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:
  57  *
  58  * == 1 if we're ready to go into guest state
  59  * <= 0 if we need to go back to the host with return value
  60  */
  61 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  62 {
  63         int r = 1;
  64
  65         WARN_ON_ONCE(!irqs_disabled());
  66         while (true) {
  67                 if (need_resched()) {
  68                         local_irq_enable();
  69                         cond_resched();
  70                         local_irq_disable();
  71                         continue;
  72                 }
  73
  74                 if (signal_pending(current)) {
  75                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
  76                         vcpu->run->exit_reason = KVM_EXIT_INTR;
  77                         r = -EINTR;
  78                         break;
  79                 }
  80
  81                 vcpu->mode = IN_GUEST_MODE;
  82
  83                 /*
  84                  * Reading vcpu->requests must happen after setting vcpu->mode,
  85                  * so we don't miss a request because the requester sees
  86                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
  87                  * before next entering the guest (and thus doesn't IPI).
  88                  */
  89                 smp_mb();
  90
  91                 if (vcpu->requests) {
  92                         /* Make sure we process requests preemptable */
  93                         local_irq_enable();
  94                         trace_kvm_check_requests(vcpu);
  95                         r = kvmppc_core_check_requests(vcpu);
  96                         local_irq_disable();
  97                         if (r > 0)
  98                                 continue;
  99                         break;
 100                 }
 101
 102                 if (kvmppc_core_prepare_to_enter(vcpu)) {
 103                         /* interrupts got enabled in between, so we
 104                            are back at square 1 */
 105                         continue;
 106                 }
 107
 108 #ifdef CONFIG_PPC64
 109                 /* lazy EE magic */
 110                 hard_irq_disable();
 111                 if (lazy_irq_pending()) {
 112                         /* Got an interrupt in between, try again */
 113                         local_irq_enable();
 114                         local_irq_disable();
 115                         kvm_guest_exit();
 116                         continue;
 117                 }
 118
 119                 trace_hardirqs_on();
 120 #endif
 121
 122                 kvm_guest_enter();
 123                 break;
 124         }
 125
 126         return r;
 127 }
 128 #endif /* CONFIG_KVM_BOOK3S_64_HV */
 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 #ifdef CONFIG_KVM_BOOK3S_64_HV
 196         /* HV KVM can only do PAPR mode for now */
 197         if (!vcpu->arch.papr_enabled)
 198                 goto out;
 199 #endif
 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
 213 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 214 {
 215         enum emulation_result er;
 216         int r;
 217
 218         er = kvmppc_emulate_instruction(run, vcpu);
 219         switch (er) {
 220         case EMULATE_DONE:
 221                 /* Future optimization: only reload non-volatiles if they were
 222                  * actually modified. */
 223                 r = RESUME_GUEST_NV;
 224                 break;
 225         case EMULATE_DO_MMIO:
 226                 run->exit_reason = KVM_EXIT_MMIO;
 227                 /* We must reload nonvolatiles because "update" load/store
 228                  * instructions modify register state. */
 229                 /* Future optimization: only reload non-volatiles if they were
 230                  * actually modified. */
 231                 r = RESUME_HOST_NV;
 232                 break;
 233         case EMULATE_FAIL:
 234                 /* XXX Deliver Program interrupt to guest. */
 235                 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
