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 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  43 {
  44         return !!(v->arch.pending_exceptions) ||
  45                v->requests;
  46 }
  47
  48 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  49 {
  50         return 1;
  51 }
  52
  53 #ifndef CONFIG_KVM_BOOK3S_64_HV
  54 /*
  55  * Common checks before entering the guest world.  Call with interrupts
  56  * disabled.
  57  *
  58  * returns:
  59  *
  60  * == 1 if we're ready to go into guest state
  61  * <= 0 if we need to go back to the host with return value
  62  */
  63 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
  64 {
  65         int r = 1;
  66
  67         WARN_ON_ONCE(!irqs_disabled());
  68         while (true) {
  69                 if (need_resched()) {
  70                         local_irq_enable();
  71                         cond_resched();
  72                         local_irq_disable();
  73                         continue;
  74                 }
  75
  76                 if (signal_pending(current)) {
  77                         kvmppc_account_exit(vcpu, SIGNAL_EXITS);
  78                         vcpu->run->exit_reason = KVM_EXIT_INTR;
  79                         r = -EINTR;
  80                         break;
  81                 }
  82
  83                 vcpu->mode = IN_GUEST_MODE;
  84
  85                 /*
  86                  * Reading vcpu->requests must happen after setting vcpu->mode,
  87                  * so we don't miss a request because the requester sees
  88                  * OUTSIDE_GUEST_MODE and assumes we'll be checking requests
  89                  * before next entering the guest (and thus doesn't IPI).
  90                  */
  91                 smp_mb();
  92
  93                 if (vcpu->requests) {
  94                         /* Make sure we process requests preemptable */
  95                         local_irq_enable();
  96                         trace_kvm_check_requests(vcpu);
  97                         r = kvmppc_core_check_requests(vcpu);
  98                         local_irq_disable();
  99                         if (r > 0)
 100                                 continue;
 101                         break;
 102                 }
 103
 104                 if (kvmppc_core_prepare_to_enter(vcpu)) {
 105                         /* interrupts got enabled in between, so we
 106                            are back at square 1 */
 107                         continue;
 108                 }
 109
 110 #ifdef CONFIG_PPC64
 111                 /* lazy EE magic */
 112                 hard_irq_disable();
 113                 if (lazy_irq_pending()) {
 114                         /* Got an interrupt in between, try again */
 115                         local_irq_enable();
 116                         local_irq_disable();
 117                         kvm_guest_exit();
 118                         continue;
 119                 }
 120
 121                 trace_hardirqs_on();
 122 #endif
 123
 124                 kvm_guest_enter();
 125                 break;
 126         }
 127
 128         return r;
 129 }
 130 #endif /* CONFIG_KVM_BOOK3S_64_HV */
 131
 132 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu)
 133 {
 134         int nr = kvmppc_get_gpr(vcpu, 11);
 135         int r;
 136         unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3);
 137         unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4);
 138         unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5);
 139         unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6);
 140         unsigned long r2 = 0;
 141
 142         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 143                 /* 32 bit mode */
 144                 param1 &= 0xffffffff;
 145                 param2 &= 0xffffffff;
 146                 param3 &= 0xffffffff;
 147                 param4 &= 0xffffffff;
 148         }
 149
 150         switch (nr) {
 151         case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE):
 152         {
 153                 vcpu->arch.magic_page_pa = param1;
 154                 vcpu->arch.magic_page_ea = param2;
 155
 156                 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7;
 157
 158                 r = EV_SUCCESS;
 159                 break;
 160         }
 161         case KVM_HCALL_TOKEN(KVM_HC_FEATURES):
