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/sched/signal.h> 27 #include <linux/fs.h> 28 #include <linux/slab.h> 29 #include <linux/file.h> 30 #include <linux/module.h> 31 #include <linux/irqbypass.h> 32 #include <linux/kvm_irqfd.h> 33 #include <asm/cputable.h> 34 #include <linux/uaccess.h> 35 #include <asm/kvm_ppc.h> 36 #include <asm/tlbflush.h> 37 #include <asm/cputhreads.h> 38 #include <asm/irqflags.h> 39 #include <asm/iommu.h> 40 #include <asm/switch_to.h> 41 #include <asm/xive.h> 42 43 #include "timing.h" 44 #include "irq.h" 45 #include "../mm/mmu_decl.h" 46 47 #define CREATE_TRACE_POINTS 48 #include "trace.h" 49 50 struct kvmppc_ops *kvmppc_hv_ops; 51 EXPORT_SYMBOL_GPL(kvmppc_hv_ops); 52 struct kvmppc_ops *kvmppc_pr_ops; 53 EXPORT_SYMBOL_GPL(kvmppc_pr_ops); 54 55 56 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) 57 { 58 return !!(v->arch.pending_exceptions) || kvm_request_pending(v); 59 } 60 61 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 62 { 63 return false; 64 } 65 66 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 67 { 68 return 1; 69 } 70 71 /* 72 * Common checks before entering the guest world. Call with interrupts 73 * disabled. 74 * 75 * returns: 76 * 77 * == 1 if we're ready to go into guest state 78 * <= 0 if we need to go back to the host with return value 79 */ 80 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu) 81 { 82 int r; 83 84 WARN_ON(irqs_disabled()); 85 hard_irq_disable(); 86 87 while (true) { 88 if (need_resched()) { 89 local_irq_enable(); 90 cond_resched(); 91 hard_irq_disable(); 92 continue; 93 } 94 95 if (signal_pending(current)) { 96 kvmppc_account_exit(vcpu, SIGNAL_EXITS); 97 vcpu->run->exit_reason = KVM_EXIT_INTR; 98 r = -EINTR; 99 break; 100 } 101 102 vcpu->mode = IN_GUEST_MODE; 103 104 /* 105 * Reading vcpu->requests must happen after setting vcpu->mode, 106 * so we don't miss a request because the requester sees 107 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests 108 * before next entering the guest (and thus doesn't IPI). 109 * This also orders the write to mode from any reads 110 * to the page tables done while the VCPU is running. 111 * Please see the comment in kvm_flush_remote_tlbs. 112 */ 113 smp_mb(); 114 115 if (kvm_request_pending(vcpu)) { 116 /* Make sure we process requests preemptable */ 117 local_irq_enable(); 118 trace_kvm_check_requests(vcpu); 119 r = kvmppc_core_check_requests(vcpu); 120 hard_irq_disable(); 121 if (r > 0) 122 continue; 123 break; 124 } 125 126 if (kvmppc_core_prepare_to_enter(vcpu)) { 127 /* interrupts got enabled in between, so we 128 are back at square 1 */ 129 continue; 130 } 131 132 guest_enter_irqoff(); 133 return 1; 134 } 135 136 /* return to host */ 137 local_irq_enable(); 138 return r; 139 } 140 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter); 141 142 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE) 143 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu) 144 { 145 struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared; 146 int i; 147 148 shared->sprg0 = swab64(shared->sprg0); 149 shared->sprg1 = swab64(shared->sprg1); 150 shared->sprg2 = swab64(shared->sprg2); 151 shared->sprg3 = swab64(shared->sprg3); 152 shared->srr0 = swab64(shared->srr0); 153 shared->srr1 = swab64(shared->srr1); 154 shared->dar = swab64(shared->dar); 155 shared->msr = swab64(shared->msr); 156 shared->dsisr = swab32(shared->dsisr); 157 shared->int_pending = swab32(shared->int_pending); 158 for (i = 0; i < ARRAY_SIZE(shared->sr); i++) 159 shared->sr[i] = swab32(shared->sr[i]); 160 } 161 #endif 162 163 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu) 164 { 165 int nr = kvmppc_get_gpr(vcpu, 11); 166 int r; 167 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3); 168 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4); 169 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5); 170 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6); 171 unsigned long r2 = 0; 172 173 if (!(kvmppc_get_msr(vcpu) & MSR_SF)) { 174 /* 32 bit mode */ 175 param1 &= 0xffffffff; 176 param2 &= 0xffffffff; 177 param3 &= 0xffffffff; 178 param4 &= 0xffffffff; 179 } 180 181 switch (nr) { 182 case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE): 183 { 184 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE) 185 /* Book3S can be little endian, find it out here */ 186 int shared_big_endian = true; 187 if (vcpu->arch.intr_msr & MSR_LE) 188 shared_big_endian = false; 189 if (shared_big_endian != vcpu->arch.shared_big_endian) 190 kvmppc_swab_shared(vcpu); 191 vcpu->arch.shared_big_endian = shared_big_endian; 192 #endif 193 194 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) { 195 /* 196 * Older versions of the Linux magic page code had 197 * a bug where they would map their trampoline code 198 * NX. If that's the case, remove !PR NX capability. 199 */ 200 vcpu->arch.disable_kernel_nx = true; 201 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 202 } 203 204 vcpu->arch.magic_page_pa = param1 & ~0xfffULL; 205 vcpu->arch.magic_page_ea = param2 & ~0xfffULL; 206 207 #ifdef CONFIG_PPC_64K_PAGES 208 /* 209 * Make sure our 4k magic page is in the same window of a 64k 210 * page within the guest and within the host's page. 211 */ 212 if ((vcpu->arch.magic_page_pa & 0xf000) != 213 ((ulong)vcpu->arch.shared & 0xf000)) { 214 void *old_shared = vcpu->arch.shared; 215 ulong shared = (ulong)vcpu->arch.shared; 216 void *new_shared; 217 218 shared &= PAGE_MASK; 219 shared |= vcpu->arch.magic_page_pa & 0xf000; 220 new_shared = (void*)shared; 221 memcpy(new_shared, old_shared, 0x1000); 222 vcpu->arch.shared = new_shared; 223 } 224 #endif 225 226 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7; 227 228 r = EV_SUCCESS; 229 break; 230 } 231 case KVM_HCALL_TOKEN(KVM_HC_FEATURES): 232 r = EV_SUCCESS; 233 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2) 234 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE); 235 #endif 236 237 /* Second return value is in r4 */ 238 break; 239 case EV_HCALL_TOKEN(EV_IDLE): 240 r = EV_SUCCESS; 241 kvm_vcpu_block(vcpu); 242 kvm_clear_request(KVM_REQ_UNHALT, vcpu); 243 break; 244 default: 245 r = EV_UNIMPLEMENTED; 246 break; 247 } 248 249 kvmppc_set_gpr(vcpu, 4, r2); 250 251 return r; 252 } 253 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv); 254 255 int kvmppc_sanity_check(struct kvm_vcpu *vcpu) 256 { 257 int r = false; 258 259 /* We have to know what CPU to virtualize */ 260 if (!vcpu->arch.pvr) 261 goto out; 262 263 /* PAPR only works with book3s_64 */ 264 if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled) 265 goto out; 266 267 /* HV KVM can only do PAPR mode for now */ 268 if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm)) 269 goto out; 270 271 #ifdef CONFIG_KVM_BOOKE_HV 272 if (!cpu_has_feature(CPU_FTR_EMB_HV)) 273 goto out; 274 #endif 275 276 r = true; 277 278 out: 279 vcpu->arch.sane = r; 280 return r ? 