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 <asm/cputable.h>
  31 #include <asm/uaccess.h>
  32 #include <asm/kvm_ppc.h>
  33 #include <asm/tlbflush.h>
  34 #include <asm/cputhreads.h>
  35 #include <asm/irqflags.h>
  36 #include "timing.h"
  37 #include "irq.h"
  38 #include "../mm/mmu_decl.h"
  39
  40 #define CREATE_TRACE_POINTS
  41 #include "trace.h"
  42
  43 struct kvmppc_ops *kvmppc_hv_ops;
  44 EXPORT_SYMBOL_GPL(kvmppc_hv_ops);
  45 struct kvmppc_ops *kvmppc_pr_ops;
  46 EXPORT_SYMBOL_GPL(kvmppc_pr_ops);
  47
  48
  49 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  50 {
  51         return !!(v->arch.pending_exceptions) ||
  52                v->requests;
  53 }
  54
  55 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  56 {
  57         return 1;
  58 }
  59
  60 /*
  61  * Common checks before entering the guest world.  Call with interrupts
  62  * disabled.
  63  *
  64  * returns:
  65  *
  66  * == 1 if we're ready to go into guest state
  67  * <= 0 if we need to go back to the host with return value
  68  */
  69 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  70 {
  71         int r;
  72
  73         WARN_ON(irqs_disabled());
  74         hard_irq_disable();
  75
  76         while (true) {
  77                 if (need_resched()) {
  78                         local_irq_enable();
  79                         cond_resched();
  80                         hard_irq_disable();
  81                         continue;
  82                 }
  83
  84                 if (signal_pending(current)) {
  85                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
  86                         vcpu->run->exit_reason = KVM_EXIT_INTR;
  87                         r = -EINTR;
  88                         break;
  89                 }
  90
  91                 vcpu->mode = IN_GUEST_MODE;
  92
  93                 /*
  94                  * Reading vcpu->requests must happen after setting vcpu->mode,
  95                  * so we don't miss a request because the requester sees
  96                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
  97                  * before next entering the guest (and thus doesn't IPI).
  98                  */
  99                 smp_mb();
 100
 101                 if (vcpu->requests) {
 102                         /* Make sure we process requests preemptable */
 103                         local_irq_enable();
 104                         trace_kvm_check_requests(vcpu);
 105                         r = kvmppc_core_check_requests(vcpu);
 106                         hard_irq_disable();
 107                         if (r > 0)
 108                                 continue;
 109                         break;
 110                 }
 111
 112                 if (kvmppc_core_prepare_to_enter(vcpu)) {
 113                         /* interrupts got enabled in between, so we
 114                            are back at square 1 */
 115                         continue;
 116                 }
 117
 118                 kvm_guest_enter();
 119                 return 1;
 120         }
 121
 122         /* return to host */
 123         local_irq_enable();
 124         return r;
 125 }
 126 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
 127
 128 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 129 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu)
 130 {
 131         struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared;
 132         int i;
 133
 134         shared->sprg0 = swab64(shared->sprg0);
 135         shared->sprg1 = swab64(shared->sprg1);
 136         shared->sprg2 = swab64(shared->sprg2);
 137         shared->sprg3 = swab64(shared->sprg3);
 138         shared->srr0 = swab64(shared->srr0);
 139         shared->srr1 = swab64(shared->srr1);
 140         shared->dar = swab64(shared->dar);
 141         shared->msr = swab64(shared->msr);
 142         shared->dsisr = swab32(shared->dsisr);
 143         shared->int_pending = swab32(shared->int_pending);
 144         for (i = 0; i < ARRAY_SIZE(shared->sr); i++)
 145                 shared->sr[i] = swab32(shared->sr[i]);
 146 }
 147 #endif
 148
 149 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
 150 {
 151         int nr = kvmppc_get_gpr(vcpu, 11);
 152         int r;
 153         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
 154         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
 155         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
 156         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
 157         unsigned long r2 = 0;
 158
 159         if (!(kvmppc_get_msr(vcpu) & MSR_SF)) {
 160                 /* 32 bit mode */
 161                 param1 &= 0xffffffff;
 162                 param2 &= 0xffffffff;
 163                 param3 &= 0xffffffff;
 164                 param4 &= 0xffffffff;
 165         }
 166
 167         switch (nr) {
 168         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
 169         {
 170 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE)
 171                 /* Book3S can be little endian, find it out here */
 172                 int shared_big_endian = true;
 173                 if (vcpu->arch.intr_msr & MSR_LE)
 174                         shared_big_endian = false;
 175                 if (shared_big_endian != vcpu->arch.shared_big_endian)
 176                         kvmppc_swab_shared(vcpu);
 177                 vcpu->arch.shared_big_endian = shared_big_endian;
 178 #endif
 179
 180                 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) {
 181                         /*
 182                          * Older versions of the Linux magic page code had
 183                          * a bug where they would map their trampoline code
 184                          * NX. If that's the case, remove !PR NX capability.