 162                 r = EV_SUCCESS;
 163 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2)
 164                 /* XXX Missing magic page on 44x */
 165                 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE);
 166 #endif
 167
 168                 /* Second return value is in r4 */
 169                 break;
 170         case EV_HCALL_TOKEN(EV_IDLE):
 171                 r = EV_SUCCESS;
 172                 kvm_vcpu_block(vcpu);
 173                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 174                 break;
 175         default:
 176                 r = EV_UNIMPLEMENTED;
 177                 break;
 178         }
 179
 180         kvmppc_set_gpr(vcpu, 4, r2);
 181
 182         return r;
 183 }
 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 #ifdef CONFIG_KVM_BOOK3S_64_HV
 198         /* HV KVM can only do PAPR mode for now */
 199         if (!vcpu->arch.papr_enabled)
 200                 goto out;
 201 #endif
 202
 203 #ifdef CONFIG_KVM_BOOKE_HV
 204         if (!cpu_has_feature(CPU_FTR_EMB_HV))
 205                 goto out;
 206 #endif
 207
 208         r = true;
 209
 210 out:
 211         vcpu->arch.sane = r;
 212         return r ? 0 : -EINVAL;
 213 }
 214
 215 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu)
 216 {
 217         enum emulation_result er;
 218         int r;
 219
 220         er = kvmppc_emulate_instruction(run, vcpu);
 221         switch (er) {
 222         case EMULATE_DONE:
 223                 /* Future optimization: only reload non-volatiles if they were
 224                  * actually modified. */
 225                 r = RESUME_GUEST_NV;
 226                 break;
 227         case EMULATE_DO_MMIO:
 228                 run->exit_reason = KVM_EXIT_MMIO;
 229                 /* We must reload nonvolatiles because "update" load/store
 230                  * instructions modify register state. */
 231                 /* Future optimization: only reload non-volatiles if they were
 232                  * actually modified. */
 233                 r = RESUME_HOST_NV;
 234                 break;
 235         case EMULATE_FAIL:
 236                 /* XXX Deliver Program interrupt to guest. */
 237                 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__,
 238                        kvmppc_get_last_inst(vcpu));
 239                 r = RESUME_HOST;
 240                 break;
 241         default:
 242                 WARN_ON(1);
 243                 r = RESUME_GUEST;
 244         }
 245
 246         return r;
 247 }
 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 #ifndef CONFIG_KVM_BOOK3S_64_HV
 326         case KVM_CAP_PPC_PAIRED_SINGLES:
 327         case KVM_CAP_PPC_OSI:
 328         case KVM_CAP_PPC_GET_PVINFO:
 329 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 330         case KVM_CAP_SW_TLB:
 331 #endif
 332 #ifdef CONFIG_KVM_MPIC
 333         case KVM_CAP_IRQ_MPIC:
 334 #endif
 335                 r = 1;
 336                 break;
 337         case KVM_CAP_COALESCED_MMIO:
 338                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 339                 break;
 340 #endif
 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_64_HV
 352         case KVM_CAP_PPC_SMT:
 353                 r = threads_per_core;
 354                 break;
 355         case KVM_CAP_PPC_RMA:
 356                 r = 1;
 357                 /* PPC970 requires an RMA */
 358                 if (cpu_has_feature(CPU_FTR_ARCH_201))
 359                         r = 2;
 360                 break;
 361 #endif
 362         case KVM_CAP_SYNC_MMU:
 363 #ifdef CONFIG_KVM_BOOK3S_64_HV
 364                 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
 365 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 366                 r = 1;
 367 #else
 368                 r = 0;
 369                 break;
 370 #endif
 371 #ifdef CONFIG_KVM_BOOK3S_64_HV
 372         case KVM_CAP_PPC_HTAB_FD:
 373                 r = 1;
 374                 break;
 375 #endif
 376                 break;
 377         case KVM_CAP_NR_VCPUS:
 378                 /*
 379                  * Recommending a number of CPUs is somewhat arbitrary; we
 380                  * return the number of present CPUs for -HV (since a host
 381                  * will have secondary threads "offline"), and for other KVM
 382                  * implementations just count online CPUs.
