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/fs.h>
  27 #include <linux/slab.h>
  28 #include <linux/file.h>
  29 #include <linux/module.h>
  30 #include <linux/irqbypass.h>
  31 #include <linux/kvm_irqfd.h>
  32 #include <asm/cputable.h>
  33 #include <linux/uaccess.h>
  34 #include <asm/kvm_ppc.h>
  35 #include <asm/tlbflush.h>
  36 #include <asm/cputhreads.h>
  37 #include <asm/irqflags.h>
  38 #include <asm/iommu.h>
  39 #include "timing.h"
  40 #include "irq.h"
  41 #include "../mm/mmu_decl.h"
  42
  43 #define CREATE_TRACE_POINTS
  44 #include "trace.h"
  45
  46 struct kvmppc_ops *kvmppc_hv_ops;
  47 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
  48 struct kvmppc_ops *kvmppc_pr_ops;
  49 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
  50
  51
  52 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  53 {
  54         return !!(v->arch.pending_exceptions) ||
  55                v->requests;
  56 }
  57
  58 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  59 {
  60         return 1;
  61 }
  62
  63 /*
  64  * Common checks before entering the guest world.  Call with interrupts
  65  * disabled.
  66  *
  67  * returns:
  68  *
  69  * == 1 if we're ready to go into guest state
  70  * <= 0 if we need to go back to the host with return value
  71  */
  72 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  73 {
  74         int r;
  75
  76         WARN_ON(irqs_disabled());
  77         hard_irq_disable();
  78
  79         while (true) {
  80                 if (need_resched()) {
  81                         local_irq_enable();
  82                         cond_resched();
  83                         hard_irq_disable();
  84                         continue;
  85                 }
  86
  87                 if (signal_pending(current)) {
  88                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
  89                         vcpu->run->exit_reason = KVM_EXIT_INTR;
  90                         r = -EINTR;
  91                         break;
  92                 }
  93
  94                 vcpu->mode = IN_GUEST_MODE;
  95
  96                 /*
  97                  * Reading vcpu->requests must happen after setting vcpu->mode,
  98                  * so we don't miss a request because the requester sees
  99                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
 100                  * before next entering the guest (and thus doesn't IPI).
 101                  * This also orders the write to mode from any reads
 102                  * to the page tables done while the VCPU is running.
 103                  * Please see the comment in kvm_flush_remote_tlbs.
 104                  */
 105                 smp_mb();
 106
 107                 if (vcpu->requests) {
 108                         /* Make sure we process requests preemptable */
 109                         local_irq_enable();
 110                         trace_kvm_check_requests(vcpu);
 111                         r = kvmppc_core_check_requests(vcpu);
 112                         hard_irq_disable();
 113                         if (r > 0)
 114                                 continue;
 115                         break;
 116                 }
 117
 118                 if (kvmppc_core_prepare_to_enter(vcpu)) {
 119                         /* interrupts got enabled in between, so we
 120                            are back at square 1 */
 121                         continue;
 122                 }
 123
 124                 guest_enter_irqoff();
 125                 return 1;
 126         }
 127
 128         /* return to host */
 129         local_irq_enable();
 130         return r;
 131 }
 132 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
 133
 134 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 135 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
 136 {
 137         struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
 138         int i;
 139
 140         shared->sprg0 = swab64(shared->sprg0);
 141         shared->sprg1 = swab64(shared->sprg1);
 142         shared->sprg2 = swab64(shared->sprg2);
 143         shared->sprg3 = swab64(shared->sprg3);
 144         shared->srr0 = swab64(shared->srr0);
 145         shared->srr1 = swab64(shared->srr1);
 146         shared->dar = swab64(shared->dar);
 147         shared->msr = swab64(shared->msr);
 148         shared->dsisr = swab32(shared->dsisr);
 149         shared->int_pending = swab32(shared->int_pending);
 150         for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
 151                 shared->sr[i] = swab32(shared->sr[i]);
 152 }
 153 #endif
 154
 155 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
 156 {
 157         int nr = kvmppc_get_gpr(vcpu, 11);
 158         int r;
 159         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
 160         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
 161         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
 162         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
 163         unsigned long r2 = 0;
 164
 165         if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
 166                 /* 32 bit mode */
 167                 param1 &= 0xffffffff;
 168                 param2 &= 0xffffffff;
 169                 param3 &= 0xffffffff;
 170                 param4 &= 0xffffffff;
 171         }
 172
 173         switch (nr) {
 174         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
 175         {
 176 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 177                 /* Book3S can be little endian, find it out here */
 178                 int shared_big_endian = true;
 179                 if (vcpu->arch.intr_msr & MSR_LE)
 180                         shared_big_endian = false;
 181                 if (shared_big_endian != vcpu->arch.shared_big_endian)
 182                         kvmppc_swab_shared(vcpu);
 183                 vcpu->arch.shared_big_endian = shared_big_endian;
 184 #endif
 185
 186                 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
 187                         /*
 188                          * Older versions of the Linux magic page code had
 189                          * a bug where they would map their trampoline code
 190                          * NX. If that's the case, remove !PR NX capability.
 191                          */
 192                         vcpu->arch.disable_kernel_nx = true;
 193                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 194                 }
 195
 196                 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
 197                 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
 198
 199 #ifdef CONFIG_PPC_64K_PAGES
 200                 /*
 201                  * Make sure our 4k magic page is in the same window of a 64k
 202                  * page within the guest and within the host's page.
