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                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 194
 195                 r = EV_SUCCESS;
 196                 break;
 197         }
 198         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 199                 r = EV_SUCCESS;
 200 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 201                 /* XXX Missing magic page on 44x */
 202                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 203 #endif
 204
 205                 /* Second return value is in r4 */
 206                 break;
 207         case EV_HCALL_TOKEN(EV_IDLE):
 208                 r = EV_SUCCESS;
 209                 kvm_vcpu_block(vcpu);
 210                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 211                 break;
 212         default:
 213                 r = EV_UNIMPLEMENTED;
 214                 break;
 215         }
 216
 217         kvmppc_set_gpr(vcpu, 4, r2);
 218
 219         return r;
 220 }
 221 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
 222
 223 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
 224 {
 225         int r = false;
 226
 227         /* We have to know what CPU to virtualize */
 228         if (!vcpu->arch.pvr)
 229                 goto out;
 230
 231         /* PAPR only works with book3s_64 */
 232         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
 233                 goto out;
 234
 235         /* HV KVM can only do PAPR mode for now */
 236         if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm))
 237                 goto out;
 238
 239 #ifdef CONFIG_KVM_BOOKE_HV
 240         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 241                 goto out;
 242 #endif
 243
 244         r = true;
 245
 246 out:
 247         vcpu->arch.sane = r;
 248         return r ? 0 : -EINVAL;
 249 }
 250 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
 251
 252 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 253 {
 254         enum emulation_result er;
 255         int r;
 256
 257         er = kvmppc_emulate_instruction(run, vcpu);
 258         switch (er) {
 259         case EMULATE_DONE:
 260                 /* Future optimization: only reload non-volatiles if they were
 261                  * actually modified. */
 262                 r = RESUME_GUEST_NV;
 263                 break;
 264         case EMULATE_DO_MMIO:
 265                 run->exit_reason = KVM_EXIT_MMIO;
 266                 /* We must reload nonvolatiles because "update" load/store
 267                  * instructions modify register state. */
 268                 /* Future optimization: only reload non-volatiles if they were
 269                  * actually modified. */
 270                 r = RESUME_HOST_NV;
 271                 break;
 272         case EMULATE_FAIL:
 273                 /* XXX Deliver Program interrupt to guest. */
 274                 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
 275                        kvmppc_get_last_inst(vcpu));
 276                 r = RESUME_HOST;
 277                 break;
 278         default:
 279                 WARN_ON(1);
 280                 r = RESUME_GUEST;
 281         }
 282
 283         return r;
 284 }
 285 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
 286
 287 int kvm_arch_hardware_enable(void *garbage)
 288 {
 289         return 0;
 290 }
 291
 292 void kvm_arch_hardware_disable(void *garbage)
 293 {
 294 }
 295
 296 int kvm_arch_hardware_setup(void)
 297 {
 298         return 0;
 299 }
 300
 301 void kvm_arch_hardware_unsetup(void)
 302 {
 303 }
 304
 305 void kvm_arch_check_processor_compat(void *rtn)
 306 {
 307         *(int *)rtn = kvmppc_core_check_processor_compat();
 308 }
 309
 310 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 311 {
 312         struct kvmppc_ops *kvm_ops = NULL;
 313         /*
 314          * if we have both HV and PR enabled, default is HV
 315          */
 316         if (type == 0) {
 317                 if (kvmppc_hv_ops)
 318                         kvm_ops = kvmppc_hv_ops;
 319                 else
 320                         kvm_ops = kvmppc_pr_ops;
 321                 if (!kvm_ops)
 322                         goto err_out;
 323         } else  if (type == KVM_VM_PPC_HV) {
 324                 if (!kvmppc_hv_ops)
 325                         goto err_out;
 326                 kvm_ops = kvmppc_hv_ops;
 327         } else if (type == KVM_VM_PPC_PR) {
 328                 if (!kvmppc_pr_ops)
 329                         goto err_out;
 330                 kvm_ops = kvmppc_pr_ops;
 331         } else
 332                 goto err_out;
 333
 334         if (kvm_ops->owner && !try_module_get(kvm_ops->owner))
 335                 return -ENOENT;
 336
 337         kvm->arch.kvm_ops = kvm_ops;
 338         return kvmppc_core_init_vm(kvm);
 339 err_out:
 340         return -EINVAL;
 341 }
 342
 343 void kvm_arch_destroy_vm(struct kvm *kvm)
 344 {
 345         unsigned int i;
 346         struct kvm_vcpu *vcpu;
 347
 348         kvm_for_each_vcpu(i, vcpu, kvm)
 349                 kvm_arch_vcpu_free(vcpu);
 350
 351         mutex_lock(&kvm->lock);
 352         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 353                 kvm->vcpus[i] = NULL;
 354
 355         atomic_set(&kvm->online_vcpus, 0);
 356
 357         kvmppc_core_destroy_vm(kvm);
 358
 359         mutex_unlock(&kvm->lock);
 360
 361         /* drop the module reference */
 362         module_put(kvm->arch.kvm_ops->owner);
 363 }
 364
 365 void kvm_arch_sync_events(struct kvm *kvm)
 366 {
 367 }
 368
 369 int kvm_dev_ioctl_check_extension(long ext)
 370 {
 371         int r;
 372         /* FIXME!!
