Pileus Git - ~andy/linux/blob - arch/powerpc/kvm/powerpc.c

   1 /*
   2  * This program is free software; you can redistribute it and/or modify
   3  * it under the terms of the GNU General Public License, version 2, as
   4  * published by the Free Software Foundation.
   5  *
   6  * This program is distributed in the hope that it will be useful,
   7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
   8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   9  * GNU General Public License for more details.
  10  *
  11  * You should have received a copy of the GNU General Public License
  12  * along with this program; if not, write to the Free Software
  13  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  14  *
  15  * Copyright IBM Corp. 2007
  16  *
  17  * Authors: Hollis Blanchard <hollisb@us.ibm.com>
  18  *          Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
  19  */
  20
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/hrtimer.h>
  26 #include <linux/fs.h>
  27 #include <linux/slab.h>
  28 #include <asm/cputable.h>
  29 #include <asm/uaccess.h>
  30 #include <asm/kvm_ppc.h>
  31 #include <asm/tlbflush.h>
  32 #include <asm/cputhreads.h>
  33 #include <asm/irqflags.h>
  34 #include "timing.h"
  35 #include "../mm/mmu_decl.h"
  36
  37 #define CREATE_TRACE_POINTS
  38 #include "trace.h"
  39
  40 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
  41 {
  42         return !!(v->arch.pending_exceptions) ||
  43                v->requests;
  44 }
  45
  46 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  47 {
  48         return 1;
  49 }
  50
  51 #ifndef CONFIG_KVM_BOOK3S_64_HV
  52 /*
  53  * Common checks before entering the guest world.  Call with interrupts
  54  * disabled.
  55  *
  56  * returns:
  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         case KVM_CAP_DEVICE_CTRL:
 321                 r = 1;
 322                 break;
 323 #ifndef CONFIG_KVM_BOOK3S_64_HV
 324         case KVM_CAP_PPC_PAIRED_SINGLES:
 325         case KVM_CAP_PPC_OSI:
 326         case KVM_CAP_PPC_GET_PVINFO:
 327 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 328         case KVM_CAP_SW_TLB:
 329 #endif
 330                 r = 1;
 331                 break;
 332         case KVM_CAP_COALESCED_MMIO:
 333                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 334                 break;
 335 #endif
 336 #ifdef CONFIG_PPC_BOOK3S_64
 337         case KVM_CAP_SPAPR_TCE:
 338         case KVM_CAP_PPC_ALLOC_HTAB:
 339                 r = 1;
 340                 break;
 341 #endif /* CONFIG_PPC_BOOK3S_64 */
 342 #ifdef CONFIG_KVM_BOOK3S_64_HV
 343         case KVM_CAP_PPC_SMT:
 344                 r = threads_per_core;
 345                 break;
 346         case KVM_CAP_PPC_RMA:
 347                 r = 1;
 348                 /* PPC970 requires an RMA */
 349                 if (cpu_has_feature(CPU_FTR_ARCH_201))
 350                         r = 2;
 351                 break;
 352 #endif
 353         case KVM_CAP_SYNC_MMU:
 354 #ifdef CONFIG_KVM_BOOK3S_64_HV
 355                 r = cpu_has_feature(CPU_FTR_ARCH_206) ? 1 : 0;
 356 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 357                 r = 1;
 358 #else
 359                 r = 0;
 360                 break;
 361 #endif
 362 #ifdef CONFIG_KVM_BOOK3S_64_HV
 363         case KVM_CAP_PPC_HTAB_FD:
 364                 r = 1;
 365                 break;
 366 #endif
 367                 break;
 368         case KVM_CAP_NR_VCPUS:
 369                 /*
 370                  * Recommending a number of CPUs is somewhat arbitrary; we
 371                  * return the number of present CPUs for -HV (since a host
 372                  * will have secondary threads "offline"), and for other KVM
 373                  * implementations just count online CPUs.
