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 "timing.h"
  41 #include "irq.h"
  42 #include "../mm/mmu_decl.h"
  43
  44 #define CREATE_TRACE_POINTS
  45 #include "trace.h"
  46
  47 struct kvmppc_ops *kvmppc_hv_ops;
  48 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
  49 struct kvmppc_ops *kvmppc_pr_ops;
  50 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
  51
  52
  53 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  54 {
  55         return !!(v->arch.pending_exceptions) ||
  56                v->requests;
  57 }
  58
  59 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  60 {
  61         return 1;
  62 }
  63
  64 /*
  65  * Common checks before entering the guest world.  Call with interrupts
  66  * disabled.
  67  *
  68  * returns:
  69  *
  70  * == 1 if we're ready to go into guest state
  71  * <= 0 if we need to go back to the host with return value
  72  */
  73 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  74 {
  75         int r;
  76
  77         WARN_ON(irqs_disabled());
  78         hard_irq_disable();
  79
  80         while (true) {
  81                 if (need_resched()) {
  82                         local_irq_enable();
  83                         cond_resched();
  84                         hard_irq_disable();
  85                         continue;
  86                 }
  87
  88                 if (signal_pending(current)) {
  89                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
  90                         vcpu->run->exit_reason = KVM_EXIT_INTR;
  91                         r = -EINTR;
  92                         break;
  93                 }
  94
  95                 vcpu->mode = IN_GUEST_MODE;
  96
  97                 /*
  98                  * Reading vcpu->requests must happen after setting vcpu->mode,
  99                  * so we don't miss a request because the requester sees
 100                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
 101                  * before next entering the guest (and thus doesn't IPI).
 102                  * This also orders the write to mode from any reads
 103                  * to the page tables done while the VCPU is running.
 104                  * Please see the comment in kvm_flush_remote_tlbs.
 105                  */
 106                 smp_mb();
 107
 108                 if (vcpu->requests) {
 109                         /* Make sure we process requests preemptable */
 110                         local_irq_enable();
 111                         trace_kvm_check_requests(vcpu);
 112                         r = kvmppc_core_check_requests(vcpu);
 113                         hard_irq_disable();
 114                         if (r > 0)
 115                                 continue;
 116                         break;
 117                 }
 118
 119                 if (kvmppc_core_prepare_to_enter(vcpu)) {
 120                         /* interrupts got enabled in between, so we
 121                            are back at square 1 */
 122                         continue;
 123                 }
 124
 125                 guest_enter_irqoff();
 126                 return 1;
 127         }
 128
 129         /* return to host */
 130         local_irq_enable();
 131         return r;
 132 }
 133 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
 134
 135 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 136 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
 137 {
 138         struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
 139         int i;
 140
 141         shared->sprg0 = swab64(shared->sprg0);
 142         shared->sprg1 = swab64(shared->sprg1);
 143         shared->sprg2 = swab64(shared->sprg2);
 144         shared->sprg3 = swab64(shared->sprg3);
 145         shared->srr0 = swab64(shared->srr0);
 146         shared->srr1 = swab64(shared->srr1);
 147         shared->dar = swab64(shared->dar);
 148         shared->msr = swab64(shared->msr);
 149         shared->dsisr = swab32(shared->dsisr);
 150         shared->int_pending = swab32(shared->int_pending);
 151         for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
 152                 shared->sr[i] = swab32(shared->sr[i]);
 153 }
 154 #endif
 155
 156 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
 157 {
 158         int nr = kvmppc_get_gpr(vcpu, 11);
 159         int r;
 160         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
 161         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
 162         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
 163         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
 164         unsigned long r2 = 0;
 165
 166         if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
 167                 /* 32 bit mode */
 168                 param1 &= 0xffffffff;
 169                 param2 &= 0xffffffff;
 170                 param3 &= 0xffffffff;
 171                 param4 &= 0xffffffff;
 172         }
 173
 174         switch (nr) {
 175         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
 176         {
 177 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 178                 /* Book3S can be little endian, find it out here */
 179                 int shared_big_endian = true;
 180                 if (vcpu->arch.intr_msr & MSR_LE)
 181                         shared_big_endian = false;
 182                 if (shared_big_endian != vcpu->arch.shared_big_endian)
 183                         kvmppc_swab_shared(vcpu);
 184                 vcpu->arch.shared_big_endian = shared_big_endian;
 185 #endif
 186
 187                 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
 188                         /*
 189                          * Older versions of the Linux magic page code had
 190                          * a bug where they would map their trampoline code
 191                          * NX. If that's the case, remove !PR NX capability.
 192                          */
 193                         vcpu->arch.disable_kernel_nx = true;
 194                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 195                 }
 196
 197                 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
 198                 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
 199
 200 #ifdef CONFIG_PPC_64K_PAGES
 201                 /*
 202                  * Make sure our 4k magic page is in the same window of a 64k
 203                  * page within the guest and within the host's page.