 383                  */
 384 #ifdef CONFIG_KVM_BOOK3S_64_HV
 385                 r = num_present_cpus();
 386 #else
 387                 r = num_online_cpus();
 388 #endif
 389                 break;
 390         case KVM_CAP_MAX_VCPUS:
 391                 r = KVM_MAX_VCPUS;
 392                 break;
 393 #ifdef CONFIG_PPC_BOOK3S_64
 394         case KVM_CAP_PPC_GET_SMMU_INFO:
 395                 r = 1;
 396                 break;
 397 #endif
 398         default:
 399                 r = 0;
 400                 break;
 401         }
 402         return r;
 403
 404 }
 405
 406 long kvm_arch_dev_ioctl(struct file *filp,
 407                         unsigned int ioctl, unsigned long arg)
 408 {
 409         return -EINVAL;
 410 }
 411
 412 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 413                            struct kvm_memory_slot *dont)
 414 {
 415         kvmppc_core_free_memslot(kvm, free, dont);
 416 }
 417
 418 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 419                             unsigned long npages)
 420 {
 421         return kvmppc_core_create_memslot(kvm, slot, npages);
 422 }
 423
 424 void kvm_arch_memslots_updated(struct kvm *kvm)
 425 {
 426 }
 427
 428 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 429                                    struct kvm_memory_slot *memslot,
 430                                    struct kvm_userspace_memory_region *mem,
 431                                    enum kvm_mr_change change)
 432 {
 433         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 434 }
 435
 436 void kvm_arch_commit_memory_region(struct kvm *kvm,
 437                                    struct kvm_userspace_memory_region *mem,
 438                                    const struct kvm_memory_slot *old,
 439                                    enum kvm_mr_change change)
 440 {
 441         kvmppc_core_commit_memory_region(kvm, mem, old);
 442 }
 443
 444 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 445 {
 446 }
 447
 448 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 449                                    struct kvm_memory_slot *slot)
 450 {
 451         kvmppc_core_flush_memslot(kvm, slot);
 452 }
 453
 454 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 455 {
 456         struct kvm_vcpu *vcpu;
 457         vcpu = kvmppc_core_vcpu_create(kvm, id);
 458         if (!IS_ERR(vcpu)) {
 459                 vcpu->arch.wqp = &vcpu->wq;
 460                 kvmppc_create_vcpu_debugfs(vcpu, id);
 461         }
 462         return vcpu;
 463 }
 464
 465 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 466 {
 467         return 0;
 468 }
 469
 470 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 471 {
 472         /* Make sure we're not using the vcpu anymore */
 473         hrtimer_cancel(&vcpu->arch.dec_timer);
 474         tasklet_kill(&vcpu->arch.tasklet);
 475
 476         kvmppc_remove_vcpu_debugfs(vcpu);
 477
 478         switch (vcpu->arch.irq_type) {
 479         case KVMPPC_IRQ_MPIC:
 480                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 481                 break;
 482         case KVMPPC_IRQ_XICS:
 483                 kvmppc_xics_free_icp(vcpu);
 484                 break;
 485         }
 486
 487         kvmppc_core_vcpu_free(vcpu);
 488 }
 489
 490 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 491 {
 492         kvm_arch_vcpu_free(vcpu);
 493 }
 494
 495 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 496 {
 497         return kvmppc_core_pending_dec(vcpu);
 498 }
 499
 500 /*
 501  * low level hrtimer wake routine. Because this runs in hardirq context
 502  * we schedule a tasklet to do the real work.
 503  */
 504 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 505 {
 506         struct kvm_vcpu *vcpu;
 507
 508         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 509         tasklet_schedule(&vcpu->arch.tasklet);
 510
 511         return HRTIMER_NORESTART;
 512 }
 513
 514 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 515 {
 516         int ret;
 517
 518         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 519         tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
 520         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 521         vcpu->arch.dec_expires = ~(u64)0;
 522
 523 #ifdef CONFIG_KVM_EXIT_TIMING
 524         mutex_init(&vcpu->arch.exit_timing_lock);
 525 #endif
 526         ret = kvmppc_subarch_vcpu_init(vcpu);
 527         return ret;
 528 }
 529
 530 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 531 {
 532         kvmppc_mmu_destroy(vcpu);
 533         kvmppc_subarch_vcpu_uninit(vcpu);
 534 }
 535
 536 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 537 {
 538 #ifdef CONFIG_BOOKE
 539         /*
 540          * vrsave (formerly usprg0) isn't used by Linux, but may
 541          * be used by the guest.
 542          *
 543          * On non-booke this is associated with Altivec and
 544          * is handled by code in book3s.c.