 185                          */
 186                         vcpu->arch.disable_kernel_nx = true;
 187                         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 188                 }
 189
 190                 vcpu->arch.magic_page_pa = param1 & ~0xfffULL;
 191                 vcpu->arch.magic_page_ea = param2 & ~0xfffULL;
 192
 193 #ifdef CONFIG_PPC_64K_PAGES
 194                 /*
 195                  * Make sure our 4k magic page is in the same window of a 64k
 196                  * page within the guest and within the host's page.
 197                  */
 198                 if ((vcpu->arch.magic_page_pa & 0xf000) !=
 199                     ((ulong)vcpu->arch.shared & 0xf000)) {
 200                         void *old_shared = vcpu->arch.shared;
 201                         ulong shared = (ulong)vcpu->arch.shared;
 202                         void *new_shared;
 203
 204                         shared &= PAGE_MASK;
 205                         shared |= vcpu->arch.magic_page_pa & 0xf000;
 206                         new_shared = (void*)shared;
 207                         memcpy(new_shared, old_shared, 0x1000);
 208                         vcpu->arch.shared = new_shared;
 209                 }
 210 #endif
 211
 212                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 213
 214                 r = EV_SUCCESS;
 215                 break;
 216         }
 217         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 218                 r = EV_SUCCESS;
 219 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 220                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 221 #endif
 222
 223                 /* Second return value is in r4 */
 224                 break;
 225         case EV_HCALL_TOKEN(EV_IDLE):
 226                 r = EV_SUCCESS;
 227                 kvm_vcpu_block(vcpu);
 228                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 229                 break;
 230         default:
 231                 r = EV_UNIMPLEMENTED;
 232                 break;
 233         }
 234
 235         kvmppc_set_gpr(vcpu, 4, r2);
 236
 237         return r;
 238 }
 239 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
 240
 241 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
 242 {
 243         int r = false;
 244
 245         /* We have to know what CPU to virtualize */
 246         if (!vcpu->arch.pvr)
 247                 goto out;
 248
 249         /* PAPR only works with book3s_64 */
 250         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
 251                 goto out;
 252
 253         /* HV KVM can only do PAPR mode for now */
 254         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
 255                 goto out;
 256
 257 #ifdef CONFIG_KVM_BOOKE_HV
 258         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 259                 goto out;
 260 #endif
 261
 262         r = true;
 263
 264 out:
 265         vcpu->arch.sane = r;
 266         return r ? 0 : -EINVAL;
 267 }
 268 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
 269
 270 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 271 {
 272         enum emulation_result er;
 273         int r;
 274
 275         er = kvmppc_emulate_loadstore(vcpu);
 276         switch (er) {
 277         case EMULATE_DONE:
 278                 /* Future optimization: only reload non-volatiles if they were
 279                  * actually modified. */
 280                 r = RESUME_GUEST_NV;
 281                 break;
 282         case EMULATE_AGAIN:
 283                 r = RESUME_GUEST;
 284                 break;
 285         case EMULATE_DO_MMIO:
 286                 run->exit_reason = KVM_EXIT_MMIO;
 287                 /* We must reload nonvolatiles because "update" load/store
 288                  * instructions modify register state. */
 289                 /* Future optimization: only reload non-volatiles if they were
 290                  * actually modified. */
 291                 r = RESUME_HOST_NV;
 292                 break;
 293         case EMULATE_FAIL:
 294         {
 295                 u32 last_inst;
 296
 297                 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
 298                 /* XXX Deliver Program interrupt to guest. */
 299                 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst);
 300                 r = RESUME_HOST;
 301                 break;
 302         }
 303         default:
 304                 WARN_ON(1);
 305                 r = RESUME_GUEST;
 306         }
 307
 308         return r;
 309 }
 310 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
 311
 312 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 313               bool data)
 314 {
 315         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 316         struct kvmppc_pte pte;
 317         int r;
 318
 319         vcpu->stat.st++;
 320
 321         r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 322                          XLATE_WRITE, &pte);
 323         if (r < 0)
 324                 return r;
 325
 326         *eaddr = pte.raddr;
 327
 328         if (!pte.may_write)
 329                 return -EPERM;
 330
 331         /* Magic page override */
 332         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 333             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 334             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 335                 void *magic = vcpu->arch.shared;
 336                 magic += pte.eaddr & 0xfff;
 337                 memcpy(magic, ptr, size);
 338                 return EMULATE_DONE;
 339         }
 340
 341         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 342                 return EMULATE_DO_MMIO;
 343
 344         return EMULATE_DONE;
 345 }
 346 EXPORT_SYMBOL_GPL(kvmppc_st);
 347
 348 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 349                       bool data)
 350 {
 351         ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK;
 352         struct kvmppc_pte pte;
 353         int rc;
 354
 355         vcpu->stat.ld++;
 356
 357         rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST,
 358                           XLATE_READ, &pte);
 359         if (rc)
 360                 return rc;
 361
 362         *eaddr = pte.raddr;
 363
