_enable_cap cap;
1987                 r = -EFAULT;
1988                 vcpu_load(vcpu);
1989                 if (copy_from_user(&cap, argp, sizeof(cap)))
1990                         goto out;
1991                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1992                 vcpu_put(vcpu);
1993                 break;
1994         }
1995
1996         case KVM_SET_ONE_REG:
1997         case KVM_GET_ONE_REG:
1998         {
1999                 struct kvm_one_reg reg;
2000                 r = -EFAULT;
2001                 if (copy_from_user(&reg, argp, sizeof(reg)))
2002                         goto out;
2003                 if (ioctl == KVM_SET_ONE_REG)
2004                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
2005                 else
2006                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
2007                 break;
2008         }
2009
2010 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
2011         case KVM_DIRTY_TLB: {
2012                 struct kvm_dirty_tlb dirty;
2013                 r = -EFAULT;
2014                 vcpu_load(vcpu);
2015                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
2016                         goto out;
2017                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
2018                 vcpu_put(vcpu);
2019                 break;
2020         }
2021 #endif
2022         default:
2023                 r = -EINVAL;
2024         }
2025
2026 out:
2027         return r;
2028 }
2029
2030 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2031 {
2032         return VM_FAULT_SIGBUS;
2033 }
2034
2035 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
2036 {
2037         u32 inst_nop = 0x60000000;
2038 #ifdef CONFIG_KVM_BOOKE_HV
2039         u32 inst_sc1 = 0x44000022;
2040         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
2041         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
2042         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
2043         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2044 #else
2045         u32 inst_lis = 0x3c000000;
2046         u32 inst_ori = 0x60000000;
2047         u32 inst_sc = 0x44000002;
2048         u32 inst_imm_mask = 0xffff;
2049
2050         /*
2051          * The hypercall to get into KVM from within guest context is as
2052          * follows:
2053          *
2054          *    lis r0, r0, KVM_SC_MAGIC_R0@h
2055          *    ori r0, KVM_SC_MAGIC_R0@l
2056          *    sc
2057          *    nop
2058          */
2059         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
2060         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
2061         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
2062         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
2063 #endif
2064
2065         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
2066
2067         return 0;
2068 }
2069
2070 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
2071                           bool line_status)
2072 {
2073         if (!irqchip_in_kernel(kvm))
2074                 return -ENXIO;
2075
2076         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2077                                         irq_event->irq, irq_event->level,
2078                                         line_status);
2079         return 0;
2080 }
2081
2082
2083 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
2084                                    struct kvm_enable_cap *cap)
2085 {
2086         int r;
2087
2088         if (cap->flags)
2089                 return -EINVAL;
2090
2091         switch (cap->cap) {
2092 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
2093         case KVM_CAP_PPC_ENABLE_HCALL: {
2094                 unsigned long hcall = cap->args[0];
2095
2096                 r = -EINVAL;
2097                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
2098                     cap->args[1] > 1)
2099                         break;
2100                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
2101                         break;
2102                 if (cap->args[1])
2103                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
2104                 else
2105                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
2106                 r = 0;
2107                 break;
2108         }
2109         case KVM_CAP_PPC_SMT: {
2110                 unsigned long mode = cap->args[0];
2111                 unsigned long flags = cap->args[1];
2112
2113                 r = -EINVAL;
2114                 if (kvm->arch.kvm_ops->set_smt_mode)
2115                         r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags);
2116                 break;
2117         }
2118 #endif
2119         default:
2120                 r = -EINVAL;
2121                 break;
2122         }
2123
2124         return r;
2125 }
2126
2127 #ifdef CONFIG_PPC_BOOK3S_64
2128 /*
2129  * These functions check whether the underlying hardware is safe
2130  * against attacks based on observing the effects of speculatively
2131  * executed instructions, and whether it supplies instructions for
2132  * use in workarounds.  The information comes from firmware, either
2133  * via the device tree on powernv platforms or from an hcall on
2134  * pseries platforms.