0 : -EINVAL; 281 } 282 EXPORT_SYMBOL_GPL(kvmppc_sanity_check); 283 284 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu) 285 { 286 enum emulation_result er; 287 int r; 288 289 er = kvmppc_emulate_loadstore(vcpu); 290 switch (er) { 291 case EMULATE_DONE: 292 /* Future optimization: only reload non-volatiles if they were 293 * actually modified. */ 294 r = RESUME_GUEST_NV; 295 break; 296 case EMULATE_AGAIN: 297 r = RESUME_GUEST; 298 break; 299 case EMULATE_DO_MMIO: 300 run->exit_reason = KVM_EXIT_MMIO; 301 /* We must reload nonvolatiles because "update" load/store 302 * instructions modify register state. */ 303 /* Future optimization: only reload non-volatiles if they were 304 * actually modified. */ 305 r = RESUME_HOST_NV; 306 break; 307 case EMULATE_FAIL: 308 { 309 u32 last_inst; 310 311 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst); 312 /* XXX Deliver Program interrupt to guest. */ 313 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst); 314 r = RESUME_HOST; 315 break; 316 } 317 default: 318 WARN_ON(1); 319 r = RESUME_GUEST; 320 } 321 322 return r; 323 } 324 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio); 325 326 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr, 327 bool data) 328 { 329 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK; 330 struct kvmppc_pte pte; 331 int r; 332 333 vcpu->stat.st++; 334 335 r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST, 336 XLATE_WRITE, &pte); 337 if (r < 0) 338 return r; 339 340 *eaddr = pte.raddr; 341 342 if (!pte.may_write) 343 return -EPERM; 344 345 /* Magic page override */ 346 if (kvmppc_supports_magic_page(vcpu) && mp_pa && 347 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) && 348 !(kvmppc_get_msr(vcpu) & MSR_PR)) { 349 void *magic = vcpu->arch.shared; 350 magic += pte.eaddr & 0xfff; 351 memcpy(magic, ptr, size); 352 return EMULATE_DONE; 353 } 354 355 if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size)) 356 return EMULATE_DO_MMIO; 357 358 return EMULATE_DONE; 359 } 360 EXPORT_SYMBOL_GPL(kvmppc_st); 361 362 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr, 363 bool data) 364 { 365 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK; 366 struct kvmppc_pte pte; 367 int rc; 368 369 vcpu->stat.ld++; 370 371 rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST, 372 XLATE_READ, &pte); 373 if (rc) 374 return rc; 375 376 *eaddr = pte.raddr; 377 378 if (!pte.may_read) 379 return -EPERM; 380 381 if (!data && !pte.may_execute) 382 return -ENOEXEC; 383 384 /* Magic page override */ 385 if (kvmppc_supports_magic_page(vcpu) && mp_pa && 386 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) && 387 !(kvmppc_get_msr(vcpu) & MSR_PR)) { 388 void *magic = vcpu->arch.shared; 389 magic += pte.eaddr & 0xfff; 390 memcpy(ptr, magic, size); 391 return EMULATE_DONE; 392 } 393 394 if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size)) 395 return EMULATE_DO_MMIO; 396 397 return EMULATE_DONE; 398 } 399 EXPORT_SYMBOL_GPL(kvmppc_ld); 400 401 int kvm_arch_hardware_enable(void) 402 { 403 return 0; 404 } 405 406 int kvm_arch_hardware_setup(void) 407 { 408 return 0; 409 } 410 411 void kvm_arch_check_processor_compat(void *rtn) 412 { 413 *(int *)rtn = kvmppc_core_check_processor_compat(); 414 } 415 416 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 417 { 418 struct kvmppc_ops *kvm_ops = NULL; 419 /* 420 * if we have both HV and PR enabled, default is HV 421 */ 422 if (type == 0) { 423 if (kvmppc_hv_ops) 424 kvm_ops = kvmppc_hv_ops; 425 else 426 kvm_ops = kvmppc_pr_ops; 427 if (!kvm_ops) 428 goto err_out; 429 } else if (type == KVM_VM_PPC_HV) { 430 if (!kvmppc_hv_ops) 431 goto err_out; 432 kvm_ops = kvmppc_hv_ops; 433 } else if (type == KVM_VM_PPC_PR) { 434 if (!kvmppc_pr_ops) 435 goto err_out; 436 kvm_ops = kvmppc_pr_ops; 437 } else 438 goto err_out; 439 440 if (kvm_ops->owner && !try_module_get(kvm_ops->owner)) 441 return -ENOENT; 442 443 kvm->arch.kvm_ops = kvm_ops; 444 return kvmppc_core_init_vm(kvm); 445 err_out: 446 return -EINVAL; 447 } 448 449 bool kvm_arch_has_vcpu_debugfs(void) 450 { 451 return false; 452 } 453 454 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu) 455 { 456 return 0; 457 } 458 459 void kvm_arch_destroy_vm(struct kvm *kvm) 460 { 461 unsigned int i; 462 struct kvm_vcpu *vcpu; 463 464 #ifdef CONFIG_KVM_XICS 465 /* 466 * We call kick_all_cpus_sync() to ensure that all 467 * CPUs have executed any pending IPIs before we 468 * continue and free VCPUs structures below. 469 */ 470 if (is_kvmppc_hv_enabled(kvm)) 471 kick_all_cpus_sync(); 472 #endif 473 474 kvm_for_each_vcpu(i, vcpu, kvm) 475 kvm_arch_vcpu_free(vcpu); 476 477 mutex_lock(&kvm->lock); 478 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) 479 kvm->vcpus[i] = NULL; 480 481 atomic_set(&kvm->online_vcpus, 0); 482 483 kvmppc_core_destroy_vm(kvm); 484 485 mutex_unlock(&kvm->lock); 486 487 /* drop the module reference */ 488 module_put(kvm->arch.kvm_ops->owner); 489 } 490 491 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 492 { 493 int r; 494 /* Assume we're using HV mode when the HV module is loaded */ 495 int hv_enabled = kvmppc_hv_ops ? 1 : 0; 496 497 if (kvm) { 498 /* 499 * Hooray - we know which VM type we're running on. Depend on 500 * that rather than the guess above. 