Pileus Git - ~andy/linux/blob - 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 <asm/cputable.h>
  29 #include <asm/uaccess.h>
  30 #include <asm/kvm_ppc.h>
  31 #include <asm/tlbflush.h>
  32 #include <asm/cputhreads.h>
  33 #include <asm/irqflags.h>
  34 #include "timing.h"
  35 #include "../mm/mmu_decl.h"
  36
  37 #define CREATE_TRACE_POINTS
  38 #include "trace.h"
  39
  40 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  41 {
  42         return !!(v->arch.pending_exceptions) ||
  43                v->requests;
  44 }
  45
  46 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  47 {
  48         return 1;
  49 }
  50
  51 #ifndef CONFIG_KVM_BOOK3S_64_HV
  52 /*
  53  * Common checks before entering the guest world.  Call with interrupts
  54  * disabled.
  55  *
  56  * returns !0 if a signal is pending and check_signal is true
  57  */
  58 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  59 {
  60         int r = 0;
  61
  62         WARN_ON_ONCE(!irqs_disabled());
  63         while (true) {
  64                 if (need_resched()) {
  65                         local_irq_enable();
  66                         cond_resched();
  67                         local_irq_disable();
  68                         continue;
  69                 }
  70
  71                 if (signal_pending(current)) {
  72                         r = 1;
  73                         break;
  74                 }
  75
  76                 smp_mb();
  77                 if (vcpu->requests) {
  78                         /* Make sure we process requests preemptable */
  79                         local_irq_enable();
  80                         trace_kvm_check_requests(vcpu);
  81                         kvmppc_core_check_requests(vcpu);
  82                         local_irq_disable();
  83                         continue;
  84                 }
  85
  86                 if (kvmppc_core_prepare_to_enter(vcpu)) {
  87                         /* interrupts got enabled in between, so we
  88                            are back at square 1 */
  89                         continue;
  90                 }
  91
  92                 if (vcpu->mode == EXITING_GUEST_MODE) {
  93                         r = 1;
  94                         break;
  95                 }
  96
  97 #ifdef CONFIG_PPC64
  98                 /* lazy EE magic */
  99                 hard_irq_disable();
 100                 if (lazy_irq_pending()) {
 101                         /* Got an interrupt in between, try again */
 102                         local_irq_enable();
 103                         local_irq_disable();
 104                         kvm_guest_exit();
 105                         continue;
 106                 }
 107
 108                 trace_hardirqs_on();
 109 #endif
 110
 111                 kvm_guest_enter();
 112
 113                 /* Going into guest context! Yay! */
 114                 vcpu->mode = IN_GUEST_MODE;
 115                 smp_wmb();
 116
 117                 break;
 118         }
 119
 120         return r;
 121 }
 122 #endif /* CONFIG_KVM_BOOK3S_64_HV */
 123
 124 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
 125 {
 126         int nr = kvmppc_get_gpr(vcpu, 11);
 127         int r;
 128         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
 129         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
 130         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
 131         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
 132         unsigned long r2 = 0;
 133
 134         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 135                 /* 32 bit mode */
 136                 param1 &= 0xffffffff;
 137                 param2 &= 0xffffffff;
 138                 param3 &= 0xffffffff;
 139                 param4 &= 0xffffffff;
 140         }
 141
 142         switch (nr) {
 143         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
 144         {
 145                 vcpu->arch.magic_page_pa = param1;
 146                 vcpu->arch.magic_page_ea = param2;
 147
 148                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 149
 150                 r = EV_SUCCESS;
 151                 break;
 152         }
 153         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 154                 r = EV_SUCCESS;
 155 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 156                 /* XXX Missing magic page on 44x */
 157                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 158 #endif
 159
 160                 /* Second return value is in r4 */
 161                 break;