 364         if (!pte.may_read)
 365                 return -EPERM;
 366
 367         if (!data && !pte.may_execute)
 368                 return -ENOEXEC;
 369
 370         /* Magic page override */
 371         if (kvmppc_supports_magic_page(vcpu) && mp_pa &&
 372             ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) &&
 373             !(kvmppc_get_msr(vcpu) & MSR_PR)) {
 374                 void *magic = vcpu->arch.shared;
 375                 magic += pte.eaddr & 0xfff;
 376                 memcpy(ptr, magic, size);
 377                 return EMULATE_DONE;
 378         }
 379
 380         if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size))
 381                 return EMULATE_DO_MMIO;
 382
 383         return EMULATE_DONE;
 384 }
 385 EXPORT_SYMBOL_GPL(kvmppc_ld);
 386
 387 int kvm_arch_hardware_enable(void)
 388 {
 389         return 0;
 390 }
 391
 392 int kvm_arch_hardware_setup(void)
 393 {
 394         return 0;
 395 }
 396
 397 void kvm_arch_check_processor_compat(void *rtn)
 398 {
 399         *(int *)rtn = kvmppc_core_check_processor_compat();
 400 }
 401
 402 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 403 {
 404         struct kvmppc_ops *kvm_ops = NULL;
 405         /*
 406          * if we have both HV and PR enabled, default is HV
 407          */
 408         if (type == 0) {
 409                 if (kvmppc_hv_ops)
 410                         kvm_ops = kvmppc_hv_ops;
 411                 else
 412                         kvm_ops = kvmppc_pr_ops;
 413                 if (!kvm_ops)
 414                         goto err_out;
 415         } else  if (type == KVM_VM_PPC_HV) {
 416                 if (!kvmppc_hv_ops)
 417                         goto err_out;
 418                 kvm_ops = kvmppc_hv_ops;
 419         } else if (type == KVM_VM_PPC_PR) {
 420                 if (!kvmppc_pr_ops)
 421                         goto err_out;
 422                 kvm_ops = kvmppc_pr_ops;
 423         } else
 424                 goto err_out;
 425
 426         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
 427                 return -ENOENT;
 428
 429         kvm->arch.kvm_ops = kvm_ops;
 430         return kvmppc_core_init_vm(kvm);
 431 err_out:
 432         return -EINVAL;
 433 }
 434
 435 void kvm_arch_destroy_vm(struct kvm *kvm)
 436 {
 437         unsigned int i;
 438         struct kvm_vcpu *vcpu;
 439
 440         kvm_for_each_vcpu(i, vcpu, kvm)
 441                 kvm_arch_vcpu_free(vcpu);
 442
 443         mutex_lock(&kvm->lock);
 444         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 445                 kvm->vcpus[i] = NULL;
 446
 447         atomic_set(&kvm->online_vcpus, 0);
 448
 449         kvmppc_core_destroy_vm(kvm);
 450
 451         mutex_unlock(&kvm->lock);
 452
 453         /* drop the module reference */
 454         module_put(kvm->arch.kvm_ops->owner);
 455 }
 456
 457 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 458 {
 459         int r;
 460         /* Assume we're using HV mode when the HV module is loaded */
 461         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
 462
 463         if (kvm) {
 464                 /*
 465                  * Hooray - we know which VM type we're running on. Depend on
 466                  * that rather than the guess above.
 467                  */
 468                 hv_enabled = is_kvmppc_hv_enabled(kvm);
 469         }
 470
 471         switch (ext) {
 472 #ifdef CONFIG_BOOKE
 473         case KVM_CAP_PPC_BOOKE_SREGS:
 474         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 475         case KVM_CAP_PPC_EPR:
 476 #else
 477         case KVM_CAP_PPC_SEGSTATE:
 478         case KVM_CAP_PPC_HIOR:
 479         case KVM_CAP_PPC_PAPR:
 480 #endif
 481         case KVM_CAP_PPC_UNSET_IRQ:
 482         case KVM_CAP_PPC_IRQ_LEVEL:
 483         case KVM_CAP_ENABLE_CAP:
 484         case KVM_CAP_ENABLE_CAP_VM:
 485         case KVM_CAP_ONE_REG:
 486         case KVM_CAP_IOEVENTFD:
 487         case KVM_CAP_DEVICE_CTRL:
 488                 r = 1;
 489                 break;
 490         case KVM_CAP_PPC_PAIRED_SINGLES:
 491         case KVM_CAP_PPC_OSI:
 492         case KVM_CAP_PPC_GET_PVINFO:
 493 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 494         case KVM_CAP_SW_TLB:
 495 #endif
 496                 /* We support this only for PR */
 497                 r = !hv_enabled;
 498                 break;
 499 #ifdef CONFIG_KVM_MMIO
 500         case KVM_CAP_COALESCED_MMIO:
 501                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 502                 break;
 503 #endif
 504 #ifdef CONFIG_KVM_MPIC
 505         case KVM_CAP_IRQ_MPIC:
 506                 r = 1;
 507                 break;
 508 #endif
 509
 510 #ifdef CONFIG_PPC_BOOK3S_64
 511         case KVM_CAP_SPAPR_TCE:
 512         case KVM_CAP_PPC_ALLOC_HTAB:
 513         case KVM_CAP_PPC_RTAS:
 514         case KVM_CAP_PPC_FIXUP_HCALL:
 515         case KVM_CAP_PPC_ENABLE_HCALL:
 516 #ifdef CONFIG_KVM_XICS
 517         case KVM_CAP_IRQ_XICS:
 518 #endif
 519                 r = 1;
 520                 break;
 521 #endif /* CONFIG_PPC_BOOK3S_64 */
 522 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 523         case KVM_CAP_PPC_SMT:
 524                 if (hv_enabled)
 525                         r = threads_per_subcore;
 526                 else
 527                         r = 0;
 528                 break;
 529         case KVM_CAP_PPC_RMA:
 530                 r = 0;
 531                 break;
 532 #endif
 533         case KVM_CAP_SYNC_MMU:
 534 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 535                 r = hv_enabled;
 536 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 537                 r = 1;
 538 #else
 539                 r = 0;
 540 #endif
 541                 break;
 542 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 543         case KVM_CAP_PPC_HTAB_FD:
 544                 r = hv_enabled;
 545                 break;
 546 #endif
 547         case KVM_CAP_NR_VCPUS:
 548                 /*
 549                  * Recommending a number of CPUs is somewhat arbitrary; we
 550                  * return the number of present CPUs for -HV (since a host
 551                  * will have secondary threads "offline"), and for other KVM
 552                  * implementations just count online CPUs.