2135  */
2136 #ifdef CONFIG_PPC_PSERIES
2137 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2138 {
2139         struct h_cpu_char_result c;
2140         unsigned long rc;
2141
2142         if (!machine_is(pseries))
2143                 return -ENOTTY;
2144
2145         rc = plpar_get_cpu_characteristics(&c);
2146         if (rc == H_SUCCESS) {
2147                 cp->character = c.character;
2148                 cp->behaviour = c.behaviour;
2149                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2150                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2151                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2152                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2153                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2154                         KVM_PPC_CPU_CHAR_BR_HINT_HONOURED |
2155                         KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF |
2156                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2157                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2158                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2159                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2160         }
2161         return 0;
2162 }
2163 #else
2164 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2165 {
2166         return -ENOTTY;
2167 }
2168 #endif
2169
2170 static inline bool have_fw_feat(struct device_node *fw_features,
2171                                 const char *state, const char *name)
2172 {
2173         struct device_node *np;
2174         bool r = false;
2175
2176         np = of_get_child_by_name(fw_features, name);
2177         if (np) {
2178                 r = of_property_read_bool(np, state);
2179                 of_node_put(np);
2180         }
2181         return r;
2182 }
2183
2184 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
2185 {
2186         struct device_node *np, *fw_features;
2187         int r;
2188
2189         memset(cp, 0, sizeof(*cp));
2190         r = pseries_get_cpu_char(cp);
2191         if (r != -ENOTTY)
2192                 return r;
2193
2194         np = of_find_node_by_name(NULL, "ibm,opal");
2195         if (np) {
2196                 fw_features = of_get_child_by_name(np, "fw-features");
2197                 of_node_put(np);
2198                 if (!fw_features)
2199                         return 0;
2200                 if (have_fw_feat(fw_features, "enabled",
2201                                  "inst-spec-barrier-ori31,31,0"))
2202                         cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31;
2203                 if (have_fw_feat(fw_features, "enabled",
2204                                  "fw-bcctrl-serialized"))
2205                         cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED;
2206                 if (have_fw_feat(fw_features, "enabled",
2207                                  "inst-l1d-flush-ori30,30,0"))
2208                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30;
2209                 if (have_fw_feat(fw_features, "enabled",
2210                                  "inst-l1d-flush-trig2"))
2211                         cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2;
2212                 if (have_fw_feat(fw_features, "enabled",
2213                                  "fw-l1d-thread-split"))
2214                         cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV;
2215                 if (have_fw_feat(fw_features, "enabled",
2216                                  "fw-count-cache-disabled"))
2217                         cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2218                 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 |
2219                         KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED |
2220                         KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 |
2221                         KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 |
2222                         KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV |
2223                         KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS;
2224
2225                 if (have_fw_feat(fw_features, "enabled",
2226                                  "speculation-policy-favor-security"))
2227                         cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY;
2228                 if (!have_fw_feat(fw_features, "disabled",
2229                                   "needs-l1d-flush-msr-pr-0-to-1"))
2230                         cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR;
2231                 if (!have_fw_feat(fw_features, "disabled",
2232                                   "needs-spec-barrier-for-bound-checks"))
2233                         cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2234                 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY |
2235                         KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR |
2236                         KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR;
2237
2238                 of_node_put(fw_features);
2239         }
2240
2241         return 0;
2242 }
2243 #endif
2244
2245 long kvm_arch_vm_ioctl(struct file *filp,
2246                        unsigned int ioctl, unsigned long arg)
2247 {
2248         struct kvm *kvm __maybe_unused = filp->private_data;
2249         void __user *argp = (void __user *)arg;
2250         long r;
2251
2252         switch (ioctl) {
2253         case KVM_PPC_GET_PVINFO: {
2254                 struct kvm_ppc_pvinfo pvinfo;
2255                 memset(&pvinfo, 0, sizeof(pvinfo));
2256                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