 162         case EV_HCALL_TOKEN(EV_IDLE):
 163                 r = EV_SUCCESS;
 164                 kvm_vcpu_block(vcpu);
 165                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 166                 break;
 167         default:
 168                 r = EV_UNIMPLEMENTED;
 169                 break;
 170         }
 171
 172         kvmppc_set_gpr(vcpu, 4, r2);
 173
 174         return r;
 175 }
 176
 177 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
 178 {
 179         int r = false;
 180
 181         /* We have to know what CPU to virtualize */
 182         if (!vcpu->arch.pvr)
 183                 goto out;
 184
 185         /* PAPR only works with book3s_64 */
 186         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
 187                 goto out;
 188
 189 #ifdef CONFIG_KVM_BOOK3S_64_HV
 190         /* HV KVM can only do PAPR mode for now */
 191         if (!vcpu->arch.papr_enabled)
 192                 goto out;
 193 #endif
 194
 195 #ifdef CONFIG_KVM_BOOKE_HV
 196         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 197                 goto out;
 198 #endif
 199
 200         r = true;
 201
 202 out:
 203         vcpu->arch.sane = r;
 204         return r ? 0 : -EINVAL;
 205 }
 206
 207 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 208 {
 209         enum emulation_result er;
 210         int r;
 211
 212         er = kvmppc_emulate_instruction(run, vcpu);
 213         switch (er) {
 214         case EMULATE_DONE:
 215                 /* Future optimization: only reload non-volatiles if they were
 216                  * actually modified. */
 217                 r = RESUME_GUEST_NV;
 218                 break;
 219         case EMULATE_DO_MMIO:
 220                 run->exit_reason = KVM_EXIT_MMIO;
 221                 /* We must reload nonvolatiles because "update" load/store
 222                  * instructions modify register state. */
 223                 /* Future optimization: only reload non-volatiles if they were
 224                  * actually modified. */
 225                 r = RESUME_HOST_NV;
 226                 break;
 227         case EMULATE_FAIL:
 228                 /* XXX Deliver Program interrupt to guest. */
 229                 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
 230                        kvmppc_get_last_inst(vcpu));
 231                 r = RESUME_HOST;
 232                 break;
 233         default:
 234                 BUG();
 235         }
 236
 237         return r;
 238 }
 239
 240 int kvm_arch_hardware_enable(void *garbage)
 241 {
 242         return 0;
 243 }
 244
 245 void kvm_arch_hardware_disable(void *garbage)
 246 {
 247 }
 248
 249 int kvm_arch_hardware_setup(void)
 250 {
 251         return 0;
 252 }
 253
 254 void kvm_arch_hardware_unsetup(void)
 255 {
 256 }
 257
 258 void kvm_arch_check_processor_compat(void *rtn)
 259 {
 260         *(int *)rtn = kvmppc_core_check_processor_compat();
 261 }
 262
 263 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 264 {
 265         if (type)
 266                 return -EINVAL;
 267
 268         return kvmppc_core_init_vm(kvm);
 269 }
 270
 271 void kvm_arch_destroy_vm(struct kvm *kvm)
 272 {
 273         unsigned int i;
 274         struct kvm_vcpu *vcpu;
 275
 276         kvm_for_each_vcpu(i, vcpu, kvm)
 277                 kvm_arch_vcpu_free(vcpu);
 278
 279         mutex_lock(&kvm->lock);
 280         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 281                 kvm->vcpus[i] = NULL;
 282
 283         atomic_set(&kvm->online_vcpus, 0);
 284
 285         kvmppc_core_destroy_vm(kvm);
 286
 287         mutex_unlock(&kvm->lock);
 288 }
 289
 290 void kvm_arch_sync_events(struct kvm *kvm)
 291 {
 292 }
 293
 294 int kvm_dev_ioctl_check_extension(long ext)
 295 {
 296         int r;
 297
 298         switch (ext) {
 299 #ifdef CONFIG_BOOKE
 300         case KVM_CAP_PPC_BOOKE_SREGS:
 301 #else
 302         case KVM_CAP_PPC_SEGSTATE:
 303         case KVM_CAP_PPC_HIOR:
 304         case KVM_CAP_PPC_PAPR:
 305 #endif
 306         case KVM_CAP_PPC_UNSET_IRQ:
 307         case KVM_CAP_PPC_IRQ_LEVEL:
 308         case KVM_CAP_ENABLE_CAP:
 309         case KVM_CAP_ONE_REG:
 310                 r = 1;
 311                 break;
 312 #ifndef CONFIG_KVM_BOOK3S_64_HV
 313         case KVM_CAP_PPC_PAIRED_SINGLES:
 314         case KVM_CAP_PPC_OSI:
 315         case KVM_CAP_PPC_GET_PVINFO:
 316 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 317         case KVM_CAP_SW_TLB:
 318 #endif
 319                 r = 1;
 320                 break;
 321         case KVM_CAP_COALESCED_MMIO:
 322                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 323                 break;
 324 #endif
 325 #ifdef CONFIG_PPC_BOOK3S_64
 326         case KVM_CAP_SPAPR_TCE:
 327         case KVM_CAP_PPC_ALLOC_HTAB:
 328                 r = 1;
 329                 break;
 330 #endif /* CONFIG_PPC_BOOK3S_64 */
 331 #ifdef CONFIG_KVM_BOOK3S_64_HV
 332         case KVM_CAP_PPC_SMT:
 333                 r = threads_per_core;
 334                 break;
 335         case KVM_CAP_PPC_RMA:
 336                 r = 1;
 337                 /* PPC970 requires an RMA */
 338                 if (cpu_has_feature(CPU_FTR_ARCH_201))
 339                         r = 2;
 340                 break;
 341 #endif
 342         case KVM_CAP_SYNC_MMU:
 343 #ifdef CONFIG_KVM_BOOK3S_64_HV
 344                 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
 345 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 346                 r = 1;
 347 #else
 348                 r = 0;
 349 #endif
 350                 break;
 351         case KVM_CAP_NR_VCPUS:
 352                 /*
 353                  * Recommending a number of CPUs is somewhat arbitrary; we
 354                  * return the number of present CPUs for -HV (since a host
 355                  * will have secondary threads "offline"), and for other KVM
 356                  * implementations just count online CPUs.
 357                  */
 358 #ifdef CONFIG_KVM_BOOK3S_64_HV
 359                 r = num_present_cpus();
 360 #else
 361                 r = num_online_cpus();
 362 #endif
 363                 break;
 364         case KVM_CAP_MAX_VCPUS:
 365                 r = KVM_MAX_VCPUS;
 366                 break;
 367 #ifdef CONFIG_PPC_BOOK3S_64
 368         case KVM_CAP_PPC_GET_SMMU_INFO:
 369                 r = 1;
 370                 break;
 371 #endif
 372         default:
 373                 r = 0;
 374                 break;
 375         }
 376         return r;
 377
 378 }
 379
 380 long kvm_arch_dev_ioctl(struct file *filp,
 381                         unsigned int ioctl, unsigned long arg)
 382 {
 383         return -EINVAL;
 384 }
 385
 386 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
 387                            struct kvm_memory_slot *dont)
 388 {
 389         if (!dont || free->arch.rmap != dont->arch.rmap) {
 390                 vfree(free->arch.rmap);
 391                 free->arch.rmap = NULL;
 392         }
 393 }
 394
 395 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
 396 {
 397         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
 398         if (!slot->arch.rmap)
 399                 return -ENOMEM;
 400
 401         return 0;
 402 }
 403
 404 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 405                                    struct kvm_memory_slot *memslot,
 406                                    struct kvm_memory_slot old,
 407                                    struct kvm_userspace_memory_region *mem,
 408                                    int user_alloc)
 409 {
 410         return kvmppc_core_prepare_memory_region(kvm, mem);
 411 }
 412
 413 void kvm_arch_commit_memory_region(struct kvm *kvm,
 414                struct kvm_userspace_memory_region *mem,
 415                struct kvm_memory_slot old,
 416                int user_alloc)
 417 {
 418         kvmppc_core_commit_memory_region(kvm, mem);
 419 }
 420
 421 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 422 {
 423 }
 424
 425 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 426                                    struct kvm_memory_slot *slot)
 427 {
 428 }
 429
 430 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 431 {
 432         struct kvm_vcpu *vcpu;
 433         vcpu = kvmppc_core_vcpu_create(kvm, id);
 434         if (!IS_ERR(vcpu)) {
 435                 vcpu->arch.wqp = &vcpu->wq;
 436                 kvmppc_create_vcpu_debugfs(vcpu, id);
 437         }
 438         return vcpu;
 439 }
 440
 441 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 442 {
 443         /* Make sure we're not using the vcpu anymore */
 444         hrtimer_cancel(&vcpu->arch.dec_timer);
 445         tasklet_kill(&vcpu->arch.tasklet);
 446
 447         kvmppc_remove_vcpu_debugfs(vcpu);
 448         kvmppc_core_vcpu_free(vcpu);
 449 }
 450
 451 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 452 {
 453         kvm_arch_vcpu_free(vcpu);
 454 }
 455
 456 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 457 {
 458         return kvmppc_core_pending_dec(vcpu);
 459 }
 460
 461 /*
 462  * low level hrtimer wake routine. Because this runs in hardirq context
 463  * we schedule a tasklet to do the real work.
 464  */
 465 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 466 {
 467         struct kvm_vcpu *vcpu;
 468
 469         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 470         tasklet_schedule(&vcpu->arch.tasklet);
 471
 472         return HRTIMER_NORESTART;
 473 }
 474
 475 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 476 {
 477         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 478         tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
 479         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 480         vcpu->arch.dec_expires = ~(u64)0;
 481
 482 #ifdef CONFIG_KVM_EXIT_TIMING
 483         mutex_init(&vcpu->arch.exit_timing_lock);
 484 #endif
 485
 486         return 0;
 487 }
 488
 489 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 490 {
 491         kvmppc_mmu_destroy(vcpu);
 492 }
 493
 494 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 495 {
 496 #ifdef CONFIG_BOOKE
 497         /*
 498          * vrsave (formerly usprg0) isn't used by Linux, but may
 499          * be used by the guest.
 500          *
 501          * On non-booke this is associated with Altivec and
 502          * is handled by code in book3s.c.
 503          */
 504         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 505 #endif
 506         kvmppc_core_vcpu_load(vcpu, cpu);
 507         vcpu->cpu = smp_processor_id();
 508 }
 509
 510 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 511 {
 512         kvmppc_core_vcpu_put(vcpu);
 513 #ifdef CONFIG_BOOKE
 514         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 515 #endif
 516         vcpu->cpu = -1;
 517 }
 518
 519 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 520                                         struct kvm_guest_debug *dbg)
 521 {
 522         return -EINVAL;
 523 }
 524
 525 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
 526                                      struct kvm_run *run)
 527 {
 528         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
 529 }
 530
 531 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 532                                       struct kvm_run *run)
 533 {
 534         u64 uninitialized_var(gpr);
 535
 536         if (run->mmio.len > sizeof(gpr)) {
 537                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 538                 return;
 539         }
 540
 541         if (vcpu->arch.mmio_is_bigendian) {
 542                 switch (run->mmio.len) {
 543                 case 8: gpr = *(u64 *)run->mmio.data; break;
 544                 case 4: gpr = *(u32 *)run->mmio.data; break;
 545                 case 2: gpr = *(u16 *)run->mmio.data; break;
 546                 case 1: gpr = *(u8 *)run->mmio.data; break;
 547                 }
 548         } else {
 549                 /* Convert BE data from userland back to LE. */
 550                 switch (run->mmio.len) {
 551                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
 552                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
 553                 case 1: gpr = *(u8 *)run->mmio.data; break;
 554                 }
 555         }
 556
 557         if (vcpu->arch.mmio_sign_extend) {
 558                 switch (run->mmio.len) {
 559 #ifdef CONFIG_PPC64
 560                 case 4:
 561                         gpr = (s64)(s32)gpr;
 562                         break;
 563 #endif
 564                 case 2:
 565                         gpr = (s64)(s16)gpr;
 566                         break;
 567                 case 1:
 568                         gpr = (s64)(s8)gpr;
 569                         break;
 570                 }
 571         }
 572
 573         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 574
 575         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 576         case KVM_MMIO_REG_GPR:
 577                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 578                 break;
 579         case KVM_MMIO_REG_FPR:
 580                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 581                 break;
 582 #ifdef CONFIG_PPC_BOOK3S
 583         case KVM_MMIO_REG_QPR:
 584                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 585                 break;
 586         case KVM_MMIO_REG_FQPR:
 587                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 588                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 589                 break;