 204                  */
 205                 if ((vcpu->arch.magic_page_pa & 0xf000) !=
 206                     ((ulong)vcpu->arch.shared & 0xf000)) {
 207                         void *old_shared = vcpu->arch.shared;
 208                         ulong shared = (ulong)vcpu->arch.shared;
 209                         void *new_shared;
 210
 211                         shared &= PAGE_MASK;
 212                         shared |= vcpu->arch.magic_page_pa & 0xf000;
 213                         new_shared = (void*)shared;
 214                         memcpy(new_shared, old_shared, 0x1000);
 215                         vcpu->arch.shared = new_shared;
 216                 }
 217 #endif
 218
 219                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 220
 221                 r = EV_SUCCESS;
 222                 break;
 223         }
 224         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 225                 r = EV_SUCCESS;
 226 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 227                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 228 #endif
 229
 230                 /* Second return value is in r4 */
 231                 break;
 232         case EV_HCALL_TOKEN(EV_IDLE):
 233                 r = EV_SUCCESS;
 234                 kvm_vcpu_block(vcpu);
 235                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 236                 break;
 237         default:
 238                 r = EV_UNIMPLEMENTED;
 239                 break;
 240         }
 241
 242         kvmppc_set_gpr(vcpu, 4, r2);
 243
 244         return r;
 245 }
 246 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
 247
 248 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
 249 {
 250         int r = false;
 251
 252         /* We have to know what CPU to virtualize */
 253         if (!vcpu->arch.pvr)
 254                 goto out;
 255
 256         /* PAPR only works with book3s_64 */
 257         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
 258                 goto out;
 259
 260         /* HV KVM can only do PAPR mode for now */
 261         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
 262                 goto out;
 263
 264 #ifdef CONFIG_KVM_BOOKE_HV
 265         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 266                 goto out;
 267 #endif
 268
 269         r = true;
 270
 271 out:
 272         vcpu->arch.sane = r;
 273         return r ? 0 : -EINVAL;
 274 }
 275 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
 276
 277 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 278 {
 279         enum emulation_result er;
 280         int r;
 281
 282         er = kvmppc_emulate_loadstore(vcpu);
 283         switch (er) {
 284         case EMULATE_DONE:
 285                 /* Future optimization: only reload non-volatiles if they were
 286                  * actually modified. */
 287                 r = RESUME_GUEST_NV;
 288                 break;
 289         case EMULATE_AGAIN:
 290                 r = RESUME_GUEST;
 291                 break;
 292         case EMULATE_DO_MMIO:
 293                 run->exit_reason = KVM_EXIT_MMIO;
 294                 /* We must reload nonvolatiles because "update" load/store
 295                  * instructions modify register state. */
 296                 /* Future optimization: only reload non-volatiles if they were
 297                  * actually modified. */
 298                 r = RESUME_HOST_NV;
 299                 break;
 300         case EMULATE_FAIL:
 301         {
 302                 u32 last_inst;
 303
 304                 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
 305                 /* XXX Deliver Program interrupt to guest. */
 306                 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
 307                 r = RESUME_HOST;
 308                 break;
 309         }
 310         default:
 311                 WARN_ON(1);
 312                 r = RESUME_GUEST;
 313         }
 314
 315         return r;
 316 }
 317 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
 318
 319 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 320               bool data)
 321 {
 322         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 323         struct kvmppc_pte pte;
 324         int r;
 325
 326         vcpu->stat.st++;
 327
 328         r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 329                          XLATE_WRITE, &pte);
 330         if (r < 0)
 331                 return r;
 332
 333         *eaddr = pte.raddr;
 334
 335         if (!pte.may_write)
 336                 return -EPERM;
 337
 338         /* Magic page override */
 339         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 340             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 341             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 342                 void *magic = vcpu->arch.shared;
 343                 magic += pte.eaddr & 0xfff;
 344                 memcpy(magic, ptr, size);
 345                 return EMULATE_DONE;
 346         }
 347
 348         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 349                 return EMULATE_DO_MMIO;
 350
 351         return EMULATE_DONE;
 352 }
 353 EXPORT_SYMBOL_GPL(kvmppc_st);
 354
 355 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 356                       bool data)
 357 {
 358         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 359         struct kvmppc_pte pte;
 360         int rc;
 361
 362         vcpu->stat.ld++;
 363
 364         rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 365                           XLATE_READ, &pte);
 366         if (rc)
 367                 return rc;
 368
 369         *eaddr = pte.raddr;
 370
 371         if (!pte.may_read)
 372                 return -EPERM;
 373
 374         if (!data && !pte.may_execute)
 375                 return -ENOEXEC;
 376
 377         /* Magic page override */
 378         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 379             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 380             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 381                 void *magic = vcpu->arch.shared;
 382                 magic += pte.eaddr & 0xfff;
 383                 memcpy(ptr, magic, size);
 384                 return EMULATE_DONE;
 385         }
 386
 387         if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
 388                 return EMULATE_DO_MMIO;
 389
 390         return EMULATE_DONE;
 391 }
 392 EXPORT_SYMBOL_GPL(kvmppc_ld);
 393
 394 int kvm_arch_hardware_enable(void)
 395 {
 396         return 0;
 397 }
 398
 399 int kvm_arch_hardware_setup(void)
 400 {
 401         return 0;
 402 }
 403
 404 void kvm_arch_check_processor_compat(void *rtn)
 405 {
 406         *(int *)rtn = kvmppc_core_check_processor_compat();
 407 }
 408
 409 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 410 {
 411         struct kvmppc_ops *kvm_ops = NULL;
 412         /*
 413          * if we have both HV and PR enabled, default is HV
 414          */
 415         if (type == 0) {
 416                 if (kvmppc_hv_ops)
 417                         kvm_ops = kvmppc_hv_ops;
 418                 else
 419                         kvm_ops = kvmppc_pr_ops;
 420                 if (!kvm_ops)
 421                         goto err_out;
 422         } else  if (type == KVM_VM_PPC_HV) {
 423                 if (!kvmppc_hv_ops)
 424                         goto err_out;
 425                 kvm_ops = kvmppc_hv_ops;
 426         } else if (type == KVM_VM_PPC_PR) {
 427                 if (!kvmppc_pr_ops)
 428                         goto err_out;
 429                 kvm_ops = kvmppc_pr_ops;
 430         } else
 431                 goto err_out;
 432
 433         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
 434                 return -ENOENT;
 435
 436         kvm->arch.kvm_ops = kvm_ops;
 437         return kvmppc_core_init_vm(kvm);
 438 err_out:
 439         return -EINVAL;
 440 }
 441
 442 bool kvm_arch_has_vcpu_debugfs(void)
 443 {
 444         return false;
 445 }
 446
 447 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 448 {
 449         return 0;
 450 }
 451
 452 void kvm_arch_destroy_vm(struct kvm *kvm)
 453 {
 454         unsigned int i;
 455         struct kvm_vcpu *vcpu;
 456
 457 #ifdef CONFIG_KVM_XICS
 458         /*
 459          * We call kick_all_cpus_sync() to ensure that all
 460          * CPUs have executed any pending IPIs before we
 461          * continue and free VCPUs structures below.