 236                        kvmppc_get_last_inst(vcpu));
 237                 r = RESUME_HOST;
 238                 break;
 239         default:
 240                 WARN_ON(1);
 241                 r = RESUME_GUEST;
 242         }
 243
 244         return r;
 245 }
 246
 247 int kvm_arch_hardware_enable(void *garbage)
 248 {
 249         return 0;
 250 }
 251
 252 void kvm_arch_hardware_disable(void *garbage)
 253 {
 254 }
 255
 256 int kvm_arch_hardware_setup(void)
 257 {
 258         return 0;
 259 }
 260
 261 void kvm_arch_hardware_unsetup(void)
 262 {
 263 }
 264
 265 void kvm_arch_check_processor_compat(void *rtn)
 266 {
 267         *(int *)rtn = kvmppc_core_check_processor_compat();
 268 }
 269
 270 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 271 {
 272         if (type)
 273                 return -EINVAL;
 274
 275         return kvmppc_core_init_vm(kvm);
 276 }
 277
 278 void kvm_arch_destroy_vm(struct kvm *kvm)
 279 {
 280         unsigned int i;
 281         struct kvm_vcpu *vcpu;
 282
 283         kvm_for_each_vcpu(i, vcpu, kvm)
 284                 kvm_arch_vcpu_free(vcpu);
 285
 286         mutex_lock(&kvm->lock);
 287         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
 288                 kvm->vcpus[i] = NULL;
 289
 290         atomic_set(&kvm->online_vcpus, 0);
 291
 292         kvmppc_core_destroy_vm(kvm);
 293
 294         mutex_unlock(&kvm->lock);
 295 }
 296
 297 void kvm_arch_sync_events(struct kvm *kvm)
 298 {
 299 }
 300
 301 int kvm_dev_ioctl_check_extension(long ext)
 302 {
 303         int r;
 304
 305         switch (ext) {
 306 #ifdef CONFIG_BOOKE
 307         case KVM_CAP_PPC_BOOKE_SREGS:
 308         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 309         case KVM_CAP_PPC_EPR:
 310 #else
 311         case KVM_CAP_PPC_SEGSTATE:
 312         case KVM_CAP_PPC_HIOR:
 313         case KVM_CAP_PPC_PAPR:
 314 #endif
 315         case KVM_CAP_PPC_UNSET_IRQ:
 316         case KVM_CAP_PPC_IRQ_LEVEL:
 317         case KVM_CAP_ENABLE_CAP:
 318         case KVM_CAP_ONE_REG:
 319         case KVM_CAP_IOEVENTFD:
 320                 r = 1;
 321                 break;
 322 #ifndef CONFIG_KVM_BOOK3S_64_HV
 323         case KVM_CAP_PPC_PAIRED_SINGLES:
 324         case KVM_CAP_PPC_OSI:
 325         case KVM_CAP_PPC_GET_PVINFO:
 326 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 327         case KVM_CAP_SW_TLB:
 328 #endif
 329                 r = 1;
 330                 break;
 331         case KVM_CAP_COALESCED_MMIO:
 332                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 333                 break;
 334 #endif
 335 #ifdef CONFIG_PPC_BOOK3S_64
 336         case KVM_CAP_SPAPR_TCE:
 337         case KVM_CAP_PPC_ALLOC_HTAB:
 338                 r = 1;
 339                 break;
 340 #endif /* CONFIG_PPC_BOOK3S_64 */
 341 #ifdef CONFIG_KVM_BOOK3S_64_HV
 342         case KVM_CAP_PPC_SMT:
 343                 r = threads_per_core;
 344                 break;
 345         case KVM_CAP_PPC_RMA:
 346                 r = 1;
 347                 /* PPC970 requires an RMA */
 348                 if (cpu_has_feature(CPU_FTR_ARCH_201))
 349                         r = 2;
 350                 break;
 351 #endif
 352         case KVM_CAP_SYNC_MMU:
 353 #ifdef CONFIG_KVM_BOOK3S_64_HV
 354                 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
 355 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 356                 r = 1;
 357 #else
 358                 r = 0;
 359                 break;
 360 #endif
 361 #ifdef CONFIG_KVM_BOOK3S_64_HV
 362         case KVM_CAP_PPC_HTAB_FD:
 363                 r = 1;
 364                 break;
 365 #endif
 366                 break;
 367         case KVM_CAP_NR_VCPUS:
 368                 /*
 369                  * Recommending a number of CPUs is somewhat arbitrary; we
 370                  * return the number of present CPUs for -HV (since a host
 371                  * will have secondary threads "offline"), and for other KVM
 372                  * implementations just count online CPUs.