 590 #endif
 591         default:
 592                 BUG();
 593         }
 594 }
 595
 596 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 597                        unsigned int rt, unsigned int bytes, int is_bigendian)
 598 {
 599         if (bytes > sizeof(run->mmio.data)) {
 600                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 601                        run->mmio.len);
 602         }
 603
 604         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 605         run->mmio.len = bytes;
 606         run->mmio.is_write = 0;
 607
 608         vcpu->arch.io_gpr = rt;
 609         vcpu->arch.mmio_is_bigendian = is_bigendian;
 610         vcpu->mmio_needed = 1;
 611         vcpu->mmio_is_write = 0;
 612         vcpu->arch.mmio_sign_extend = 0;
 613
 614         return EMULATE_DO_MMIO;
 615 }
 616
 617 /* Same as above, but sign extends */
 618 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 619                         unsigned int rt, unsigned int bytes, int is_bigendian)
 620 {
 621         int r;
 622
 623         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
 624         vcpu->arch.mmio_sign_extend = 1;
 625
 626         return r;
 627 }
 628
 629 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 630                         u64 val, unsigned int bytes, int is_bigendian)
 631 {
 632         void *data = run->mmio.data;
 633
 634         if (bytes > sizeof(run->mmio.data)) {
 635                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 636                        run->mmio.len);
 637         }
 638
 639         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 640         run->mmio.len = bytes;
 641         run->mmio.is_write = 1;
 642         vcpu->mmio_needed = 1;
 643         vcpu->mmio_is_write = 1;
 644
 645         /* Store the value at the lowest bytes in 'data'. */
 646         if (is_bigendian) {
 647                 switch (bytes) {
 648                 case 8: *(u64 *)data = val; break;
 649                 case 4: *(u32 *)data = val; break;
 650                 case 2: *(u16 *)data = val; break;
 651                 case 1: *(u8  *)data = val; break;
 652                 }
 653         } else {
 654                 /* Store LE value into 'data'. */
 655                 switch (bytes) {
 656                 case 4: st_le32(data, val); break;
 657                 case 2: st_le16(data, val); break;
 658                 case 1: *(u8 *)data = val; break;
 659                 }
 660         }
 661
 662         return EMULATE_DO_MMIO;
 663 }
 664
 665 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 666 {
 667         int r;
 668         sigset_t sigsaved;
 669
 670         if (vcpu->sigset_active)
 671                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 672
 673         if (vcpu->mmio_needed) {
 674                 if (!vcpu->mmio_is_write)
 675                         kvmppc_complete_mmio_load(vcpu, run);
 676                 vcpu->mmio_needed = 0;
 677         } else if (vcpu->arch.dcr_needed) {
 678                 if (!vcpu->arch.dcr_is_write)
 679                         kvmppc_complete_dcr_load(vcpu, run);
 680                 vcpu->arch.dcr_needed = 0;
 681         } else if (vcpu->arch.osi_needed) {
 682                 u64 *gprs = run->osi.gprs;
 683                 int i;
 684
 685                 for (i = 0; i < 32; i++)
 686                         kvmppc_set_gpr(vcpu, i, gprs[i]);
 687                 vcpu->arch.osi_needed = 0;
 688         } else if (vcpu->arch.hcall_needed) {
 689                 int i;
 690
 691                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
 692                 for (i = 0; i < 9; ++i)
 693                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
 694                 vcpu->arch.hcall_needed = 0;
 695         }
 696
 697         r = kvmppc_vcpu_run(run, vcpu);
 698
 699         if (vcpu->sigset_active)
 700                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 701
 702         return r;
 703 }
 704
 705 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
 706 {
 707         if (irq->irq == KVM_INTERRUPT_UNSET) {
 708                 kvmppc_core_dequeue_external(vcpu, irq);
 709                 return 0;
 710         }
 711
 712         kvmppc_core_queue_external(vcpu, irq);
 713
 714         kvm_vcpu_kick(vcpu);
 715
 716         return 0;
 717 }
 718
 719 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