 373          * Should some of this be vm ioctl ? is it possible now ?
 374          */
 375         int hv_enabled = kvmppc_hv_ops ? 1 : 0;
 376
 377         switch (ext) {
 378 #ifdef CONFIG_BOOKE
 379         case KVM_CAP_PPC_BOOKE_SREGS:
 380         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 381         case KVM_CAP_PPC_EPR:
 382 #else
 383         case KVM_CAP_PPC_SEGSTATE:
 384         case KVM_CAP_PPC_HIOR:
 385         case KVM_CAP_PPC_PAPR:
 386 #endif
 387         case KVM_CAP_PPC_UNSET_IRQ:
 388         case KVM_CAP_PPC_IRQ_LEVEL:
 389         case KVM_CAP_ENABLE_CAP:
 390         case KVM_CAP_ONE_REG:
 391         case KVM_CAP_IOEVENTFD:
 392         case KVM_CAP_DEVICE_CTRL:
 393                 r = 1;
 394                 break;
 395         case KVM_CAP_PPC_PAIRED_SINGLES:
 396         case KVM_CAP_PPC_OSI:
 397         case KVM_CAP_PPC_GET_PVINFO:
 398 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 399         case KVM_CAP_SW_TLB:
 400 #endif
 401                 /* We support this only for PR */
 402                 r = !hv_enabled;
 403                 break;
 404 #ifdef CONFIG_KVM_MMIO
 405         case KVM_CAP_COALESCED_MMIO:
 406                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 407                 break;
 408 #endif
 409 #ifdef CONFIG_KVM_MPIC
 410         case KVM_CAP_IRQ_MPIC:
 411                 r = 1;
 412                 break;
 413 #endif
 414
 415 #ifdef CONFIG_PPC_BOOK3S_64
 416         case KVM_CAP_SPAPR_TCE:
 417         case KVM_CAP_PPC_ALLOC_HTAB:
 418         case KVM_CAP_PPC_RTAS:
 419         case KVM_CAP_PPC_FIXUP_HCALL:
 420 #ifdef CONFIG_KVM_XICS
 421         case KVM_CAP_IRQ_XICS:
 422 #endif
 423                 r = 1;
 424                 break;
 425 #endif /* CONFIG_PPC_BOOK3S_64 */
 426 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 427         case KVM_CAP_PPC_SMT:
 428                 if (hv_enabled)
 429                         r = threads_per_subcore;
 430                 else
 431                         r = 0;
 432                 break;
 433         case KVM_CAP_PPC_RMA:
 434                 r = hv_enabled;
 435                 /* PPC970 requires an RMA */
 436                 if (r && cpu_has_feature(CPU_FTR_ARCH_201))
 437                         r = 2;
 438                 break;
 439 #endif
 440         case KVM_CAP_SYNC_MMU:
 441 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 442                 if (hv_enabled)
 443                         r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
 444                 else
 445                         r = 0;
 446 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 447                 r = 1;
 448 #else
 449                 r = 0;
 450 #endif
 451                 break;
 452 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 453         case KVM_CAP_PPC_HTAB_FD:
 454                 r = hv_enabled;
 455                 break;
 456 #endif
 457         case KVM_CAP_NR_VCPUS:
 458                 /*
 459                  * Recommending a number of CPUs is somewhat arbitrary; we
 460                  * return the number of present CPUs for -HV (since a host
 461                  * will have secondary threads "offline"), and for other KVM
 462                  * implementations just count online CPUs.