 553                  */
 554                 if (hv_enabled)
 555                         r = num_present_cpus();
 556                 else
 557                         r = num_online_cpus();
 558                 break;
 559         case KVM_CAP_MAX_VCPUS:
 560                 r = KVM_MAX_VCPUS;
 561                 break;
 562 #ifdef CONFIG_PPC_BOOK3S_64
 563         case KVM_CAP_PPC_GET_SMMU_INFO:
 564                 r = 1;
 565                 break;
 566 #endif
 567         default:
 568                 r = 0;
 569                 break;
 570         }
 571         return r;
 572
 573 }
 574
 575 long kvm_arch_dev_ioctl(struct file *filp,
 576                         unsigned int ioctl, unsigned long arg)
 577 {
 578         return -EINVAL;
 579 }
 580
 581 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 582                            struct kvm_memory_slot *dont)
 583 {
 584         kvmppc_core_free_memslot(kvm, free, dont);
 585 }
 586
 587 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 588                             unsigned long npages)
 589 {
 590         return kvmppc_core_create_memslot(kvm, slot, npages);
 591 }
 592
 593 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 594                                    struct kvm_memory_slot *memslot,
 595                                    struct kvm_userspace_memory_region *mem,
 596                                    enum kvm_mr_change change)
 597 {
 598         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 599 }
 600
 601 void kvm_arch_commit_memory_region(struct kvm *kvm,
 602                                    struct kvm_userspace_memory_region *mem,
 603                                    const struct kvm_memory_slot *old,
 604                                    enum kvm_mr_change change)
 605 {
 606         kvmppc_core_commit_memory_region(kvm, mem, old);
 607 }
 608
 609 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 610                                    struct kvm_memory_slot *slot)
 611 {
 612         kvmppc_core_flush_memslot(kvm, slot);
 613 }
 614
 615 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 616 {
 617         struct kvm_vcpu *vcpu;
 618         vcpu = kvmppc_core_vcpu_create(kvm, id);
 619         if (!IS_ERR(vcpu)) {
 620                 vcpu->arch.wqp = &vcpu->wq;
 621                 kvmppc_create_vcpu_debugfs(vcpu, id);
 622         }
 623         return vcpu;
 624 }
 625
 626 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 627 {
 628 }
 629
 630 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 631 {
 632         /* Make sure we're not using the vcpu anymore */
 633         hrtimer_cancel(&vcpu->arch.dec_timer);
 634
 635         kvmppc_remove_vcpu_debugfs(vcpu);
 636
 637         switch (vcpu->arch.irq_type) {
 638         case KVMPPC_IRQ_MPIC:
 639                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 640                 break;
 641         case KVMPPC_IRQ_XICS:
 642                 kvmppc_xics_free_icp(vcpu);
 643                 break;
 644         }
 645
 646         kvmppc_core_vcpu_free(vcpu);
 647 }
 648
 649 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 650 {
 651         kvm_arch_vcpu_free(vcpu);
 652 }
 653
 654 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 655 {
 656         return kvmppc_core_pending_dec(vcpu);
 657 }
 658
 659 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 660 {
 661         struct kvm_vcpu *vcpu;
 662
 663         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 664         kvmppc_decrementer_func(vcpu);
 665
 666         return HRTIMER_NORESTART;
 667 }
 668
 669 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 670 {
 671         int ret;
 672
 673         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 674         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 675         vcpu->arch.dec_expires = ~(u64)0;
 676
 677 #ifdef CONFIG_KVM_EXIT_TIMING
 678         mutex_init(&vcpu->arch.exit_timing_lock);
 679 #endif
 680         ret = kvmppc_subarch_vcpu_init(vcpu);
 681         return ret;
 682 }
 683
 684 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 685 {
 686         kvmppc_mmu_destroy(vcpu);
 687         kvmppc_subarch_vcpu_uninit(vcpu);
 688 }
 689
 690 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 691 {
 692 #ifdef CONFIG_BOOKE
 693         /*
 694          * vrsave (formerly usprg0) isn't used by Linux, but may
 695          * be used by the guest.