 462          */
 463         if (is_kvmppc_hv_enabled(kvm))
 464                 kick_all_cpus_sync();
 465 #endif
 466
 467         kvm_for_each_vcpu(i, vcpu, kvm)
 468                 kvm_arch_vcpu_free(vcpu);
 469
 470         mutex_lock(&kvm->lock);
 471         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 472                 kvm->vcpus[i] = NULL;
 473
 474         atomic_set(&kvm->online_vcpus, 0);
 475
 476         kvmppc_core_destroy_vm(kvm);
 477
 478         mutex_unlock(&kvm->lock);
 479
 480         /* drop the module reference */
 481         module_put(kvm->arch.kvm_ops->owner);
 482 }
 483
 484 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 485 {
 486         int r;
 487         /* Assume we're using HV mode when the HV module is loaded */
 488         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
 489
 490         if (kvm) {
 491                 /*
 492                  * Hooray - we know which VM type we're running on. Depend on
 493                  * that rather than the guess above.
 494                  */
 495                 hv_enabled = is_kvmppc_hv_enabled(kvm);
 496         }
 497
 498         switch (ext) {
 499 #ifdef CONFIG_BOOKE
 500         case KVM_CAP_PPC_BOOKE_SREGS:
 501         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 502         case KVM_CAP_PPC_EPR:
 503 #else
 504         case KVM_CAP_PPC_SEGSTATE:
 505         case KVM_CAP_PPC_HIOR:
 506         case KVM_CAP_PPC_PAPR:
 507 #endif
 508         case KVM_CAP_PPC_UNSET_IRQ:
 509         case KVM_CAP_PPC_IRQ_LEVEL:
 510         case KVM_CAP_ENABLE_CAP:
 511         case KVM_CAP_ENABLE_CAP_VM:
 512         case KVM_CAP_ONE_REG:
 513         case KVM_CAP_IOEVENTFD:
 514         case KVM_CAP_DEVICE_CTRL:
 515         case KVM_CAP_IMMEDIATE_EXIT:
 516                 r = 1;
 517                 break;
 518         case KVM_CAP_PPC_PAIRED_SINGLES:
 519         case KVM_CAP_PPC_OSI:
 520         case KVM_CAP_PPC_GET_PVINFO:
 521 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 522         case KVM_CAP_SW_TLB:
 523 #endif
 524                 /* We support this only for PR */
 525                 r = !hv_enabled;
 526                 break;
 527 #ifdef CONFIG_KVM_MMIO
 528         case KVM_CAP_COALESCED_MMIO:
 529                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 530                 break;
 531 #endif
 532 #ifdef CONFIG_KVM_MPIC
 533         case KVM_CAP_IRQ_MPIC:
 534                 r = 1;
 535                 break;
 536 #endif
 537
 538 #ifdef CONFIG_PPC_BOOK3S_64
 539         case KVM_CAP_SPAPR_TCE:
 540         case KVM_CAP_SPAPR_TCE_64:
 541         case KVM_CAP_PPC_RTAS:
 542         case KVM_CAP_PPC_FIXUP_HCALL:
 543         case KVM_CAP_PPC_ENABLE_HCALL:
 544 #ifdef CONFIG_KVM_XICS
 545         case KVM_CAP_IRQ_XICS:
 546 #endif
 547                 r = 1;
 548                 break;
 549
 550         case KVM_CAP_PPC_ALLOC_HTAB:
 551                 r = hv_enabled;
 552                 break;
 553 #endif /* CONFIG_PPC_BOOK3S_64 */
 554 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 555         case KVM_CAP_PPC_SMT:
 556                 r = 0;
 557                 if (hv_enabled) {
 558                         if (cpu_has_feature(CPU_FTR_ARCH_300))
 559                                 r = 1;
 560                         else
 561                                 r = threads_per_subcore;
 562                 }
 563                 break;
 564         case KVM_CAP_PPC_RMA:
 565                 r = 0;
 566                 break;
 567         case KVM_CAP_PPC_HWRNG:
 568                 r = kvmppc_hwrng_present();
 569                 break;
 570         case KVM_CAP_PPC_MMU_RADIX:
 571                 r = !!(hv_enabled && radix_enabled());
 572                 break;
 573         case KVM_CAP_PPC_MMU_HASH_V3:
 574                 r = !!(hv_enabled && !radix_enabled() &&
 575                        cpu_has_feature(CPU_FTR_ARCH_300));
 576                 break;
 577 #endif
 578         case KVM_CAP_SYNC_MMU:
 579 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 580                 r = hv_enabled;
 581 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 582                 r = 1;
 583 #else
 584                 r = 0;
 585 #endif
 586                 break;
 587 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 588         case KVM_CAP_PPC_HTAB_FD:
 589                 r = hv_enabled;
 590                 break;
 591 #endif
 592         case KVM_CAP_NR_VCPUS:
 593                 /*
 594                  * Recommending a number of CPUs is somewhat arbitrary; we
 595                  * return the number of present CPUs for -HV (since a host
 596                  * will have secondary threads "offline"), and for other KVM
 597                  * implementations just count online CPUs.