 463                  */
 464                 if (hv_enabled)
 465                         r = num_present_cpus();
 466                 else
 467                         r = num_online_cpus();
 468                 break;
 469         case KVM_CAP_MAX_VCPUS:
 470                 r = KVM_MAX_VCPUS;
 471                 break;
 472 #ifdef CONFIG_PPC_BOOK3S_64
 473         case KVM_CAP_PPC_GET_SMMU_INFO:
 474                 r = 1;
 475                 break;
 476 #endif
 477         default:
 478                 r = 0;
 479                 break;
 480         }
 481         return r;
 482
 483 }
 484
 485 long kvm_arch_dev_ioctl(struct file *filp,
 486                         unsigned int ioctl, unsigned long arg)
 487 {
 488         return -EINVAL;
 489 }
 490
 491 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 492                            struct kvm_memory_slot *dont)
 493 {
 494         kvmppc_core_free_memslot(kvm, free, dont);
 495 }
 496
 497 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 498                             unsigned long npages)
 499 {
 500         return kvmppc_core_create_memslot(kvm, slot, npages);
 501 }
 502
 503 void kvm_arch_memslots_updated(struct kvm *kvm)
 504 {
 505 }
 506
 507 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 508                                    struct kvm_memory_slot *memslot,
 509                                    struct kvm_userspace_memory_region *mem,
 510                                    enum kvm_mr_change change)
 511 {
 512         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 513 }
 514
 515 void kvm_arch_commit_memory_region(struct kvm *kvm,
 516                                    struct kvm_userspace_memory_region *mem,
 517                                    const struct kvm_memory_slot *old,
 518                                    enum kvm_mr_change change)
 519 {
 520         kvmppc_core_commit_memory_region(kvm, mem, old);
 521 }
 522
 523 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 524 {
 525 }
 526
 527 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 528                                    struct kvm_memory_slot *slot)
 529 {
 530         kvmppc_core_flush_memslot(kvm, slot);
 531 }
 532
 533 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 534 {
 535         struct kvm_vcpu *vcpu;
 536         vcpu = kvmppc_core_vcpu_create(kvm, id);
 537         if (!IS_ERR(vcpu)) {
 538                 vcpu->arch.wqp = &vcpu->wq;
 539                 kvmppc_create_vcpu_debugfs(vcpu, id);
 540         }
 541         return vcpu;
 542 }
 543
 544 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 545 {
 546         return 0;
 547 }
 548
 549 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 550 {
 551         /* Make sure we're not using the vcpu anymore */
 552         hrtimer_cancel(&vcpu->arch.dec_timer);
 553         tasklet_kill(&vcpu->arch.tasklet);
 554
 555         kvmppc_remove_vcpu_debugfs(vcpu);
 556
 557         switch (vcpu->arch.irq_type) {
 558         case KVMPPC_IRQ_MPIC:
 559                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 560                 break;
 561         case KVMPPC_IRQ_XICS:
 562                 kvmppc_xics_free_icp(vcpu);
 563                 break;
 564         }
 565
 566         kvmppc_core_vcpu_free(vcpu);
 567 }
 568
 569 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 570 {
 571         kvm_arch_vcpu_free(vcpu);
 572 }
 573
 574 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 575 {
 576         return kvmppc_core_pending_dec(vcpu);
 577 }
 578
 579 /*
 580  * low level hrtimer wake routine. Because this runs in hardirq context
 581  * we schedule a tasklet to do the real work.
 582  */
 583 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 584 {
 585         struct kvm_vcpu *vcpu;
 586
 587         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 588         tasklet_schedule(&vcpu->arch.tasklet);
 589
 590         return HRTIMER_NORESTART;
 591 }
 592
 593 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 594 {
 595         int ret;
 596
 597         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 598         tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
 599         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 600         vcpu->arch.dec_expires = ~(u64)0;
 601
 602 #ifdef CONFIG_KVM_EXIT_TIMING
 603         mutex_init(&vcpu->arch.exit_timing_lock);
 604 #endif
 605         ret = kvmppc_subarch_vcpu_init(vcpu);
 606         return ret;
 607 }
 608
 609 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 610 {
 611         kvmppc_mmu_destroy(vcpu);
 612         kvmppc_subarch_vcpu_uninit(vcpu);
 613 }
 614
 615 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 616 {
 617 #ifdef CONFIG_BOOKE
 618         /*
 619          * vrsave (formerly usprg0) isn't used by Linux, but may
 620          * be used by the guest.