 696          *
 697          * On non-booke this is associated with Altivec and
 698          * is handled by code in book3s.c.
 699          */
 700         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 701 #endif
 702         kvmppc_core_vcpu_load(vcpu, cpu);
 703 }
 704
 705 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 706 {
 707         kvmppc_core_vcpu_put(vcpu);
 708 #ifdef CONFIG_BOOKE
 709         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 710 #endif
 711 }
 712
 713 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 714                                       struct kvm_run *run)
 715 {
 716         u64 uninitialized_var(gpr);
 717
 718         if (run->mmio.len > sizeof(gpr)) {
 719                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 720                 return;
 721         }
 722
 723         if (vcpu->arch.mmio_is_bigendian) {
 724                 switch (run->mmio.len) {
 725                 case 8: gpr = *(u64 *)run->mmio.data; break;
 726                 case 4: gpr = *(u32 *)run->mmio.data; break;
 727                 case 2: gpr = *(u16 *)run->mmio.data; break;
 728                 case 1: gpr = *(u8 *)run->mmio.data; break;
 729                 }
 730         } else {
 731                 /* Convert BE data from userland back to LE. */
 732                 switch (run->mmio.len) {
 733                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
 734                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
 735                 case 1: gpr = *(u8 *)run->mmio.data; break;
 736                 }
 737         }
 738
 739         if (vcpu->arch.mmio_sign_extend) {
 740                 switch (run->mmio.len) {
 741 #ifdef CONFIG_PPC64
 742                 case 4:
 743                         gpr = (s64)(s32)gpr;
 744                         break;
 745 #endif
 746                 case 2:
 747                         gpr = (s64)(s16)gpr;
 748                         break;
 749                 case 1:
 750                         gpr = (s64)(s8)gpr;
 751                         break;
 752                 }
 753         }
 754
 755         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 756
 757         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 758         case KVM_MMIO_REG_GPR:
 759                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 760                 break;
 761         case KVM_MMIO_REG_FPR:
 762                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
 763                 break;
 764 #ifdef CONFIG_PPC_BOOK3S
 765         case KVM_MMIO_REG_QPR:
 766                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 767                 break;
 768         case KVM_MMIO_REG_FQPR:
 769                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
 770                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 771                 break;
 772 #endif
 773         default:
 774                 BUG();
 775         }
 776 }
 777
 778 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 779                        unsigned int rt, unsigned int bytes,
 780                        int is_default_endian)
 781 {
 782         int idx, ret;
 783         int is_bigendian;
 784
 785         if (kvmppc_need_byteswap(vcpu)) {
 786                 /* Default endianness is "little endian". */
 787                 is_bigendian = !is_default_endian;
 788         } else {
 789                 /* Default endianness is "big endian". */
 790                 is_bigendian = is_default_endian;
 791         }
 792
 793         if (bytes > sizeof(run->mmio.data)) {
 794                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 795                        run->mmio.len);
 796         }
 797
 798         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 799         run->mmio.len = bytes;
 800         run->mmio.is_write = 0;
 801
 802         vcpu->arch.io_gpr = rt;
 803         vcpu->arch.mmio_is_bigendian = is_bigendian;
 804         vcpu->mmio_needed = 1;
 805         vcpu->mmio_is_write = 0;
 806         vcpu->arch.mmio_sign_extend = 0;
 807
 808         idx = srcu_read_lock(&vcpu->kvm->srcu);
 809
 810         ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
 811                               bytes, &run->mmio.data);
 812
 813         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 814
 815         if (!ret) {
 816                 kvmppc_complete_mmio_load(vcpu, run);
 817                 vcpu->mmio_needed = 0;
 818                 return EMULATE_DONE;
 819         }
 820
 821         return EMULATE_DO_MMIO;
 822 }
 823 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
 824
 825 /* Same as above, but sign extends */
 826 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 827                         unsigned int rt, unsigned int bytes,
 828                         int is_default_endian)
 829 {
 830         int r;
 831
 832         vcpu->arch.mmio_sign_extend = 1;
 833         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian);
 834
 835         return r;
 836 }
 837
 838 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 839                         u64 val, unsigned int bytes, int is_default_endian)
 840 {
 841         void *data = run->mmio.data;
 842         int idx, ret;
 843         int is_bigendian;
 844
 845         if (kvmppc_need_byteswap(vcpu)) {
 846                 /* Default endianness is "little endian". */
 847                 is_bigendian = !is_default_endian;
 848         } else {
 849                 /* Default endianness is "big endian". */
 850                 is_bigendian = is_default_endian;
 851         }
 852
 853         if (bytes > sizeof(run->mmio.data)) {
 854                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 855                        run->mmio.len);
 856         }
 857
 858         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 859         run->mmio.len = bytes;
 860         run->mmio.is_write = 1;
 861         vcpu->mmio_needed = 1;
 862         vcpu->mmio_is_write = 1;
 863
 864         /* Store the value at the lowest bytes in 'data'. */
 865         if (is_bigendian) {
 866                 switch (bytes) {
 867                 case 8: *(u64 *)data = val; break;