 203                  */
 204                 if ((vcpu->arch.magic_page_pa & 0xf000) !=
 205                     ((ulong)vcpu->arch.shared & 0xf000)) {
 206                         void *old_shared = vcpu->arch.shared;
 207                         ulong shared = (ulong)vcpu->arch.shared;
 208                         void *new_shared;
 209
 210                         shared &= PAGE_MASK;
 211                         shared |= vcpu->arch.magic_page_pa & 0xf000;
 212                         new_shared = (void*)shared;
 213                         memcpy(new_shared, old_shared, 0x1000);
 214                         vcpu->arch.shared = new_shared;
 215                 }
 216 #endif
 217
 218                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 219
 220                 r = EV_SUCCESS;
 221                 break;
 222         }
 223         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 224                 r = EV_SUCCESS;
 225 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 226                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 227 #endif
 228
 229                 /* Second return value is in r4 */
 230                 break;
 231         case EV_HCALL_TOKEN(EV_IDLE):
 232                 r = EV_SUCCESS;
 233                 kvm_vcpu_block(vcpu);
 234                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 235                 break;
 236         default:
 237                 r = EV_UNIMPLEMENTED;
 238                 break;
 239         }
 240
 241         kvmppc_set_gpr(vcpu, 4, r2);
 242
 243         return r;
 244 }
 245 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
 246
 247 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
 248 {
 249         int r = false;
 250
 251         /* We have to know what CPU to virtualize */
 252         if (!vcpu->arch.pvr)
 253                 goto out;
 254
 255         /* PAPR only works with book3s_64 */
 256         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
 257                 goto out;
 258
 259         /* HV KVM can only do PAPR mode for now */
 260         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
 261                 goto out;
 262
 263 #ifdef CONFIG_KVM_BOOKE_HV
 264         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 265                 goto out;
 266 #endif
 267
 268         r = true;
 269
 270 out:
 271         vcpu->arch.sane = r;
 272         return r ? 0 : -EINVAL;
 273 }
 274 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
 275
 276 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 277 {
 278         enum emulation_result er;
 279         int r;
 280
 281         er = kvmppc_emulate_loadstore(vcpu);
 282         switch (er) {
 283         case EMULATE_DONE:
 284                 /* Future optimization: only reload non-volatiles if they were
 285                  * actually modified. */
 286                 r = RESUME_GUEST_NV;
 287                 break;
 288         case EMULATE_AGAIN:
 289                 r = RESUME_GUEST;
 290                 break;
 291         case EMULATE_DO_MMIO:
 292                 run->exit_reason = KVM_EXIT_MMIO;
 293                 /* We must reload nonvolatiles because "update" load/store
 294                  * instructions modify register state. */
 295                 /* Future optimization: only reload non-volatiles if they were
 296                  * actually modified. */
 297                 r = RESUME_HOST_NV;
 298                 break;
 299         case EMULATE_FAIL:
 300         {
 301                 u32 last_inst;
 302
 303                 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
 304                 /* XXX Deliver Program interrupt to guest. */
 305                 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
 306                 r = RESUME_HOST;
 307                 break;
 308         }
 309         default:
 310                 WARN_ON(1);
 311                 r = RESUME_GUEST;
 312         }
 313
 314         return r;
 315 }
 316 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
 317
 318 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 319               bool data)
 320 {
 321         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 322         struct kvmppc_pte pte;
 323         int r;
 324
 325         vcpu->stat.st++;
 326
 327         r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 328                          XLATE_WRITE, &pte);
 329         if (r < 0)
 330                 return r;
 331
 332         *eaddr = pte.raddr;
 333
 334         if (!pte.may_write)
 335                 return -EPERM;
 336
 337         /* Magic page override */
 338         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 339             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 340             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 341                 void *magic = vcpu->arch.shared;
 342                 magic += pte.eaddr & 0xfff;
 343                 memcpy(magic, ptr, size);
 344                 return EMULATE_DONE;
 345         }
 346
 347         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 348                 return EMULATE_DO_MMIO;
 349
 350         return EMULATE_DONE;
 351 }
 352 EXPORT_SYMBOL_GPL(kvmppc_st);
 353
 354 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 355                       bool data)
 356 {
 357         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 358         struct kvmppc_pte pte;
 359         int rc;
 360
 361         vcpu->stat.ld++;
 362
 363         rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 364                           XLATE_READ, &pte);
 365         if (rc)
 366                 return rc;
 367
 368         *eaddr = pte.raddr;
 369
 370         if (!pte.may_read)
 371                 return -EPERM;
 372
 373         if (!data && !pte.may_execute)
 374                 return -ENOEXEC;
 375
 376         /* Magic page override */
 377         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 378             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 379             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 380                 void *magic = vcpu->arch.shared;
 381                 magic += pte.eaddr & 0xfff;
 382                 memcpy(ptr, magic, size);
 383                 return EMULATE_DONE;
 384         }
 385
 386         if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
 387                 return EMULATE_DO_MMIO;
 388
 389         return EMULATE_DONE;
 390 }
 391 EXPORT_SYMBOL_GPL(kvmppc_ld);
 392
 393 int kvm_arch_hardware_enable(void)
 394 {
 395         return 0;
 396 }
 397
 398 int kvm_arch_hardware_setup(void)
 399 {
 400         return 0;
 401 }
 402
 403 void kvm_arch_check_processor_compat(void *rtn)
 404 {
 405         *(int *)rtn = kvmppc_core_check_processor_compat();
 406 }
 407
 408 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 409 {
 410         struct kvmppc_ops *kvm_ops = NULL;
 411         /*
 412          * if we have both HV and PR enabled, default is HV
 413          */
 414         if (type == 0) {
 415                 if (kvmppc_hv_ops)
 416                         kvm_ops = kvmppc_hv_ops;
 417                 else
 418                         kvm_ops = kvmppc_pr_ops;
 419                 if (!kvm_ops)
 420                         goto err_out;
 421         } else  if (type == KVM_VM_PPC_HV) {
 422                 if (!kvmppc_hv_ops)
 423                         goto err_out;
 424                 kvm_ops = kvmppc_hv_ops;
 425         } else if (type == KVM_VM_PPC_PR) {
 426                 if (!kvmppc_pr_ops)
 427                         goto err_out;
 428                 kvm_ops = kvmppc_pr_ops;
 429         } else
 430                 goto err_out;
 431
 432         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
 433                 return -ENOENT;
 434
 435         kvm->arch.kvm_ops = kvm_ops;
 436         return kvmppc_core_init_vm(kvm);
 437 err_out:
 438         return -EINVAL;
 439 }
 440
 441 bool kvm_arch_has_vcpu_debugfs(void)
 442 {
 443         return false;
 444 }
 445
 446 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 447 {
 448         return 0;
 449 }
 450
 451 void kvm_arch_destroy_vm(struct kvm *kvm)
 452 {
 453         unsigned int i;
 454         struct kvm_vcpu *vcpu;
 455
 456 #ifdef CONFIG_KVM_XICS
 457         /*
 458          * We call kick_all_cpus_sync() to ensure that all
 459          * CPUs have executed any pending IPIs before we
 460          * continue and free VCPUs structures below.