 545          */
 546         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 547 #endif
 548         kvmppc_core_vcpu_load(vcpu, cpu);
 549 }
 550
 551 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 552 {
 553         kvmppc_core_vcpu_put(vcpu);
 554 #ifdef CONFIG_BOOKE
 555         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 556 #endif
 557 }
 558
 559 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
 560                                      struct kvm_run *run)
 561 {
 562         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
 563 }
 564
 565 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 566                                       struct kvm_run *run)
 567 {
 568         u64 uninitialized_var(gpr);
 569
 570         if (run->mmio.len > sizeof(gpr)) {
 571                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 572                 return;
 573         }
 574
 575         if (vcpu->arch.mmio_is_bigendian) {
 576                 switch (run->mmio.len) {
 577                 case 8: gpr = *(u64 *)run->mmio.data; break;
 578                 case 4: gpr = *(u32 *)run->mmio.data; break;
 579                 case 2: gpr = *(u16 *)run->mmio.data; break;
 580                 case 1: gpr = *(u8 *)run->mmio.data; break;
 581                 }
 582         } else {
 583                 /* Convert BE data from userland back to LE. */
 584                 switch (run->mmio.len) {
 585                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
 586                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
 587                 case 1: gpr = *(u8 *)run->mmio.data; break;
 588                 }
 589         }
 590
 591         if (vcpu->arch.mmio_sign_extend) {
 592                 switch (run->mmio.len) {
 593 #ifdef CONFIG_PPC64
 594                 case 4:
 595                         gpr = (s64)(s32)gpr;
 596                         break;
 597 #endif
 598                 case 2:
 599                         gpr = (s64)(s16)gpr;
 600                         break;
 601                 case 1:
 602                         gpr = (s64)(s8)gpr;
 603                         break;
 604                 }
 605         }
 606
 607         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 608
 609         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 610         case KVM_MMIO_REG_GPR:
 611                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 612                 break;
 613         case KVM_MMIO_REG_FPR:
 614                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 615                 break;
 616 #ifdef CONFIG_PPC_BOOK3S
 617         case KVM_MMIO_REG_QPR:
 618                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 619                 break;
 620         case KVM_MMIO_REG_FQPR:
 621                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 622                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 623                 break;
 624 #endif
 625         default:
 626                 BUG();
 627         }
 628 }
 629
 630 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 631                        unsigned int rt, unsigned int bytes, int is_bigendian)
 632 {
 633         int idx, ret;
 634
 635         if (bytes > sizeof(run->mmio.data)) {
 636                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 637                        run->mmio.len);
 638         }
 639
 640         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 641         run->mmio.len = bytes;
 642         run->mmio.is_write = 0;
 643
 644         vcpu->arch.io_gpr = rt;
 645         vcpu->arch.mmio_is_bigendian = is_bigendian;
 646         vcpu->mmio_needed = 1;
 647         vcpu->mmio_is_write = 0;
 648         vcpu->arch.mmio_sign_extend = 0;
 649
 650         idx = srcu_read_lock(&vcpu->kvm->srcu);
 651
 652         ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 653                               bytes, &run->mmio.data);
 654
 655         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 656
 657         if (!ret) {
 658                 kvmppc_complete_mmio_load(vcpu, run);
 659                 vcpu->mmio_needed = 0;
 660                 return EMULATE_DONE;
 661         }
 662
 663         return EMULATE_DO_MMIO;
 664 }
 665
 666 /* Same as above, but sign extends */
 667 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 668                         unsigned int rt, unsigned int bytes, int is_bigendian)
 669 {
 670         int r;
 671
 672         vcpu->arch.mmio_sign_extend = 1;
 673         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
 674
 675         return r;
 676 }
 677
 678 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 679                         u64 val, unsigned int bytes, int is_bigendian)
 680 {
 681         void *data = run->mmio.data;
 682         int idx, ret;
 683
 684         if (bytes > sizeof(run->mmio.data)) {
 685                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 686                        run->mmio.len);