 868                 case 4: *(u32 *)data = val; break;
 869                 case 2: *(u16 *)data = val; break;
 870                 case 1: *(u8  *)data = val; break;
 871                 }
 872         } else {
 873                 /* Store LE value into 'data'. */
 874                 switch (bytes) {
 875                 case 4: st_le32(data, val); break;
 876                 case 2: st_le16(data, val); break;
 877                 case 1: *(u8 *)data = val; break;
 878                 }
 879         }
 880
 881         idx = srcu_read_lock(&vcpu->kvm->srcu);
 882
 883         ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr,
 884                                bytes, &run->mmio.data);
 885
 886         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 887
 888         if (!ret) {
 889                 vcpu->mmio_needed = 0;
 890                 return EMULATE_DONE;
 891         }
 892
 893         return EMULATE_DO_MMIO;
 894 }
 895 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
 896
 897 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
 898 {
 899         int r = 0;
 900         union kvmppc_one_reg val;
 901         int size;
 902
 903         size = one_reg_size(reg->id);
 904         if (size > sizeof(val))
 905                 return -EINVAL;
 906
 907         r = kvmppc_get_one_reg(vcpu, reg->id, &val);
 908         if (r == -EINVAL) {
 909                 r = 0;
 910                 switch (reg->id) {
 911 #ifdef CONFIG_ALTIVEC
 912                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
 913                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
 914                                 r = -ENXIO;
 915                                 break;
 916                         }
 917                         vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval;
 918                         break;
 919                 case KVM_REG_PPC_VSCR:
 920                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
 921                                 r = -ENXIO;
 922                                 break;
 923                         }
 924                         vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val);
 925                         break;
 926                 case KVM_REG_PPC_VRSAVE:
 927                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
 928                                 r = -ENXIO;
 929                                 break;
 930                         }
 931                         vcpu->arch.vrsave = set_reg_val(reg->id, val);
 932                         break;
 933 #endif /* CONFIG_ALTIVEC */
 934                 default:
 935                         r = -EINVAL;
 936                         break;
 937                 }
 938         }
 939
 940         if (r)
 941                 return r;
 942
 943         if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size))
 944                 r = -EFAULT;
 945
 946         return r;
 947 }
 948
 949 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg)
 950 {
 951         int r;
 952         union kvmppc_one_reg val;
 953         int size;
 954
 955         size = one_reg_size(reg->id);
 956         if (size > sizeof(val))
 957                 return -EINVAL;
 958
 959         if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size))
 960                 return -EFAULT;
 961
 962         r = kvmppc_set_one_reg(vcpu, reg->id, &val);
 963         if (r == -EINVAL) {
 964                 r = 0;
 965                 switch (reg->id) {
 966 #ifdef CONFIG_ALTIVEC
 967                 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31:
 968                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
 969                                 r = -ENXIO;
 970                                 break;
 971                         }
 972                         val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0];
 973                         break;
 974                 case KVM_REG_PPC_VSCR:
 975                         if (!cpu_has_feature(CPU_FTR_ALTIVEC)) {
 976                                 r = -ENXIO;
 977                                 break;
 978                         }
 979                         val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]);
 980                         break;
 981                 case KVM_REG_PPC_VRSAVE:
 982                         val = get_reg_val(reg->id, vcpu->arch.vrsave);
 983                         break;
 984 #endif /* CONFIG_ALTIVEC */
 985                 default:
 986                         r = -EINVAL;
 987                         break;
 988                 }
 989         }
 990
 991         return r;
 992 }
 993
 994 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 995 {
 996         int r;
 997         sigset_t sigsaved;
 998
 999         if (vcpu->sigset_active)
1000                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1001
1002         if (vcpu->mmio_needed) {
1003                 if (!vcpu->mmio_is_write)
1004                         kvmppc_complete_mmio_load(vcpu, run);
1005                 vcpu->mmio_needed = 0;
1006         } else if (vcpu->arch.osi_needed) {
1007                 u64 *gprs = run->osi.gprs;
1008                 int i;
1009
1010                 for (i = 0; i < 32; i++)
1011                         kvmppc_set_gpr(vcpu, i, gprs[i]);
1012                 vcpu->arch.osi_needed = 0;
1013         } else if (vcpu->arch.hcall_needed) {
1014                 int i;
1015
1016                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
1017                 for (i = 0; i < 9; ++i)
1018                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
1019                 vcpu->arch.hcall_needed = 0;
1020 #ifdef CONFIG_BOOKE
1021         } else if (vcpu->arch.epr_needed) {
1022                 kvmppc_set_epr(vcpu, run->epr.epr);
1023                 vcpu->arch.epr_needed = 0;
1024 #endif
1025         }
1026
1027         r = kvmppc_vcpu_run(run, vcpu);
1028
1029         if (vcpu->sigset_active)
1030                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1031
1032         return r;
1033 }
1034
1035 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1036 {