 621          *
 622          * On non-booke this is associated with Altivec and
 623          * is handled by code in book3s.c.
 624          */
 625         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 626 #endif
 627         kvmppc_core_vcpu_load(vcpu, cpu);
 628 }
 629
 630 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 631 {
 632         kvmppc_core_vcpu_put(vcpu);
 633 #ifdef CONFIG_BOOKE
 634         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 635 #endif
 636 }
 637
 638 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
 639                                      struct kvm_run *run)
 640 {
 641         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
 642 }
 643
 644 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 645                                       struct kvm_run *run)
 646 {
 647         u64 uninitialized_var(gpr);
 648
 649         if (run->mmio.len > sizeof(gpr)) {
 650                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 651                 return;
 652         }
 653
 654         if (vcpu->arch.mmio_is_bigendian) {
 655                 switch (run->mmio.len) {
 656                 case 8: gpr = *(u64 *)run->mmio.data; break;
 657                 case 4: gpr = *(u32 *)run->mmio.data; break;
 658                 case 2: gpr = *(u16 *)run->mmio.data; break;
 659                 case 1: gpr = *(u8 *)run->mmio.data; break;
 660                 }
 661         } else {
 662                 /* Convert BE data from userland back to LE. */
 663                 switch (run->mmio.len) {
 664                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
 665                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
 666                 case 1: gpr = *(u8 *)run->mmio.data; break;
 667                 }
 668         }
 669
 670         if (vcpu->arch.mmio_sign_extend) {
 671                 switch (run->mmio.len) {
 672 #ifdef CONFIG_PPC64
 673                 case 4:
 674                         gpr = (s64)(s32)gpr;
 675                         break;
 676 #endif
 677                 case 2:
 678                         gpr = (s64)(s16)gpr;
 679                         break;
 680                 case 1:
 681                         gpr = (s64)(s8)gpr;
 682                         break;
 683                 }
 684         }
 685
 686         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 687
 688         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 689         case KVM_MMIO_REG_GPR:
 690                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 691                 break;
 692         case KVM_MMIO_REG_FPR:
 693                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
 694                 break;
 695 #ifdef CONFIG_PPC_BOOK3S
 696         case KVM_MMIO_REG_QPR:
 697                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 698                 break;
 699         case KVM_MMIO_REG_FQPR:
 700                 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr;
 701                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 702                 break;
 703 #endif
 704         default:
 705                 BUG();
 706         }
 707 }
 708
 709 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 710                        unsigned int rt, unsigned int bytes,
 711                        int is_default_endian)
 712 {
 713         int idx, ret;
 714         int is_bigendian;
 715
 716         if (kvmppc_need_byteswap(vcpu)) {
 717                 /* Default endianness is "little endian". */
 718                 is_bigendian = !is_default_endian;
 719         } else {
 720                 /* Default endianness is "big endian". */
 721                 is_bigendian = is_default_endian;
 722         }
 723
 724         if (bytes > sizeof(run->mmio.data)) {
 725                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 726                        run->mmio.len);
 727         }
 728
 729         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 730         run->mmio.len = bytes;
 731         run->mmio.is_write = 0;
 732
 733         vcpu->arch.io_gpr = rt;
 734         vcpu->arch.mmio_is_bigendian = is_bigendian;
 735         vcpu->mmio_needed = 1;
 736         vcpu->mmio_is_write = 0;
 737         vcpu->arch.mmio_sign_extend = 0;
 738
 739         idx = srcu_read_lock(&vcpu->kvm->srcu);
 740
 741         ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 742                               bytes, &run->mmio.data);
 743
 744         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 745
 746         if (!ret) {
 747                 kvmppc_complete_mmio_load(vcpu, run);
 748                 vcpu->mmio_needed = 0;
 749                 return EMULATE_DONE;
 750         }
 751
 752         return EMULATE_DO_MMIO;
 753 }
 754 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
 755
 756 /* Same as above, but sign extends */
 757 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 758                         unsigned int rt, unsigned int bytes,