 598                  */
 599                 if (hv_enabled)
 600                         r = num_present_cpus();
 601                 else
 602                         r = num_online_cpus();
 603                 break;
 604         case KVM_CAP_NR_MEMSLOTS:
 605                 r = KVM_USER_MEM_SLOTS;
 606                 break;
 607         case KVM_CAP_MAX_VCPUS:
 608                 r = KVM_MAX_VCPUS;
 609                 break;
 610 #ifdef CONFIG_PPC_BOOK3S_64
 611         case KVM_CAP_PPC_GET_SMMU_INFO:
 612                 r = 1;
 613                 break;
 614         case KVM_CAP_SPAPR_MULTITCE:
 615                 r = 1;
 616                 break;
 617         case KVM_CAP_SPAPR_RESIZE_HPT:
 618                 /* Disable this on POWER9 until code handles new HPTE format */
 619                 r = !!hv_enabled && !cpu_has_feature(CPU_FTR_ARCH_300);
 620                 break;
 621 #endif
 622         case KVM_CAP_PPC_HTM:
 623                 r = cpu_has_feature(CPU_FTR_TM_COMP) &&
 624                     is_kvmppc_hv_enabled(kvm);
 625                 break;
 626         default:
 627                 r = 0;
 628                 break;
 629         }
 630         return r;
 631
 632 }
 633
 634 long kvm_arch_dev_ioctl(struct file *filp,
 635                         unsigned int ioctl, unsigned long arg)
 636 {
 637         return -EINVAL;
 638 }
 639
 640 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 641                            struct kvm_memory_slot *dont)
 642 {
 643         kvmppc_core_free_memslot(kvm, free, dont);
 644 }
 645
 646 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 647                             unsigned long npages)
 648 {
 649         return kvmppc_core_create_memslot(kvm, slot, npages);
 650 }
 651
 652 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 653                                    struct kvm_memory_slot *memslot,
 654                                    const struct kvm_userspace_memory_region *mem,
 655                                    enum kvm_mr_change change)
 656 {
 657         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 658 }
 659
 660 void kvm_arch_commit_memory_region(struct kvm *kvm,
 661                                    const struct kvm_userspace_memory_region *mem,
 662                                    const struct kvm_memory_slot *old,
 663                                    const struct kvm_memory_slot *new,
 664                                    enum kvm_mr_change change)
 665 {
 666         kvmppc_core_commit_memory_region(kvm, mem, old, new);
 667 }
 668
 669 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 670                                    struct kvm_memory_slot *slot)
 671 {
 672         kvmppc_core_flush_memslot(kvm, slot);
 673 }
 674
 675 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 676 {
 677         struct kvm_vcpu *vcpu;
 678         vcpu = kvmppc_core_vcpu_create(kvm, id);
 679         if (!IS_ERR(vcpu)) {
 680                 vcpu->arch.wqp = &vcpu->wq;
 681                 kvmppc_create_vcpu_debugfs(vcpu, id);
 682         }
 683         return vcpu;
 684 }
 685
 686 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 687 {
 688 }
 689
 690 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 691 {
 692         /* Make sure we're not using the vcpu anymore */
 693         hrtimer_cancel(&vcpu->arch.dec_timer);
 694
 695         kvmppc_remove_vcpu_debugfs(vcpu);
 696
 697         switch (vcpu->arch.irq_type) {
 698         case KVMPPC_IRQ_MPIC:
 699                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 700                 break;
 701         case KVMPPC_IRQ_XICS:
 702                 kvmppc_xics_free_icp(vcpu);
 703                 break;
 704         }
 705
 706         kvmppc_core_vcpu_free(vcpu);
 707 }
 708
 709 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 710 {
 711         kvm_arch_vcpu_free(vcpu);
 712 }
 713
 714 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 715 {
 716         return kvmppc_core_pending_dec(vcpu);
 717 }
 718
 719 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 720 {
 721         struct kvm_vcpu *vcpu;
 722
 723         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 724         kvmppc_decrementer_func(vcpu);
 725
 726         return HRTIMER_NORESTART;
 727 }
 728
 729 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 730 {
 731         int ret;
 732
 733         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 734         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 735         vcpu->arch.dec_expires = ~(u64)0;
 736
 737 #ifdef CONFIG_KVM_EXIT_TIMING
 738         mutex_init(&vcpu->arch.exit_timing_lock);
 739 #endif
 740         ret = kvmppc_subarch_vcpu_init(vcpu);
 741         return ret;
 742 }
 743
 744 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 745 {
 746         kvmppc_mmu_destroy(vcpu);
 747         kvmppc_subarch_vcpu_uninit(vcpu);
 748 }
 749
 750 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 751 {
 752 #ifdef CONFIG_BOOKE
 753         /*
 754          * vrsave (formerly usprg0) isn't used by Linux, but may
 755          * be used by the guest.
 756          *
 757          * On non-booke this is associated with Altivec and
 758          * is handled by code in book3s.c.
 759          */
 760         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 761 #endif
 762         kvmppc_core_vcpu_load(vcpu, cpu);
 763 }
 764
 765 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 766 {
 767         kvmppc_core_vcpu_put(vcpu);
 768 #ifdef CONFIG_BOOKE
 769         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 770 #endif
 771 }
 772
 773 /*
 774  * irq_bypass_add_producer and irq_bypass_del_producer are only
 775  * useful if the architecture supports PCI passthrough.
 776  * irq_bypass_stop and irq_bypass_start are not needed and so
 777  * kvm_ops are not defined for them.
 778  */
 779 bool kvm_arch_has_irq_bypass(void)
 780 {
 781         return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) ||
 782                 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer));
 783 }
 784
 785 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
 786                                      struct irq_bypass_producer *prod)
 787 {
 788         struct kvm_kernel_irqfd *irqfd =
 789                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 790         struct kvm *kvm = irqfd->kvm;
 791
 792         if (kvm->arch.kvm_ops->irq_bypass_add_producer)
 793                 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod);
 794
 795         return 0;
 796 }
 797
 798 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
 799                                       struct irq_bypass_producer *prod)
 800 {
 801         struct kvm_kernel_irqfd *irqfd =
 802                 container_of(cons, struct kvm_kernel_irqfd, consumer);
 803         struct kvm *kvm = irqfd->kvm;
 804
 805         if (kvm->arch.kvm_ops->irq_bypass_del_producer)
 806                 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod);
 807 }
 808
 809 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 810                                       struct kvm_run *run)
 811 {
 812         u64 uninitialized_var(gpr);
 813
 814         if (run->mmio.len > sizeof(gpr)) {
 815                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 816                 return;
 817         }
 818
 819         if (!vcpu->arch.mmio_host_swabbed) {
 820                 switch (run->mmio.len) {
 821                 case 8: gpr = *(u64 *)run->mmio.data; break;
 822                 case 4: gpr = *(u32 *)run->mmio.data; break;
 823                 case 2: gpr = *(u16 *)run->mmio.data; break;
 824                 case 1: gpr = *(u8 *)run->mmio.data; break;
 825                 }
 826         } else {
 827                 switch (run->mmio.len) {
 828                 case 8: gpr = swab64(*(u64 *)run->mmio.data); break;
 829                 case 4: gpr = swab32(*(u32 *)run->mmio.data); break;
 830                 case 2: gpr = swab16(*(u16 *)run->mmio.data); break;
 831                 case 1: gpr = *(u8 *)run->mmio.data; break;
 832                 }
 833         }
 834
 835         if (vcpu->arch.mmio_sign_extend) {
 836                 switch (run->mmio.len) {
 837 #ifdef CONFIG_PPC64
 838                 case 4:
 839                         gpr = (s64)(s32)gpr;
 840                         break;
 841 #endif
 842                 case 2:
 843                         gpr = (s64)(s16)gpr;
 844                         break;
 845                 case 1:
 846                         gpr = (s64)(s8)gpr;
 847                         break;
 848                 }
 849         }
 850
 851         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 852
 853         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 854         case KVM_MMIO_REG_GPR:
 855                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 856                 break;
 857         case KVM_MMIO_REG_FPR:
 858                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
 859                 break;
 860 #ifdef CONFIG_PPC_BOOK3S
 861         case KVM_MMIO_REG_QPR:
 862                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 863                 break;
 864         case KVM_MMIO_REG_FQPR:
 865                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
 866                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 867                 break;
 868 #endif
 869         default:
 870                 BUG();
 871         }
 872 }
 873
 874 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 875                                 unsigned int rt, unsigned int bytes,
 876                                 int is_default_endian, int sign_extend)
 877 {
 878         int idx, ret;
 879         bool host_swabbed;
 880
 881         /* Pity C doesn't have a logical XOR operator */
 882         if (kvmppc_need_byteswap(vcpu)) {
 883                 host_swabbed = is_default_endian;
 884         } else {
 885                 host_swabbed = !is_default_endian;
 886         }
 887
 888         if (bytes > sizeof(run->mmio.data)) {
 889                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 890                        run->mmio.len);
 891         }
 892
 893         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 894         run->mmio.len = bytes;
 895         run->mmio.is_write = 0;
 896
 897         vcpu->arch.io_gpr = rt;
 898         vcpu->arch.mmio_host_swabbed = host_swabbed;
 899         vcpu->mmio_needed = 1;
 900         vcpu->mmio_is_write = 0;
 901         vcpu->arch.mmio_sign_extend = sign_extend;
 902
 903         idx = srcu_read_lock(&vcpu->kvm->srcu);
 904
 905         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
 906                               bytes, &run->mmio.data);
 907
 908         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 909
 910         if (!ret) {
 911                 kvmppc_complete_mmio_load(vcpu, run);
 912                 vcpu->mmio_needed = 0;
 913                 return EMULATE_DONE;
 914         }
 915
 916         return EMULATE_DO_MMIO;
 917 }
 918
 919 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 920                        unsigned int rt, unsigned int bytes,
 921                        int is_default_endian)
 922 {
 923         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0);
 924 }
 925 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
 926
 927 /* Same as above, but sign extends */
 928 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 929                         unsigned int rt, unsigned int bytes,
 930                         int is_default_endian)
 931 {
 932         return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1);
 933 }
 934
 935 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 936                         u64 val, unsigned int bytes, int is_default_endian)
 937 {
 938         void *data = run->mmio.data;
 939         int idx, ret;
 940         bool host_swabbed;
 941
 942         /* Pity C doesn't have a logical XOR operator */
 943         if (kvmppc_need_byteswap(vcpu)) {
 944                 host_swabbed = is_default_endian;
 945         } else {
 946                 host_swabbed = !is_default_endian;
 947         }
 948
 949         if (bytes > sizeof(run->mmio.data)) {
 950                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 951                        run->mmio.len);
 952         }
 953
 954         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 955         run->mmio.len = bytes;
 956         run->mmio.is_write = 1;
 957         vcpu->mmio_needed = 1;
 958         vcpu->mmio_is_write = 1;
 959
 960         /* Store the value at the lowest bytes in 'data'. */
 961         if (!host_swabbed) {
 962                 switch (bytes) {
 963                 case 8: *(u64 *)data = val; break;
 964                 case 4: *(u32 *)data = val; break;
 965                 case 2: *(u16 *)data = val; break;
 966                 case 1: *(u8  *)data = val; break;
 967                 }
 968         } else {
 969                 switch (bytes) {
 970                 case 8: *(u64 *)data = swab64(val); break;
 971                 case 4: *(u32 *)data = swab32(val); break;
 972                 case 2: *(u16 *)data = swab16(val); break;
 973                 case 1: *(u8  *)data = val; break;
 974                 }
 975         }
 976
 977         idx = srcu_read_lock(&vcpu->kvm->srcu);
 978
 979         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
 980                                bytes, &run->mmio.data);
 981
 982         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 983
 984         if (!ret) {
 985                 vcpu->mmio_needed = 0;
 986                 return EMULATE_DONE;
 987         }
 988
 989         return EMULATE_DO_MMIO;
 990 }
 991 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
 992
 993 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
 994 {
 995         int r = 0;
 996         union kvmppc_one_reg val;
 997         int size;
 998
 999         size = one_reg_size(reg->id);
1000         if (size > sizeof(val))
1001                 return -EINVAL;
1002
1003         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
1004         if (r == -EINVAL) {
1005                 r = 0;
1006                 switch (reg->id) {
1007 #ifdef CONFIG_ALTIVEC
1008                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1009                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1010                                 r = -ENXIO;
1011                                 break;
1012                         }
1013                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
1014                         break;
1015                 case KVM_REG_PPC_VSCR:
1016                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1017                                 r = -ENXIO;
1018                                 break;
1019                         }
1020                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
1021                         break;
1022                 case KVM_REG_PPC_VRSAVE:
1023                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
1024                         break;
1025 #endif /* CONFIG_ALTIVEC */
1026                 default:
1027                         r = -EINVAL;
1028                         break;
1029                 }
1030         }
1031
1032         if (r)
1033                 return r;
1034
1035         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
1036                 r = -EFAULT;
1037
1038         return r;
1039 }
1040
1041 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
1042 {
1043         int r;
1044         union kvmppc_one_reg val;
1045         int size;
1046
1047         size = one_reg_size(reg->id);
1048         if (size > sizeof(val))
1049                 return -EINVAL;
1050
1051         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
1052                 return -EFAULT;
1053
1054         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
1055         if (r == -EINVAL) {
1056                 r = 0;
1057                 switch (reg->id) {
1058 #ifdef CONFIG_ALTIVEC
1059                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
1060                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1061                                 r = -ENXIO;
1062                                 break;
1063                         }
1064                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
1065                         break;
1066                 case KVM_REG_PPC_VSCR:
1067                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1068                                 r = -ENXIO;
1069                                 break;
1070                         }
1071                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
1072                         break;
1073                 case KVM_REG_PPC_VRSAVE:
1074                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
1075                                 r = -ENXIO;
1076                                 break;
1077                         }
1078                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
1079                         break;
1080 #endif /* CONFIG_ALTIVEC */
1081                 default:
1082                         r = -EINVAL;
1083                         break;
1084                 }
1085         }
1086
1087         return r;
1088 }
1089
1090 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
1091 {
1092         int r;
1093         sigset_t sigsaved;
1094
1095         if (vcpu->sigset_active)
1096                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1097
1098         if (vcpu->mmio_needed) {
1099                 if (!vcpu->mmio_is_write)
1100                         kvmppc_complete_mmio_load(vcpu, run);
1101                 vcpu->mmio_needed = 0;
1102         } else if (vcpu->arch.osi_needed) {
1103                 u64 *gprs = run->osi.gprs;
1104                 int i;
1105
1106                 for (i = 0; i < 32; i++)
1107                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1108                 vcpu->arch.osi_needed = 0;
1109         } else if (vcpu->arch.hcall_needed) {
1110                 int i;
1111
1112                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1113                 for (i = 0; i < 9; ++i)
1114                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1115                 vcpu->arch.hcall_needed = 0;
1116 #ifdef CONFIG_BOOKE
1117         } else if (vcpu->arch.epr_needed) {
1118                 kvmppc_set_epr(vcpu, run->epr.epr);
1119                 vcpu->arch.epr_needed = 0;
1120 #endif
1121         }
1122
1123         if (run->immediate_exit)
1124                 r = -EINTR;
1125         else
1126                 r = kvmppc_vcpu_run(run, vcpu);
1127
1128         if (vcpu->sigset_active)
1129                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1130
1131         return r;
1132 }
1133
1134 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1135 {
1136         if (irq->irq == KVM_INTERRUPT_UNSET) {
1137                 kvmppc_core_dequeue_external(vcpu);
1138                 return 0;
1139         }
1140
1141         kvmppc_core_queue_external(vcpu, irq);
1142
1143         kvm_vcpu_kick(vcpu);
1144
1145         return 0;
1146 }
1147
1148 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1149                                      struct kvm_enable_cap *cap)
1150 {
1151         int r;
1152
1153         if (cap->flags)