 461          */
 462         if (is_kvmppc_hv_enabled(kvm))
 463                 kick_all_cpus_sync();
 464 #endif
 465
 466         kvm_for_each_vcpu(i, vcpu, kvm)
 467                 kvm_arch_vcpu_free(vcpu);
 468
 469         mutex_lock(&kvm->lock);
 470         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 471                 kvm->vcpus[i] = NULL;
 472
 473         atomic_set(&kvm->online_vcpus, 0);
 474
 475         kvmppc_core_destroy_vm(kvm);
 476
 477         mutex_unlock(&kvm->lock);
 478
 479         /* drop the module reference */
 480         module_put(kvm->arch.kvm_ops->owner);
 481 }
 482
 483 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 484 {
 485         int r;
 486         /* Assume we're using HV mode when the HV module is loaded */
 487         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
 488
 489         if (kvm) {
 490                 /*
 491                  * Hooray - we know which VM type we're running on. Depend on
 492                  * that rather than the guess above.
 493                  */
 494                 hv_enabled = is_kvmppc_hv_enabled(kvm);
 495         }
 496
 497         switch (ext) {
 498 #ifdef CONFIG_BOOKE
 499         case KVM_CAP_PPC_BOOKE_SREGS:
 500         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 501         case KVM_CAP_PPC_EPR:
 502 #else
 503         case KVM_CAP_PPC_SEGSTATE:
 504         case KVM_CAP_PPC_HIOR:
 505         case KVM_CAP_PPC_PAPR:
 506 #endif
 507         case KVM_CAP_PPC_UNSET_IRQ:
 508         case KVM_CAP_PPC_IRQ_LEVEL:
 509         case KVM_CAP_ENABLE_CAP:
 510         case KVM_CAP_ENABLE_CAP_VM:
 511         case KVM_CAP_ONE_REG:
 512         case KVM_CAP_IOEVENTFD:
 513         case KVM_CAP_DEVICE_CTRL:
 514         case KVM_CAP_IMMEDIATE_EXIT:
 515                 r = 1;
 516                 break;
 517         case KVM_CAP_PPC_PAIRED_SINGLES:
 518         case KVM_CAP_PPC_OSI:
 519         case KVM_CAP_PPC_GET_PVINFO:
 520 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 521         case KVM_CAP_SW_TLB:
 522 #endif
 523                 /* We support this only for PR */
 524                 r = !hv_enabled;
 525                 break;
 526 #ifdef CONFIG_KVM_MMIO
 527         case KVM_CAP_COALESCED_MMIO:
 528                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 529                 break;
 530 #endif
 531 #ifdef CONFIG_KVM_MPIC
 532         case KVM_CAP_IRQ_MPIC:
 533                 r = 1;
 534                 break;
 535 #endif
 536
 537 #ifdef CONFIG_PPC_BOOK3S_64
 538         case KVM_CAP_SPAPR_TCE:
 539         case KVM_CAP_SPAPR_TCE_64:
 540         case KVM_CAP_PPC_RTAS:
 541         case KVM_CAP_PPC_FIXUP_HCALL:
 542         case KVM_CAP_PPC_ENABLE_HCALL:
 543 #ifdef CONFIG_KVM_XICS
 544         case KVM_CAP_IRQ_XICS:
 545 #endif
 546                 r = 1;
 547                 break;
 548
 549         case KVM_CAP_PPC_ALLOC_HTAB:
 550                 r = hv_enabled;
 551                 break;
 552 #endif /* CONFIG_PPC_BOOK3S_64 */
 553 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 554         case KVM_CAP_PPC_SMT:
 555                 r = 0;
 556                 if (hv_enabled) {
 557                         if (cpu_has_feature(CPU_FTR_ARCH_300))
 558                                 r = 1;
 559                         else
 560                                 r = threads_per_subcore;
 561                 }
 562                 break;
 563         case KVM_CAP_PPC_RMA:
 564                 r = 0;
 565                 break;
 566         case KVM_CAP_PPC_HWRNG:
 567                 r = kvmppc_hwrng_present();
 568                 break;
 569         case KVM_CAP_PPC_MMU_RADIX:
 570                 r = !!(hv_enabled && radix_enabled());
 571                 break;
 572         case KVM_CAP_PPC_MMU_HASH_V3:
 573                 r = !!(hv_enabled && !radix_enabled() &&
 574                        cpu_has_feature(CPU_FTR_ARCH_300));
 575                 break;
 576 #endif
 577         case KVM_CAP_SYNC_MMU:
 578 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 579                 r = hv_enabled;
 580 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 581                 r = 1;
 582 #else
 583                 r = 0;
 584 #endif
 585                 break;
 586 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 587         case KVM_CAP_PPC_HTAB_FD:
 588                 r = hv_enabled;
 589                 break;
 590 #endif
 591         case KVM_CAP_NR_VCPUS:
 592                 /*
 593                  * Recommending a number of CPUs is somewhat arbitrary; we
 594                  * return the number of present CPUs for -HV (since a host
 595                  * will have secondary threads "offline"), and for other KVM
 596                  * implementations just count online CPUs.
 597                  */
 598                 if (hv_enabled)
 599                         r = num_present_cpus();
 600                 else
 601                         r = num_online_cpus();
 602                 break;
 603         case KVM_CAP_NR_MEMSLOTS:
 604                 r = KVM_USER_MEM_SLOTS;
 605                 break;
 606         case KVM_CAP_MAX_VCPUS:
 607                 r = KVM_MAX_VCPUS;
 608                 break;
 609 #ifdef CONFIG_PPC_BOOK3S_64
 610         case KVM_CAP_PPC_GET_SMMU_INFO:
 611                 r = 1;
 612                 break;
 613         case KVM_CAP_SPAPR_MULTITCE:
 614                 r = 1;
 615                 break;
 616         case KVM_CAP_SPAPR_RESIZE_HPT:
 617                 /* Disable this on POWER9 until code handles new HPTE format */
 618                 r = !!hv_enabled && !cpu_has_feature(CPU_FTR_ARCH_300);
 619                 break;
 620 #endif
 621         case KVM_CAP_PPC_HTM:
 622                 r = cpu_has_feature(CPU_FTR_TM_COMP) &&
 623                     is_kvmppc_hv_enabled(kvm);
 624                 break;
 625         default:
 626                 r = 0;
 627                 break;
 628         }
 629         return r;
 630
 631 }
 632
 633 long kvm_arch_dev_ioctl(struct file *filp,
 634                         unsigned int ioctl, unsigned long arg)
 635 {
 636         return -EINVAL;
 637 }
 638
 639 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 640                            struct kvm_memory_slot *dont)
 641 {
 642         kvmppc_core_free_memslot(kvm, free, dont);
 643 }
 644
 645 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 646                             unsigned long npages)
 647 {
 648         return kvmppc_core_create_memslot(kvm, slot, npages);
 649 }
 650
 651 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 652                                    struct kvm_memory_slot *memslot,
 653                                    const struct kvm_userspace_memory_region *mem,
 654                                    enum kvm_mr_change change)
 655 {
 656         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 657 }
 658
 659 void kvm_arch_commit_memory_region(struct kvm *kvm,
 660                                    const struct kvm_userspace_memory_region *mem,
 661                                    const struct kvm_memory_slot *old,
 662                                    const struct kvm_memory_slot *new,
 663                                    enum kvm_mr_change change)
 664 {
 665         kvmppc_core_commit_memory_region(kvm, mem, old, new);
 666 }
 667
 668 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 669                                    struct kvm_memory_slot *slot)
 670 {
 671         kvmppc_core_flush_memslot(kvm, slot);
 672 }
 673
 674 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 675 {
 676         struct kvm_vcpu *vcpu;
 677         vcpu = kvmppc_core_vcpu_create(kvm, id);
 678         if (!IS_ERR(vcpu)) {
 679                 vcpu->arch.wqp = &vcpu->wq;
 680                 kvmppc_create_vcpu_debugfs(vcpu, id);
 681         }
 682         return vcpu;
 683 }
 684
 685 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 686 {
 687 }
 688
 689 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 690 {
 691         /* Make sure we're not using the vcpu anymore */
 692         hrtimer_cancel(&vcpu->arch.dec_timer);
 693
 694         kvmppc_remove_vcpu_debugfs(vcpu);
 695
 696         switch (vcpu->arch.irq_type) {
 697         case KVMPPC_IRQ_MPIC:
 698                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 699                 break;
 700         case KVMPPC_IRQ_XICS:
 701                 kvmppc_xics_free_icp(vcpu);
 702                 break;
 703         }
 704
 705         kvmppc_core_vcpu_free(vcpu);
 706 }
 707
 708 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 709 {
 710         kvm_arch_vcpu_free(vcpu);
 711 }
 712
 713 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 714 {
 715         return kvmppc_core_pending_dec(vcpu);
 716 }
 717
 718 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 719 {
 720         struct kvm_vcpu *vcpu;
 721
 722         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 723         kvmppc_decrementer_func(vcpu);
 724
 725         return HRTIMER_NORESTART;
 726 }
 727
 728 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 729 {
 730         int ret;
 731
 732         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 733         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 734         vcpu->arch.dec_expires = ~(u64)0;
 735
 736 #ifdef CONFIG_KVM_EXIT_TIMING
 737         mutex_init(&vcpu->arch.exit_timing_lock);
 738 #endif
 739         ret = kvmppc_subarch_vcpu_init(vcpu);
 740         return ret;
 741 }
 742
 743 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 744 {
 745         kvmppc_mmu_destroy(vcpu);
 746         kvmppc_subarch_vcpu_uninit(vcpu);
 747 }
 748
 749 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 750 {
 751 #ifdef CONFIG_BOOKE
 752         /*
 753          * vrsave (formerly usprg0) isn't used by Linux, but may
 754          * be used by the guest.
 755          *
 756          * On non-booke this is associated with Altivec and
 757          * is handled by code in book3s.c.
 758          */
 759         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 760 #endif
 761         kvmppc_core_vcpu_load(vcpu, cpu);
 762 }
 763
 764 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 765 {
 766         kvmppc_core_vcpu_put(vcpu);
 767 #ifdef CONFIG_BOOKE
 768         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 769 #endif
 770 }
 771
 772 /*
 773  * irq_bypass_add_producer and irq_bypass_del_producer are only
 774  * useful if the architecture supports PCI passthrough.
 775  * irq_bypass_stop and irq_bypass_start are not needed and so
 776  * kvm_ops are not defined for them.
 777  */
 778 bool kvm_arch_has_irq_bypass(void)
 779 {
 780         return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
 781                 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
 782 }
 783
 784 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
 785                                      struct irq_bypass_producer *prod)
 786 {
 787         struct kvm_kernel_irqfd *irqfd =
 788                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 789         struct kvm *kvm = irqfd->kvm;
 790
 791         if (kvm->arch.kvm_ops->irq_bypass_add_producer)
 792                 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
 793
 794         return 0;
 795 }
 796
 797 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
 798                                       struct irq_bypass_producer *prod)
 799 {
 800         struct kvm_kernel_irqfd *irqfd =
 801                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 802         struct kvm *kvm = irqfd->kvm;
 803
 804         if (kvm->arch.kvm_ops->irq_bypass_del_producer)
 805                 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
 806 }
 807
 808 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 809                                       struct kvm_run *run)
 810 {
 811         u64 uninitialized_var(gpr);
 812
 813         if (run->mmio.len > sizeof(gpr)) {
 814                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 815                 return;
 816         }
 817
 818         if (!vcpu->arch.mmio_host_swabbed) {
 819                 switch (run->mmio.len) {
 820                 case 8: gpr = *(u64 *)run->mmio.data; break;
 821                 case 4: gpr = *(u32 *)run->mmio.data; break;
 822                 case 2: gpr = *(u16 *)run->mmio.data; break;
 823                 case 1: gpr = *(u8 *)run->mmio.data; break;
 824                 }
 825         } else {
 826                 switch (run->mmio.len) {
 827                 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
 828                 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
 829                 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
 830                 case 1: gpr = *(u8 *)run->mmio.data; break;
 831                 }
 832         }
 833
 834         if (vcpu->arch.mmio_sign_extend) {
 835                 switch (run->mmio.len) {
 836 #ifdef CONFIG_PPC64
 837                 case 4:
 838                         gpr = (s64)(s32)gpr;
 839                         break;
 840 #endif
 841                 case 2:
 842                         gpr = (s64)(s16)gpr;
 843                         break;
 844                 case 1:
 845                         gpr = (s64)(s8)gpr;
 846                         break;
 847                 }
 848         }
 849
 850         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 851
 852         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 853         case KVM_MMIO_REG_GPR:
 854                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 855                 break;
 856         case KVM_MMIO_REG_FPR:
 857                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
 858                 break;
 859 #ifdef CONFIG_PPC_BOOK3S
 860         case KVM_MMIO_REG_QPR:
 861                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 862                 break;
 863         case KVM_MMIO_REG_FQPR:
 864                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
 865                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 866                 break;
 867 #endif
 868         default:
 869                 BUG();
 870         }
 871 }
 872
 873 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 874                                 unsigned int rt, unsigned int bytes,
 875                                 int is_default_endian, int sign_extend)
 876 {
 877         int idx, ret;
 878         bool host_swabbed;
 879
 880         /* Pity C doesn't have a logical XOR operator */
 881         if (kvmppc_need_byteswap(vcpu)) {
 882                 host_swabbed = is_default_endian;
 883         } else {
 884                 host_swabbed = !is_default_endian;
 885         }
 886
 887         if (bytes > sizeof(run->mmio.data)) {
 888                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 889                        run->mmio.len);
 890         }
 891
 892         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 893         run->mmio.len = bytes;
 894         run->mmio.is_write = 0;
 895
 896         vcpu->arch.io_gpr = rt;
 897         vcpu->arch.mmio_host_swabbed = host_swabbed;
 898         vcpu->mmio_needed = 1;
 899         vcpu->mmio_is_write = 0;
 900         vcpu->arch.mmio_sign_extend = sign_extend;
 901
 902         idx = srcu_read_lock(&vcpu->kvm->srcu);
 903
 904         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
 905                               bytes, &run->mmio.data);
 906
 907         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 908
 909         if (!ret) {
 910                 kvmppc_complete_mmio_load(vcpu, run);
 911                 vcpu->mmio_needed = 0;
 912                 return EMULATE_DONE;
 913         }
 914
 915         return EMULATE_DO_MMIO;
 916 }
 917
 918 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 919                        unsigned int rt, unsigned int bytes,
 920                        int is_default_endian)
 921 {
 922         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
 923 }
 924 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
 925
 926 /* Same as above, but sign extends */
 927 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 928                         unsigned int rt, unsigned int bytes,
 929                         int is_default_endian)
 930 {
 931         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
 932 }
 933
 934 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 935                         u64 val, unsigned int bytes, int is_default_endian)
 936 {
 937         void *data = run->mmio.data;
 938         int idx, ret;
 939         bool host_swabbed;
 940
 941         /* Pity C doesn't have a logical XOR operator */
 942         if (kvmppc_need_byteswap(vcpu)) {
 943                 host_swabbed = is_default_endian;
 944         } else {
 945                 host_swabbed = !is_default_endian;
 946         }
 947
 948         if (bytes > sizeof(run->mmio.data)) {
 949                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 950                        run->mmio.len);
 951         }
 952
 953         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 954         run->mmio.len = bytes;
 955         run->mmio.is_write = 1;
 956         vcpu->mmio_needed = 1;
 957         vcpu->mmio_is_write = 1;
 958
 959         /* Store the value at the lowest bytes in 'data'. */
 960         if (!host_swabbed) {
 961                 switch (bytes) {
 962                 case 8: *(u64 *)data = val; break;
 963                 case 4: *(u32 *)data = val; break;
 964                 case 2: *(u16 *)data = val; break;
 965                 case 1: *(u8  *)data = val; break;
 966                 }
 967         } else {
 968                 switch (bytes) {
 969                 case 8: *(u64 *)data = swab64(val); break;
 970                 case 4: *(u32 *)data = swab32(val); break;
 971                 case 2: *(u16 *)data = swab16(val); break;
 972                 case 1: *(u8  *)data = val; break;
 973                 }
 974         }
 975
 976         idx = srcu_read_lock(&vcpu->kvm->srcu);
 977
 978         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
 979                                bytes, &run->mmio.data);
 980
 981         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 982
 983         if (!ret) {
 984                 vcpu->mmio_needed = 0;
 985                 return EMULATE_DONE;
 986         }
 987
 988         return EMULATE_DO_MMIO;
 989 }
 990 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
 991
 992 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
 993 {
 994         int r = 0;
 995         union kvmppc_one_reg val;
 996         int size;
 997
 998         size = one_reg_size(reg->id);
 999         if (size > sizeof(val))
1000                 return -EINVAL;
1001
1002         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1003         if (r == -EINVAL) {
1004                 r = 0;
1005                 switch (reg->id) {
1006 #ifdef CONFIG_ALTIVEC
1007                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1008                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1009                                 r = -ENXIO;
1010                                 break;
1011                         }
1012                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1013                         break;
1014                 case KVM_REG_PPC_VSCR:
1015                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1016                                 r = -ENXIO;
1017                                 break;
1018                         }
1019                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1020                         break;
1021                 case KVM_REG_PPC_VRSAVE:
1022                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
1023                         break;
1024 #endif /* CONFIG_ALTIVEC */
1025                 default:
1026                         r = -EINVAL;
1027                         break;
1028                 }
1029         }
1030
1031         if (r)
1032                 return r;
1033
1034         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1035                 r = -EFAULT;
1036
1037         return r;
1038 }
1039
1040 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1041 {
1042         int r;
1043         union kvmppc_one_reg val;
1044         int size;
1045
1046         size = one_reg_size(reg->id);
1047         if (size > sizeof(val))
1048                 return -EINVAL;
1049
1050         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1051                 return -EFAULT;
1052
1053         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1054         if (r == -EINVAL) {
1055                 r = 0;
1056                 switch (reg->id) {
1057 #ifdef CONFIG_ALTIVEC
1058                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1059                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1060                                 r = -ENXIO;
1061                                 break;
1062                         }
1063                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1064                         break;
1065                 case KVM_REG_PPC_VSCR:
1066                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1067                                 r = -ENXIO;
1068                                 break;
1069                         }
1070                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1071                         break;
1072                 case KVM_REG_PPC_VRSAVE:
1073                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1074                                 r = -ENXIO;
1075                                 break;
1076                         }
1077                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
1078                         break;
1079 #endif /* CONFIG_ALTIVEC */
1080                 default:
1081                         r = -EINVAL;
1082                         break;
1083                 }
1084         }
1085
1086         return r;
1087 }
1088
1089 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1090 {
1091         int r;
1092         sigset_t sigsaved;
1093
1094         if (vcpu->sigset_active)
1095                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1096
1097         if (vcpu->mmio_needed) {
1098                 if (!vcpu->mmio_is_write)
1099                         kvmppc_complete_mmio_load(vcpu, run);
1100                 vcpu->mmio_needed = 0;
1101         } else if (vcpu->arch.osi_needed) {
1102                 u64 *gprs = run->osi.gprs;
1103                 int i;
1104
1105                 for (i = 0; i < 32; i++)
1106                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1107                 vcpu->arch.osi_needed = 0;
1108         } else if (vcpu->arch.hcall_needed) {
1109                 int i;
1110
1111                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1112                 for (i = 0; i < 9; ++i)
1113                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1114                 vcpu->arch.hcall_needed = 0;
1115 #ifdef CONFIG_BOOKE
1116         } else if (vcpu->arch.epr_needed) {
1117                 kvmppc_set_epr(vcpu, run->epr.epr);
1118                 vcpu->arch.epr_needed = 0;
1119 #endif
1120         }
1121
1122         if (run->immediate_exit)
1123                 r = -EINTR;
1124         else
1125                 r = kvmppc_vcpu_run(run, vcpu);
1126
1127         if (vcpu->sigset_active)
1128                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1129
1130         return r;
1131 }
1132
1133 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1134 {
1135         if (irq->irq == KVM_INTERRUPT_UNSET) {
1136                 kvmppc_core_dequeue_external(vcpu);
1137                 return 0;
1138         }
1139
1140         kvmppc_core_queue_external(vcpu, irq);
1141
1142         kvm_vcpu_kick(vcpu);
1143
1144         return 0;
1145 }
1146
1147 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1148                                      struct kvm_enable_cap *cap)
1149 {