 373                  */
 374 #ifdef CONFIG_KVM_BOOK3S_64_HV
 375                 r = num_present_cpus();
 376 #else
 377                 r = num_online_cpus();
 378 #endif
 379                 break;
 380         case KVM_CAP_MAX_VCPUS:
 381                 r = KVM_MAX_VCPUS;
 382                 break;
 383 #ifdef CONFIG_PPC_BOOK3S_64
 384         case KVM_CAP_PPC_GET_SMMU_INFO:
 385                 r = 1;
 386                 break;
 387 #endif
 388         default:
 389                 r = 0;
 390                 break;
 391         }
 392         return r;
 393
 394 }
 395
 396 long kvm_arch_dev_ioctl(struct file *filp,
 397                         unsigned int ioctl, unsigned long arg)
 398 {
 399         return -EINVAL;
 400 }
 401
 402 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
 403                            struct kvm_memory_slot *dont)
 404 {
 405         kvmppc_core_free_memslot(free, dont);
 406 }
 407
 408 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
 409 {
 410         return kvmppc_core_create_memslot(slot, npages);
 411 }
 412
 413 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 414                                    struct kvm_memory_slot *memslot,
 415                                    struct kvm_memory_slot old,
 416                                    struct kvm_userspace_memory_region *mem,
 417                                    bool user_alloc)
 418 {
 419         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 420 }
 421
 422 void kvm_arch_commit_memory_region(struct kvm *kvm,
 423                struct kvm_userspace_memory_region *mem,
 424                struct kvm_memory_slot old,
 425                bool user_alloc)
 426 {
 427         kvmppc_core_commit_memory_region(kvm, mem, old);
 428 }
 429
 430 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 431 {
 432 }
 433
 434 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 435                                    struct kvm_memory_slot *slot)
 436 {
 437         kvmppc_core_flush_memslot(kvm, slot);
 438 }
 439
 440 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 441 {
 442         struct kvm_vcpu *vcpu;
 443         vcpu = kvmppc_core_vcpu_create(kvm, id);
 444         if (!IS_ERR(vcpu)) {
 445                 vcpu->arch.wqp = &vcpu->wq;
 446                 kvmppc_create_vcpu_debugfs(vcpu, id);
 447         }
 448         return vcpu;
 449 }
 450
 451 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 452 {
 453         return 0;
 454 }
 455
 456 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 457 {
 458         /* Make sure we're not using the vcpu anymore */
 459         hrtimer_cancel(&vcpu->arch.dec_timer);
 460         tasklet_kill(&vcpu->arch.tasklet);
 461
 462         kvmppc_remove_vcpu_debugfs(vcpu);
 463         kvmppc_core_vcpu_free(vcpu);
 464 }
 465
 466 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 467 {
 468         kvm_arch_vcpu_free(vcpu);
 469 }
 470
 471 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 472 {
 473         return kvmppc_core_pending_dec(vcpu);
 474 }
 475
 476 /*
 477  * low level hrtimer wake routine. Because this runs in hardirq context
 478  * we schedule a tasklet to do the real work.
 479  */
 480 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 481 {
 482         struct kvm_vcpu *vcpu;
 483
 484         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 485         tasklet_schedule(&vcpu->arch.tasklet);
 486
 487         return HRTIMER_NORESTART;
 488 }
 489
 490 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 491 {
 492         int ret;
 493
 494         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 495         tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
 496         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 497         vcpu->arch.dec_expires = ~(u64)0;
 498
 499 #ifdef CONFIG_KVM_EXIT_TIMING
 500         mutex_init(&vcpu->arch.exit_timing_lock);
 501 #endif
 502         ret = kvmppc_subarch_vcpu_init(vcpu);
 503         return ret;
 504 }
 505
 506 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 507 {
 508         kvmppc_mmu_destroy(vcpu);
 509         kvmppc_subarch_vcpu_uninit(vcpu);
 510 }
 511
 512 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 513 {
 514 #ifdef CONFIG_BOOKE
 515         /*
 516          * vrsave (formerly usprg0) isn't used by Linux, but may
 517          * be used by the guest.