1154                 return -EINVAL;
1155
1156         switch (cap->cap) {
1157         case KVM_CAP_PPC_OSI:
1158                 r = 0;
1159                 vcpu->arch.osi_enabled = true;
1160                 break;
1161         case KVM_CAP_PPC_PAPR:
1162                 r = 0;
1163                 vcpu->arch.papr_enabled = true;
1164                 break;
1165         case KVM_CAP_PPC_EPR:
1166                 r = 0;
1167                 if (cap->args[0])
1168                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1169                 else
1170                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1171                 break;
1172 #ifdef CONFIG_BOOKE
1173         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1174                 r = 0;
1175                 vcpu->arch.watchdog_enabled = true;
1176                 break;
1177 #endif
1178 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1179         case KVM_CAP_SW_TLB: {
1180                 struct kvm_config_tlb cfg;
1181                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1182
1183                 r = -EFAULT;
1184                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1185                         break;
1186
1187                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1188                 break;
1189         }
1190 #endif
1191 #ifdef CONFIG_KVM_MPIC
1192         case KVM_CAP_IRQ_MPIC: {
1193                 struct fd f;
1194                 struct kvm_device *dev;
1195
1196                 r = -EBADF;
1197                 f = fdget(cap->args[0]);
1198                 if (!f.file)
1199                         break;
1200
1201                 r = -EPERM;
1202                 dev = kvm_device_from_filp(f.file);
1203                 if (dev)
1204                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1205
1206                 fdput(f);
1207                 break;
1208         }
1209 #endif
1210 #ifdef CONFIG_KVM_XICS
1211         case KVM_CAP_IRQ_XICS: {
1212                 struct fd f;
1213                 struct kvm_device *dev;
1214
1215                 r = -EBADF;
1216                 f = fdget(cap->args[0]);
1217                 if (!f.file)
1218                         break;
1219
1220                 r = -EPERM;
1221                 dev = kvm_device_from_filp(f.file);
1222                 if (dev)
1223                         r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1224
1225                 fdput(f);
1226                 break;
1227         }
1228 #endif /* CONFIG_KVM_XICS */
1229         default:
1230                 r = -EINVAL;
1231                 break;
1232         }
1233
1234         if (!r)
1235                 r = kvmppc_sanity_check(vcpu);
1236
1237         return r;
1238 }
1239
1240 bool kvm_arch_intc_initialized(struct kvm *kvm)
1241 {
1242 #ifdef CONFIG_KVM_MPIC
1243         if (kvm->arch.mpic)
1244                 return true;
1245 #endif
1246 #ifdef CONFIG_KVM_XICS
1247         if (kvm->arch.xics)
1248                 return true;
1249 #endif
1250         return false;
1251 }
1252
1253 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1254                                     struct kvm_mp_state *mp_state)
1255 {
1256         return -EINVAL;
1257 }
1258
1259 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1260                                     struct kvm_mp_state *mp_state)
1261 {
1262         return -EINVAL;
1263 }
1264
1265 long kvm_arch_vcpu_ioctl(struct file *filp,
1266                          unsigned int ioctl, unsigned long arg)
1267 {
1268         struct kvm_vcpu *vcpu = filp->private_data;
1269         void __user *argp = (void __user *)arg;
1270         long r;
1271
1272         switch (ioctl) {
1273         case KVM_INTERRUPT: {
1274                 struct kvm_interrupt irq;
1275                 r = -EFAULT;
1276                 if (copy_from_user(&irq, argp, sizeof(irq)))
1277                         goto out;
1278                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1279                 goto out;
1280         }
1281
1282         case KVM_ENABLE_CAP:
1283         {
1284                 struct kvm_enable_cap cap;
1285                 r = -EFAULT;
1286                 if (copy_from_user(&cap, argp, sizeof(cap)))
1287                         goto out;
1288                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1289                 break;
1290         }
1291
1292         case KVM_SET_ONE_REG:
1293         case KVM_GET_ONE_REG:
1294         {
1295                 struct kvm_one_reg reg;
1296                 r = -EFAULT;
1297                 if (copy_from_user(&reg, argp, sizeof(reg)))
1298                         goto out;
1299                 if (ioctl == KVM_SET_ONE_REG)
1300                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1301                 else
1302                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1303                 break;
1304         }
1305
1306 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1307         case KVM_DIRTY_TLB: {
1308                 struct kvm_dirty_tlb dirty;
1309                 r = -EFAULT;
1310                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1311                         goto out;
1312                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1313                 break;
1314         }
1315 #endif
1316         default:
1317                 r = -EINVAL;
1318         }
1319
1320 out:
1321         return r;
1322 }
1323
1324 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1325 {
1326         return VM_FAULT_SIGBUS;
1327 }
1328
1329 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1330 {
1331         u32 inst_nop = 0x60000000;
1332 #ifdef CONFIG_KVM_BOOKE_HV
1333         u32 inst_sc1 = 0x44000022;
1334         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1335         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1336         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1337         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1338 #else
1339         u32 inst_lis = 0x3c000000;
1340         u32 inst_ori = 0x60000000;
1341         u32 inst_sc = 0x44000002;
1342         u32 inst_imm_mask = 0xffff;
1343
1344         /*
1345          * The hypercall to get into KVM from within guest context is as
1346          * follows:
1347          *
1348          *    lis r0, r0, KVM_SC_MAGIC_R0@h
1349          *    ori r0, KVM_SC_MAGIC_R0@l
1350          *    sc
1351          *    nop
1352          */
1353         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1354         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1355         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1356         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1357 #endif
1358
1359         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1360
1361         return 0;
1362 }
1363
1364 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1365                           bool line_status)
1366 {
1367         if (!irqchip_in_kernel(kvm))
1368                 return -ENXIO;
1369
1370         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1371                                         irq_event->irq, irq_event->level,
