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 <linux/file.h>
  29 #include <asm/cputable.h>
  30 #include <asm/uaccess.h>
  31 #include <asm/kvm_ppc.h>
  32 #include <asm/tlbflush.h>
  33 #include <asm/cputhreads.h>
  34 #include <asm/irqflags.h>
  35 #include "timing.h"
  36 #include "irq.h"
  37 #include "../mm/mmu_decl.h"
  38
  39 #define CREATE_TRACE_POINTS
  40 #include "trace.h"
  41
  42 struct kvmppc_ops *kvmppc_ops;
  43
  44 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  45 {
  46         return !!(v->arch.pending_exceptions) ||
  47                v->requests;
  48 }
  49
  50 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  51 {
  52         return 1;
  53 }
  54
  55 /*
  56  * Common checks before entering the guest world.  Call with interrupts
  57  * disabled.
  58  *
  59  * returns:
  60  *
  61  * == 1 if we're ready to go into guest state
  62  * <= 0 if we need to go back to the host with return value
  63  */
  64 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  65 {
  66         int r = 1;
  67
  68         WARN_ON_ONCE(!irqs_disabled());
  69         while (true) {
  70                 if (need_resched()) {
  71                         local_irq_enable();
  72                         cond_resched();
  73                         local_irq_disable();
  74                         continue;
  75                 }
  76
  77                 if (signal_pending(current)) {
  78                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
  79                         vcpu->run->exit_reason = KVM_EXIT_INTR;
  80                         r = -EINTR;
  81                         break;
  82                 }
  83
  84                 vcpu->mode = IN_GUEST_MODE;
  85
  86                 /*
  87                  * Reading vcpu->requests must happen after setting vcpu->mode,
  88                  * so we don't miss a request because the requester sees
  89                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
  90                  * before next entering the guest (and thus doesn't IPI).
  91                  */
  92                 smp_mb();
  93
  94                 if (vcpu->requests) {
  95                         /* Make sure we process requests preemptable */
  96                         local_irq_enable();
  97                         trace_kvm_check_requests(vcpu);
  98                         r = kvmppc_core_check_requests(vcpu);
  99                         local_irq_disable();
 100                         if (r > 0)
 101                                 continue;
 102                         break;
 103                 }
 104
 105                 if (kvmppc_core_prepare_to_enter(vcpu)) {
 106                         /* interrupts got enabled in between, so we
 107                            are back at square 1 */
 108                         continue;
 109                 }
 110
 111 #ifdef CONFIG_PPC64
 112                 /* lazy EE magic */
 113                 hard_irq_disable();
 114                 if (lazy_irq_pending()) {
 115                         /* Got an interrupt in between, try again */
 116                         local_irq_enable();
 117                         local_irq_disable();
 118                         kvm_guest_exit();
 119                         continue;
 120                 }
 121 #endif
 122
 123                 kvm_guest_enter();
 124                 break;
 125         }
 126
 127         return r;
 128 }
 129 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter);
 130
 131 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
 132 {
 133         int nr = kvmppc_get_gpr(vcpu, 11);
 134         int r;
 135         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
 136         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
 137         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
 138         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
 139         unsigned long r2 = 0;
 140
 141         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 142                 /* 32 bit mode */
 143                 param1 &= 0xffffffff;
 144                 param2 &= 0xffffffff;
 145                 param3 &= 0xffffffff;
 146                 param4 &= 0xffffffff;
 147         }
 148
 149         switch (nr) {
 150         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
 151         {
 152                 vcpu->arch.magic_page_pa = param1;
 153                 vcpu->arch.magic_page_ea = param2;
 154
 155                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 156
 157                 r = EV_SUCCESS;
 158                 break;
 159         }
 160         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 161                 r = EV_SUCCESS;
 162 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 163                 /* XXX Missing magic page on 44x */
 164                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 165 #endif
 166
 167                 /* Second return value is in r4 */
 168                 break;
 169         case EV_HCALL_TOKEN(EV_IDLE):
 170                 r = EV_SUCCESS;
 171                 kvm_vcpu_block(vcpu);
 172                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 173                 break;
 174         default:
 175                 r = EV_UNIMPLEMENTED;
 176                 break;
 177         }
 178
 179         kvmppc_set_gpr(vcpu, 4, r2);
 180
 181         return r;
 182 }
 183 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv);
 184
 185 int kvmppc_sanity_check(struct kvm_vcpu *vcpu)
 186 {
 187         int r = false;
 188
 189         /* We have to know what CPU to virtualize */
 190         if (!vcpu->arch.pvr)
 191                 goto out;
 192
 193         /* PAPR only works with book3s_64 */
 194         if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled)
 195                 goto out;
 196
 197         /* HV KVM can only do PAPR mode for now */
 198         if (!vcpu->arch.papr_enabled && kvmppc_ops->is_hv_enabled)
 199                 goto out;
 200
 201 #ifdef CONFIG_KVM_BOOKE_HV
 202         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 203                 goto out;
 204 #endif
 205
 206         r = true;
 207
 208 out:
 209         vcpu->arch.sane = r;
 210         return r ? 0 : -EINVAL;
 211 }
 212 EXPORT_SYMBOL_GPL(kvmppc_sanity_check);
 213
 214 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 215 {
 216         enum emulation_result er;
 217         int r;
 218
 219         er = kvmppc_emulate_instruction(run, vcpu);
 220         switch (er) {
 221         case EMULATE_DONE:
 222                 /* Future optimization: only reload non-volatiles if they were
 223                  * actually modified. */
 224                 r = RESUME_GUEST_NV;
 225                 break;
 226         case EMULATE_DO_MMIO:
 227                 run->exit_reason = KVM_EXIT_MMIO;
 228                 /* We must reload nonvolatiles because "update" load/store
 229                  * instructions modify register state. */
 230                 /* Future optimization: only reload non-volatiles if they were
 231                  * actually modified. */
 232                 r = RESUME_HOST_NV;
 233                 break;
 234         case EMULATE_FAIL:
 235                 /* XXX Deliver Program interrupt to guest. */
 236                 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
 237                        kvmppc_get_last_inst(vcpu));
 238                 r = RESUME_HOST;
 239                 break;
 240         default:
 241                 WARN_ON(1);
 242                 r = RESUME_GUEST;
 243         }
 244
 245         return r;
 246 }
 247 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio);
 248
 249 int kvm_arch_hardware_enable(void *garbage)
 250 {
 251         return 0;
 252 }
 253
 254 void kvm_arch_hardware_disable(void *garbage)
 255 {
 256 }
 257
 258 int kvm_arch_hardware_setup(void)
 259 {
 260         return 0;
 261 }
 262
 263 void kvm_arch_hardware_unsetup(void)
 264 {
 265 }
 266
 267 void kvm_arch_check_processor_compat(void *rtn)
 268 {
 269         *(int *)rtn = kvmppc_core_check_processor_compat();
 270 }
 271
 272 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 273 {
 274         if (type)
 275                 return -EINVAL;
 276
 277         return kvmppc_core_init_vm(kvm);
 278 }
 279
 280 void kvm_arch_destroy_vm(struct kvm *kvm)
 281 {
 282         unsigned int i;
 283         struct kvm_vcpu *vcpu;
 284
 285         kvm_for_each_vcpu(i, vcpu, kvm)
 286                 kvm_arch_vcpu_free(vcpu);
 287
 288         mutex_lock(&kvm->lock);
 289         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 290                 kvm->vcpus[i] = NULL;
 291
 292         atomic_set(&kvm->online_vcpus, 0);
 293
 294         kvmppc_core_destroy_vm(kvm);
 295
 296         mutex_unlock(&kvm->lock);
 297 }
 298
 299 void kvm_arch_sync_events(struct kvm *kvm)
 300 {
 301 }
 302
 303 int kvm_dev_ioctl_check_extension(long ext)
 304 {
 305         int r;
 306
 307         switch (ext) {
 308 #ifdef CONFIG_BOOKE
 309         case KVM_CAP_PPC_BOOKE_SREGS:
 310         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 311         case KVM_CAP_PPC_EPR:
 312 #else
 313         case KVM_CAP_PPC_SEGSTATE:
 314         case KVM_CAP_PPC_HIOR:
 315         case KVM_CAP_PPC_PAPR:
 316 #endif
 317         case KVM_CAP_PPC_UNSET_IRQ:
 318         case KVM_CAP_PPC_IRQ_LEVEL:
 319         case KVM_CAP_ENABLE_CAP:
 320         case KVM_CAP_ONE_REG:
 321         case KVM_CAP_IOEVENTFD:
 322         case KVM_CAP_DEVICE_CTRL:
 323                 r = 1;
 324                 break;
 325         case KVM_CAP_PPC_PAIRED_SINGLES:
 326         case KVM_CAP_PPC_OSI:
 327         case KVM_CAP_PPC_GET_PVINFO:
 328 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 329         case KVM_CAP_SW_TLB:
 330 #endif
 331                 /* We support this only for PR */
 332                 r = !kvmppc_ops->is_hv_enabled;
 333                 break;
 334 #ifdef CONFIG_KVM_MMIO
 335         case KVM_CAP_COALESCED_MMIO:
 336                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 337                 break;
 338 #endif
 339 #ifdef CONFIG_KVM_MPIC
 340         case KVM_CAP_IRQ_MPIC:
 341                 r = 1;
 342                 break;
 343 #endif
 344
 345 #ifdef CONFIG_PPC_BOOK3S_64
 346         case KVM_CAP_SPAPR_TCE:
 347         case KVM_CAP_PPC_ALLOC_HTAB:
 348         case KVM_CAP_PPC_RTAS:
 349 #ifdef CONFIG_KVM_XICS
 350         case KVM_CAP_IRQ_XICS:
 351 #endif
 352                 r = 1;
 353                 break;
 354 #endif /* CONFIG_PPC_BOOK3S_64 */
 355 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 356         case KVM_CAP_PPC_SMT:
 357                 if (kvmppc_ops->is_hv_enabled)
 358                         r = threads_per_core;
 359                 else
 360                         r = 0;
 361                 break;
 362         case KVM_CAP_PPC_RMA:
 363                 r = kvmppc_ops->is_hv_enabled;
 364                 /* PPC970 requires an RMA */
 365                 if (r && cpu_has_feature(CPU_FTR_ARCH_201))
 366                         r = 2;
 367                 break;
 368 #endif
 369         case KVM_CAP_SYNC_MMU:
 370 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 371                 if (kvmppc_ops->is_hv_enabled)
 372                         r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
 373                 else
 374                         r = 0;
 375 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 376                 r = 1;
 377 #else
 378                 r = 0;
 379 #endif
 380                 break;
 381 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 382         case KVM_CAP_PPC_HTAB_FD:
 383                 r = kvmppc_ops->is_hv_enabled;
 384                 break;
 385 #endif
 386         case KVM_CAP_NR_VCPUS:
 387                 /*
 388                  * Recommending a number of CPUs is somewhat arbitrary; we
 389                  * return the number of present CPUs for -HV (since a host
 390                  * will have secondary threads "offline"), and for other KVM
 391                  * implementations just count online CPUs.
 392                  */
 393                 if (kvmppc_ops->is_hv_enabled)
 394                         r = num_present_cpus();
 395                 else
 396                         r = num_online_cpus();
 397                 break;
 398         case KVM_CAP_MAX_VCPUS:
 399                 r = KVM_MAX_VCPUS;
 400                 break;
 401 #ifdef CONFIG_PPC_BOOK3S_64
 402         case KVM_CAP_PPC_GET_SMMU_INFO:
 403                 r = 1;
 404                 break;
 405 #endif
 406         default:
 407                 r = 0;
 408                 break;
 409         }
 410         return r;
 411
 412 }
 413
 414 long kvm_arch_dev_ioctl(struct file *filp,
 415                         unsigned int ioctl, unsigned long arg)
 416 {
 417         return -EINVAL;
 418 }
 419
 420 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 421                            struct kvm_memory_slot *dont)
 422 {
 423         kvmppc_core_free_memslot(kvm, free, dont);
 424 }
 425
 426 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 427                             unsigned long npages)
 428 {
 429         return kvmppc_core_create_memslot(kvm, slot, npages);
 430 }
 431
 432 void kvm_arch_memslots_updated(struct kvm *kvm)
 433 {
 434 }
 435
 436 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 437                                    struct kvm_memory_slot *memslot,
 438                                    struct kvm_userspace_memory_region *mem,
 439                                    enum kvm_mr_change change)
 440 {
 441         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 442 }
 443
 444 void kvm_arch_commit_memory_region(struct kvm *kvm,
 445                                    struct kvm_userspace_memory_region *mem,
 446                                    const struct kvm_memory_slot *old,
 447                                    enum kvm_mr_change change)
 448 {
 449         kvmppc_core_commit_memory_region(kvm, mem, old);
 450 }
 451
 452 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 453 {
 454 }
 455
 456 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 457                                    struct kvm_memory_slot *slot)
 458 {
 459         kvmppc_core_flush_memslot(kvm, slot);
 460 }
 461
 462 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 463 {
 464         struct kvm_vcpu *vcpu;
 465         vcpu = kvmppc_core_vcpu_create(kvm, id);
 466         if (!IS_ERR(vcpu)) {
 467                 vcpu->arch.wqp = &vcpu->wq;
 468                 kvmppc_create_vcpu_debugfs(vcpu, id);
 469         }
 470         return vcpu;
 471 }
 472
 473 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 474 {
 475         return 0;
 476 }
 477
 478 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 479 {
 480         /* Make sure we're not using the vcpu anymore */
 481         hrtimer_cancel(&vcpu->arch.dec_timer);
 482         tasklet_kill(&vcpu->arch.tasklet);
 483
 484         kvmppc_remove_vcpu_debugfs(vcpu);
 485
 486         switch (vcpu->arch.irq_type) {
 487         case KVMPPC_IRQ_MPIC:
 488                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 489                 break;
 490         case KVMPPC_IRQ_XICS:
 491                 kvmppc_xics_free_icp(vcpu);
 492                 break;
 493         }
 494
 495         kvmppc_core_vcpu_free(vcpu);
 496 }
 497
 498 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 499 {
 500         kvm_arch_vcpu_free(vcpu);
 501 }
 502
 503 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 504 {
 505         return kvmppc_core_pending_dec(vcpu);
 506 }
 507
 508 /*
 509  * low level hrtimer wake routine. Because this runs in hardirq context
 510  * we schedule a tasklet to do the real work.
 511  */
 512 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 513 {
 514         struct kvm_vcpu *vcpu;
 515
 516         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 517         tasklet_schedule(&vcpu->arch.tasklet);
 518
 519         return HRTIMER_NORESTART;
 520 }
 521
 522 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 523 {
 524         int ret;
 525
 526         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 527         tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
 528         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 529         vcpu->arch.dec_expires = ~(u64)0;
 530
 531 #ifdef CONFIG_KVM_EXIT_TIMING
 532         mutex_init(&vcpu->arch.exit_timing_lock);
 533 #endif
 534         ret = kvmppc_subarch_vcpu_init(vcpu);
 535         return ret;
 536 }
 537
 538 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 539 {
 540         kvmppc_mmu_destroy(vcpu);
 541         kvmppc_subarch_vcpu_uninit(vcpu);
 542 }
 543
 544 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 545 {
 546 #ifdef CONFIG_BOOKE
 547         /*
 548          * vrsave (formerly usprg0) isn't used by Linux, but may
 549          * be used by the guest.
 550          *
 551          * On non-booke this is associated with Altivec and
 552          * is handled by code in book3s.c.
 553          */
 554         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 555 #endif
 556         kvmppc_core_vcpu_load(vcpu, cpu);
 557 }
 558
 559 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 560 {
 561         kvmppc_core_vcpu_put(vcpu);
 562 #ifdef CONFIG_BOOKE
 563         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 564 #endif
 565 }
 566
 567 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
 568                                      struct kvm_run *run)
 569 {
 570         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
 571 }
 572
 573 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 574                                       struct kvm_run *run)
 575 {
 576         u64 uninitialized_var(gpr);
 577
 578         if (run->mmio.len > sizeof(gpr)) {
 579                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 580                 return;
 581         }
 582
 583         if (vcpu->arch.mmio_is_bigendian) {
 584                 switch (run->mmio.len) {
 585                 case 8: gpr = *(u64 *)run->mmio.data; break;
 586                 case 4: gpr = *(u32 *)run->mmio.data; break;
 587                 case 2: gpr = *(u16 *)run->mmio.data; break;
 588                 case 1: gpr = *(u8 *)run->mmio.data; break;
 589                 }
 590         } else {
 591                 /* Convert BE data from userland back to LE. */
 592                 switch (run->mmio.len) {
 593                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
 594                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
 595                 case 1: gpr = *(u8 *)run->mmio.data; break;
 596                 }
 597         }
 598
 599         if (vcpu->arch.mmio_sign_extend) {
 600                 switch (run->mmio.len) {
 601 #ifdef CONFIG_PPC64
 602                 case 4:
 603                         gpr = (s64)(s32)gpr;
 604                         break;
 605 #endif
 606                 case 2:
 607                         gpr = (s64)(s16)gpr;
 608                         break;
 609                 case 1:
 610                         gpr = (s64)(s8)gpr;
 611                         break;
 612                 }
 613         }
 614
 615         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 616
 617         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 618         case KVM_MMIO_REG_GPR:
 619                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 620                 break;
 621         case KVM_MMIO_REG_FPR:
 622                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 623                 break;
 624 #ifdef CONFIG_PPC_BOOK3S
 625         case KVM_MMIO_REG_QPR:
 626                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 627                 break;
 628         case KVM_MMIO_REG_FQPR:
 629                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 630                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 631                 break;
 632 #endif
 633         default:
 634                 BUG();
 635         }
 636 }
 637
 638 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 639                        unsigned int rt, unsigned int bytes, int is_bigendian)
 640 {
 641         int idx, ret;
 642
 643         if (bytes > sizeof(run->mmio.data)) {
 644                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 645                        run->mmio.len);
 646         }
 647
 648         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 649         run->mmio.len = bytes;
 650         run->mmio.is_write = 0;
 651
 652         vcpu->arch.io_gpr = rt;
 653         vcpu->arch.mmio_is_bigendian = is_bigendian;
 654         vcpu->mmio_needed = 1;
 655         vcpu->mmio_is_write = 0;
 656         vcpu->arch.mmio_sign_extend = 0;
 657
 658         idx = srcu_read_lock(&vcpu->kvm->srcu);
 659
 660         ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 661                               bytes, &run->mmio.data);
 662
 663         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 664
 665         if (!ret) {
 666                 kvmppc_complete_mmio_load(vcpu, run);
 667                 vcpu->mmio_needed = 0;
 668                 return EMULATE_DONE;
 669         }
 670
 671         return EMULATE_DO_MMIO;
 672 }
 673 EXPORT_SYMBOL_GPL(kvmppc_handle_load);
 674
 675 /* Same as above, but sign extends */
 676 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 677                         unsigned int rt, unsigned int bytes, int is_bigendian)
 678 {
 679         int r;
 680
 681         vcpu->arch.mmio_sign_extend = 1;
 682         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
 683
 684         return r;
 685 }
 686
 687 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 688                         u64 val, unsigned int bytes, int is_bigendian)
 689 {
 690         void *data = run->mmio.data;
 691         int idx, ret;
 692
 693         if (bytes > sizeof(run->mmio.data)) {
 694                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 695                        run->mmio.len);
 696         }
 697
 698         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 699         run->mmio.len = bytes;
 700         run->mmio.is_write = 1;
 701         vcpu->mmio_needed = 1;
 702         vcpu->mmio_is_write = 1;
 703
 704         /* Store the value at the lowest bytes in 'data'. */
 705         if (is_bigendian) {
 706                 switch (bytes) {
 707                 case 8: *(u64 *)data = val; break;
 708                 case 4: *(u32 *)data = val; break;
 709                 case 2: *(u16 *)data = val; break;
 710                 case 1: *(u8  *)data = val; break;
 711                 }
 712         } else {
 713                 /* Store LE value into 'data'. */
 714                 switch (bytes) {
 715                 case 4: st_le32(data, val); break;
 716                 case 2: st_le16(data, val); break;
 717                 case 1: *(u8 *)data = val; break;
 718                 }
 719         }
 720
 721         idx = srcu_read_lock(&vcpu->kvm->srcu);
 722
 723         ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 724                                bytes, &run->mmio.data);
 725
 726         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 727
 728         if (!ret) {
 729                 vcpu->mmio_needed = 0;
 730                 return EMULATE_DONE;
 731         }
 732
 733         return EMULATE_DO_MMIO;
 734 }
 735 EXPORT_SYMBOL_GPL(kvmppc_handle_store);
 736
 737 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 738 {
 739         int r;
 740         sigset_t sigsaved;
 741
 742         if (vcpu->sigset_active)
 743                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 744
 745         if (vcpu->mmio_needed) {
 746                 if (!vcpu->mmio_is_write)
 747                         kvmppc_complete_mmio_load(vcpu, run);
 748                 vcpu->mmio_needed = 0;
 749         } else if (vcpu->arch.dcr_needed) {
 750                 if (!vcpu->arch.dcr_is_write)
 751                         kvmppc_complete_dcr_load(vcpu, run);
 752                 vcpu->arch.dcr_needed = 0;
 753         } else if (vcpu->arch.osi_needed) {
 754                 u64 *gprs = run->osi.gprs;
 755                 int i;
 756
 757                 for (i = 0; i < 32; i++)
 758                         kvmppc_set_gpr(vcpu, i, gprs[i]);
 759                 vcpu->arch.osi_needed = 0;
 760         } else if (vcpu->arch.hcall_needed) {
 761                 int i;
 762
 763                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
 764                 for (i = 0; i < 9; ++i)
 765                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
 766                 vcpu->arch.hcall_needed = 0;
 767 #ifdef CONFIG_BOOKE
 768         } else if (vcpu->arch.epr_needed) {
 769                 kvmppc_set_epr(vcpu, run->epr.epr);
 770                 vcpu->arch.epr_needed = 0;
 771 #endif
 772         }
 773
 774         r = kvmppc_vcpu_run(run, vcpu);
 775
 776         if (vcpu->sigset_active)
 777                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 778
 779         return r;
 780 }
 781
 782 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
 783 {
 784         if (irq->irq == KVM_INTERRUPT_UNSET) {
 785                 kvmppc_core_dequeue_external(vcpu);
 786                 return 0;
 787         }
 788
 789         kvmppc_core_queue_external(vcpu, irq);
 790
 791         kvm_vcpu_kick(vcpu);
 792
 793         return 0;
 794 }
 795
 796 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
 797                                      struct kvm_enable_cap *cap)
 798 {
 799         int r;
 800
 801         if (cap->flags)
 802                 return -EINVAL;
 803
 804         switch (cap->cap) {
 805         case KVM_CAP_PPC_OSI:
 806                 r = 0;
 807                 vcpu->arch.osi_enabled = true;
 808                 break;
 809         case KVM_CAP_PPC_PAPR:
 810                 r = 0;
 811                 vcpu->arch.papr_enabled = true;
 812                 break;
 813         case KVM_CAP_PPC_EPR:
 814                 r = 0;
 815                 if (cap->args[0])
 816                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
 817                 else
 818                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
 819                 break;
 820 #ifdef CONFIG_BOOKE
 821         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 822                 r = 0;
 823                 vcpu->arch.watchdog_enabled = true;
 824                 break;
