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_memory_slot *free,
 413                            struct kvm_memory_slot *dont)
 414 {
 415         kvmppc_core_free_memslot(free, dont);
 416 }
 417
 418 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
 419 {
 420         return kvmppc_core_create_memslot(slot, npages);
 421 }
 422
 423 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 424                                    struct kvm_memory_slot *memslot,
 425                                    struct kvm_userspace_memory_region *mem,
 426                                    enum kvm_mr_change change)
 427 {
 428         return kvmppc_core_prepare_memory_region(kvm, memslot, mem);
 429 }
 430
 431 void kvm_arch_commit_memory_region(struct kvm *kvm,
 432                                    struct kvm_userspace_memory_region *mem,
 433                                    const struct kvm_memory_slot *old,
 434                                    enum kvm_mr_change change)
 435 {
 436         kvmppc_core_commit_memory_region(kvm, mem, old);
 437 }
 438
 439 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 440 {
 441 }
 442
 443 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 444                                    struct kvm_memory_slot *slot)
 445 {
 446         kvmppc_core_flush_memslot(kvm, slot);
 447 }
 448
 449 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 450 {
 451         struct kvm_vcpu *vcpu;
 452         vcpu = kvmppc_core_vcpu_create(kvm, id);
 453         if (!IS_ERR(vcpu)) {
 454                 vcpu->arch.wqp = &vcpu->wq;
 455                 kvmppc_create_vcpu_debugfs(vcpu, id);
 456         }
 457         return vcpu;
 458 }
 459
 460 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 461 {
 462         return 0;
 463 }
 464
 465 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 466 {
 467         /* Make sure we're not using the vcpu anymore */
 468         hrtimer_cancel(&vcpu->arch.dec_timer);
 469         tasklet_kill(&vcpu->arch.tasklet);
 470
 471         kvmppc_remove_vcpu_debugfs(vcpu);
 472
 473         switch (vcpu->arch.irq_type) {
 474         case KVMPPC_IRQ_MPIC:
 475                 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu);
 476                 break;
 477         case KVMPPC_IRQ_XICS:
 478                 kvmppc_xics_free_icp(vcpu);
 479                 break;
 480         }
 481
 482         kvmppc_core_vcpu_free(vcpu);
 483 }
 484
 485 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 486 {
 487         kvm_arch_vcpu_free(vcpu);
 488 }
 489
 490 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 491 {
 492         return kvmppc_core_pending_dec(vcpu);
 493 }
 494
 495 /*
 496  * low level hrtimer wake routine. Because this runs in hardirq context
 497  * we schedule a tasklet to do the real work.
 498  */
 499 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer)
 500 {
 501         struct kvm_vcpu *vcpu;
 502
 503         vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer);
 504         tasklet_schedule(&vcpu->arch.tasklet);
 505
 506         return HRTIMER_NORESTART;
 507 }
 508
 509 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 510 {
 511         int ret;
 512
 513         hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 514         tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu);
 515         vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup;
 516         vcpu->arch.dec_expires = ~(u64)0;
 517
 518 #ifdef CONFIG_KVM_EXIT_TIMING
 519         mutex_init(&vcpu->arch.exit_timing_lock);
 520 #endif
 521         ret = kvmppc_subarch_vcpu_init(vcpu);
 522         return ret;
 523 }
 524
 525 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 526 {
 527         kvmppc_mmu_destroy(vcpu);
 528         kvmppc_subarch_vcpu_uninit(vcpu);
 529 }
 530
 531 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 532 {
 533 #ifdef CONFIG_BOOKE
 534         /*
 535          * vrsave (formerly usprg0) isn't used by Linux, but may
 536          * be used by the guest.
 537          *
 538          * On non-booke this is associated with Altivec and
 539          * is handled by code in book3s.c.
 540          */
 541         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
 542 #endif
 543         kvmppc_core_vcpu_load(vcpu, cpu);
 544 }
 545
 546 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 547 {
 548         kvmppc_core_vcpu_put(vcpu);
 549 #ifdef CONFIG_BOOKE
 550         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
 551 #endif
 552 }
 553
 554 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu,
 555                                      struct kvm_run *run)
 556 {
 557         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data);
 558 }
 559
 560 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu,
 561                                       struct kvm_run *run)
 562 {
 563         u64 uninitialized_var(gpr);
 564
 565         if (run->mmio.len > sizeof(gpr)) {
 566                 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len);
 567                 return;
 568         }
 569
 570         if (vcpu->arch.mmio_is_bigendian) {
 571                 switch (run->mmio.len) {
 572                 case 8: gpr = *(u64 *)run->mmio.data; break;
 573                 case 4: gpr = *(u32 *)run->mmio.data; break;
 574                 case 2: gpr = *(u16 *)run->mmio.data; break;
 575                 case 1: gpr = *(u8 *)run->mmio.data; break;
 576                 }
 577         } else {
 578                 /* Convert BE data from userland back to LE. */
 579                 switch (run->mmio.len) {
 580                 case 4: gpr = ld_le32((u32 *)run->mmio.data); break;
 581                 case 2: gpr = ld_le16((u16 *)run->mmio.data); break;
 582                 case 1: gpr = *(u8 *)run->mmio.data; break;
 583                 }
 584         }
 585
 586         if (vcpu->arch.mmio_sign_extend) {
 587                 switch (run->mmio.len) {
 588 #ifdef CONFIG_PPC64
 589                 case 4:
 590                         gpr = (s64)(s32)gpr;
 591                         break;
 592 #endif
 593                 case 2:
 594                         gpr = (s64)(s16)gpr;
 595                         break;
 596                 case 1:
 597                         gpr = (s64)(s8)gpr;
 598                         break;
 599                 }
 600         }
 601
 602         kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 603
 604         switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) {
 605         case KVM_MMIO_REG_GPR:
 606                 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr);
 607                 break;
 608         case KVM_MMIO_REG_FPR:
 609                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 610                 break;
 611 #ifdef CONFIG_PPC_BOOK3S
 612         case KVM_MMIO_REG_QPR:
 613                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 614                 break;
 615         case KVM_MMIO_REG_FQPR:
 616                 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 617                 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr;
 618                 break;
 619 #endif
 620         default:
 621                 BUG();
 622         }
 623 }
 624
 625 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
 626                        unsigned int rt, unsigned int bytes, int is_bigendian)
 627 {
 628         int idx, ret;
 629
 630         if (bytes > sizeof(run->mmio.data)) {
 631                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 632                        run->mmio.len);
 633         }
 634
 635         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 636         run->mmio.len = bytes;
 637         run->mmio.is_write = 0;
 638
 639         vcpu->arch.io_gpr = rt;
 640         vcpu->arch.mmio_is_bigendian = is_bigendian;
 641         vcpu->mmio_needed = 1;
 642         vcpu->mmio_is_write = 0;
 643         vcpu->arch.mmio_sign_extend = 0;
 644
 645         idx = srcu_read_lock(&vcpu->kvm->srcu);
 646
 647         ret = kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 648                               bytes, &run->mmio.data);
 649
 650         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 651
 652         if (!ret) {
 653                 kvmppc_complete_mmio_load(vcpu, run);
 654                 vcpu->mmio_needed = 0;
 655                 return EMULATE_DONE;
 656         }
 657
 658         return EMULATE_DO_MMIO;
 659 }
 660
 661 /* Same as above, but sign extends */
 662 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu,
 663                         unsigned int rt, unsigned int bytes, int is_bigendian)
 664 {
 665         int r;
 666
 667         vcpu->arch.mmio_sign_extend = 1;
 668         r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian);
 669
 670         return r;
 671 }
 672
 673 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu,
 674                         u64 val, unsigned int bytes, int is_bigendian)
 675 {
 676         void *data = run->mmio.data;
 677         int idx, ret;
 678
 679         if (bytes > sizeof(run->mmio.data)) {
 680                 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__,
 681                        run->mmio.len);
 682         }
 683
 684         run->mmio.phys_addr = vcpu->arch.paddr_accessed;
 685         run->mmio.len = bytes;
 686         run->mmio.is_write = 1;
 687         vcpu->mmio_needed = 1;
 688         vcpu->mmio_is_write = 1;
 689
 690         /* Store the value at the lowest bytes in 'data'. */
 691         if (is_bigendian) {
 692                 switch (bytes) {
 693                 case 8: *(u64 *)data = val; break;
 694                 case 4: *(u32 *)data = val; break;
 695                 case 2: *(u16 *)data = val; break;
 696                 case 1: *(u8  *)data = val; break;
 697                 }
 698         } else {
 699                 /* Store LE value into 'data'. */
 700                 switch (bytes) {
 701                 case 4: st_le32(data, val); break;
 702                 case 2: st_le16(data, val); break;
 703                 case 1: *(u8 *)data = val; break;
 704                 }
 705         }
 706
 707         idx = srcu_read_lock(&vcpu->kvm->srcu);
 708
 709         ret = kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, run->mmio.phys_addr,
 710                                bytes, &run->mmio.data);
 711
 712         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 713
 714         if (!ret) {
 715                 vcpu->mmio_needed = 0;
 716                 return EMULATE_DONE;
 717         }
 718
 719         return EMULATE_DO_MMIO;
 720 }
 721
 722 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 723 {
 724         int r;
 725         sigset_t sigsaved;
 726
 727         if (vcpu->sigset_active)
 728                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 729
 730         if (vcpu->mmio_needed) {
 731                 if (!vcpu->mmio_is_write)
 732                         kvmppc_complete_mmio_load(vcpu, run);
 733                 vcpu->mmio_needed = 0;
 734         } else if (vcpu->arch.dcr_needed) {
 735                 if (!vcpu->arch.dcr_is_write)
 736                         kvmppc_complete_dcr_load(vcpu, run);
 737                 vcpu->arch.dcr_needed = 0;
 738         } else if (vcpu->arch.osi_needed) {
 739                 u64 *gprs = run->osi.gprs;
 740                 int i;
 741
 742                 for (i = 0; i < 32; i++)
 743                         kvmppc_set_gpr(vcpu, i, gprs[i]);
 744                 vcpu->arch.osi_needed = 0;
 745         } else if (vcpu->arch.hcall_needed) {
 746                 int i;
 747
 748                 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret);
 749                 for (i = 0; i < 9; ++i)
 750                         kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]);
 751                 vcpu->arch.hcall_needed = 0;
 752 #ifdef CONFIG_BOOKE
 753         } else if (vcpu->arch.epr_needed) {
 754                 kvmppc_set_epr(vcpu, run->epr.epr);
 755                 vcpu->arch.epr_needed = 0;
 756 #endif
 757         }
 758
 759         r = kvmppc_vcpu_run(run, vcpu);
 760
 761         if (vcpu->sigset_active)
 762                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 763
 764         return r;
 765 }
 766
 767 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
 768 {
 769         if (irq->irq == KVM_INTERRUPT_UNSET) {
 770                 kvmppc_core_dequeue_external(vcpu);
 771                 return 0;
 772         }
 773
 774         kvmppc_core_queue_external(vcpu, irq);
 775
 776         kvm_vcpu_kick(vcpu);
 777
 778         return 0;
 779 }
 780
 781 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
 782                                      struct kvm_enable_cap *cap)
 783 {
 784         int r;
 785
 786         if (cap->flags)
 787                 return -EINVAL;
 788
 789         switch (cap->cap) {
 790         case KVM_CAP_PPC_OSI:
 791                 r = 0;
 792                 vcpu->arch.osi_enabled = true;
 793                 break;
 794         case KVM_CAP_PPC_PAPR:
 795                 r = 0;
 796                 vcpu->arch.papr_enabled = true;
 797                 break;
 798         case KVM_CAP_PPC_EPR:
 799                 r = 0;
 800                 if (cap->args[0])
 801                         vcpu->arch.epr_flags |= KVMPPC_EPR_USER;
 802                 else
 803                         vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER;
 804                 break;
 805 #ifdef CONFIG_BOOKE
 806         case KVM_CAP_PPC_BOOKE_WATCHDOG:
 807                 r = 0;
 808                 vcpu->arch.watchdog_enabled = true;
 809                 break;
 810 #endif
 811 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 812         case KVM_CAP_SW_TLB: {
 813                 struct kvm_config_tlb cfg;
 814                 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0];
 815
 816                 r = -EFAULT;
 817                 if (copy_from_user(&cfg, user_ptr, sizeof(cfg)))
 818                         break;
 819
 820                 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg);
 821                 break;
 822         }
 823 #endif
 824 #ifdef CONFIG_KVM_MPIC
 825         case KVM_CAP_IRQ_MPIC: {
 826                 struct file *filp;
 827                 struct kvm_device *dev;
 828
 829                 r = -EBADF;
 830                 filp = fget(cap->args[0]);
 831                 if (!filp)
 832                         break;
 833
 834                 r = -EPERM;
 835                 dev = kvm_device_from_filp(filp);
 836                 if (dev)
 837                         r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]);
 838
 839                 fput(filp);
 840                 break;
 841         }
 842 #endif
 843 #ifdef CONFIG_KVM_XICS
 844         case KVM_CAP_IRQ_XICS: {
 845                 struct file *filp;
 846                 struct kvm_device *dev;
 847
 848                 r = -EBADF;
 849                 filp = fget(cap->args[0]);
 850                 if (!filp)
 851                         break;
 852
 853                 r = -EPERM;
 854                 dev = kvm_device_from_filp(filp);
 855                 if (dev)
 856                         r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]);
 857
 858                 fput(filp);
 859                 break;
 860         }
 861 #endif /* CONFIG_KVM_XICS */
 862         default:
 863                 r = -EINVAL;
 864                 break;
 865         }
 866
 867         if (!r)
 868                 r = kvmppc_sanity_check(vcpu);
 869
 870         return r;
 871 }
 872
 873 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 874                                     struct kvm_mp_state *mp_state)
 875 {
 876         return -EINVAL;
 877 }
 878
 879 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 880                                     struct kvm_mp_state *mp_state)
 881 {
 882         return -EINVAL;
 883 }
 884
 885 long kvm_arch_vcpu_ioctl(struct file *filp,
 886                          unsigned int ioctl, unsigned long arg)
 887 {
 888         struct kvm_vcpu *vcpu = filp->private_data;
 889         void __user *argp = (void __user *)arg;
 890         long r;
 891
 892         switch (ioctl) {
 893         case KVM_INTERRUPT: {
 894                 struct kvm_interrupt irq;
 895                 r = -EFAULT;
 896                 if (copy_from_user(&irq, argp, sizeof(irq)))
 897                         goto out;
 898                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 899                 goto out;
 900         }
 901
 902         case KVM_ENABLE_CAP:
 903         {
 904                 struct kvm_enable_cap cap;
 905                 r = -EFAULT;
 906                 if (copy_from_user(&cap, argp, sizeof(cap)))
 907                         goto out;
 908                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
 909                 break;
 910         }
 911
 912         case KVM_SET_ONE_REG:
 913         case KVM_GET_ONE_REG:
 914         {
 915                 struct kvm_one_reg reg;
 916                 r = -EFAULT;
 917                 if (copy_from_user(&reg, argp, sizeof(reg)))
 918                         goto out;
 919                 if (ioctl == KVM_SET_ONE_REG)
 920                         r = kvm_vcpu_ioctl_set_one_reg(vcpu, &reg);
 921                 else
 922                         r = kvm_vcpu_ioctl_get_one_reg(vcpu, &reg);
 923                 break;
 924         }
 925
 926 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC)
 927         case KVM_DIRTY_TLB: {
 928                 struct kvm_dirty_tlb dirty;
 929                 r = -EFAULT;
 930                 if (copy_from_user(&dirty, argp, sizeof(dirty)))
 931                         goto out;
 932                 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty);
 933                 break;
 934         }
 935 #endif
 936         default:
 937                 r = -EINVAL;
 938         }
 939
 940 out:
 941         return r;
 942 }
 943
 944 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 945 {
 946         return VM_FAULT_SIGBUS;
 947 }
 948
 949 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo)
 950 {
 951         u32 inst_nop = 0x60000000;
 952 #ifdef CONFIG_KVM_BOOKE_HV
 953         u32 inst_sc1 = 0x44000022;
 954         pvinfo->hcall[0] = inst_sc1;
 955         pvinfo->hcall[1] = inst_nop;
 956         pvinfo->hcall[2] = inst_nop;
 957         pvinfo->hcall[3] = inst_nop;
 958 #else
 959         u32 inst_lis = 0x3c000000;
 960         u32 inst_ori = 0x60000000;
 961         u32 inst_sc = 0x44000002;
 962         u32 inst_imm_mask = 0xffff;
 963
 964         /*
 965          * The hypercall to get into KVM from within guest context is as
 966          * follows:
 967          *
 968          *    lis r0, r0, KVM_SC_MAGIC_R0@h
 969          *    ori r0, KVM_SC_MAGIC_R0@l
 970          *    sc
 971          *    nop
 972          */
 973         pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask);
 974         pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask);
 975         pvinfo->hcall[2] = inst_sc;