 518          *
 519          * On non-booke this is associated with Altivec and
 520          * is handled by code in book3s.c.
 521          */
 522         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 523 #endif
 524         kvmppc_core_vcpu_load(vcpu, cpu);
 525 }
 526
 527 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 528 {
 529         kvmppc_core_vcpu_put(vcpu);
 530 #ifdef CONFIG_BOOKE
 531         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 532 #endif
 533 }
 534
 535 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 536                                         struct kvm_guest_debug *dbg)
 537 {
 538         return -EINVAL;
 539 }
 540
 541 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
 542                                      struct kvm_run *run)
 543 {
 544         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
 545 }
 546
 547 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 548                                       struct kvm_run *run)
 549 {
 550         u64 uninitialized_var(gpr);
 551
 552         if (run->mmio.len > sizeof(gpr)) {
 553                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 554                 return;
 555         }
 556
 557         if (vcpu->arch.mmio_is_bigendian) {
 558                 switch (run->mmio.len) {
 559                 case 8: gpr = *(u64 *)run->mmio.data; break;
 560                 case 4: gpr = *(u32 *)run->mmio.data; break;
 561                 case 2: gpr = *(u16 *)run->mmio.data; break;
 562                 case 1: gpr = *(u8 *)run->mmio.data; break;
 563                 }
 564         } else {
 565                 /* Convert BE data from userland back to LE. */
 566                 switch (run->mmio.len) {
 567                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
 568                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
 569                 case 1: gpr = *(u8 *)run->mmio.data; break;
 570                 }
 571         }
 572
 573         if (vcpu->arch.mmio_sign_extend) {
 574                 switch (run->mmio.len) {
 575 #ifdef CONFIG_PPC64
 576                 case 4:
 577                         gpr = (s64)(s32)gpr;
 578                         break;
 579 #endif
 580                 case 2:
 581                         gpr = (s64)(s16)gpr;
 582                         break;
 583                 case 1:
 584                         gpr = (s64)(s8)gpr;
 585                         break;
 586                 }
 587         }
 588
 589         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 590
 591         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 592         case KVM_MMIO_REG_GPR:
 593                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 594                 break;
 595         case KVM_MMIO_REG_FPR:
 596                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 597                 break;
 598 #ifdef CONFIG_PPC_BOOK3S
 599         case KVM_MMIO_REG_QPR:
 600                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 601                 break;
 602         case KVM_MMIO_REG_FQPR:
 603                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 604                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 605                 break;
 606 #endif
 607         default:
 608                 BUG();
 609         }
 610 }
 611
 612 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 613                        unsigned int rt, unsigned int bytes, int is_bigendian)
 614 {
 615         if (bytes > sizeof(run->mmio.data)) {
 616                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 617                        run->mmio.len);
 618         }
 619
 620         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 621         run->mmio.len = bytes;
 622         run->mmio.is_write = 0;
 623
 624         vcpu->arch.io_gpr = rt;
 625         vcpu->arch.mmio_is_bigendian = is_bigendian;
 626         vcpu->mmio_needed = 1;
 627         vcpu->mmio_is_write = 0;
 628         vcpu->arch.mmio_sign_extend = 0;
 629
 630         if (!kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 631                              bytes, &run->mmio.data)) {
 632                 kvmppc_complete_mmio_load(vcpu, run);
 633                 vcpu->mmio_needed = 0;
 634                 return EMULATE_DONE;
 635         }
 636
 637         return EMULATE_DO_MMIO;
 638 }
 639
 640 /* Same as above, but sign extends */