 825 #endif
 826 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 827         case KVM_CAP_SW_TLB: {
 828                 struct kvm_config_tlb cfg;
 829                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
 830
 831                 r = -EFAULT;
 832                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
 833                         break;
 834
 835                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
 836                 break;
 837         }
 838 #endif
 839 #ifdef CONFIG_KVM_MPIC
 840         case KVM_CAP_IRQ_MPIC: {
 841                 struct fd f;
 842                 struct kvm_device *dev;
 843
 844                 r = -EBADF;
 845                 f = fdget(cap->args[0]);
 846                 if (!f.file)
 847                         break;
 848
 849                 r = -EPERM;
 850                 dev = kvm_device_from_filp(f.file);
 851                 if (dev)
 852                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
 853
 854                 fdput(f);
 855                 break;
 856         }
 857 #endif
 858 #ifdef CONFIG_KVM_XICS
 859         case KVM_CAP_IRQ_XICS: {
 860                 struct fd f;
 861                 struct kvm_device *dev;
 862
 863                 r = -EBADF;
 864                 f = fdget(cap->args[0]);
 865                 if (!f.file)
 866                         break;
 867
 868                 r = -EPERM;
 869                 dev = kvm_device_from_filp(f.file);
 870                 if (dev)
 871                         r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
 872
 873                 fdput(f);
 874                 break;
 875         }
 876 #endif /* CONFIG_KVM_XICS */
 877         default:
 878                 r = -EINVAL;
 879                 break;
 880         }
 881
 882         if (!r)
 883                 r = kvmppc_sanity_check(vcpu);
 884
 885         return r;
 886 }
 887
 888 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 889                                     struct kvm_mp_state *mp_state)
 890 {
 891         return -EINVAL;
 892 }
 893
 894 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 895                                     struct kvm_mp_state *mp_state)
 896 {
 897         return -EINVAL;
 898 }
 899
 900 long kvm_arch_vcpu_ioctl(struct file *filp,
 901                          unsigned int ioctl, unsigned long arg)
 902 {
 903         struct kvm_vcpu *vcpu = filp->private_data;
 904         void __user *argp = (void __user *)arg;
 905         long r;
 906
 907         switch (ioctl) {
 908         case KVM_INTERRUPT: {
 909                 struct kvm_interrupt irq;
 910                 r = -EFAULT;
 911                 if (copy_from_user(&irq, argp, sizeof(irq)))
 912                         goto out;
 913                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 914                 goto out;
 915         }
 916
 917         case KVM_ENABLE_CAP:
 918         {
 919                 struct kvm_enable_cap cap;
 920                 r = -EFAULT;
 921                 if (copy_from_user(&cap, argp, sizeof(cap)))
 922                         goto out;
 923                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
 924                 break;
 925         }
 926
 927         case KVM_SET_ONE_REG:
 928         case KVM_GET_ONE_REG:
 929         {
 930                 struct kvm_one_reg reg;
 931                 r = -EFAULT;
 932                 if (copy_from_user(&reg, argp, sizeof(reg)))
 933                         goto out;
 934                 if (ioctl == KVM_SET_ONE_REG)
 935                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
 936                 else
 937                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
 938                 break;
 939         }
 940
 941 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 942         case KVM_DIRTY_TLB: {
 943                 struct kvm_dirty_tlb dirty;
 944                 r = -EFAULT;
 945                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
 946                         goto out;
 947                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
 948                 break;
 949         }
 950 #endif
 951         default:
 952                 r = -EINVAL;
 953         }
 954
 955 out:
 956         return r;
 957 }
 958
 959 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 960 {
 961         return VM_FAULT_SIGBUS;
 962 }
 963
 964 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
 965 {
 966         u32 inst_nop = 0x60000000;
 967 #ifdef CONFIG_KVM_BOOKE_HV
 968         u32 inst_sc1 = 0x44000022;
 969         pvinfo->hcall[0] = inst_sc1;
 970         pvinfo->hcall[1] = inst_nop;
 971         pvinfo->hcall[2] = inst_nop;
 972         pvinfo->hcall[3] = inst_nop;
