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