1372                                         line_status);
1373         return 0;
1374 }
1375
1376
1377 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1378                                    struct kvm_enable_cap *cap)
1379 {
1380         int r;
1381
1382         if (cap->flags)
1383                 return -EINVAL;
1384
1385         switch (cap->cap) {
1386 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1387         case KVM_CAP_PPC_ENABLE_HCALL: {
1388                 unsigned long hcall = cap->args[0];
1389
1390                 r = -EINVAL;
1391                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1392                     cap->args[1] > 1)
1393                         break;
1394                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1395                         break;
1396                 if (cap->args[1])
1397                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1398                 else
1399                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1400                 r = 0;
1401                 break;
1402         }
1403 #endif
1404         default:
1405                 r = -EINVAL;
1406                 break;
1407         }
1408
1409         return r;
1410 }
1411
1412 long kvm_arch_vm_ioctl(struct file *filp,
1413                        unsigned int ioctl, unsigned long arg)
1414 {
1415         struct kvm *kvm __maybe_unused = filp->private_data;
1416         void __user *argp = (void __user *)arg;
1417         long r;
1418
1419         switch (ioctl) {
1420         case KVM_PPC_GET_PVINFO: {
1421                 struct kvm_ppc_pvinfo pvinfo;
1422                 memset(&pvinfo, 0, sizeof(pvinfo));
1423                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1424                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1425                         r = -EFAULT;
1426                         goto out;
1427                 }
1428
1429                 break;
1430         }
1431         case KVM_ENABLE_CAP:
1432         {
1433                 struct kvm_enable_cap cap;
1434                 r = -EFAULT;
1435                 if (copy_from_user(&cap, argp, sizeof(cap)))
1436                         goto out;
1437                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1438                 break;
1439         }
1440 #ifdef CONFIG_PPC_BOOK3S_64
1441         case KVM_CREATE_SPAPR_TCE_64: {
1442                 struct kvm_create_spapr_tce_64 create_tce_64;
1443
1444                 r = -EFAULT;
1445                 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64)))
1446                         goto out;
1447                 if (create_tce_64.flags) {
1448                         r = -EINVAL;
1449                         goto out;
1450                 }
1451                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1452                 goto out;
1453         }
1454         case KVM_CREATE_SPAPR_TCE: {
1455                 struct kvm_create_spapr_tce create_tce;
1456                 struct kvm_create_spapr_tce_64 create_tce_64;
1457
1458                 r = -EFAULT;
1459                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1460                         goto out;
1461
1462                 create_tce_64.liobn = create_tce.liobn;
1463                 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K;
1464                 create_tce_64.offset = 0;
1465                 create_tce_64.size = create_tce.window_size >>
1466                                 IOMMU_PAGE_SHIFT_4K;
1467                 create_tce_64.flags = 0;
1468                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64);
1469                 goto out;
1470         }
1471         case KVM_PPC_GET_SMMU_INFO: {
1472                 struct kvm_ppc_smmu_info info;
1473                 struct kvm *kvm = filp->private_data;
1474
1475                 memset(&info, 0, sizeof(info));
1476                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1477                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1478                         r = -EFAULT;
1479                 break;
1480         }
1481         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1482                 struct kvm *kvm = filp->private_data;
1483
1484                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1485                 break;
1486         }
1487         case KVM_PPC_CONFIGURE_V3_MMU: {
1488                 struct kvm *kvm = filp->private_data;
1489                 struct kvm_ppc_mmuv3_cfg cfg;
1490
1491                 r = -EINVAL;
1492                 if (!kvm->arch.kvm_ops->configure_mmu)
1493                         goto out;
1494                 r = -EFAULT;
1495                 if (copy_from_user(&cfg, argp, sizeof(cfg)))
1496                         goto out;
1497                 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg);
1498                 break;
1499         }
1500         case KVM_PPC_GET_RMMU_INFO: {
1501                 struct kvm *kvm = filp->private_data;
1502                 struct kvm_ppc_rmmu_info info;
1503
1504                 r = -EINVAL;
1505                 if (!kvm->arch.kvm_ops->get_rmmu_info)
1506                         goto out;
1507                 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info);
1508                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1509                         r = -EFAULT;
1510                 break;
1511         }
1512         default: {
1513                 struct kvm *kvm = filp->private_data;
1514                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1515         }
1516 #else /* CONFIG_PPC_BOOK3S_64 */
1517         default:
1518                 r = -ENOTTY;
1519 #endif
1520         }
1521 out:
1522         return r;
1523 }
1524
1525 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1526 static unsigned long nr_lpids;
1527
1528 long kvmppc_alloc_lpid(void)
1529 {
1530         long lpid;
1531
1532         do {
1533                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1534                 if (lpid >= nr_lpids) {
1535                         pr_err("%s: No LPIDs free\n", __func__);
1536                         return -ENOMEM;
1537                 }
1538         } while (test_and_set_bit(lpid, lpid_inuse));
1539
1540         return lpid;
1541 }
1542 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1543
1544 void kvmppc_claim_lpid(long lpid)
1545 {
1546         set_bit(lpid, lpid_inuse);
1547 }
1548 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1549
1550 void kvmppc_free_lpid(long lpid)
1551 {
1552         clear_bit(lpid, lpid_inuse);
1553 }
1554 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1555
1556 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1557 {
1558         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1559         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1560 }
1561 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1562
1563 int kvm_arch_init(void *opaque)
1564 {
1565         return 0;
1566 }
1567
1568 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);