 687         }
 688
 689         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 690         run->mmio.len = bytes;
 691         run->mmio.is_write = 1;
 692         vcpu->mmio_needed = 1;
 693         vcpu->mmio_is_write = 1;
 694
 695         /* Store the value at the lowest bytes in 'data'. */
 696         if (is_bigendian) {
 697                 switch (bytes) {
 698                 case 8: *(u64 *)data = val; break;
 699                 case 4: *(u32 *)data = val; break;
 700                 case 2: *(u16 *)data = val; break;
 701                 case 1: *(u8  *)data = val; break;
 702                 }
 703         } else {
 704                 /* Store LE value into 'data'. */
 705                 switch (bytes) {
 706                 case 4: st_le32(data, val); break;
 707                 case 2: st_le16(data, val); break;
 708                 case 1: *(u8 *)data = val; break;
 709                 }
 710         }
 711
 712         idx = srcu_read_lock(&vcpu->kvm->srcu);
 713
 714         ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 715                                bytes, &run->mmio.data);
 716
 717         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 718
 719         if (!ret) {
 720                 vcpu->mmio_needed = 0;
 721                 return EMULATE_DONE;
 722         }
 723
 724         return EMULATE_DO_MMIO;
 725 }
 726
 727 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 728 {
 729         int r;
 730         sigset_t sigsaved;
 731
 732         if (vcpu->sigset_active)
 733                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 734
 735         if (vcpu->mmio_needed) {
 736                 if (!vcpu->mmio_is_write)
 737                         kvmppc_complete_mmio_load(vcpu, run);
 738                 vcpu->mmio_needed = 0;
 739         } else if (vcpu->arch.dcr_needed) {
 740                 if (!vcpu->arch.dcr_is_write)
 741                         kvmppc_complete_dcr_load(vcpu, run);
 742                 vcpu->arch.dcr_needed = 0;
 743         } else if (vcpu->arch.osi_needed) {
 744                 u64 *gprs = run->osi.gprs;
 745                 int i;
 746
 747                 for (i = 0; i < 32; i++)
 748                         kvmppc_set_gpr(vcpu, i, gprs[i]);
 749                 vcpu->arch.osi_needed = 0;
 750         } else if (vcpu->arch.hcall_needed) {
 751                 int i;
 752
 753                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
 754                 for (i = 0; i < 9; ++i)
 755                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
 756                 vcpu->arch.hcall_needed = 0;
 757 #ifdef CONFIG_BOOKE
 758         } else if (vcpu->arch.epr_needed) {
 759                 kvmppc_set_epr(vcpu, run->epr.epr);
 760                 vcpu->arch.epr_needed = 0;
 761 #endif
 762         }
 763
 764         r = kvmppc_vcpu_run(run, vcpu);
 765
 766         if (vcpu->sigset_active)
 767                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 768
 769         return r;
 770 }
 771
 772 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
 773 {
 774         if (irq->irq == KVM_INTERRUPT_UNSET) {
 775                 kvmppc_core_dequeue_external(vcpu);
 776                 return 0;
 777         }
 778
 779         kvmppc_core_queue_external(vcpu, irq);
 780
 781         kvm_vcpu_kick(vcpu);
 782
 783         return 0;
 784 }
 785
 786 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
 787                                      struct kvm_enable_cap *cap)
 788 {
 789         int r;
 790
 791         if (cap->flags)
 792                 return -EINVAL;
 793
 794         switch (cap->cap) {
 795         case KVM_CAP_PPC_OSI:
 796                 r = 0;
 797                 vcpu->arch.osi_enabled = true;
 798                 break;
 799         case KVM_CAP_PPC_PAPR:
 800                 r = 0;
 801                 vcpu->arch.papr_enabled = true;
 802                 break;
 803         case KVM_CAP_PPC_EPR:
 804                 r = 0;
 805                 if (cap->args[0])
 806                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
 807                 else
 808                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
 809                 break;
 810 #ifdef CONFIG_BOOKE
 811         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 812                 r = 0;
 813                 vcpu->arch.watchdog_enabled = true;
 814                 break;
 815 #endif
 816 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 817         case KVM_CAP_SW_TLB: {
 818                 struct kvm_config_tlb cfg;
 819                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
 820
 821                 r = -EFAULT;
 822                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
 823                         break;
 824
 825                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
 826                 break;
 827         }
 828 #endif
 829 #ifdef CONFIG_KVM_MPIC
 830         case KVM_CAP_IRQ_MPIC: {
 831                 struct file *filp;
 832                 struct kvm_device *dev;
 833
 834                 r = -EBADF;
 835                 filp = fget(cap->args[0]);
 836                 if (!filp)
 837                         break;
 838
 839                 r = -EPERM;
 840                 dev = kvm_device_from_filp(filp);
 841                 if (dev)
 842                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
 843
 844                 fput(filp);
 845                 break;
 846         }
 847 #endif
 848 #ifdef CONFIG_KVM_XICS
 849         case KVM_CAP_IRQ_XICS: {
 850                 struct file *filp;
 851                 struct kvm_device *dev;
 852
 853                 r = -EBADF;
 854                 filp = fget(cap->args[0]);
 855                 if (!filp)
 856                         break;
 857
 858                 r = -EPERM;
 859                 dev = kvm_device_from_filp(filp);
 860                 if (dev)
 861                         r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
 862
 863                 fput(filp);
 864                 break;
 865         }
 866 #endif /* CONFIG_KVM_XICS */
 867         default:
 868                 r = -EINVAL;
 869                 break;
 870         }
 871
 872         if (!r)
 873                 r = kvmppc_sanity_check(vcpu);
 874
 875         return r;
 876 }
 877
 878 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 879                                     struct kvm_mp_state *mp_state)
 880 {
 881         return -EINVAL;
 882 }
 883
 884 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 885                                     struct kvm_mp_state *mp_state)
 886 {
 887         return -EINVAL;
 888 }
 889
 890 long kvm_arch_vcpu_ioctl(struct file *filp,
 891                          unsigned int ioctl, unsigned long arg)
 892 {
 893         struct kvm_vcpu *vcpu = filp->private_data;
 894         void __user *argp = (void __user *)arg;
 895         long r;
 896
 897         switch (ioctl) {
 898         case KVM_INTERRUPT: {
 899                 struct kvm_interrupt irq;
 900                 r = -EFAULT;
 901                 if (copy_from_user(&irq, argp, sizeof(irq)))
 902                         goto out;
 903                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 904                 goto out;
 905         }
 906
 907         case KVM_ENABLE_CAP:
 908         {
 909                 struct kvm_enable_cap cap;
 910                 r = -EFAULT;
 911                 if (copy_from_user(&cap, argp, sizeof(cap)))
 912                         goto out;
 913                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
 914                 break;
 915         }
 916
 917         case KVM_SET_ONE_REG:
 918         case KVM_GET_ONE_REG:
 919         {
 920                 struct kvm_one_reg reg;
 921                 r = -EFAULT;
 922                 if (copy_from_user(&reg, argp, sizeof(reg)))
 923                         goto out;
 924                 if (ioctl == KVM_SET_ONE_REG)
 925                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
 926                 else
 927                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
 928                 break;
 929         }
 930
 931 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 932         case KVM_DIRTY_TLB: {
 933                 struct kvm_dirty_tlb dirty;
 934                 r = -EFAULT;
 935                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
 936                         goto out;
 937                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
 938                 break;
 939         }
 940 #endif
 941         default:
 942                 r = -EINVAL;
 943         }
 944
 945 out:
 946         return r;
 947 }
 948
 949 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 950 {
 951         return VM_FAULT_SIGBUS;
 952 }
 953
 954 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
 955 {
 956         u32 inst_nop = 0x60000000;
 957 #ifdef CONFIG_KVM_BOOKE_HV
 958         u32 inst_sc1 = 0x44000022;
 959         pvinfo->hcall[0] = inst_sc1;
 960         pvinfo->hcall[1] = inst_nop;
 961         pvinfo->hcall[2] = inst_nop;
 962         pvinfo->hcall[3] = inst_nop;
 963 #else
 964         u32 inst_lis = 0x3c000000;
 965         u32 inst_ori = 0x60000000;
 966         u32 inst_sc = 0x44000002;
 967         u32 inst_imm_mask = 0xffff;
 968
 969         /*
 970          * The hypercall to get into KVM from within guest context is as
 971          * follows:
 972          *
 973          *    lis r0, r0, KVM_SC_MAGIC_R0@h
 974          *    ori r0, KVM_SC_MAGIC_R0@l
 975          *    sc
 976          *    nop
 977          */
 978         pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
 979         pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
 980         pvinfo->hcall[2] = inst_sc;
 981         pvinfo->hcall[3] = inst_nop;
 982 #endif