 374                  */
 375 #ifdef CONFIG_KVM_BOOK3S_64_HV
 376                 r = num_present_cpus();
 377 #else
 378                 r = num_online_cpus();
 379 #endif
 380                 break;
 381         case KVM_CAP_MAX_VCPUS:
 382                 r = KVM_MAX_VCPUS;
 383                 break;
 384 #ifdef CONFIG_PPC_BOOK3S_64
 385         case KVM_CAP_PPC_GET_SMMU_INFO:
 386                 r = 1;
 387                 break;
 388 #endif
 389         default:
 390                 r = 0;
 391                 break;
 392         }
 393         return r;
 394
 395 }
 396
 397 long kvm_arch_dev_ioctl(struct file *filp,
 398                         unsigned int ioctl, unsigned long arg)
 399 {
 400         return -EINVAL;
 401 }
 402
 403 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
 404                            struct kvm_memory_slot *dont)
 405 {
 406         kvmppc_core_free_memslot(free, dont);
 407 }
 408
 409 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
 410 {
 411         return kvmppc_core_create_memslot(slot, npages);
 412 }
 413
 414 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 415                                    struct kvm_memory_slot *memslot,
 416                                    struct kvm_userspace_memory_region *mem,
 417                                    enum kvm_mr_change change)
 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                                    const struct kvm_memory_slot *old,
 425                                    enum kvm_mr_change change)
 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 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
 536                                      struct kvm_run *run)
 537 {
 538         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
 539 }
 540
 541 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 542                                       struct kvm_run *run)
 543 {
 544         u64 uninitialized_var(gpr);
 545
 546         if (run->mmio.len > sizeof(gpr)) {
 547                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 548                 return;
 549         }
 550
 551         if (vcpu->arch.mmio_is_bigendian) {
 552                 switch (run->mmio.len) {
 553                 case 8: gpr = *(u64 *)run->mmio.data; break;
 554                 case 4: gpr = *(u32 *)run->mmio.data; break;
 555                 case 2: gpr = *(u16 *)run->mmio.data; break;
 556                 case 1: gpr = *(u8 *)run->mmio.data; break;
 557                 }
 558         } else {
 559                 /* Convert BE data from userland back to LE. */
 560                 switch (run->mmio.len) {
 561                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
 562                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
 563                 case 1: gpr = *(u8 *)run->mmio.data; break;
 564                 }
 565         }
 566
 567         if (vcpu->arch.mmio_sign_extend) {
 568                 switch (run->mmio.len) {
 569 #ifdef CONFIG_PPC64
 570                 case 4:
 571                         gpr = (s64)(s32)gpr;
 572                         break;
 573 #endif
 574                 case 2:
 575                         gpr = (s64)(s16)gpr;
 576                         break;
 577                 case 1:
 578                         gpr = (s64)(s8)gpr;
 579                         break;
 580                 }
 581         }
 582
 583         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 584
 585         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 586         case KVM_MMIO_REG_GPR:
 587                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 588                 break;
 589         case KVM_MMIO_REG_FPR:
 590                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 591                 break;
 592 #ifdef CONFIG_PPC_BOOK3S
 593         case KVM_MMIO_REG_QPR:
 594                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 595                 break;
 596         case KVM_MMIO_REG_FQPR:
 597                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 598                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 599                 break;
 600 #endif
 601         default:
 602                 BUG();
 603         }
 604 }
 605
 606 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 607                        unsigned int rt, unsigned int bytes, int is_bigendian)
 608 {
 609         if (bytes > sizeof(run->mmio.data)) {
 610                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 611                        run->mmio.len);
 612         }
 613
 614         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 615         run->mmio.len = bytes;
 616         run->mmio.is_write = 0;
 617
 618         vcpu->arch.io_gpr = rt;
 619         vcpu->arch.mmio_is_bigendian = is_bigendian;
 620         vcpu->mmio_needed = 1;
 621         vcpu->mmio_is_write = 0;
 622         vcpu->arch.mmio_sign_extend = 0;
 623
 624         if (!kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 625                              bytes, &run->mmio.data)) {
 626                 kvmppc_complete_mmio_load(vcpu, run);
 627                 vcpu->mmio_needed = 0;
 628                 return EMULATE_DONE;
 629         }
 630
 631         return EMULATE_DO_MMIO;
 632 }
 633
 634 /* Same as above, but sign extends */
 635 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 636                         unsigned int rt, unsigned int bytes, int is_bigendian)
 637 {
 638         int r;
 639
 640         vcpu->arch.mmio_sign_extend = 1;
 641         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
 642
 643         return r;
 644 }
 645
 646 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 647                         u64 val, unsigned int bytes, int is_bigendian)
 648 {
 649         void *data = run->mmio.data;
 650
 651         if (bytes > sizeof(run->mmio.data)) {
 652                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 653                        run->mmio.len);
 654         }
 655
 656         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 657         run->mmio.len = bytes;
 658         run->mmio.is_write = 1;
 659         vcpu->mmio_needed = 1;
 660         vcpu->mmio_is_write = 1;
 661
 662         /* Store the value at the lowest bytes in 'data'. */
 663         if (is_bigendian) {
 664                 switch (bytes) {
 665                 case 8: *(u64 *)data = val; break;
 666                 case 4: *(u32 *)data = val; break;
 667                 case 2: *(u16 *)data = val; break;
 668                 case 1: *(u8  *)data = val; break;
 669                 }
 670         } else {
 671                 /* Store LE value into 'data'. */
 672                 switch (bytes) {
 673                 case 4: st_le32(data, val); break;
 674                 case 2: st_le16(data, val); break;
 675                 case 1: *(u8 *)data = val; break;
 676                 }
 677         }
 678
 679         if (!kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 680                               bytes, &run->mmio.data)) {
 681                 vcpu->mmio_needed = 0;
 682                 return EMULATE_DONE;
 683         }
 684
 685         return EMULATE_DO_MMIO;
 686 }
 687
 688 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 689 {
 690         int r;
 691         sigset_t sigsaved;
 692
 693         if (vcpu->sigset_active)
 694                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 695
 696         if (vcpu->mmio_needed) {
 697                 if (!vcpu->mmio_is_write)
 698                         kvmppc_complete_mmio_load(vcpu, run);
 699                 vcpu->mmio_needed = 0;
 700         } else if (vcpu->arch.dcr_needed) {
 701                 if (!vcpu->arch.dcr_is_write)
 702                         kvmppc_complete_dcr_load(vcpu, run);
 703                 vcpu->arch.dcr_needed = 0;
 704         } else if (vcpu->arch.osi_needed) {
 705                 u64 *gprs = run->osi.gprs;
 706                 int i;
 707
 708                 for (i = 0; i < 32; i++)
 709                         kvmppc_set_gpr(vcpu, i, gprs[i]);
 710                 vcpu->arch.osi_needed = 0;
 711         } else if (vcpu->arch.hcall_needed) {
 712                 int i;
 713
 714                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
 715                 for (i = 0; i < 9; ++i)
 716                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
 717                 vcpu->arch.hcall_needed = 0;
 718 #ifdef CONFIG_BOOKE
 719         } else if (vcpu->arch.epr_needed) {
 720                 kvmppc_set_epr(vcpu, run->epr.epr);
 721                 vcpu->arch.epr_needed = 0;
 722 #endif
 723         }
 724
 725         r = kvmppc_vcpu_run(run, vcpu);
 726
 727         if (vcpu->sigset_active)
 728                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 729
 730         return r;
 731 }
 732
 733 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
 734 {
 735         if (irq->irq == KVM_INTERRUPT_UNSET) {
 736                 kvmppc_core_dequeue_external(vcpu);
 737                 return 0;
 738         }
 739
 740         kvmppc_core_queue_external(vcpu, irq);
 741
 742         kvm_vcpu_kick(vcpu);
 743
 744         return 0;
 745 }
 746
 747 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
 748                                      struct kvm_enable_cap *cap)
 749 {
 750         int r;
 751
 752         if (cap->flags)
 753                 return -EINVAL;
 754
 755         switch (cap->cap) {
 756         case KVM_CAP_PPC_OSI:
 757                 r = 0;
 758                 vcpu->arch.osi_enabled = true;
 759                 break;
 760         case KVM_CAP_PPC_PAPR:
 761                 r = 0;
 762                 vcpu->arch.papr_enabled = true;
 763                 break;
 764         case KVM_CAP_PPC_EPR:
 765                 r = 0;
 766                 if (cap->args[0])
 767                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
 768                 else
 769                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
 770                 break;
 771 #ifdef CONFIG_BOOKE
 772         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 773                 r = 0;
 774                 vcpu->arch.watchdog_enabled = true;
 775                 break;
 776 #endif
 777 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 778         case KVM_CAP_SW_TLB: {
 779                 struct kvm_config_tlb cfg;
 780                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
 781
 782                 r = -EFAULT;
 783                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
 784                         break;
 785
 786                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
 787                 break;
 788         }
 789 #endif
 790         default:
 791                 r = -EINVAL;
 792                 break;
 793         }
 794
 795         if (!r)
 796                 r = kvmppc_sanity_check(vcpu);
 797
 798         return r;
 799 }
 800
 801 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 802                                     struct kvm_mp_state *mp_state)
 803 {
 804         return -EINVAL;
 805 }
 806
 807 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 808                                     struct kvm_mp_state *mp_state)
 809 {
 810         return -EINVAL;
 811 }
 812
 813 long kvm_arch_vcpu_ioctl(struct file *filp,
 814                          unsigned int ioctl, unsigned long arg)
 815 {
 816         struct kvm_vcpu *vcpu = filp->private_data;
 817         void __user *argp = (void __user *)arg;
 818         long r;
 819
 820         switch (ioctl) {
 821         case KVM_INTERRUPT: {
 822                 struct kvm_interrupt irq;
 823                 r = -EFAULT;
 824                 if (copy_from_user(&irq, argp, sizeof(irq)))
 825                         goto out;
 826                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 827                 goto out;
 828         }
 829
 830         case KVM_ENABLE_CAP:
 831         {
 832                 struct kvm_enable_cap cap;
 833                 r = -EFAULT;
 834                 if (copy_from_user(&cap, argp, sizeof(cap)))
 835                         goto out;
 836                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
 837                 break;
 838         }
 839
 840         case KVM_SET_ONE_REG:
 841         case KVM_GET_ONE_REG:
 842         {
 843                 struct kvm_one_reg reg;
 844                 r = -EFAULT;
 845                 if (copy_from_user(&reg, argp, sizeof(reg)))
 846                         goto out;
 847                 if (ioctl == KVM_SET_ONE_REG)
 848                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
 849                 else
 850                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
 851                 break;
 852         }
 853
 854 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 855         case KVM_DIRTY_TLB: {
 856                 struct kvm_dirty_tlb dirty;
 857                 r = -EFAULT;
 858                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
 859                         goto out;
 860                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
 861                 break;
 862         }
 863 #endif
 864         default:
 865                 r = -EINVAL;
 866         }
 867
 868 out:
 869         return r;
 870 }
 871
 872 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 873 {
 874         return VM_FAULT_SIGBUS;
 875 }
 876
 877 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
 878 {
 879         u32 inst_nop = 0x60000000;
 880 #ifdef CONFIG_KVM_BOOKE_HV
 881         u32 inst_sc1 = 0x44000022;
 882         pvinfo->hcall[0] = inst_sc1;
 883         pvinfo->hcall[1] = inst_nop;
 884         pvinfo->hcall[2] = inst_nop;
 885         pvinfo->hcall[3] = inst_nop;
 886 #else
 887         u32 inst_lis = 0x3c000000;
 888         u32 inst_ori = 0x60000000;
 889         u32 inst_sc = 0x44000002;
 890         u32 inst_imm_mask = 0xffff;
 891
 892         /*
 893          * The hypercall to get into KVM from within guest context is as
 894          * follows:
 895          *
 896          *    lis r0, r0, KVM_SC_MAGIC_R0@h
 897          *    ori r0, KVM_SC_MAGIC_R0@l
 898          *    sc
 899          *    nop
 900          */
 901         pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
 902         pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
 903         pvinfo->hcall[2] = inst_sc;
 904         pvinfo->hcall[3] = inst_nop;
 905 #endif
 906
 907         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
 908
 909         return 0;
 910 }
 911
 912 long kvm_arch_vm_ioctl(struct file *filp,
 913                        unsigned int ioctl, unsigned long arg)
 914 {
 915         struct kvm *kvm __maybe_unused = filp->private_data;
 916         void __user *argp = (void __user *)arg;
 917         long r;
 918
 919         switch (ioctl) {
 920         case KVM_PPC_GET_PVINFO: {
 921                 struct kvm_ppc_pvinfo pvinfo;
 922                 memset(&pvinfo, 0, sizeof(pvinfo));
 923                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
 924                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
 925                         r = -EFAULT;
 926                         goto out;
 927                 }
 928
 929                 break;
 930         }
 931 #ifdef CONFIG_PPC_BOOK3S_64
 932         case KVM_CREATE_SPAPR_TCE: {
 933                 struct kvm_create_spapr_tce create_tce;
 934
 935                 r = -EFAULT;
 936                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
 937                         goto out;
 938                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
 939                 goto out;
 940         }
 941 #endif /* CONFIG_PPC_BOOK3S_64 */
 942
 943 #ifdef CONFIG_KVM_BOOK3S_64_HV
 944         case KVM_ALLOCATE_RMA: {
 945                 struct kvm_allocate_rma rma;
 946
 947                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
 948                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
 949                         r = -EFAULT;
 950                 break;
 951         }
 952
 953         case KVM_PPC_ALLOCATE_HTAB: {
 954                 u32 htab_order;
 955
 956                 r = -EFAULT;
 957                 if (get_user(htab_order, (u32 __user *)argp))
 958                         break;
 959                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
 960                 if (r)
 961                         break;
 962                 r = -EFAULT;
 963                 if (put_user(htab_order, (u32 __user *)argp))
 964                         break;
 965                 r = 0;
 966                 break;
 967         }
 968
 969         case KVM_PPC_GET_HTAB_FD: {
 970                 struct kvm_get_htab_fd ghf;
 971
 972                 r = -EFAULT;
 973                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
 974                         break;
 975                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
 976                 break;
 977         }
 978 #endif /* CONFIG_KVM_BOOK3S_64_HV */
 979
 980 #ifdef CONFIG_PPC_BOOK3S_64
 981         case KVM_PPC_GET_SMMU_INFO: {
 982                 struct kvm_ppc_smmu_info info;
 983
 984                 memset(&info, 0, sizeof(info));
 985                 r = kvm_vm_ioctl_get_smmu_info(kvm, &info);
 986                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
 987                         r = -EFAULT;
 988                 break;
 989         }
 990 #endif /* CONFIG_PPC_BOOK3S_64 */
 991         default:
 992                 r = -ENOTTY;
 993         }
 994
 995 out:
 996         return r;
 997 }
 998
 999 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1000 static unsigned long nr_lpids;
1001
1002 long kvmppc_alloc_lpid(void)
1003 {
1004         long lpid;
1005
1006         do {
1007                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1008                 if (lpid >= nr_lpids) {
1009                         pr_err("%s: No LPIDs free\n", __func__);
1010                         return -ENOMEM;
1011                 }
1012         } while (test_and_set_bit(lpid, lpid_inuse));
1013
1014         return lpid;
1015 }
1016
1017 void kvmppc_claim_lpid(long lpid)
1018 {
1019         set_bit(lpid, lpid_inuse);
1020 }
1021
1022 void kvmppc_free_lpid(long lpid)
1023 {
1024         clear_bit(lpid, lpid_inuse);
1025 }
1026
1027 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1028 {
1029         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1030         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1031 }
1032
1033 int kvm_arch_init(void *opaque)
1034 {
1035         return 0;
1036 }
1037
1038 void kvm_arch_exit(void)
1039 {
1040 }