 759                         int is_default_endian)
 760 {
 761         int r;
 762
 763         vcpu->arch.mmio_sign_extend = 1;
 764         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian);
 765
 766         return r;
 767 }
 768
 769 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 770                         u64 val, unsigned int bytes, int is_default_endian)
 771 {
 772         void *data = run->mmio.data;
 773         int idx, ret;
 774         int is_bigendian;
 775
 776         if (kvmppc_need_byteswap(vcpu)) {
 777                 /* Default endianness is "little endian". */
 778                 is_bigendian = !is_default_endian;
 779         } else {
 780                 /* Default endianness is "big endian". */
 781                 is_bigendian = is_default_endian;
 782         }
 783
 784         if (bytes > sizeof(run->mmio.data)) {
 785                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 786                        run->mmio.len);
 787         }
 788
 789         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 790         run->mmio.len = bytes;
 791         run->mmio.is_write = 1;
 792         vcpu->mmio_needed = 1;
 793         vcpu->mmio_is_write = 1;
 794
 795         /* Store the value at the lowest bytes in 'data'. */
 796         if (is_bigendian) {
 797                 switch (bytes) {
 798                 case 8: *(u64 *)data = val; break;
 799                 case 4: *(u32 *)data = val; break;
 800                 case 2: *(u16 *)data = val; break;
 801                 case 1: *(u8  *)data = val; break;
 802                 }
 803         } else {
 804                 /* Store LE value into 'data'. */
 805                 switch (bytes) {
 806                 case 4: st_le32(data, val); break;
 807                 case 2: st_le16(data, val); break;
 808                 case 1: *(u8 *)data = val; break;
 809                 }
 810         }
 811
 812         idx = srcu_read_lock(&vcpu->kvm->srcu);
 813
 814         ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 815                                bytes, &run->mmio.data);
 816
 817         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 818
 819         if (!ret) {
 820                 vcpu->mmio_needed = 0;
 821                 return EMULATE_DONE;
 822         }
 823
 824         return EMULATE_DO_MMIO;
 825 }
 826 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
 827
 828 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 829 {
 830         int r;
 831         sigset_t sigsaved;
 832
 833         if (vcpu->sigset_active)
 834                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 835
 836         if (vcpu->mmio_needed) {
 837                 if (!vcpu->mmio_is_write)
 838                         kvmppc_complete_mmio_load(vcpu, run);
 839                 vcpu->mmio_needed = 0;
 840         } else if (vcpu->arch.dcr_needed) {
 841                 if (!vcpu->arch.dcr_is_write)
 842                         kvmppc_complete_dcr_load(vcpu, run);
 843                 vcpu->arch.dcr_needed = 0;
 844         } else if (vcpu->arch.osi_needed) {
 845                 u64 *gprs = run->osi.gprs;
 846                 int i;
 847
 848                 for (i = 0; i < 32; i++)
 849                         kvmppc_set_gpr(vcpu, i, gprs[i]);
 850                 vcpu->arch.osi_needed = 0;
 851         } else if (vcpu->arch.hcall_needed) {
 852                 int i;
 853
 854                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
 855                 for (i = 0; i < 9; ++i)
 856                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
 857                 vcpu->arch.hcall_needed = 0;
 858 #ifdef CONFIG_BOOKE
 859         } else if (vcpu->arch.epr_needed) {
 860                 kvmppc_set_epr(vcpu, run->epr.epr);
 861                 vcpu->arch.epr_needed = 0;
 862 #endif
 863         }
 864
 865         r = kvmppc_vcpu_run(run, vcpu);
 866
 867         if (vcpu->sigset_active)
 868                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 869
 870         return r;
 871 }
 872
 873 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
 874 {
 875         if (irq->irq == KVM_INTERRUPT_UNSET) {
 876                 kvmppc_core_dequeue_external(vcpu);
 877                 return 0;
 878         }
 879
 880         kvmppc_core_queue_external(vcpu, irq);
 881
 882         kvm_vcpu_kick(vcpu);
 883
 884         return 0;
 885 }
 886
 887 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
 888                                      struct kvm_enable_cap *cap)
 889 {
 890         int r;
 891
 892         if (cap->flags)
 893                 return -EINVAL;
 894
 895         switch (cap->cap) {
 896         case KVM_CAP_PPC_OSI:
 897                 r = 0;
 898                 vcpu->arch.osi_enabled = true;
 899                 break;
 900         case KVM_CAP_PPC_PAPR:
 901                 r = 0;
 902                 vcpu->arch.papr_enabled = true;
 903                 break;
 904         case KVM_CAP_PPC_EPR:
 905                 r = 0;
 906                 if (cap->args[0])
 907                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
 908                 else
 909                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
 910                 break;
 911 #ifdef CONFIG_BOOKE
 912         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 913                 r = 0;
 914                 vcpu->arch.watchdog_enabled = true;
 915                 break;
 916 #endif
 917 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 918         case KVM_CAP_SW_TLB: {
 919                 struct kvm_config_tlb cfg;
 920                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
 921
 922                 r = -EFAULT;
 923                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
 924                         break;
 925
 926                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
 927                 break;
 928         }
 929 #endif
 930 #ifdef CONFIG_KVM_MPIC
 931         case KVM_CAP_IRQ_MPIC: {
 932                 struct fd f;
 933                 struct kvm_device *dev;
 934
 935                 r = -EBADF;
 936                 f = fdget(cap->args[0]);
 937                 if (!f.file)
 938                         break;
 939
 940                 r = -EPERM;
 941                 dev = kvm_device_from_filp(f.file);
 942                 if (dev)
 943                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
 944
 945                 fdput(f);
 946                 break;
 947         }
 948 #endif
 949 #ifdef CONFIG_KVM_XICS
 950         case KVM_CAP_IRQ_XICS: {
 951                 struct fd f;
 952                 struct kvm_device *dev;
 953
 954                 r = -EBADF;
 955                 f = fdget(cap->args[0]);
 956                 if (!f.file)
 957                         break;
 958
 959                 r = -EPERM;
 960                 dev = kvm_device_from_filp(f.file);
 961                 if (dev)
 962                         r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
 963
 964                 fdput(f);
 965                 break;
 966         }
 967 #endif /* CONFIG_KVM_XICS */
 968         default:
 969                 r = -EINVAL;
 970                 break;
 971         }
 972
 973         if (!r)
 974                 r = kvmppc_sanity_check(vcpu);
 975
 976         return r;
 977 }
 978
 979 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 980                                     struct kvm_mp_state *mp_state)
 981 {
 982         return -EINVAL;
 983 }
 984
 985 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 986                                     struct kvm_mp_state *mp_state)
 987 {
 988         return -EINVAL;
 989 }
 990
 991 long kvm_arch_vcpu_ioctl(struct file *filp,
 992                          unsigned int ioctl, unsigned long arg)
 993 {
 994         struct kvm_vcpu *vcpu = filp->private_data;
 995         void __user *argp = (void __user *)arg;
 996         long r;
 997
 998         switch (ioctl) {
 999         case KVM_INTERRUPT: {
1000                 struct kvm_interrupt irq;
1001                 r = -EFAULT;
1002                 if (copy_from_user(&irq, argp, sizeof(irq)))
1003                         goto out;
1004                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1005                 goto out;
1006         }
1007
1008         case KVM_ENABLE_CAP:
1009         {
1010                 struct kvm_enable_cap cap;
1011                 r = -EFAULT;
1012                 if (copy_from_user(&cap, argp, sizeof(cap)))
1013                         goto out;
1014                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1015                 break;
1016         }
1017
1018         case KVM_SET_ONE_REG:
1019         case KVM_GET_ONE_REG:
1020         {
1021                 struct kvm_one_reg reg;
1022                 r = -EFAULT;
1023                 if (copy_from_user(&reg, argp, sizeof(reg)))
1024                         goto out;
1025                 if (ioctl == KVM_SET_ONE_REG)
1026                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
1027                 else
1028                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
1029                 break;
1030         }
1031
1032 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
1033         case KVM_DIRTY_TLB: {
1034                 struct kvm_dirty_tlb dirty;
1035                 r = -EFAULT;
1036                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
1037                         goto out;
1038                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
1039                 break;
1040         }
1041 #endif
1042         default:
1043                 r = -EINVAL;
1044         }
1045
1046 out:
1047         return r;
1048 }
1049
1050 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1051 {
1052         return VM_FAULT_SIGBUS;
1053 }
1054
1055 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
1056 {
1057         u32 inst_nop = 0x60000000;