 641 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 642                         unsigned int rt, unsigned int bytes, int is_bigendian)
 643 {
 644         int r;
 645
 646         vcpu->arch.mmio_sign_extend = 1;
 647         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
 648
 649         return r;
 650 }
 651
 652 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 653                         u64 val, unsigned int bytes, int is_bigendian)
 654 {
 655         void *data = run->mmio.data;
 656
 657         if (bytes > sizeof(run->mmio.data)) {
 658                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 659                        run->mmio.len);
 660         }
 661
 662         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 663         run->mmio.len = bytes;
 664         run->mmio.is_write = 1;
 665         vcpu->mmio_needed = 1;
 666         vcpu->mmio_is_write = 1;
 667
 668         /* Store the value at the lowest bytes in 'data'. */
 669         if (is_bigendian) {
 670                 switch (bytes) {
 671                 case 8: *(u64 *)data = val; break;
 672                 case 4: *(u32 *)data = val; break;
 673                 case 2: *(u16 *)data = val; break;
 674                 case 1: *(u8  *)data = val; break;
 675                 }
 676         } else {
 677                 /* Store LE value into 'data'. */
 678                 switch (bytes) {
 679                 case 4: st_le32(data, val); break;
 680                 case 2: st_le16(data, val); break;
 681                 case 1: *(u8 *)data = val; break;
 682                 }
 683         }
 684
 685         if (!kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 686                               bytes, &run->mmio.data)) {
 687                 kvmppc_complete_mmio_load(vcpu, run);
 688                 vcpu->mmio_needed = 0;
 689                 return EMULATE_DONE;
 690         }
 691
 692         return EMULATE_DO_MMIO;
 693 }
 694
 695 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 696 {
 697         int r;
 698         sigset_t sigsaved;
 699
 700         if (vcpu->sigset_active)
 701                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 702
 703         if (vcpu->mmio_needed) {
 704                 if (!vcpu->mmio_is_write)
 705                         kvmppc_complete_mmio_load(vcpu, run);
 706                 vcpu->mmio_needed = 0;
 707         } else if (vcpu->arch.dcr_needed) {
 708                 if (!vcpu->arch.dcr_is_write)
 709                         kvmppc_complete_dcr_load(vcpu, run);
 710                 vcpu->arch.dcr_needed = 0;
 711         } else if (vcpu->arch.osi_needed) {
 712                 u64 *gprs = run->osi.gprs;
 713                 int i;
 714
 715                 for (i = 0; i < 32; i++)
 716                         kvmppc_set_gpr(vcpu, i, gprs[i]);
 717                 vcpu->arch.osi_needed = 0;
 718         } else if (vcpu->arch.hcall_needed) {
 719                 int i;
 720
 721                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
 722                 for (i = 0; i < 9; ++i)
 723                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
 724                 vcpu->arch.hcall_needed = 0;
 725 #ifdef CONFIG_BOOKE
 726         } else if (vcpu->arch.epr_needed) {
 727                 kvmppc_set_epr(vcpu, run->epr.epr);
 728                 vcpu->arch.epr_needed = 0;
 729 #endif
 730         }
 731
 732         r = kvmppc_vcpu_run(run, vcpu);
 733
 734         if (vcpu->sigset_active)
 735                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 736
 737         return r;
 738 }
 739
 740 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
 741 {
 742         if (irq->irq == KVM_INTERRUPT_UNSET) {
 743                 kvmppc_core_dequeue_external(vcpu, irq);
 744                 return 0;
 745         }
 746
 747         kvmppc_core_queue_external(vcpu, irq);
 748
 749         kvm_vcpu_kick(vcpu);
 750
 751         return 0;
 752 }
 753
 754 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
 755                                      struct kvm_enable_cap *cap)
 756 {
 757         int r;
 758
 759         if (cap->flags)
 760                 return -EINVAL;
 761
 762         switch (cap->cap) {
 763         case KVM_CAP_PPC_OSI:
 764                 r = 0;
 765                 vcpu->arch.osi_enabled = true;
 766                 break;
 767         case KVM_CAP_PPC_PAPR:
 768                 r = 0;
 769                 vcpu->arch.papr_enabled = true;
 770                 break;
 771         case KVM_CAP_PPC_EPR:
 772                 r = 0;
 773                 vcpu->arch.epr_enabled = cap->args[0];
 774                 break;
 775 #ifdef CONFIG_BOOKE
 776         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 777                 r = 0;
 778                 vcpu->arch.watchdog_enabled = true;
 779                 break;
 780 #endif
 781 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 782         case KVM_CAP_SW_TLB: {
 783                 struct kvm_config_tlb cfg;
 784                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
 785
 786                 r = -EFAULT;
 787                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
 788                         break;
 789
 790                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
 791                 break;
 792         }
 793 #endif
 794         default:
 795                 r = -EINVAL;
 796                 break;
 797         }
 798
 799         if (!r)
 800                 r = kvmppc_sanity_check(vcpu);
 801
 802         return r;
 803 }
 804
 805 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 806                                     struct kvm_mp_state *mp_state)
 807 {
 808         return -EINVAL;
 809 }
 810
 811 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 812                                     struct kvm_mp_state *mp_state)
 813 {
 814         return -EINVAL;
 815 }
 816
 817 long kvm_arch_vcpu_ioctl(struct file *filp,
 818                          unsigned int ioctl, unsigned long arg)
 819 {
 820         struct kvm_vcpu *vcpu = filp->private_data;
 821         void __user *argp = (void __user *)arg;
 822         long r;
 823
 824         switch (ioctl) {
 825         case KVM_INTERRUPT: {
 826                 struct kvm_interrupt irq;
 827                 r = -EFAULT;
 828                 if (copy_from_user(&irq, argp, sizeof(irq)))
 829                         goto out;
 830                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 831                 goto out;
 832         }
 833
 834         case KVM_ENABLE_CAP:
 835         {
 836                 struct kvm_enable_cap cap;
 837                 r = -EFAULT;
 838                 if (copy_from_user(&cap, argp, sizeof(cap)))
 839                         goto out;
 840                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
 841                 break;
 842         }
 843
 844         case KVM_SET_ONE_REG:
 845         case KVM_GET_ONE_REG:
 846         {
 847                 struct kvm_one_reg reg;
 848                 r = -EFAULT;
 849                 if (copy_from_user(&reg, argp, sizeof(reg)))
 850                         goto out;
 851                 if (ioctl == KVM_SET_ONE_REG)
 852                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
 853                 else
 854                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
 855                 break;
 856         }
 857
 858 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 859         case KVM_DIRTY_TLB: {
 860                 struct kvm_dirty_tlb dirty;
 861                 r = -EFAULT;
 862                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
 863                         goto out;
 864                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
 865                 break;
 866         }
 867 #endif
 868         default:
 869                 r = -EINVAL;
 870         }
 871
 872 out:
 873         return r;
 874 }
 875
 876 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 877 {
 878         return VM_FAULT_SIGBUS;
 879 }
 880
 881 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
 882 {
 883         u32 inst_nop = 0x60000000;
 884 #ifdef CONFIG_KVM_BOOKE_HV
 885         u32 inst_sc1 = 0x44000022;
 886         pvinfo->hcall[0] = inst_sc1;
 887         pvinfo->hcall[1] = inst_nop;
 888         pvinfo->hcall[2] = inst_nop;
 889         pvinfo->hcall[3] = inst_nop;
 890 #else
 891         u32 inst_lis = 0x3c000000;
 892         u32 inst_ori = 0x60000000;
 893         u32 inst_sc = 0x44000002;
 894         u32 inst_imm_mask = 0xffff;
 895
 896         /*
 897          * The hypercall to get into KVM from within guest context is as
 898          * follows:
 899          *
 900          *    lis r0, r0, KVM_SC_MAGIC_R0@h
 901          *    ori r0, KVM_SC_MAGIC_R0@l
 902          *    sc
 903          *    nop
 904          */
 905         pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
 906         pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
 907         pvinfo->hcall[2] = inst_sc;