2257                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
2258                         r = -EFAULT;
2259                         goto out;
2260                 }
2261
2262                 break;
2263         }
2264         case KVM_ENABLE_CAP:
2265         {
2266                 struct kvm_enable_cap cap;
2267                 r = -EFAULT;
2268                 if (copy_from_user(&cap, argp, sizeof(cap)))
2269                         goto out;
2270                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
2271                 break;
2272         }
2273 #ifdef CONFIG_SPAPR_TCE_IOMMU
2274         case KVM_CREATE_SPAPR_TCE_64: {
2275                 struct kvm_create_spapr_tce_64 create_tce_64;
2276
2277                 r = -EFAULT;
2278                 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
2279                         goto out;
2280                 if (create_tce_64.flags) {
2281                         r = -EINVAL;
2282                         goto out;
2283                 }
2284                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2285                 goto out;
2286         }
2287         case KVM_CREATE_SPAPR_TCE: {
2288                 struct kvm_create_spapr_tce create_tce;
2289                 struct kvm_create_spapr_tce_64 create_tce_64;
2290
2291                 r = -EFAULT;
2292                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
2293                         goto out;
2294
2295                 create_tce_64.liobn = create_tce.liobn;
2296                 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
2297                 create_tce_64.offset = 0;
2298                 create_tce_64.size = create_tce.window_size >>
2299                                 IOMMU_PAGE_SHIFT_4K;
2300                 create_tce_64.flags = 0;
2301                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
2302                 goto out;
2303         }
2304 #endif
2305 #ifdef CONFIG_PPC_BOOK3S_64
2306         case KVM_PPC_GET_SMMU_INFO: {
2307                 struct kvm_ppc_smmu_info info;
2308                 struct kvm *kvm = filp->private_data;
2309
2310                 memset(&info, 0, sizeof(info));
2311                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
2312                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2313                         r = -EFAULT;
2314                 break;
2315         }
2316         case KVM_PPC_RTAS_DEFINE_TOKEN: {
2317                 struct kvm *kvm = filp->private_data;
2318
2319                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
2320                 break;
2321         }
2322         case KVM_PPC_CONFIGURE_V3_MMU: {
2323                 struct kvm *kvm = filp->private_data;
2324                 struct kvm_ppc_mmuv3_cfg cfg;
2325
2326                 r = -EINVAL;
2327                 if (!kvm->arch.kvm_ops->configure_mmu)
2328                         goto out;
2329                 r = -EFAULT;
2330                 if (copy_from_user(&cfg, argp, sizeof(cfg)))
2331                         goto out;
2332                 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
2333                 break;
2334         }
2335         case KVM_PPC_GET_RMMU_INFO: {
2336                 struct kvm *kvm = filp->private_data;
2337                 struct kvm_ppc_rmmu_info info;
2338
2339                 r = -EINVAL;
2340                 if (!kvm->arch.kvm_ops->get_rmmu_info)
2341                         goto out;
2342                 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
2343                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
2344                         r = -EFAULT;
2345                 break;
2346         }
2347         case KVM_PPC_GET_CPU_CHAR: {
2348                 struct kvm_ppc_cpu_char cpuchar;
2349
2350                 r = kvmppc_get_cpu_char(&cpuchar);
2351                 if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar)))
2352                         r = -EFAULT;
2353                 break;
2354         }
2355         default: {
2356                 struct kvm *kvm = filp->private_data;
2357                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
2358         }
2359 #else /* CONFIG_PPC_BOOK3S_64 */
2360         default:
2361                 r = -ENOTTY;
2362 #endif
2363         }
2364 out:
2365         return r;
2366 }
2367
2368 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
2369 static unsigned long nr_lpids;
2370
2371 long kvmppc_alloc_lpid(void)
2372 {
2373         long lpid;
2374
2375         do {
2376                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
2377                 if (lpid >= nr_lpids) {
2378                         pr_err("%s: No LPIDs free\n", __func__);
2379                         return -ENOMEM;
2380                 }
2381         } while (test_and_set_bit(lpid, lpid_inuse));
2382
2383         return lpid;
2384 }
2385 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
2386
2387 void kvmppc_claim_lpid(long lpid)
2388 {
2389         set_bit(lpid, lpid_inuse);
2390 }
2391 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
2392
2393 void kvmppc_free_lpid(long lpid)
2394 {
2395         clear_bit(lpid, lpid_inuse);
2396 }
2397 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
2398
2399 void kvmppc_init_lpid(unsigned long nr_lpids_param)
2400 {
2401         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
2402         memset(lpid_inuse, 0, sizeof(lpid_inuse));
2403 }
2404 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
2405
2406 int kvm_arch_init(void *opaque)
2407 {
2408         return 0;
2409 }
2410
2411 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);