1058 #ifdef CONFIG_KVM_BOOKE_HV
1059         u32 inst_sc1 = 0x44000022;
1060         pvinfo->hcall[0] = cpu_to_be32(inst_sc1);
1061         pvinfo->hcall[1] = cpu_to_be32(inst_nop);
1062         pvinfo->hcall[2] = cpu_to_be32(inst_nop);
1063         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1064 #else
1065         u32 inst_lis = 0x3c000000;
1066         u32 inst_ori = 0x60000000;
1067         u32 inst_sc = 0x44000002;
1068         u32 inst_imm_mask = 0xffff;
1069
1070         /*
1071          * The hypercall to get into KVM from within guest context is as
1072          * follows:
1073          *
1074          *    lis r0, r0, KVM_SC_MAGIC_R0@h
1075          *    ori r0, KVM_SC_MAGIC_R0@l
1076          *    sc
1077          *    nop
1078          */
1079         pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask));
1080         pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask));
1081         pvinfo->hcall[2] = cpu_to_be32(inst_sc);
1082         pvinfo->hcall[3] = cpu_to_be32(inst_nop);
1083 #endif
1084
1085         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
1086
1087         return 0;
1088 }
1089
1090 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1091                           bool line_status)
1092 {
1093         if (!irqchip_in_kernel(kvm))
1094                 return -ENXIO;
1095
1096         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1097                                         irq_event->irq, irq_event->level,
1098                                         line_status);
1099         return 0;
1100 }
1101
1102 long kvm_arch_vm_ioctl(struct file *filp,
1103                        unsigned int ioctl, unsigned long arg)
1104 {
1105         struct kvm *kvm __maybe_unused = filp->private_data;
1106         void __user *argp = (void __user *)arg;
1107         long r;
1108
1109         switch (ioctl) {
1110         case KVM_PPC_GET_PVINFO: {
1111                 struct kvm_ppc_pvinfo pvinfo;
1112                 memset(&pvinfo, 0, sizeof(pvinfo));
1113                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1114                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1115                         r = -EFAULT;
1116                         goto out;
1117                 }
1118
1119                 break;
1120         }
1121 #ifdef CONFIG_PPC_BOOK3S_64
1122         case KVM_CREATE_SPAPR_TCE: {
1123                 struct kvm_create_spapr_tce create_tce;
1124
1125                 r = -EFAULT;
1126                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1127                         goto out;
1128                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1129                 goto out;
1130         }
1131         case KVM_PPC_GET_SMMU_INFO: {
1132                 struct kvm_ppc_smmu_info info;
1133                 struct kvm *kvm = filp->private_data;
1134
1135                 memset(&info, 0, sizeof(info));
1136                 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info);
1137                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1138                         r = -EFAULT;
1139                 break;
1140         }
1141         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1142                 struct kvm *kvm = filp->private_data;
1143
1144                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1145                 break;
1146         }
1147         default: {
1148                 struct kvm *kvm = filp->private_data;
1149                 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg);
1150         }
1151 #else /* CONFIG_PPC_BOOK3S_64 */
1152         default:
1153                 r = -ENOTTY;
1154 #endif
1155         }
1156 out:
1157         return r;
1158 }
1159
1160 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1161 static unsigned long nr_lpids;
1162
1163 long kvmppc_alloc_lpid(void)
1164 {
1165         long lpid;
1166
1167         do {
1168                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1169                 if (lpid >= nr_lpids) {
1170                         pr_err("%s: No LPIDs free\n", __func__);
1171                         return -ENOMEM;
1172                 }
1173         } while (test_and_set_bit(lpid, lpid_inuse));
1174
1175         return lpid;
1176 }
1177 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1178
1179 void kvmppc_claim_lpid(long lpid)
1180 {
1181         set_bit(lpid, lpid_inuse);
1182 }
1183 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1184
1185 void kvmppc_free_lpid(long lpid)
1186 {
1187         clear_bit(lpid, lpid_inuse);
1188 }
1189 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1190
1191 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1192 {
1193         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1194         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1195 }
1196 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1197
1198 int kvm_arch_init(void *opaque)
1199 {
1200         return 0;
1201 }
1202
1203 void kvm_arch_exit(void)
1204 {
1205
1206 }