 720                                      struct kvm_enable_cap *cap)
 721 {
 722         int r;
 723
 724         if (cap->flags)
 725                 return -EINVAL;
 726
 727         switch (cap->cap) {
 728         case KVM_CAP_PPC_OSI:
 729                 r = 0;
 730                 vcpu->arch.osi_enabled = true;
 731                 break;
 732         case KVM_CAP_PPC_PAPR:
 733                 r = 0;
 734                 vcpu->arch.papr_enabled = true;
 735                 break;
 736 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 737         case KVM_CAP_SW_TLB: {
 738                 struct kvm_config_tlb cfg;
 739                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
 740
 741                 r = -EFAULT;
 742                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
 743                         break;
 744
 745                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
 746                 break;
 747         }
 748 #endif
 749         default:
 750                 r = -EINVAL;
 751                 break;
 752         }
 753
 754         if (!r)
 755                 r = kvmppc_sanity_check(vcpu);
 756
 757         return r;
 758 }
 759
 760 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 761                                     struct kvm_mp_state *mp_state)
 762 {
 763         return -EINVAL;
 764 }
 765
 766 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 767                                     struct kvm_mp_state *mp_state)
 768 {
 769         return -EINVAL;
 770 }
 771
 772 long kvm_arch_vcpu_ioctl(struct file *filp,
 773                          unsigned int ioctl, unsigned long arg)
 774 {
 775         struct kvm_vcpu *vcpu = filp->private_data;
 776         void __user *argp = (void __user *)arg;
 777         long r;
 778
 779         switch (ioctl) {
 780         case KVM_INTERRUPT: {
 781                 struct kvm_interrupt irq;
 782                 r = -EFAULT;
 783                 if (copy_from_user(&irq, argp, sizeof(irq)))
 784                         goto out;
 785                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 786                 goto out;
 787         }
 788
 789         case KVM_ENABLE_CAP:
 790         {
 791                 struct kvm_enable_cap cap;
 792                 r = -EFAULT;
 793                 if (copy_from_user(&cap, argp, sizeof(cap)))
 794                         goto out;
 795                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
 796                 break;
 797         }
 798
 799         case KVM_SET_ONE_REG:
 800         case KVM_GET_ONE_REG:
 801         {
 802                 struct kvm_one_reg reg;
 803                 r = -EFAULT;
 804                 if (copy_from_user(&reg, argp, sizeof(reg)))
 805                         goto out;
 806                 if (ioctl == KVM_SET_ONE_REG)
 807                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
 808                 else
 809                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
 810                 break;
 811         }
 812
 813 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 814         case KVM_DIRTY_TLB: {
 815                 struct kvm_dirty_tlb dirty;
 816                 r = -EFAULT;
 817                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
 818                         goto out;
 819                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
 820                 break;
 821         }
 822 #endif
 823         default:
 824                 r = -EINVAL;
 825         }
 826
 827 out:
 828         return r;
 829 }
 830
 831 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 832 {
 833         return VM_FAULT_SIGBUS;
 834 }
 835
 836 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
 837 {
 838         u32 inst_nop = 0x60000000;
 839 #ifdef CONFIG_KVM_BOOKE_HV
 840         u32 inst_sc1 = 0x44000022;
 841         pvinfo->hcall[0] = inst_sc1;
 842         pvinfo->hcall[1] = inst_nop;
 843         pvinfo->hcall[2] = inst_nop;
 844         pvinfo->hcall[3] = inst_nop;
 845 #else
 846         u32 inst_lis = 0x3c000000;
 847         u32 inst_ori = 0x60000000;
 848         u32 inst_sc = 0x44000002;
 849         u32 inst_imm_mask = 0xffff;
 850
 851         /*
 852          * The hypercall to get into KVM from within guest context is as
 853          * follows:
 854          *
 855          *    lis r0, r0, KVM_SC_MAGIC_R0@h