1150         int r;
1151
1152         if (cap->flags)
1153                 return -EINVAL;
1154
1155         switch (cap->cap) {
1156         case KVM_CAP_PPC_OSI:
1157                 r = 0;
1158                 vcpu->arch.osi_enabled = true;
1159                 break;
1160         case KVM_CAP_PPC_PAPR:
1161                 r = 0;
1162                 vcpu->arch.papr_enabled = true;
1163                 break;
1164         case KVM_CAP_PPC_EPR:
1165                 r = 0;
1166                 if (cap->args[0])
1167                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1168                 else
1169                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1170                 break;
1171 #ifdef CONFIG_BOOKE
1172         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1173                 r = 0;
1174                 vcpu->arch.watchdog_enabled = true;
1175                 break;
1176 #endif
1177 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1178         case KVM_CAP_SW_TLB: {
1179                 struct kvm_config_tlb cfg;
1180                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1181
1182                 r = -EFAULT;
1183                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1184                         break;
1185
1186                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1187                 break;
1188         }
1189 #endif
1190 #ifdef CONFIG_KVM_MPIC
1191         case KVM_CAP_IRQ_MPIC: {
1192                 struct fd f;
1193                 struct kvm_device *dev;
1194
1195                 r = -EBADF;
1196                 f = fdget(cap->args[0]);
1197                 if (!f.file)
1198                         break;
1199
1200                 r = -EPERM;
1201                 dev = kvm_device_from_filp(f.file);
1202                 if (dev)
1203                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1204
1205                 fdput(f);
1206                 break;
1207         }
1208 #endif
1209 #ifdef CONFIG_KVM_XICS
1210         case KVM_CAP_IRQ_XICS: {
1211                 struct fd f;
1212                 struct kvm_device *dev;
1213
1214                 r = -EBADF;
1215                 f = fdget(cap->args[0]);
1216                 if (!f.file)
1217                         break;
1218
1219                 r = -EPERM;
1220                 dev = kvm_device_from_filp(f.file);
1221                 if (dev)
1222                         r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1223
1224                 fdput(f);
1225                 break;
1226         }
1227 #endif /* CONFIG_KVM_XICS */
1228         default:
1229                 r = -EINVAL;
1230                 break;
1231         }
1232
1233         if (!r)
1234                 r = kvmppc_sanity_check(vcpu);
1235
1236         return r;
1237 }
1238
1239 bool kvm_arch_intc_initialized(struct kvm *kvm)
1240 {
1241 #ifdef CONFIG_KVM_MPIC
1242         if (kvm->arch.mpic)
1243                 return true;
1244 #endif
1245 #ifdef CONFIG_KVM_XICS
1246         if (kvm->arch.xics)
1247                 return true;
1248 #endif
1249         return false;
1250 }
1251
1252 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1253                                     struct kvm_mp_state *mp_state)
1254 {
1255         return -EINVAL;
1256 }
1257
1258 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1259                                     struct kvm_mp_state *mp_state)
1260 {
1261         return -EINVAL;
1262 }
1263
1264 long kvm_arch_vcpu_ioctl(struct file *filp,
1265                          unsigned int ioctl, unsigned long arg)
1266 {
1267         struct kvm_vcpu *vcpu = filp->private_data;
1268         void __user *argp = (void __user *)arg;
1269         long r;
1270
1271         switch (ioctl) {
1272         case KVM_INTERRUPT: {
1273                 struct kvm_interrupt irq;
1274                 r = -EFAULT;
1275                 if (copy_from_user(&irq, argp, sizeof(irq)))
1276                         goto out;
1277                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1278                 goto out;
1279         }
1280
1281         case KVM_ENABLE_CAP:
1282         {
1283                 struct kvm_enable_cap cap;
1284                 r = -EFAULT;
1285                 if (copy_from_user(&cap, argp, sizeof(cap)))
1286                         goto out;
1287                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1288                 break;
1289         }
1290
1291         case KVM_SET_ONE_REG:
1292         case KVM_GET_ONE_REG:
1293         {
1294                 struct kvm_one_reg reg;
1295                 r = -EFAULT;
1296                 if (copy_from_user(&reg, argp, sizeof(reg)))
1297                         goto out;
1298                 if (ioctl == KVM_SET_ONE_REG)
1299                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1300                 else
1301                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1302                 break;
1303         }
1304
1305 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1306         case KVM_DIRTY_TLB: {
1307                 struct kvm_dirty_tlb dirty;
1308                 r = -EFAULT;
1309                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1310                         goto out;
1311                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1312                 break;
1313         }
1314 #endif
1315         default:
1316                 r = -EINVAL;
1317         }
1318
1319 out:
1320         return r;
1321 }
1322
1323 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1324 {
1325         return VM_FAULT_SIGBUS;
1326 }
1327
1328 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1329 {
1330         u32 inst_nop = 0x60000000;
1331 #ifdef CONFIG_KVM_BOOKE_HV
1332         u32 inst_sc1 = 0x44000022;
1333         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1334         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1335         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1336         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1337 #else
1338         u32 inst_lis = 0x3c000000;
1339         u32 inst_ori = 0x60000000;
1340         u32 inst_sc = 0x44000002;
1341         u32 inst_imm_mask = 0xffff;
1342
1343         /*
1344          * The hypercall to get into KVM from within guest context is as
1345          * follows:
1346          *
1347          *    lis r0, r0, KVM_SC_MAGIC_R0@h
1348          *    ori r0, KVM_SC_MAGIC_R0@l
1349          *    sc
1350          *    nop
1351          */
1352         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1353         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1354         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1355         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1356 #endif
1357
1358         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1359
1360         return 0;
1361 }
1362
1363 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1364                           bool line_status)
1365 {
1366         if (!irqchip_in_kernel(kvm))
1367                 return -ENXIO;
1368