1037         if (irq->irq == KVM_INTERRUPT_UNSET) {
1038                 kvmppc_core_dequeue_external(vcpu);
1039                 return 0;
1040         }
1041
1042         kvmppc_core_queue_external(vcpu, irq);
1043
1044         kvm_vcpu_kick(vcpu);
1045
1046         return 0;
1047 }
1048
1049 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1050                                      struct kvm_enable_cap *cap)
1051 {
1052         int r;
1053
1054         if (cap->flags)
1055                 return -EINVAL;
1056
1057         switch (cap->cap) {
1058         case KVM_CAP_PPC_OSI:
1059                 r = 0;
1060                 vcpu->arch.osi_enabled = true;
1061                 break;
1062         case KVM_CAP_PPC_PAPR:
1063                 r = 0;
1064                 vcpu->arch.papr_enabled = true;
1065                 break;
1066         case KVM_CAP_PPC_EPR:
1067                 r = 0;
1068                 if (cap->args[0])
1069                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
1070                 else
1071                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
1072                 break;
1073 #ifdef CONFIG_BOOKE
1074         case KVM_CAP_PPC_BOOKE_WATCHDOG:
1075                 r = 0;
1076                 vcpu->arch.watchdog_enabled = true;
1077                 break;
1078 #endif
1079 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1080         case KVM_CAP_SW_TLB: {
1081                 struct kvm_config_tlb cfg;
1082                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
1083
1084                 r = -EFAULT;
1085                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
1086                         break;
1087
1088                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
1089                 break;
1090         }
1091 #endif
1092 #ifdef CONFIG_KVM_MPIC
1093         case KVM_CAP_IRQ_MPIC: {
1094                 struct fd f;
1095                 struct kvm_device *dev;
1096
1097                 r = -EBADF;
1098                 f = fdget(cap->args[0]);
1099                 if (!f.file)
1100                         break;
1101
1102                 r = -EPERM;
1103                 dev = kvm_device_from_filp(f.file);
1104                 if (dev)
1105                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
1106
1107                 fdput(f);
1108                 break;
1109         }
1110 #endif
1111 #ifdef CONFIG_KVM_XICS
1112         case KVM_CAP_IRQ_XICS: {
1113                 struct fd f;
1114                 struct kvm_device *dev;
1115
1116                 r = -EBADF;
1117                 f = fdget(cap->args[0]);
1118                 if (!f.file)
1119                         break;
1120
1121                 r = -EPERM;
1122                 dev = kvm_device_from_filp(f.file);
1123                 if (dev)
1124                         r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
1125
1126                 fdput(f);
1127                 break;
1128         }
1129 #endif /* CONFIG_KVM_XICS */
1130         default:
1131                 r = -EINVAL;
1132                 break;
1133         }
1134
1135         if (!r)
1136                 r = kvmppc_sanity_check(vcpu);
1137
1138         return r;
1139 }
1140
1141 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1142                                     struct kvm_mp_state *mp_state)
1143 {
1144         return -EINVAL;
1145 }
1146
1147 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1148                                     struct kvm_mp_state *mp_state)
1149 {
1150         return -EINVAL;
1151 }
1152
1153 long kvm_arch_vcpu_ioctl(struct file *filp,
1154                          unsigned int ioctl, unsigned long arg)
1155 {
1156         struct kvm_vcpu *vcpu = filp->private_data;
1157         void __user *argp = (void __user *)arg;
1158         long r;
1159
1160         switch (ioctl) {
1161         case KVM_INTERRUPT: {
1162                 struct kvm_interrupt irq;
1163                 r = -EFAULT;
1164                 if (copy_from_user(&irq, argp, sizeof(irq)))
1165                         goto out;
1166                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1167                 goto out;
1168         }
1169
1170         case KVM_ENABLE_CAP:
1171         {
1172                 struct kvm_enable_cap cap;
1173                 r = -EFAULT;
1174                 if (copy_from_user(&cap, argp, sizeof(cap)))
1175                         goto out;
1176                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1177                 break;
1178         }
1179
1180         case KVM_SET_ONE_REG:
1181         case KVM_GET_ONE_REG:
1182         {
1183                 struct kvm_one_reg reg;
1184                 r = -EFAULT;
1185                 if (copy_from_user(&reg, argp, sizeof(reg)))
1186                         goto out;
1187                 if (ioctl == KVM_SET_ONE_REG)
1188                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1189                 else
1190                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1191                 break;
1192         }
1193
1194 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1195         case KVM_DIRTY_TLB: {
1196                 struct kvm_dirty_tlb dirty;
1197                 r = -EFAULT;
1198                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1199                         goto out;
1200                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1201                 break;
1202         }
1203 #endif
1204         default:
1205                 r = -EINVAL;
1206         }
1207
1208 out:
1209         return r;
1210 }
1211
1212 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1213 {
1214         return VM_FAULT_SIGBUS;
1215 }
1216
1217 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1218 {
1219         u32 inst_nop = 0x60000000;
1220 #ifdef CONFIG_KVM_BOOKE_HV
1221         u32 inst_sc1 = 0x44000022;
1222         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1223         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1224         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1225         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1226 #else
1227         u32 inst_lis = 0x3c000000;