 856          *    ori r0, KVM_SC_MAGIC_R0@l
 857          *    sc
 858          *    nop
 859          */
 860         pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
 861         pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
 862         pvinfo->hcall[2] = inst_sc;
 863         pvinfo->hcall[3] = inst_nop;
 864 #endif
 865
 866         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
 867
 868         return 0;
 869 }
 870
 871 long kvm_arch_vm_ioctl(struct file *filp,
 872                        unsigned int ioctl, unsigned long arg)
 873 {
 874         void __user *argp = (void __user *)arg;
 875         long r;
 876
 877         switch (ioctl) {
 878         case KVM_PPC_GET_PVINFO: {
 879                 struct kvm_ppc_pvinfo pvinfo;
 880                 memset(&pvinfo, 0, sizeof(pvinfo));
 881                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
 882                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
 883                         r = -EFAULT;
 884                         goto out;
 885                 }
 886
 887                 break;
 888         }
 889 #ifdef CONFIG_PPC_BOOK3S_64
 890         case KVM_CREATE_SPAPR_TCE: {
 891                 struct kvm_create_spapr_tce create_tce;
 892                 struct kvm *kvm = filp->private_data;
 893
 894                 r = -EFAULT;
 895                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
 896                         goto out;
 897                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
 898                 goto out;
 899         }
 900 #endif /* CONFIG_PPC_BOOK3S_64 */
 901
 902 #ifdef CONFIG_KVM_BOOK3S_64_HV
 903         case KVM_ALLOCATE_RMA: {
 904                 struct kvm *kvm = filp->private_data;
 905                 struct kvm_allocate_rma rma;
 906
 907                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
 908                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
 909                         r = -EFAULT;
 910                 break;
 911         }
 912
 913         case KVM_PPC_ALLOCATE_HTAB: {
 914                 struct kvm *kvm = filp->private_data;
 915                 u32 htab_order;
 916
 917                 r = -EFAULT;
 918                 if (get_user(htab_order, (u32 __user *)argp))
 919                         break;
 920                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
 921                 if (r)
 922                         break;
 923                 r = -EFAULT;
 924                 if (put_user(htab_order, (u32 __user *)argp))
 925                         break;
 926                 r = 0;
 927                 break;
 928         }
 929 #endif /* CONFIG_KVM_BOOK3S_64_HV */
 930
 931 #ifdef CONFIG_PPC_BOOK3S_64
 932         case KVM_PPC_GET_SMMU_INFO: {
 933                 struct kvm *kvm = filp->private_data;
 934                 struct kvm_ppc_smmu_info info;
 935
 936                 memset(&info, 0, sizeof(info));
 937                 r = kvm_vm_ioctl_get_smmu_info(kvm, &info);
 938                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
 939                         r = -EFAULT;
 940                 break;
 941         }
 942 #endif /* CONFIG_PPC_BOOK3S_64 */
 943         default:
 944                 r = -ENOTTY;
 945         }
 946
 947 out:
 948         return r;
 949 }
 950
 951 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
 952 static unsigned long nr_lpids;
 953
 954 long kvmppc_alloc_lpid(void)
 955 {
 956         long lpid;
 957
 958         do {
 959                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
 960                 if (lpid >= nr_lpids) {
 961                         pr_err("%s: No LPIDs free\n", __func__);
 962                         return -ENOMEM;
 963                 }
 964         } while (test_and_set_bit(lpid, lpid_inuse));
 965
 966         return lpid;
 967 }
 968
 969 void kvmppc_claim_lpid(long lpid)
 970 {
 971         set_bit(lpid, lpid_inuse);
 972 }
 973
 974 void kvmppc_free_lpid(long lpid)
 975 {
 976         clear_bit(lpid, lpid_inuse);
 977 }
 978
 979 void kvmppc_init_lpid(unsigned long nr_lpids_param)
 980 {
 981         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
 982         memset(lpid_inuse, 0, sizeof(lpid_inuse));
 983 }
 984
 985 int kvm_arch_init(void *opaque)
 986 {
 987         return 0;
 988 }
 989
 990 void kvm_arch_exit(void)
 991 {
 992 }