 983
 984         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
 985
 986         return 0;
 987 }
 988
 989 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
 990                           bool line_status)
 991 {
 992         if (!irqchip_in_kernel(kvm))
 993                 return -ENXIO;
 994
 995         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
 996                                         irq_event->irq, irq_event->level,
 997                                         line_status);
 998         return 0;
 999 }
1000
1001 long kvm_arch_vm_ioctl(struct file *filp,
1002                        unsigned int ioctl, unsigned long arg)
1003 {
1004         struct kvm *kvm __maybe_unused = filp->private_data;
1005         void __user *argp = (void __user *)arg;
1006         long r;
1007
1008         switch (ioctl) {
1009         case KVM_PPC_GET_PVINFO: {
1010                 struct kvm_ppc_pvinfo pvinfo;
1011                 memset(&pvinfo, 0, sizeof(pvinfo));
1012                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1013                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1014                         r = -EFAULT;
1015                         goto out;
1016                 }
1017
1018                 break;
1019         }
1020 #ifdef CONFIG_PPC_BOOK3S_64
1021         case KVM_CREATE_SPAPR_TCE: {
1022                 struct kvm_create_spapr_tce create_tce;
1023
1024                 r = -EFAULT;
1025                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1026                         goto out;
1027                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1028                 goto out;
1029         }
1030 #endif /* CONFIG_PPC_BOOK3S_64 */
1031
1032 #ifdef CONFIG_KVM_BOOK3S_64_HV
1033         case KVM_ALLOCATE_RMA: {
1034                 struct kvm_allocate_rma rma;
1035                 struct kvm *kvm = filp->private_data;
1036
1037                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
1038                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
1039                         r = -EFAULT;
1040                 break;
1041         }
1042
1043         case KVM_PPC_ALLOCATE_HTAB: {
1044                 u32 htab_order;
1045
1046                 r = -EFAULT;
1047                 if (get_user(htab_order, (u32 __user *)argp))
1048                         break;
1049                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
1050                 if (r)
1051                         break;
1052                 r = -EFAULT;
1053                 if (put_user(htab_order, (u32 __user *)argp))
1054                         break;
1055                 r = 0;
1056                 break;
1057         }
1058
1059         case KVM_PPC_GET_HTAB_FD: {
1060                 struct kvm_get_htab_fd ghf;
1061
1062                 r = -EFAULT;
1063                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
1064                         break;
1065                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
1066                 break;
1067         }
1068 #endif /* CONFIG_KVM_BOOK3S_64_HV */
1069
1070 #ifdef CONFIG_PPC_BOOK3S_64
1071         case KVM_PPC_GET_SMMU_INFO: {
1072                 struct kvm_ppc_smmu_info info;
1073
1074                 memset(&info, 0, sizeof(info));
1075                 r = kvm_vm_ioctl_get_smmu_info(kvm, &info);
1076                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1077                         r = -EFAULT;
1078                 break;
1079         }
1080         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1081                 struct kvm *kvm = filp->private_data;
1082
1083                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1084                 break;
1085         }
1086 #endif /* CONFIG_PPC_BOOK3S_64 */
1087         default:
1088                 r = -ENOTTY;
1089         }
1090
1091 out:
1092         return r;
1093 }
1094
1095 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1096 static unsigned long nr_lpids;
1097
1098 long kvmppc_alloc_lpid(void)
1099 {
1100         long lpid;
1101
1102         do {
1103                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1104                 if (lpid >= nr_lpids) {
1105                         pr_err("%s: No LPIDs free\n", __func__);
1106                         return -ENOMEM;
1107                 }
1108         } while (test_and_set_bit(lpid, lpid_inuse));
1109
1110         return lpid;
1111 }
1112
1113 void kvmppc_claim_lpid(long lpid)
1114 {
1115         set_bit(lpid, lpid_inuse);
1116 }
1117
1118 void kvmppc_free_lpid(long lpid)
1119 {
1120         clear_bit(lpid, lpid_inuse);
1121 }
1122
1123 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1124 {
1125         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1126         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1127 }
1128
1129 int kvm_arch_init(void *opaque)
1130 {
1131         return 0;
1132 }
1133
1134 void kvm_arch_exit(void)
1135 {
1136 }