 973 #else
 974         u32 inst_lis = 0x3c000000;
 975         u32 inst_ori = 0x60000000;
 976         u32 inst_sc = 0x44000002;
 977         u32 inst_imm_mask = 0xffff;
 978
 979         /*
 980          * The hypercall to get into KVM from within guest context is as
 981          * follows:
 982          *
 983          *    lis r0, r0, KVM_SC_MAGIC_R0@h
 984          *    ori r0, KVM_SC_MAGIC_R0@l
 985          *    sc
 986          *    nop
 987          */
 988         pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
 989         pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
 990         pvinfo->hcall[2] = inst_sc;
 991         pvinfo->hcall[3] = inst_nop;
 992 #endif
 993
 994         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
 995
 996         return 0;
 997 }
 998
 999 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
1000                           bool line_status)
1001 {
1002         if (!irqchip_in_kernel(kvm))
1003                 return -ENXIO;
1004
1005         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1006                                         irq_event->irq, irq_event->level,
1007                                         line_status);
1008         return 0;
1009 }
1010
1011 long kvm_arch_vm_ioctl(struct file *filp,
1012                        unsigned int ioctl, unsigned long arg)
1013 {
1014         struct kvm *kvm __maybe_unused = filp->private_data;
1015         void __user *argp = (void __user *)arg;
1016         long r;
1017
1018         switch (ioctl) {
1019         case KVM_PPC_GET_PVINFO: {
1020                 struct kvm_ppc_pvinfo pvinfo;
1021                 memset(&pvinfo, 0, sizeof(pvinfo));
1022                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1023                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1024                         r = -EFAULT;
1025                         goto out;
1026                 }
1027
1028                 break;
1029         }
1030 #ifdef CONFIG_PPC_BOOK3S_64
1031         case KVM_CREATE_SPAPR_TCE: {
1032                 struct kvm_create_spapr_tce create_tce;
1033
1034                 r = -EFAULT;
1035                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1036                         goto out;
1037                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1038                 goto out;
1039         }
1040         case KVM_PPC_GET_SMMU_INFO: {
1041                 struct kvm_ppc_smmu_info info;
1042
1043                 memset(&info, 0, sizeof(info));
1044                 r = kvmppc_ops->get_smmu_info(kvm, &info);
1045                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1046                         r = -EFAULT;
1047                 break;
1048         }
1049         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1050                 struct kvm *kvm = filp->private_data;
1051
1052                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1053                 break;
1054         }
1055         default:
1056                 r = kvmppc_ops->arch_vm_ioctl(filp, ioctl, arg);
1057
1058 #else /* CONFIG_PPC_BOOK3S_64 */
1059         default:
1060                 r = -ENOTTY;
1061 #endif
1062         }
1063 out:
1064         return r;
1065 }
1066
1067 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1068 static unsigned long nr_lpids;
1069
1070 long kvmppc_alloc_lpid(void)
1071 {
1072         long lpid;
1073
1074         do {
1075                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1076                 if (lpid >= nr_lpids) {
1077                         pr_err("%s: No LPIDs free\n", __func__);
1078                         return -ENOMEM;
1079                 }
1080         } while (test_and_set_bit(lpid, lpid_inuse));
1081
1082         return lpid;
1083 }
1084 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid);
1085
1086 void kvmppc_claim_lpid(long lpid)
1087 {
1088         set_bit(lpid, lpid_inuse);
1089 }
1090 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid);
1091
1092 void kvmppc_free_lpid(long lpid)
1093 {
1094         clear_bit(lpid, lpid_inuse);
1095 }
1096 EXPORT_SYMBOL_GPL(kvmppc_free_lpid);
1097
1098 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1099 {
1100         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1101         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1102 }
1103 EXPORT_SYMBOL_GPL(kvmppc_init_lpid);
1104
1105 int kvm_arch_init(void *opaque)
1106 {
1107         if (kvmppc_ops) {
1108                 printk(KERN_ERR "kvm: already loaded the other module\n");
1109                 return -EEXIST;
1110         }
1111         kvmppc_ops = (struct kvmppc_ops *)opaque;
1112         return 0;
1113 }
1114
1115 void kvm_arch_exit(void)
1116 {
1117         kvmppc_ops = NULL;
1118 }