 976         pvinfo->hcall[3] = inst_nop;
 977 #endif
 978
 979         pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE;
 980
 981         return 0;
 982 }
 983
 984 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
 985                           bool line_status)
 986 {
 987         if (!irqchip_in_kernel(kvm))
 988                 return -ENXIO;
 989
 990         irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
 991                                         irq_event->irq, irq_event->level,
 992                                         line_status);
 993         return 0;
 994 }
 995
 996 long kvm_arch_vm_ioctl(struct file *filp,
 997                        unsigned int ioctl, unsigned long arg)
 998 {
 999         struct kvm *kvm __maybe_unused = filp->private_data;
1000         void __user *argp = (void __user *)arg;
1001         long r;
1002
1003         switch (ioctl) {
1004         case KVM_PPC_GET_PVINFO: {
1005                 struct kvm_ppc_pvinfo pvinfo;
1006                 memset(&pvinfo, 0, sizeof(pvinfo));
1007                 r = kvm_vm_ioctl_get_pvinfo(&pvinfo);
1008                 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) {
1009                         r = -EFAULT;
1010                         goto out;
1011                 }
1012
1013                 break;
1014         }
1015 #ifdef CONFIG_PPC_BOOK3S_64
1016         case KVM_CREATE_SPAPR_TCE: {
1017                 struct kvm_create_spapr_tce create_tce;
1018
1019                 r = -EFAULT;
1020                 if (copy_from_user(&create_tce, argp, sizeof(create_tce)))
1021                         goto out;
1022                 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce);
1023                 goto out;
1024         }
1025 #endif /* CONFIG_PPC_BOOK3S_64 */
1026
1027 #ifdef CONFIG_KVM_BOOK3S_64_HV
1028         case KVM_ALLOCATE_RMA: {
1029                 struct kvm_allocate_rma rma;
1030                 struct kvm *kvm = filp->private_data;
1031
1032                 r = kvm_vm_ioctl_allocate_rma(kvm, &rma);
1033                 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma)))
1034                         r = -EFAULT;
1035                 break;
1036         }
1037
1038         case KVM_PPC_ALLOCATE_HTAB: {
1039                 u32 htab_order;
1040
1041                 r = -EFAULT;
1042                 if (get_user(htab_order, (u32 __user *)argp))
1043                         break;
1044                 r = kvmppc_alloc_reset_hpt(kvm, &htab_order);
1045                 if (r)
1046                         break;
1047                 r = -EFAULT;
1048                 if (put_user(htab_order, (u32 __user *)argp))
1049                         break;
1050                 r = 0;
1051                 break;
1052         }
1053
1054         case KVM_PPC_GET_HTAB_FD: {
1055                 struct kvm_get_htab_fd ghf;
1056
1057                 r = -EFAULT;
1058                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
1059                         break;
1060                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
1061                 break;
1062         }
1063 #endif /* CONFIG_KVM_BOOK3S_64_HV */
1064
1065 #ifdef CONFIG_PPC_BOOK3S_64
1066         case KVM_PPC_GET_SMMU_INFO: {
1067                 struct kvm_ppc_smmu_info info;
1068
1069                 memset(&info, 0, sizeof(info));
1070                 r = kvm_vm_ioctl_get_smmu_info(kvm, &info);
1071                 if (r >= 0 && copy_to_user(argp, &info, sizeof(info)))
1072                         r = -EFAULT;
1073                 break;
1074         }
1075         case KVM_PPC_RTAS_DEFINE_TOKEN: {
1076                 struct kvm *kvm = filp->private_data;
1077
1078                 r = kvm_vm_ioctl_rtas_define_token(kvm, argp);
1079                 break;
1080         }
1081 #endif /* CONFIG_PPC_BOOK3S_64 */
1082         default:
1083                 r = -ENOTTY;
1084         }
1085
1086 out:
1087         return r;
1088 }
1089
1090 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)];
1091 static unsigned long nr_lpids;
1092
1093 long kvmppc_alloc_lpid(void)
1094 {
1095         long lpid;
1096
1097         do {
1098                 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS);
1099                 if (lpid >= nr_lpids) {
1100                         pr_err("%s: No LPIDs free\n", __func__);
1101                         return -ENOMEM;
1102                 }
1103         } while (test_and_set_bit(lpid, lpid_inuse));
1104
1105         return lpid;
1106 }
1107
1108 void kvmppc_claim_lpid(long lpid)
1109 {
1110         set_bit(lpid, lpid_inuse);
1111 }
1112
1113 void kvmppc_free_lpid(long lpid)
1114 {
1115         clear_bit(lpid, lpid_inuse);
1116 }
1117
1118 void kvmppc_init_lpid(unsigned long nr_lpids_param)
1119 {
1120         nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param);
1121         memset(lpid_inuse, 0, sizeof(lpid_inuse));
1122 }
1123
1124 int kvm_arch_init(void *opaque)
1125 {
1126         return 0;
1127 }
1128
1129 void kvm_arch_exit(void)
1130 {
1131 }