 908         pvinfo->hcall[3] = inst_nop;
 909 #endif
 910
 911         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
 912
 913         return 0;
 914 }
 915
 916 long kvm_arch_vm_ioctl(struct file *filp,
 917                        unsigned int ioctl, unsigned long arg)
 918 {
 919         void __user *argp = (void __user *)arg;
 920         long r;
 921
 922         switch (ioctl) {
 923         case KVM_PPC_GET_PVINFO: {
 924                 struct kvm_ppc_pvinfo pvinfo;
 925                 memset(&pvinfo, 0, sizeof(pvinfo));
 926                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
 927                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
 928                         r = -EFAULT;
 929                         goto out;
 930                 }
 931
 932                 break;
 933         }
 934 #ifdef CONFIG_PPC_BOOK3S_64
 935         case KVM_CREATE_SPAPR_TCE: {
 936                 struct kvm_create_spapr_tce create_tce;
 937                 struct kvm *kvm = filp->private_data;
 938
 939                 r = -EFAULT;
 940                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
 941                         goto out;
 942                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
 943                 goto out;
 944         }
 945 #endif /* CONFIG_PPC_BOOK3S_64 */
 946
 947 #ifdef CONFIG_KVM_BOOK3S_64_HV
 948         case KVM_ALLOCATE_RMA: {
 949                 struct kvm *kvm = filp->private_data;
 950                 struct kvm_allocate_rma rma;
 951
 952                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
 953                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
 954                         r = -EFAULT;
 955                 break;
 956         }
 957
 958         case KVM_PPC_ALLOCATE_HTAB: {
 959                 struct kvm *kvm = filp->private_data;
 960                 u32 htab_order;
 961
 962                 r = -EFAULT;
 963                 if (get_user(htab_order, (u32 __user *)argp))
 964                         break;
 965                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
 966                 if (r)
 967                         break;
 968                 r = -EFAULT;
 969                 if (put_user(htab_order, (u32 __user *)argp))
 970                         break;
 971                 r = 0;
 972                 break;
 973         }
 974
 975         case KVM_PPC_GET_HTAB_FD: {
 976                 struct kvm *kvm = filp->private_data;
 977                 struct kvm_get_htab_fd ghf;
 978
 979                 r = -EFAULT;
 980                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
 981                         break;
 982                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
 983                 break;
 984         }
 985 #endif /* CONFIG_KVM_BOOK3S_64_HV */
 986
 987 #ifdef CONFIG_PPC_BOOK3S_64
 988         case KVM_PPC_GET_SMMU_INFO: {
 989                 struct kvm *kvm = filp->private_data;
 990                 struct kvm_ppc_smmu_info info;
 991
 992                 memset(&info, 0, sizeof(info));
 993                 r = kvm_vm_ioctl_get_smmu_info(kvm, &info);
 994                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
 995                         r = -EFAULT;
 996                 break;
 997         }
 998 #endif /* CONFIG_PPC_BOOK3S_64 */
 999         default:
1000                 r = -ENOTTY;
1001         }
1002
1003 out:
1004         return r;
1005 }
1006
1007 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1008 static unsigned long nr_lpids;
1009
1010 long kvmppc_alloc_lpid(void)
1011 {
1012         long lpid;
1013
1014         do {
1015                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1016                 if (lpid >= nr_lpids) {
1017                         pr_err("%s: No LPIDs free\n", __func__);
1018                         return -ENOMEM;
1019                 }
1020         } while (test_and_set_bit(lpid, lpid_inuse));
1021
1022         return lpid;
1023 }
1024
1025 void kvmppc_claim_lpid(long lpid)
1026 {
1027         set_bit(lpid, lpid_inuse);
1028 }
1029
1030 void kvmppc_free_lpid(long lpid)
1031 {
1032         clear_bit(lpid, lpid_inuse);
1033 }
1034
1035 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1036 {
1037         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1038         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1039 }
1040
1041 int kvm_arch_init(void *opaque)
1042 {
1043         return 0;
1044 }
1045
1046 void kvm_arch_exit(void)
1047 {
1048 }