1369         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1370                                         irq_event->irq, irq_event->level,
1371                                         line_status);
1372         return 0;
1373 }
1374
1375
1376 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1377                                    struct kvm_enable_cap *cap)
1378 {
1379         int r;
1380
1381         if (cap->flags)
1382                 return -EINVAL;
1383
1384         switch (cap->cap) {
1385 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1386         case KVM_CAP_PPC_ENABLE_HCALL: {
1387                 unsigned long hcall = cap->args[0];
1388
1389                 r = -EINVAL;
1390                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1391                     cap->args[1] > 1)
1392                         break;
1393                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1394                         break;
1395                 if (cap->args[1])
1396                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1397                 else
1398                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1399                 r = 0;
1400                 break;
1401         }
1402 #endif
1403         default:
1404                 r = -EINVAL;
1405                 break;
1406         }
1407
1408         return r;
1409 }
1410
1411 long kvm_arch_vm_ioctl(struct file *filp,
1412                        unsigned int ioctl, unsigned long arg)
1413 {
1414         struct kvm *kvm __maybe_unused = filp->private_data;
1415         void __user *argp = (void __user *)arg;
1416         long r;
1417
1418         switch (ioctl) {
1419         case KVM_PPC_GET_PVINFO: {
1420                 struct kvm_ppc_pvinfo pvinfo;
1421                 memset(&pvinfo, 0, sizeof(pvinfo));
1422                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1423                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1424                         r = -EFAULT;
1425                         goto out;
1426                 }
1427
1428                 break;
1429         }
1430         case KVM_ENABLE_CAP:
1431         {
1432                 struct kvm_enable_cap cap;
1433                 r = -EFAULT;
1434                 if (copy_from_user(&cap, argp, sizeof(cap)))
1435                         goto out;
1436                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1437                 break;
1438         }
1439 #ifdef CONFIG_PPC_BOOK3S_64
1440         case KVM_CREATE_SPAPR_TCE_64: {
1441                 struct kvm_create_spapr_tce_64 create_tce_64;
1442
1443                 r = -EFAULT;
1444                 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
1445                         goto out;
1446                 if (create_tce_64.flags) {
1447                         r = -EINVAL;
1448                         goto out;
1449                 }
1450                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1451                 goto out;
1452         }
1453         case KVM_CREATE_SPAPR_TCE: {
1454                 struct kvm_create_spapr_tce create_tce;
1455                 struct kvm_create_spapr_tce_64 create_tce_64;
1456
1457                 r = -EFAULT;
1458                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1459                         goto out;
1460
1461                 create_tce_64.liobn = create_tce.liobn;
1462                 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
1463                 create_tce_64.offset = 0;
1464                 create_tce_64.size = create_tce.window_size >>
1465                                 IOMMU_PAGE_SHIFT_4K;
1466                 create_tce_64.flags = 0;
1467                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1468                 goto out;
1469         }
1470         case KVM_PPC_GET_SMMU_INFO: {
1471                 struct kvm_ppc_smmu_info info;
1472                 struct kvm *kvm = filp->private_data;
1473
1474                 memset(&info, 0, sizeof(info));
1475                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1476                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1477                         r = -EFAULT;
1478                 break;
1479         }
1480         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1481                 struct kvm *kvm = filp->private_data;
1482
1483                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1484                 break;
1485         }
1486         case KVM_PPC_CONFIGURE_V3_MMU: {
1487                 struct kvm *kvm = filp->private_data;
1488                 struct kvm_ppc_mmuv3_cfg cfg;
1489
1490                 r = -EINVAL;
1491                 if (!kvm->arch.kvm_ops->configure_mmu)
1492                         goto out;
1493                 r = -EFAULT;
1494                 if (copy_from_user(&cfg, argp, sizeof(cfg)))
1495                         goto out;
1496                 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
1497                 break;
1498         }
1499         case KVM_PPC_GET_RMMU_INFO: {
1500                 struct kvm *kvm = filp->private_data;
1501                 struct kvm_ppc_rmmu_info info;
1502
1503                 r = -EINVAL;
1504                 if (!kvm->arch.kvm_ops->get_rmmu_info)
1505                         goto out;
1506                 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
1507                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1508                         r = -EFAULT;
1509                 break;
1510         }
1511         default: {
1512                 struct kvm *kvm = filp->private_data;
1513                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1514         }
1515 #else /* CONFIG_PPC_BOOK3S_64 */
1516         default:
1517                 r = -ENOTTY;
1518 #endif
1519         }
1520 out:
1521         return r;
1522 }
1523
1524 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1525 static unsigned long nr_lpids;
1526
1527 long kvmppc_alloc_lpid(void)
1528 {
1529         long lpid;
1530
1531         do {
1532                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1533                 if (lpid >= nr_lpids) {
1534                         pr_err("%s: No LPIDs free\n", __func__);
1535                         return -ENOMEM;
1536                 }
1537         } while (test_and_set_bit(lpid, lpid_inuse));
1538
1539         return lpid;
1540 }
1541 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1542
1543 void kvmppc_claim_lpid(long lpid)
1544 {
1545         set_bit(lpid, lpid_inuse);
1546 }
1547 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1548
1549 void kvmppc_free_lpid(long lpid)
1550 {
1551         clear_bit(lpid, lpid_inuse);
1552 }
1553 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1554
1555 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1556 {
1557         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1558         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1559 }
1560 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1561
1562 int kvm_arch_init(void *opaque)
1563 {
1564         return 0;
1565 }
1566
1567 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);