501 */ 502 hv_enabled = is_kvmppc_hv_enabled(kvm); 503 } 504 505 switch (ext) { 506 #ifdef CONFIG_BOOKE 507 case KVM_CAP_PPC_BOOKE_SREGS: 508 case KVM_CAP_PPC_BOOKE_WATCHDOG: 509 case KVM_CAP_PPC_EPR: 510 #else 511 case KVM_CAP_PPC_SEGSTATE: 512 case KVM_CAP_PPC_HIOR: 513 case KVM_CAP_PPC_PAPR: 514 #endif 515 case KVM_CAP_PPC_UNSET_IRQ: 516 case KVM_CAP_PPC_IRQ_LEVEL: 517 case KVM_CAP_ENABLE_CAP: 518 case KVM_CAP_ENABLE_CAP_VM: 519 case KVM_CAP_ONE_REG: 520 case KVM_CAP_IOEVENTFD: 521 case KVM_CAP_DEVICE_CTRL: 522 case KVM_CAP_IMMEDIATE_EXIT: 523 r = 1; 524 break; 525 case KVM_CAP_PPC_PAIRED_SINGLES: 526 case KVM_CAP_PPC_OSI: 527 case KVM_CAP_PPC_GET_PVINFO: 528 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC) 529 case KVM_CAP_SW_TLB: 530 #endif 531 /* We support this only for PR */ 532 r = !hv_enabled; 533 break; 534 #ifdef CONFIG_KVM_MPIC 535 case KVM_CAP_IRQ_MPIC: 536 r = 1; 537 break; 538 #endif 539 540 #ifdef CONFIG_PPC_BOOK3S_64 541 case KVM_CAP_SPAPR_TCE: 542 case KVM_CAP_SPAPR_TCE_64: 543 /* fallthrough */ 544 case KVM_CAP_SPAPR_TCE_VFIO: 545 case KVM_CAP_PPC_RTAS: 546 case KVM_CAP_PPC_FIXUP_HCALL: 547 case KVM_CAP_PPC_ENABLE_HCALL: 548 #ifdef CONFIG_KVM_XICS 549 case KVM_CAP_IRQ_XICS: 550 #endif 551 r = 1; 552 break; 553 554 case KVM_CAP_PPC_ALLOC_HTAB: 555 r = hv_enabled; 556 break; 557 #endif /* CONFIG_PPC_BOOK3S_64 */ 558 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 559 case KVM_CAP_PPC_SMT: 560 r = 0; 561 if (kvm) { 562 if (kvm->arch.emul_smt_mode > 1) 563 r = kvm->arch.emul_smt_mode; 564 else 565 r = kvm->arch.smt_mode; 566 } else if (hv_enabled) { 567 if (cpu_has_feature(CPU_FTR_ARCH_300)) 568 r = 1; 569 else 570 r = threads_per_subcore; 571 } 572 break; 573 case KVM_CAP_PPC_SMT_POSSIBLE: 574 r = 1; 575 if (hv_enabled) { 576 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 577 r = ((threads_per_subcore << 1) - 1); 578 else 579 /* P9 can emulate dbells, so allow any mode */ 580 r = 8 | 4 | 2 | 1; 581 } 582 break; 583 case KVM_CAP_PPC_RMA: 584 r = 0; 585 break; 586 case KVM_CAP_PPC_HWRNG: 587 r = kvmppc_hwrng_present(); 588 break; 589 case KVM_CAP_PPC_MMU_RADIX: 590 r = !!(hv_enabled && radix_enabled()); 591 break; 592 case KVM_CAP_PPC_MMU_HASH_V3: 593 r = !!(hv_enabled && cpu_has_feature(CPU_FTR_ARCH_300)); 594 break; 595 #endif 596 case KVM_CAP_SYNC_MMU: 597 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 598 r = hv_enabled; 599 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER) 600 r = 1; 601 #else 602 r = 0; 603 #endif 604 break; 605 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 606 case KVM_CAP_PPC_HTAB_FD: 607 r = hv_enabled; 608 break; 609 #endif 610 case KVM_CAP_NR_VCPUS: 611 /* 612 * Recommending a number of CPUs is somewhat arbitrary; we 613 * return the number of present CPUs for -HV (since a host 614 * will have secondary threads "offline"), and for other KVM 615 * implementations just count online CPUs. 616 */ 617 if (hv_enabled) 618 r = num_present_cpus(); 619 else 620 r = num_online_cpus(); 621 break; 622 case KVM_CAP_NR_MEMSLOTS: 623 r = KVM_USER_MEM_SLOTS; 624 break; 625 case KVM_CAP_MAX_VCPUS: 626 r = KVM_MAX_VCPUS; 627 break; 628 #ifdef CONFIG_PPC_BOOK3S_64 629 case KVM_CAP_PPC_GET_SMMU_INFO: 630 r = 1; 631 break; 632 case KVM_CAP_SPAPR_MULTITCE: 633 r = 1; 634 break; 635 case KVM_CAP_SPAPR_RESIZE_HPT: 636 /* Disable this on POWER9 until code handles new HPTE format */ 637 r = !!hv_enabled && !cpu_has_feature(CPU_FTR_ARCH_300); 638 break; 639 #endif 640 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 641 case KVM_CAP_PPC_FWNMI: 642 r = hv_enabled; 643 break; 644 #endif 645 case KVM_CAP_PPC_HTM: 646 r = hv_enabled && 647 (cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM_COMP); 648 break; 649 default: 650 r = 0; 651 break; 652 } 653 return r; 654 655 } 656 657 long kvm_arch_dev_ioctl(struct file *filp, 658 unsigned int ioctl, unsigned long arg) 659 { 660 return -EINVAL; 661 } 662 663 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, 664 struct kvm_memory_slot *dont) 665 { 666 kvmppc_core_free_memslot(kvm, free, dont); 667 } 668 669 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, 670 unsigned long npages) 671 { 672 return kvmppc_core_create_memslot(kvm, slot, npages); 673 } 674 675 int kvm_arch_prepare_memory_region(struct kvm *kvm, 676 struct kvm_memory_slot *memslot, 677 const struct kvm_userspace_memory_region *mem, 678 enum kvm_mr_change change) 679 { 680 return kvmppc_core_prepare_memory_region(kvm, memslot, mem); 681 } 682 683 void kvm_arch_commit_memory_region(struct kvm *kvm, 684 const struct kvm_userspace_memory_region *mem, 685 const struct kvm_memory_slot *old, 686 const struct kvm_memory_slot *new, 687 enum kvm_mr_change change) 688 { 689 kvmppc_core_commit_memory_region(kvm, mem, old, new); 690 } 691 692 void kvm_arch_flush_shadow_memslot(struct kvm *kvm, 693 struct kvm_memory_slot *slot) 694 { 695 kvmppc_core_flush_memslot(kvm, slot); 696 } 697 698 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) 699 { 700 struct kvm_vcpu *vcpu; 701 vcpu = kvmppc_core_vcpu_create(kvm, id); 702 if (!IS_ERR(vcpu)) { 703 vcpu->arch.wqp = &vcpu->wq; 704 kvmppc_create_vcpu_debugfs(vcpu, id); 705 } 706 return vcpu; 707 } 708 709 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 710 { 711 } 712 713 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) 714 { 715 /* Make sure we're not using the vcpu anymore */ 716 hrtimer_cancel(&vcpu->arch.dec_timer); 717 718 kvmppc_remove_vcpu_debugfs(vcpu); 719 720 switch (vcpu->arch.irq_type) { 721 case KVMPPC_IRQ_MPIC: 722 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu); 723 break; 724 case KVMPPC_IRQ_XICS: 725 if (xive_enabled()) 726 kvmppc_xive_cleanup_vcpu(vcpu); 727 else 728 kvmppc_xics_free_icp(vcpu); 729 break; 730 } 731 732 kvmppc_core_vcpu_free(vcpu); 733 } 734 735 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 736 { 737 kvm_arch_vcpu_free(vcpu); 738 } 739 740 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) 741 { 742 return kvmppc_core_pending_dec(vcpu); 743 } 744 745 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer) 746 { 747 struct kvm_vcpu *vcpu; 748 749 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer); 750 kvmppc_decrementer_func(vcpu); 751 752 return HRTIMER_NORESTART; 753 } 754 755 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) 756 { 757 int ret; 758 759 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS); 760 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup; 761 vcpu->arch.dec_expires = ~(u64)0; 762 763 #ifdef CONFIG_KVM_EXIT_TIMING 764 mutex_init(&vcpu->arch.exit_timing_lock); 765 #endif 766 ret = kvmppc_subarch_vcpu_init(vcpu); 767 return ret; 768 } 769 770 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) 771 { 772 kvmppc_mmu_destroy(vcpu); 773 kvmppc_subarch_vcpu_uninit(vcpu); 774 } 775 776 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 777 { 778 #ifdef CONFIG_BOOKE 779 /* 780 * vrsave (formerly usprg0) isn't used by Linux, but may 781 * be used by the guest. 782 * 783 * On non-booke this is associated with Altivec and 784 * is handled by code in book3s.c. 785 */ 786 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave); 787 #endif 788 kvmppc_core_vcpu_load(vcpu, cpu); 789 } 790 791 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 792 { 793 kvmppc_core_vcpu_put(vcpu); 794 #ifdef CONFIG_BOOKE 795 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE); 796 #endif 797 } 798 799 /* 800 * irq_bypass_add_producer and irq_bypass_del_producer are only 801 * useful if the architecture supports PCI passthrough. 802 * irq_bypass_stop and irq_bypass_start are not needed and so 803 * kvm_ops are not defined for them. 804 */ 805 bool kvm_arch_has_irq_bypass(void) 806 { 807 return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) || 808 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer)); 809 } 810 811 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, 812 struct irq_bypass_producer *prod) 813 { 814 struct kvm_kernel_irqfd *irqfd = 815 container_of(cons, struct kvm_kernel_irqfd, consumer); 816 struct kvm *kvm = irqfd->kvm; 817 818 if (kvm->arch.kvm_ops->irq_bypass_add_producer) 819 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod); 820 821 return 0; 822 } 823 824 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, 825 struct irq_bypass_producer *prod) 826 { 827 struct kvm_kernel_irqfd *irqfd = 828 container_of(cons, struct kvm_kernel_irqfd, consumer); 829 struct kvm *kvm = irqfd->kvm; 830 831 if (kvm->arch.kvm_ops->irq_bypass_del_producer) 832 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod); 833 } 834 835 #ifdef CONFIG_VSX 836 static inline int kvmppc_get_vsr_dword_offset(int index) 837 { 838 int offset; 839 840 if ((index != 0) && (index != 1)) 841 return -1; 842 843 #ifdef __BIG_ENDIAN 844 offset = index; 845 #else 846 offset = 1 - index; 847 #endif 848 849 return offset; 850 } 851 852 static inline int kvmppc_get_vsr_word_offset(int index) 853 { 854 int offset; 855 856 if ((index > 3) || (index < 0)) 857 return -1; 858 859 #ifdef __BIG_ENDIAN 860 offset = index; 861 #else 862 offset = 3 - index; 863 #endif 864 return offset; 865 } 866 867 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu, 868 u64 gpr) 869 { 870 union kvmppc_one_reg val; 871 int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset); 872 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 873 874 if (offset == -1) 875 return; 876 877 if (vcpu->arch.mmio_vsx_tx_sx_enabled) { 878 val.vval = VCPU_VSX_VR(vcpu, index); 879 val.vsxval[offset] = gpr; 880 VCPU_VSX_VR(vcpu, index) = val.vval; 881 } else { 882 VCPU_VSX_FPR(vcpu, index, offset) = gpr; 883 } 884 } 885 886 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu, 887 u64 gpr) 888 { 889 union kvmppc_one_reg val; 890 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 891 892 if (vcpu->arch.mmio_vsx_tx_sx_enabled) { 893 val.vval = VCPU_VSX_VR(vcpu, index); 894 val.vsxval[0] = gpr; 895 val.vsxval[1] = gpr; 896 VCPU_VSX_VR(vcpu, index) = val.vval; 897 } else { 898 VCPU_VSX_FPR(vcpu, index, 0) = gpr; 899 VCPU_VSX_FPR(vcpu, index, 1) = gpr; 900 } 901 } 902 903 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu, 904 u32 gpr32) 905 { 906 union kvmppc_one_reg val; 907 int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset); 908 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 909 int dword_offset, word_offset; 910 911 if (offset == -1) 912 return; 913 914 if (vcpu->arch.mmio_vsx_tx_sx_enabled) { 915 val.vval = VCPU_VSX_VR(vcpu, index); 916 val.vsx32val[offset] = gpr32; 917 VCPU_VSX_VR(vcpu, index) = val.vval; 918 } else { 919 dword_offset = offset / 2; 920 word_offset = offset % 2; 921 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset); 922 val.vsx32val[word_offset] = gpr32; 923 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0]; 924 } 925 } 926 #endif /* CONFIG_VSX */ 927 928 #ifdef CONFIG_PPC_FPU 929 static inline u64 sp_to_dp(u32 fprs) 930 { 931 u64 fprd; 932 933 preempt_disable(); 934 enable_kernel_fp(); 935 asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs) 936 : "fr0"); 937 preempt_enable(); 938 return fprd; 939 } 940 941 static inline u32 dp_to_sp(u64 fprd) 942 { 943 u32 fprs; 944 945 preempt_disable(); 946 enable_kernel_fp(); 947 asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd) 948 : "fr0"); 949 preempt_enable(); 950 return fprs; 951 } 952 953 #else 954 #define sp_to_dp(x) (x) 955 #define dp_to_sp(x) (x) 956 #endif /* CONFIG_PPC_FPU */ 957 958 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu, 959 struct kvm_run *run) 960 { 961 u64 uninitialized_var(gpr); 962 963 if (run->mmio.len > sizeof(gpr)) { 964 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len); 965 return; 966 } 967 968 if (!vcpu->arch.mmio_host_swabbed) { 969 switch (run->mmio.len) { 970 case 8: gpr = *(u64 *)run->mmio.data; break; 971 case 4: gpr = *(u32 *)run->mmio.data; break; 972 case 2: gpr = *(u16 *)run->mmio.data; break; 973 case 1: gpr = *(u8 *)run->mmio.data; break; 974 } 975 } else { 976 switch (run->mmio.len) { 977 case 8: gpr = swab64(*(u64 *)run->mmio.data); break; 978 case 4: gpr = swab32(*(u32 *)run->mmio.data); break; 979 case 2: gpr = swab16(*(u16 *)run->mmio.data); break; 980 case 1: gpr = *(u8 *)run->mmio.data; break; 981 } 982 } 983 984 /* conversion between single and double precision */ 985 if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4)) 986 gpr = sp_to_dp(gpr); 987 988 if (vcpu->arch.mmio_sign_extend) { 989 switch (run->mmio.len) { 990 #ifdef CONFIG_PPC64 991 case 4: 992 gpr = (s64)(s32)gpr; 993 break; 994 #endif 995 case 2: 996 gpr = (s64)(s16)gpr; 997 break; 998 case 1: 999 gpr = (s64)(s8)gpr; 1000 break; 1001 } 1002 } 1003 1004 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) { 1005 case KVM_MMIO_REG_GPR: 1006 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr); 1007 break; 1008 case KVM_MMIO_REG_FPR: 1009 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr; 1010 break; 1011 #ifdef CONFIG_PPC_BOOK3S 1012 case KVM_MMIO_REG_QPR: 1013 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr; 1014 break; 1015 case KVM_MMIO_REG_FQPR: 1016 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr; 1017 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr; 1018 break; 1019 #endif 1020 #ifdef CONFIG_VSX 1021 case KVM_MMIO_REG_VSX: 1022 if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_DWORD) 1023 kvmppc_set_vsr_dword(vcpu, gpr); 1024 else if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_WORD) 1025 kvmppc_set_vsr_word(vcpu, gpr); 1026 else if (vcpu->arch.mmio_vsx_copy_type == 1027 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP) 1028 kvmppc_set_vsr_dword_dump(vcpu, gpr); 1029 break; 1030 #endif 1031 default: 1032 BUG(); 1033 } 1034 } 1035 1036 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu, 1037 unsigned int rt, unsigned int bytes, 1038 int is_default_endian, int sign_extend) 1039 { 1040 int idx, ret; 1041 bool host_swabbed; 1042 1043 /* Pity C doesn't have a logical XOR operator */ 1044 if (kvmppc_need_byteswap(vcpu)) { 1045 host_swabbed = is_default_endian; 1046 } else { 1047 host_swabbed = !is_default_endian; 1048 } 1049 1050 if (bytes > sizeof(run->mmio.data)) { 1051 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__, 1052 run->mmio.len); 1053 } 1054 1055 run->mmio.phys_addr = vcpu->arch.paddr_accessed; 1056 run->mmio.len = bytes; 1057 run->mmio.is_write = 0; 1058 1059 vcpu->arch.io_gpr = rt; 1060 vcpu->arch.mmio_host_swabbed = host_swabbed; 1061 vcpu->mmio_needed = 1; 1062 vcpu->mmio_is_write = 0; 1063 vcpu->arch.mmio_sign_extend = sign_extend; 1064 1065 idx = srcu_read_lock(&vcpu->kvm->srcu); 1066 1067 ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr, 1068 bytes, &run->mmio.data); 1069 1070 srcu_read_unlock(&vcpu->kvm->srcu, idx); 1071 1072 if (!ret) { 1073 kvmppc_complete_mmio_load(vcpu, run); 1074 vcpu->mmio_needed = 0; 1075 return EMULATE_DONE; 1076 } 1077 1078 return EMULATE_DO_MMIO; 1079 } 1080 1081 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu, 1082 unsigned int rt, unsigned int bytes, 1083 int is_default_endian) 1084 { 1085 return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0); 1086 } 1087 EXPORT_SYMBOL_GPL(kvmppc_handle_load); 1088 1089 /* Same as above, but sign extends */ 1090 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu, 1091 unsigned int rt, unsigned int bytes, 1092 int is_default_endian) 1093 { 1094 return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1); 1095 } 1096 1097 #ifdef CONFIG_VSX 1098 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu, 1099 unsigned int rt, unsigned int bytes, 1100 int is_default_endian, int mmio_sign_extend) 1101 { 1102 enum emulation_result emulated = EMULATE_DONE; 1103 1104 /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */ 1105 if ( (vcpu->arch.mmio_vsx_copy_nums > 4) || 1106 (vcpu->arch.mmio_vsx_copy_nums < 0) ) { 1107 return EMULATE_FAIL; 1108 } 1109 1110 while (vcpu->arch.mmio_vsx_copy_nums) { 1111 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes, 1112 is_default_endian, mmio_sign_extend); 1113 1114 if (emulated != EMULATE_DONE) 1115 break; 1116 1117 vcpu->arch.paddr_accessed += run->mmio.len; 1118 1119 vcpu->arch.mmio_vsx_copy_nums--; 1120 vcpu->arch.mmio_vsx_offset++; 1121 } 1122 return emulated; 1123 } 1124 #endif /* CONFIG_VSX */ 1125 1126 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu, 1127 u64 val, unsigned int bytes, int is_default_endian) 1128 { 1129 void *data = run->mmio.data; 1130 int idx, ret; 1131 bool host_swabbed; 1132 1133 /* Pity C doesn't have a logical XOR operator */ 1134 if (kvmppc_need_byteswap(vcpu)) { 1135 host_swabbed = is_default_endian; 1136 } else { 1137 host_swabbed = !is_default_endian; 1138 } 1139 1140 if (bytes > sizeof(run->mmio.data)) { 1141 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__, 1142 run->mmio.len); 1143 } 1144 1145 run->mmio.phys_addr = vcpu->arch.paddr_accessed; 1146 run->mmio.len = bytes; 1147 run->mmio.is_write = 1; 1148 vcpu->mmio_needed = 1; 1149 vcpu->mmio_is_write = 1; 1150 1151 if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4)) 1152 val = dp_to_sp(val); 1153 1154 /* Store the value at the lowest bytes in 'data'. */ 1155 if (!host_swabbed) { 1156 switch (bytes) { 1157 case 8: *(u64 *)data = val; break; 1158 case 4: *(u32 *)data = val; break; 1159 case 2: *(u16 *)data = val; break; 1160 case 1: *(u8 *)data = val; break; 1161 } 1162 } else { 1163 switch (bytes) { 1164 case 8: *(u64 *)data = swab64(val); break; 1165 case 4: *(u32 *)data = swab32(val); break; 1166 case 2: *(u16 *)data = swab16(val); break; 1167 case 1: *(u8 *)data = val; break; 1168 } 1169 } 1170 1171 idx = srcu_read_lock(&vcpu->kvm->srcu); 1172 1173 ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr, 1174 bytes, &run->mmio.data); 1175 1176 srcu_read_unlock(&vcpu->kvm->srcu, idx); 1177 1178 if (!ret) { 1179 vcpu->mmio_needed = 0; 1180 return EMULATE_DONE; 1181 } 1182 1183 return EMULATE_DO_MMIO; 1184 } 1185 EXPORT_SYMBOL_GPL(kvmppc_handle_store); 1186 1187 #ifdef CONFIG_VSX 1188 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val) 1189 { 1190 u32 dword_offset, word_offset; 1191 union kvmppc_one_reg reg; 1192 int vsx_offset = 0; 1193 int copy_type = vcpu->arch.mmio_vsx_copy_type; 1194 int result = 0; 1195 1196 switch (copy_type) { 1197 case KVMPPC_VSX_COPY_DWORD: 1198 vsx_offset = 1199 kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset); 1200 1201 if (vsx_offset == -1) { 1202 result = -1; 1203 break; 1204 } 1205 1206 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) { 1207 *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset); 1208 } else { 1209 reg.vval = VCPU_VSX_VR(vcpu, rs); 1210 *val = reg.vsxval[vsx_offset]; 1211 } 1212 break; 1213 1214 case KVMPPC_VSX_COPY_WORD: 1215 vsx_offset = 1216 kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset); 1217 1218 if (vsx_offset == -1) { 1219 result = -1; 1220 break; 1221 } 1222 1223 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) { 1224 dword_offset = vsx_offset / 2; 1225 word_offset = vsx_offset % 2; 1226 reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset); 1227 *val = reg.vsx32val[word_offset]; 1228 } else { 1229 reg.vval = VCPU_VSX_VR(vcpu, rs); 1230 *val = reg.vsx32val[vsx_offset]; 1231 } 1232 break; 1233 1234 default: 1235 result = -1; 1236 break; 1237 } 1238 1239 return result; 1240 } 1241 1242 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu, 1243 int rs, unsigned int bytes, int is_default_endian) 1244 { 1245 u64 val; 1246 enum emulation_result emulated = EMULATE_DONE; 1247 1248 vcpu->arch.io_gpr = rs; 1249 1250 /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */ 1251 if ( (vcpu->arch.mmio_vsx_copy_nums > 4) || 1252 (vcpu->arch.mmio_vsx_copy_nums < 0) ) { 1253 return EMULATE_FAIL; 1254 } 1255 1256 while (vcpu->arch.mmio_vsx_copy_nums) { 1257 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1) 1258 return EMULATE_FAIL; 1259 1260 emulated = kvmppc_handle_store(run, vcpu, 1261 val, bytes, is_default_endian); 1262 1263 if (emulated != EMULATE_DONE) 1264 break; 1265 1266 vcpu->arch.paddr_accessed += run->mmio.len; 1267 1268 vcpu->arch.mmio_vsx_copy_nums--; 1269 vcpu->arch.mmio_vsx_offset++; 1270 } 1271 1272 return emulated; 1273 } 1274 1275 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu, 1276 struct kvm_run *run) 1277 { 1278 enum emulation_result emulated = EMULATE_FAIL; 1279 int r; 1280 1281 vcpu->arch.paddr_accessed += run->mmio.len; 1282 1283 if (!vcpu->mmio_is_write) { 1284 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr, 1285 run->mmio.len, 1, vcpu->arch.mmio_sign_extend); 1286 } else { 1287 emulated = kvmppc_handle_vsx_store(run, vcpu, 1288 vcpu->arch.io_gpr, run->mmio.len, 1); 1289 } 1290 1291 switch (emulated) { 1292 case EMULATE_DO_MMIO: 1293 run->exit_reason = KVM_EXIT_MMIO; 1294 r = RESUME_HOST; 1295 break; 1296 case EMULATE_FAIL: 1297 pr_info("KVM: MMIO emulation failed (VSX repeat)\n"); 1298 run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 1299 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; 1300 r = RESUME_HOST; 1301 break; 1302 default: 1303 r = RESUME_GUEST; 1304 break; 1305 } 1306 return r; 1307 } 1308 #endif /* CONFIG_VSX */ 1309 1310 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg) 1311 { 1312 int r = 0; 1313 union kvmppc_one_reg val; 1314 int size; 1315 1316 size = one_reg_size(reg->id); 1317 if (size > sizeof(val)) 1318 return -EINVAL; 1319 1320 r = kvmppc_get_one_reg(vcpu, reg->id, &val); 1321 if (r == -EINVAL) { 1322 r = 0; 1323 switch (reg->id) { 1324 #ifdef CONFIG_ALTIVEC 1325 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31: 1326 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1327 r = -ENXIO; 1328 break; 1329 } 1330 val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0]; 1331 break; 1332 case KVM_REG_PPC_VSCR: 1333 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1334 r = -ENXIO; 1335 break; 1336 } 1337 val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]); 1338 break; 1339 case KVM_REG_PPC_VRSAVE: 1340 val = get_reg_val(reg->id, vcpu->arch.vrsave); 1341 break; 1342 #endif /* CONFIG_ALTIVEC */ 1343 default: 1344 r = -EINVAL; 1345 break; 1346 } 1347 } 1348 1349 if (r) 1350 return r; 1351 1352 if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size)) 1353 r = -EFAULT; 1354 1355 return r; 1356 } 1357 1358 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg) 1359 { 1360 int r; 1361 union kvmppc_one_reg val; 1362 int size; 1363 1364 size = one_reg_size(reg->id); 1365 if (size > sizeof(val)) 1366 return -EINVAL; 1367 1368 if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size)) 1369 return -EFAULT; 1370 1371 r = kvmppc_set_one_reg(vcpu, reg->id, &val); 1372 if (r == -EINVAL) { 1373 r = 0; 1374 switch (reg->id) { 1375 #ifdef CONFIG_ALTIVEC 1376 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31: 1377 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1378 r = -ENXIO; 1379 break; 1380 } 1381 vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval; 1382 break; 1383 case KVM_REG_PPC_VSCR: 1384 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1385 r = -ENXIO; 1386 break; 1387 } 1388 vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val); 1389 break; 1390 case KVM_REG_PPC_VRSAVE: 1391 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1392 r = -ENXIO; 1393 break; 1394 } 1395 vcpu->arch.vrsave = set_reg_val(reg->id, val); 1396 break; 1397 #endif /* CONFIG_ALTIVEC */ 1398 default: 1399 r = -EINVAL; 1400 break; 1401 } 1402 } 1403 1404 return r; 1405 } 1406 1407 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) 1408 { 1409 int r; 1410 sigset_t sigsaved; 1411 1412 if (vcpu->mmio_needed) { 1413 vcpu->mmio_needed = 0; 1414 if (!vcpu->mmio_is_write) 1415 kvmppc_complete_mmio_load(vcpu, run); 1416 #ifdef CONFIG_VSX 1417 if (vcpu->arch.mmio_vsx_copy_nums > 0) { 1418 vcpu->arch.mmio_vsx_copy_nums--; 1419 vcpu->arch.mmio_vsx_offset++; 1420 } 1421 1422 if (vcpu->arch.mmio_vsx_copy_nums > 0) { 1423 r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run); 1424 if (r == RESUME_HOST) { 1425 vcpu->mmio_needed = 1; 1426 return r; 1427 } 1428 } 1429 #endif 1430 } else if (vcpu->arch.osi_needed) { 1431 u64 *gprs = run->osi.gprs; 1432 int i; 1433 1434 for (i = 0; i < 32; i++) 1435 kvmppc_set_gpr(vcpu, i, gprs[i]); 1436 vcpu->arch.osi_needed = 0; 1437 } else if (vcpu->arch.hcall_needed) { 1438 int i; 1439 1440 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret); 1441 for (i = 0; i < 9; ++i) 1442 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]); 1443 vcpu->arch.hcall_needed = 0; 1444 #ifdef CONFIG_BOOKE 1445 } else if (vcpu->arch.epr_needed) { 1446 kvmppc_set_epr(vcpu, run->epr.epr); 1447 vcpu->arch.epr_needed = 0; 1448 #endif 1449 } 1450 1451 if (vcpu->sigset_active) 1452 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); 1453 1454 if (run->immediate_exit) 1455 r = -EINTR; 1456 else 1457 r = kvmppc_vcpu_run(run, vcpu); 1458 1459 if (vcpu->sigset_active) 1460 sigprocmask(SIG_SETMASK, &sigsaved, NULL); 1461 1462 return r; 1463 } 1464 1465 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq) 1466 { 1467 if (irq->irq == KVM_INTERRUPT_UNSET) { 1468 kvmppc_core_dequeue_external(vcpu); 1469 return 0; 1470 } 1471 1472 kvmppc_core_queue_external(vcpu, irq); 1473 1474 kvm_vcpu_kick(vcpu); 1475 1476 return 0; 1477 } 1478 1479 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, 1480 struct kvm_enable_cap *cap) 1481 { 1482 int r; 1483 1484 if (cap->flags) 1485 return -EINVAL; 1486 1487 switch (cap->cap) { 1488 case KVM_CAP_PPC_OSI: 1489 r = 0; 1490 vcpu->arch.osi_enabled = true; 1491 break; 1492 case KVM_CAP_PPC_PAPR: 1493 r = 0; 1494 vcpu->arch.papr_enabled = true; 1495 break; 1496 case KVM_CAP_PPC_EPR: 1497 r = 0; 1498 if (cap->args[0]) 1499 vcpu->arch.epr_flags |= KVMPPC_EPR_USER; 1500 else 1501 vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER; 1502 break; 1503 #ifdef CONFIG_BOOKE 1504 case KVM_CAP_PPC_BOOKE_WATCHDOG: 1505 r = 0; 1506 vcpu->arch.watchdog_enabled = true; 1507 break; 1508 #endif 1509 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC) 1510 case KVM_CAP_SW_TLB: { 1511 struct kvm_config_tlb cfg; 1512 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0]; 1513 1514 r = -EFAULT; 1515 if (copy_from_user(&cfg, user_ptr, sizeof(cfg))) 1516 break; 1517 1518 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg); 1519 break; 1520 } 1521 #endif 1522 #ifdef CONFIG_KVM_MPIC 1523 case KVM_CAP_IRQ_MPIC: { 1524 struct fd f; 1525 struct kvm_device *dev; 1526 1527 r = -EBADF; 1528 f = fdget(cap->args[0]); 1529 if (!f.file) 1530 break; 1531 1532 r = -EPERM; 1533 dev = kvm_device_from_filp(f.file); 1534 if (dev) 1535 r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]); 1536 1537 fdput(f); 1538 break; 1539 } 1540 #endif 1541 #ifdef CONFIG_KVM_XICS 1542 case KVM_CAP_IRQ_XICS: { 1543 struct fd f; 1544 struct kvm_device *dev; 1545 1546 r = -EBADF; 1547 f = fdget(cap->args[0]); 1548 if (!f.file) 1549 break; 1550 1551 r = -EPERM; 1552 dev = kvm_device_from_filp(f.file); 1553 if (dev) { 1554 if (xive_enabled()) 1555 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]); 1556 else 1557 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]); 1558 } 1559 1560 fdput(f); 1561 break; 1562 } 1563 #endif /* CONFIG_KVM_XICS */ 1564 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 1565 case KVM_CAP_PPC_FWNMI: 1566 r = -EINVAL; 1567 if (!is_kvmppc_hv_enabled(vcpu->kvm)) 1568 break; 1569 r = 0; 1570 vcpu->kvm->arch.fwnmi_enabled = true; 1571 break; 1572 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */ 1573 default: 1574 r = -EINVAL; 1575 break; 1576 } 1577 1578 if (!r) 1579 r = kvmppc_sanity_check(vcpu); 1580 1581 return r; 1582 } 1583 1584 bool kvm_arch_intc_initialized(struct kvm *kvm) 1585 { 1586 #ifdef CONFIG_KVM_MPIC 1587 if (kvm->arch.mpic) 1588 return true; 1589 #endif 1590 #ifdef CONFIG_KVM_XICS 1591 if (kvm->arch.xics || kvm->arch.xive) 1592 return true; 1593 #endif 1594 return false; 1595 } 1596 1597 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 1598 struct kvm_mp_state *mp_state) 1599 { 1600 return -EINVAL; 1601 } 1602 1603 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 1604 struct kvm_mp_state *mp_state) 1605 { 1606 return -EINVAL; 1607 } 1608 1609 long kvm_arch_vcpu_ioctl(struct file *filp, 1610 unsigned int ioctl, unsigned long arg) 1611 { 1612 struct kvm_vcpu *vcpu = filp->private_data; 1613 void __user *argp = (void __user *)arg; 1614 long r; 1615 1616 switch (ioctl) { 1617 case KVM_INTERRUPT: { 1618 struct kvm_interrupt irq; 1619 r = -EFAULT; 1620 if (copy_from_user(&irq, argp, sizeof(irq))) 1621 goto out; 1622 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq); 1623 goto out; 1624 } 1625 1626 case KVM_ENABLE_CAP: 1627 { 1628 struct kvm_enable_cap cap; 1629 r = -EFAULT; 1630 if (copy_from_user(&cap, argp, sizeof(cap))) 1631 goto out; 1632 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); 1633 break; 1634 } 1635 1636 case KVM_SET_ONE_REG: 1637 case KVM_GET_ONE_REG: 1638 { 1639 struct kvm_one_reg reg; 1640 r = -EFAULT; 1641 if (copy_from_user(®, argp, sizeof(reg))) 1642 goto out; 1643 if (ioctl == KVM_SET_ONE_REG) 1644 r = kvm_vcpu_ioctl_set_one_reg(vcpu, ®); 1645 else 1646 r = kvm_vcpu_ioctl_get_one_reg(vcpu, ®); 1647 break; 1648 } 1649 1650 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC) 1651 case KVM_DIRTY_TLB: { 1652 struct kvm_dirty_tlb dirty; 1653 r = -EFAULT; 1654 if (copy_from_user(&dirty, argp, sizeof(dirty))) 1655 goto out; 1656 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty); 1657 break; 1658 } 1659 #endif 1660 default: 1661 r = -EINVAL; 1662 } 1663 1664 out: 1665 return r; 1666 } 1667 1668 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 1669 { 1670 return VM_FAULT_SIGBUS; 1671 } 1672 1673 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo) 1674 { 1675 u32 inst_nop = 0x60000000; 1676 #ifdef CONFIG_KVM_BOOKE_HV 1677 u32 inst_sc1 = 0x44000022; 1678 pvinfo->hcall[0] = cpu_to_be32(inst_sc1); 1679 pvinfo->hcall[1] = cpu_to_be32(inst_nop); 1680 pvinfo->hcall[2] = cpu_to_be32(inst_nop); 1681 pvinfo->hcall[3] = cpu_to_be32(inst_nop); 1682 #else 1683 u32 inst_lis = 0x3c000000; 1684 u32 inst_ori = 0x60000000; 1685 u32 inst_sc = 0x44000002; 1686 u32 inst_imm_mask = 0xffff; 1687 1688 /* 1689 * The hypercall to get into KVM from within guest context is as 1690 * follows: 1691 * 1692 * lis r0, r0, KVM_SC_MAGIC_R0@h 1693 * ori r0, KVM_SC_MAGIC_R0@l 1694 * sc 1695 * nop 1696 */ 1697 pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask)); 1698 pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask)); 1699 pvinfo->hcall[2] = cpu_to_be32(inst_sc); 1700 pvinfo->hcall[3] = cpu_to_be32(inst_nop); 1701 #endif 1702 1703 pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE; 1704 1705 return 0; 1706 } 1707 1708 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event, 1709 bool line_status) 1710 { 1711 if (!irqchip_in_kernel(kvm)) 1712 return -ENXIO; 1713 1714 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, 1715 irq_event->irq, irq_event->level, 1716 line_status); 1717 return 0; 1718 } 1719 1720 1721 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, 1722 struct kvm_enable_cap *cap) 1723 { 1724 int r; 1725 1726 if (cap->flags) 1727 return -EINVAL; 1728 1729 switch (cap->cap) { 1730 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER 1731 case KVM_CAP_PPC_ENABLE_HCALL: { 1732 unsigned long hcall = cap->args[0]; 1733 1734 r = -EINVAL; 1735 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) || 1736 cap->args[1] > 1) 1737 break; 1738 if (!kvmppc_book3s_hcall_implemented(kvm, hcall)) 1739 break; 1740 if (cap->args[1]) 1741 set_bit(hcall / 4, kvm->arch.enabled_hcalls); 1742 else 1743 clear_bit(hcall / 4, kvm->arch.enabled_hcalls); 1744 r = 0; 1745 break; 1746 } 1747 case KVM_CAP_PPC_SMT: { 1748 unsigned long mode = cap->args[0]; 1749 unsigned long flags = cap->args[1]; 1750 1751 r = -EINVAL; 1752 if (kvm->arch.kvm_ops->set_smt_mode) 1753 r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags); 1754 break; 1755 } 1756 #endif 1757 default: 1758 r = -EINVAL; 1759 break; 1760 } 1761 1762 return r; 1763 } 1764 1765 long kvm_arch_vm_ioctl(struct file *filp, 1766 unsigned int ioctl, unsigned long arg) 1767 { 1768 struct kvm *kvm __maybe_unused = filp->private_data; 1769 void __user *argp = (void __user *)arg; 1770 long r; 1771 1772 switch (ioctl) { 1773 case KVM_PPC_GET_PVINFO: { 1774 struct kvm_ppc_pvinfo pvinfo; 1775 memset(&pvinfo, 0, sizeof(pvinfo)); 1776 r = kvm_vm_ioctl_get_pvinfo(&pvinfo); 1777 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) { 1778 r = -EFAULT; 1779 goto out; 1780 } 1781 1782 break; 1783 } 1784 case KVM_ENABLE_CAP: 1785 { 1786 struct kvm_enable_cap cap; 1787 r = -EFAULT; 1788 if (copy_from_user(&cap, argp, sizeof(cap))) 1789 goto out; 1790 r = kvm_vm_ioctl_enable_cap(kvm, &cap); 1791 break; 1792 } 1793 #ifdef CONFIG_SPAPR_TCE_IOMMU 1794 case KVM_CREATE_SPAPR_TCE_64: { 1795 struct kvm_create_spapr_tce_64 create_tce_64; 1796 1797 r = -EFAULT; 1798 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64))) 1799 goto out; 1800 if (create_tce_64.flags) { 1801 r = -EINVAL; 1802 goto out; 1803 } 1804 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64); 1805 goto out; 1806 } 1807 case KVM_CREATE_SPAPR_TCE: { 1808 struct kvm_create_spapr_tce create_tce; 1809 struct kvm_create_spapr_tce_64 create_tce_64; 1810 1811 r = -EFAULT; 1812 if (copy_from_user(&create_tce, argp, sizeof(create_tce))) 1813 goto out; 1814 1815 create_tce_64.liobn = create_tce.liobn; 1816 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K; 1817 create_tce_64.offset = 0; 1818 create_tce_64.size = create_tce.window_size >> 1819 IOMMU_PAGE_SHIFT_4K; 1820 create_tce_64.flags = 0; 1821 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64); 1822 goto out; 1823 } 1824 #endif 1825 #ifdef CONFIG_PPC_BOOK3S_64 1826 case KVM_PPC_GET_SMMU_INFO: { 1827 struct kvm_ppc_smmu_info info; 1828 struct kvm *kvm = filp->private_data; 1829 1830 memset(&info, 0, sizeof(info)); 1831 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info); 1832 if (r >= 0 && copy_to_user(argp, &info, sizeof(info))) 1833 r = -EFAULT; 1834 break; 1835 } 1836 case KVM_PPC_RTAS_DEFINE_TOKEN: { 1837 struct kvm *kvm = filp->private_data; 1838 1839 r = kvm_vm_ioctl_rtas_define_token(kvm, argp); 1840 break; 1841 } 1842 case KVM_PPC_CONFIGURE_V3_MMU: { 1843 struct kvm *kvm = filp->private_data; 1844 struct kvm_ppc_mmuv3_cfg cfg; 1845 1846 r = -EINVAL; 1847 if (!kvm->arch.kvm_ops->configure_mmu) 1848 goto out; 1849 r = -EFAULT; 1850 if (copy_from_user(&cfg, argp, sizeof(cfg))) 1851 goto out; 1852 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg); 1853 break; 1854 } 1855 case KVM_PPC_GET_RMMU_INFO: { 1856 struct kvm *kvm = filp->private_data; 1857 struct kvm_ppc_rmmu_info info; 1858 1859 r = -EINVAL; 1860 if (!kvm->arch.kvm_ops->get_rmmu_info) 1861 goto out; 1862 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info); 1863 if (r >= 0 && copy_to_user(argp, &info, sizeof(info))) 1864 r = -EFAULT; 1865 break; 1866 } 1867 default: { 1868 struct kvm *kvm = filp->private_data; 1869 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg); 1870 } 1871 #else /* CONFIG_PPC_BOOK3S_64 */ 1872 default: 1873 r = -ENOTTY; 1874 #endif 1875 } 1876 out: 1877 return r; 1878 } 1879 1880 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)]; 1881 static unsigned long nr_lpids; 1882 1883 long kvmppc_alloc_lpid(void) 1884 { 1885 long lpid; 1886 1887 do { 1888 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS); 1889 if (lpid >= nr_lpids) { 1890 pr_err("%s: No LPIDs free\n", __func__); 1891 return -ENOMEM; 1892 } 1893 } while (test_and_set_bit(lpid, lpid_inuse)); 1894 1895 return lpid; 1896 } 1897 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid); 1898 1899 void kvmppc_claim_lpid(long lpid) 1900 { 1901 set_bit(lpid, lpid_inuse); 1902 } 1903 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid); 1904 1905 void kvmppc_free_lpid(long lpid) 1906 { 1907 clear_bit(lpid, lpid_inuse); 1908 } 1909 EXPORT_SYMBOL_GPL(kvmppc_free_lpid); 1910 1911 void kvmppc_init_lpid(unsigned long nr_lpids_param) 1912 { 1913 nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param); 1914 memset(lpid_inuse, 0, sizeof(lpid_inuse)); 1915 } 1916 EXPORT_SYMBOL_GPL(kvmppc_init_lpid); 1917 1918 int kvm_arch_init(void *opaque) 1919 { 1920 return 0; 1921 } 1922 1923 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr); 1924