1228         u32 inst_ori = 0x60000000;
1229         u32 inst_sc = 0x44000002;
1230         u32 inst_imm_mask = 0xffff;
1231
1232         /*
1233          * The hypercall to get into KVM from within guest context is as
1234          * follows:
1235          *
1236          *    lis r0, r0, KVM_SC_MAGIC_R0@h
1237          *    ori r0, KVM_SC_MAGIC_R0@l
1238          *    sc
1239          *    nop
1240          */
1241         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1242         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1243         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1244         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1245 #endif
1246
1247         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1248
1249         return 0;
1250 }
1251
1252 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1253                           bool line_status)
1254 {
1255         if (!irqchip_in_kernel(kvm))
1256                 return -ENXIO;
1257
1258         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1259                                         irq_event->irq, irq_event->level,
1260                                         line_status);
1261         return 0;
1262 }
1263
1264
1265 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
1266                                    struct kvm_enable_cap *cap)
1267 {
1268         int r;
1269
1270         if (cap->flags)
1271                 return -EINVAL;
1272
1273         switch (cap->cap) {
1274 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER
1275         case KVM_CAP_PPC_ENABLE_HCALL: {
1276                 unsigned long hcall = cap->args[0];
1277
1278                 r = -EINVAL;
1279                 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) ||
1280                     cap->args[1] > 1)
1281                         break;
1282                 if (!kvmppc_book3s_hcall_implemented(kvm, hcall))
1283                         break;
1284                 if (cap->args[1])
1285                         set_bit(hcall / 4, kvm->arch.enabled_hcalls);
1286                 else
1287                         clear_bit(hcall / 4, kvm->arch.enabled_hcalls);
1288                 r = 0;
1289                 break;
1290         }
1291 #endif
1292         default:
1293                 r = -EINVAL;
1294                 break;
1295         }
1296
1297         return r;
1298 }
1299
1300 long kvm_arch_vm_ioctl(struct file *filp,
1301                        unsigned int ioctl, unsigned long arg)
1302 {
1303         struct kvm *kvm __maybe_unused = filp->private_data;
1304         void __user *argp = (void __user *)arg;
1305         long r;
1306
1307         switch (ioctl) {
1308         case KVM_PPC_GET_PVINFO: {
1309                 struct kvm_ppc_pvinfo pvinfo;
1310                 memset(&pvinfo, 0, sizeof(pvinfo));
1311                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1312                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1313                         r = -EFAULT;
1314                         goto out;
1315                 }
1316
1317                 break;
1318         }
1319         case KVM_ENABLE_CAP:
1320         {
1321                 struct kvm_enable_cap cap;
1322                 r = -EFAULT;
1323                 if (copy_from_user(&cap, argp, sizeof(cap)))
1324                         goto out;
1325                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1326                 break;
1327         }
1328 #ifdef CONFIG_PPC_BOOK3S_64
1329         case KVM_CREATE_SPAPR_TCE: {
1330                 struct kvm_create_spapr_tce create_tce;
1331
1332                 r = -EFAULT;
1333                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1334                         goto out;
1335                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1336                 goto out;
1337         }
1338         case KVM_PPC_GET_SMMU_INFO: {
1339                 struct kvm_ppc_smmu_info info;
1340                 struct kvm *kvm = filp->private_data;
1341
1342                 memset(&info, 0, sizeof(info));
1343                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1344                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1345                         r = -EFAULT;
1346                 break;
1347         }
1348         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1349                 struct kvm *kvm = filp->private_data;
1350
1351                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1352                 break;
1353         }
1354         default: {
1355                 struct kvm *kvm = filp->private_data;
1356                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1357         }
1358 #else /* CONFIG_PPC_BOOK3S_64 */
1359         default:
1360                 r = -ENOTTY;
1361 #endif
1362         }
1363 out:
1364         return r;
1365 }
1366
1367 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1368 static unsigned long nr_lpids;
1369
1370 long kvmppc_alloc_lpid(void)
1371 {
1372         long lpid;
1373
1374         do {
1375                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1376                 if (lpid >= nr_lpids) {
1377                         pr_err("%s: No LPIDs free\n", __func__);
1378                         return -ENOMEM;
1379                 }
1380         } while (test_and_set_bit(lpid, lpid_inuse));
1381
1382         return lpid;
1383 }
1384 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1385
1386 void kvmppc_claim_lpid(long lpid)
1387 {
1388         set_bit(lpid, lpid_inuse);
1389 }
1390 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1391
1392 void kvmppc_free_lpid(long lpid)
1393 {
1394         clear_bit(lpid, lpid_inuse);
1395 }
1396 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1397
1398 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1399 {
1400         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1401         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1402 }
1403 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1404
1405 int kvm_arch_init(void *opaque)
1406 {
1407         return 0;
1408 }
1409
1410 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr);