1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * 4 * Copyright IBM Corp. 2007 5 * 6 * Authors: Hollis Blanchard <hollisb@us.ibm.com> 7 * Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com> 8 */ 9 10 #include <linux/errno.h> 11 #include <linux/err.h> 12 #include <linux/kvm_host.h> 13 #include <linux/vmalloc.h> 14 #include <linux/hrtimer.h> 15 #include <linux/sched/signal.h> 16 #include <linux/fs.h> 17 #include <linux/slab.h> 18 #include <linux/file.h> 19 #include <linux/module.h> 20 #include <linux/irqbypass.h> 21 #include <linux/kvm_irqfd.h> 22 #include <linux/of.h> 23 #include <asm/cputable.h> 24 #include <linux/uaccess.h> 25 #include <asm/kvm_ppc.h> 26 #include <asm/cputhreads.h> 27 #include <asm/irqflags.h> 28 #include <asm/iommu.h> 29 #include <asm/switch_to.h> 30 #include <asm/xive.h> 31 #ifdef CONFIG_PPC_PSERIES 32 #include <asm/hvcall.h> 33 #include <asm/plpar_wrappers.h> 34 #endif 35 #include <asm/ultravisor.h> 36 #include <asm/setup.h> 37 38 #include "timing.h" 39 #include "../mm/mmu_decl.h" 40 41 #define CREATE_TRACE_POINTS 42 #include "trace.h" 43 44 struct kvmppc_ops *kvmppc_hv_ops; 45 EXPORT_SYMBOL_GPL(kvmppc_hv_ops); 46 struct kvmppc_ops *kvmppc_pr_ops; 47 EXPORT_SYMBOL_GPL(kvmppc_pr_ops); 48 49 50 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) 51 { 52 return !!(v->arch.pending_exceptions) || kvm_request_pending(v); 53 } 54 55 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) 56 { 57 return kvm_arch_vcpu_runnable(vcpu); 58 } 59 60 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 61 { 62 return false; 63 } 64 65 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 66 { 67 return 1; 68 } 69 70 /* 71 * Common checks before entering the guest world. Call with interrupts 72 * disabled. 73 * 74 * returns: 75 * 76 * == 1 if we're ready to go into guest state 77 * <= 0 if we need to go back to the host with return value 78 */ 79 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu) 80 { 81 int r; 82 83 WARN_ON(irqs_disabled()); 84 hard_irq_disable(); 85 86 while (true) { 87 if (need_resched()) { 88 local_irq_enable(); 89 cond_resched(); 90 hard_irq_disable(); 91 continue; 92 } 93 94 if (signal_pending(current)) { 95 kvmppc_account_exit(vcpu, SIGNAL_EXITS); 96 vcpu->run->exit_reason = KVM_EXIT_INTR; 97 r = -EINTR; 98 break; 99 } 100 101 vcpu->mode = IN_GUEST_MODE; 102 103 /* 104 * Reading vcpu->requests must happen after setting vcpu->mode, 105 * so we don't miss a request because the requester sees 106 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests 107 * before next entering the guest (and thus doesn't IPI). 108 * This also orders the write to mode from any reads 109 * to the page tables done while the VCPU is running. 110 * Please see the comment in kvm_flush_remote_tlbs. 111 */ 112 smp_mb(); 113 114 if (kvm_request_pending(vcpu)) { 115 /* Make sure we process requests preemptable */ 116 local_irq_enable(); 117 trace_kvm_check_requests(vcpu); 118 r = kvmppc_core_check_requests(vcpu); 119 hard_irq_disable(); 120 if (r > 0) 121 continue; 122 break; 123 } 124 125 if (kvmppc_core_prepare_to_enter(vcpu)) { 126 /* interrupts got enabled in between, so we 127 are back at square 1 */ 128 continue; 129 } 130 131 guest_enter_irqoff(); 132 return 1; 133 } 134 135 /* return to host */ 136 local_irq_enable(); 137 return r; 138 } 139 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter); 140 141 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE) 142 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu) 143 { 144 struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared; 145 int i; 146 147 shared->sprg0 = swab64(shared->sprg0); 148 shared->sprg1 = swab64(shared->sprg1); 149 shared->sprg2 = swab64(shared->sprg2); 150 shared->sprg3 = swab64(shared->sprg3); 151 shared->srr0 = swab64(shared->srr0); 152 shared->srr1 = swab64(shared->srr1); 153 shared->dar = swab64(shared->dar); 154 shared->msr = swab64(shared->msr); 155 shared->dsisr = swab32(shared->dsisr); 156 shared->int_pending = swab32(shared->int_pending); 157 for (i = 0; i < ARRAY_SIZE(shared->sr); i++) 158 shared->sr[i] = swab32(shared->sr[i]); 159 } 160 #endif 161 162 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu) 163 { 164 int nr = kvmppc_get_gpr(vcpu, 11); 165 int r; 166 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3); 167 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4); 168 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5); 169 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6); 170 unsigned long r2 = 0; 171 172 if (!(kvmppc_get_msr(vcpu) & MSR_SF)) { 173 /* 32 bit mode */ 174 param1 &= 0xffffffff; 175 param2 &= 0xffffffff; 176 param3 &= 0xffffffff; 177 param4 &= 0xffffffff; 178 } 179 180 switch (nr) { 181 case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE): 182 { 183 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE) 184 /* Book3S can be little endian, find it out here */ 185 int shared_big_endian = true; 186 if (vcpu->arch.intr_msr & MSR_LE) 187 shared_big_endian = false; 188 if (shared_big_endian != vcpu->arch.shared_big_endian) 189 kvmppc_swab_shared(vcpu); 190 vcpu->arch.shared_big_endian = shared_big_endian; 191 #endif 192 193 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) { 194 /* 195 * Older versions of the Linux magic page code had 196 * a bug where they would map their trampoline code 197 * NX. If that's the case, remove !PR NX capability. 198 */ 199 vcpu->arch.disable_kernel_nx = true; 200 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 201 } 202 203 vcpu->arch.magic_page_pa = param1 & ~0xfffULL; 204 vcpu->arch.magic_page_ea = param2 & ~0xfffULL; 205 206 #ifdef CONFIG_PPC_64K_PAGES 207 /* 208 * Make sure our 4k magic page is in the same window of a 64k 209 * page within the guest and within the host's page. 210 */ 211 if ((vcpu->arch.magic_page_pa & 0xf000) != 212 ((ulong)vcpu->arch.shared & 0xf000)) { 213 void *old_shared = vcpu->arch.shared; 214 ulong shared = (ulong)vcpu->arch.shared; 215 void *new_shared; 216 217 shared &= PAGE_MASK; 218 shared |= vcpu->arch.magic_page_pa & 0xf000; 219 new_shared = (void*)shared; 220 memcpy(new_shared, old_shared, 0x1000); 221 vcpu->arch.shared = new_shared; 222 } 223 #endif 224 225 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7; 226 227 r = EV_SUCCESS; 228 break; 229 } 230 case KVM_HCALL_TOKEN(KVM_HC_FEATURES): 231 r = EV_SUCCESS; 232 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2) 233 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE); 234 #endif 235 236 /* Second return value is in r4 */ 237 break; 238 case EV_HCALL_TOKEN(EV_IDLE): 239 r = EV_SUCCESS; 240 kvm_vcpu_halt(vcpu); 241 break; 242 default: 243 r = EV_UNIMPLEMENTED; 244 break; 245 } 246 247 kvmppc_set_gpr(vcpu, 4, r2); 248 249 return r; 250 } 251 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv); 252 253 int kvmppc_sanity_check(struct kvm_vcpu *vcpu) 254 { 255 int r = false; 256 257 /* We have to know what CPU to virtualize */ 258 if (!vcpu->arch.pvr) 259 goto out; 260 261 /* PAPR only works with book3s_64 */ 262 if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled) 263 goto out; 264 265 /* HV KVM can only do PAPR mode for now */ 266 if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm)) 267 goto out; 268 269 #ifdef CONFIG_KVM_BOOKE_HV 270 if (!cpu_has_feature(CPU_FTR_EMB_HV)) 271 goto out; 272 #endif 273 274 r = true; 275 276 out: 277 vcpu->arch.sane = r; 278 return r ? 0 : -EINVAL; 279 } 280 EXPORT_SYMBOL_GPL(kvmppc_sanity_check); 281 282 int kvmppc_emulate_mmio(struct kvm_vcpu *vcpu) 283 { 284 enum emulation_result er; 285 int r; 286 287 er = kvmppc_emulate_loadstore(vcpu); 288 switch (er) { 289 case EMULATE_DONE: 290 /* Future optimization: only reload non-volatiles if they were 291 * actually modified. */ 292 r = RESUME_GUEST_NV; 293 break; 294 case EMULATE_AGAIN: 295 r = RESUME_GUEST; 296 break; 297 case EMULATE_DO_MMIO: 298 vcpu->run->exit_reason = KVM_EXIT_MMIO; 299 /* We must reload nonvolatiles because "update" load/store 300 * instructions modify register state. */ 301 /* Future optimization: only reload non-volatiles if they were 302 * actually modified. */ 303 r = RESUME_HOST_NV; 304 break; 305 case EMULATE_FAIL: 306 { 307 u32 last_inst; 308 309 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst); 310 kvm_debug_ratelimited("Guest access to device memory using unsupported instruction (opcode: %#08x)\n", 311 last_inst); 312 313 /* 314 * Injecting a Data Storage here is a bit more 315 * accurate since the instruction that caused the 316 * access could still be a valid one. 317 */ 318 if (!IS_ENABLED(CONFIG_BOOKE)) { 319 ulong dsisr = DSISR_BADACCESS; 320 321 if (vcpu->mmio_is_write) 322 dsisr |= DSISR_ISSTORE; 323 324 kvmppc_core_queue_data_storage(vcpu, vcpu->arch.vaddr_accessed, dsisr); 325 } else { 326 /* 327 * BookE does not send a SIGBUS on a bad 328 * fault, so use a Program interrupt instead 329 * to avoid a fault loop. 330 */ 331 kvmppc_core_queue_program(vcpu, 0); 332 } 333 334 r = RESUME_GUEST; 335 break; 336 } 337 default: 338 WARN_ON(1); 339 r = RESUME_GUEST; 340 } 341 342 return r; 343 } 344 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio); 345 346 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr, 347 bool data) 348 { 349 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK; 350 struct kvmppc_pte pte; 351 int r = -EINVAL; 352 353 vcpu->stat.st++; 354 355 if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->store_to_eaddr) 356 r = vcpu->kvm->arch.kvm_ops->store_to_eaddr(vcpu, eaddr, ptr, 357 size); 358 359 if ((!r) || (r == -EAGAIN)) 360 return r; 361 362 r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST, 363 XLATE_WRITE, &pte); 364 if (r < 0) 365 return r; 366 367 *eaddr = pte.raddr; 368 369 if (!pte.may_write) 370 return -EPERM; 371 372 /* Magic page override */ 373 if (kvmppc_supports_magic_page(vcpu) && mp_pa && 374 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) && 375 !(kvmppc_get_msr(vcpu) & MSR_PR)) { 376 void *magic = vcpu->arch.shared; 377 magic += pte.eaddr & 0xfff; 378 memcpy(magic, ptr, size); 379 return EMULATE_DONE; 380 } 381 382 if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size)) 383 return EMULATE_DO_MMIO; 384 385 return EMULATE_DONE; 386 } 387 EXPORT_SYMBOL_GPL(kvmppc_st); 388 389 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr, 390 bool data) 391 { 392 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK; 393 struct kvmppc_pte pte; 394 int rc = -EINVAL; 395 396 vcpu->stat.ld++; 397 398 if (vcpu->kvm->arch.kvm_ops && vcpu->kvm->arch.kvm_ops->load_from_eaddr) 399 rc = vcpu->kvm->arch.kvm_ops->load_from_eaddr(vcpu, eaddr, ptr, 400 size); 401 402 if ((!rc) || (rc == -EAGAIN)) 403 return rc; 404 405 rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST, 406 XLATE_READ, &pte); 407 if (rc) 408 return rc; 409 410 *eaddr = pte.raddr; 411 412 if (!pte.may_read) 413 return -EPERM; 414 415 if (!data && !pte.may_execute) 416 return -ENOEXEC; 417 418 /* Magic page override */ 419 if (kvmppc_supports_magic_page(vcpu) && mp_pa && 420 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) && 421 !(kvmppc_get_msr(vcpu) & MSR_PR)) { 422 void *magic = vcpu->arch.shared; 423 magic += pte.eaddr & 0xfff; 424 memcpy(ptr, magic, size); 425 return EMULATE_DONE; 426 } 427 428 kvm_vcpu_srcu_read_lock(vcpu); 429 rc = kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size); 430 kvm_vcpu_srcu_read_unlock(vcpu); 431 if (rc) 432 return EMULATE_DO_MMIO; 433 434 return EMULATE_DONE; 435 } 436 EXPORT_SYMBOL_GPL(kvmppc_ld); 437 438 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 439 { 440 struct kvmppc_ops *kvm_ops = NULL; 441 int r; 442 443 /* 444 * if we have both HV and PR enabled, default is HV 445 */ 446 if (type == 0) { 447 if (kvmppc_hv_ops) 448 kvm_ops = kvmppc_hv_ops; 449 else 450 kvm_ops = kvmppc_pr_ops; 451 if (!kvm_ops) 452 goto err_out; 453 } else if (type == KVM_VM_PPC_HV) { 454 if (!kvmppc_hv_ops) 455 goto err_out; 456 kvm_ops = kvmppc_hv_ops; 457 } else if (type == KVM_VM_PPC_PR) { 458 if (!kvmppc_pr_ops) 459 goto err_out; 460 kvm_ops = kvmppc_pr_ops; 461 } else 462 goto err_out; 463 464 if (!try_module_get(kvm_ops->owner)) 465 return -ENOENT; 466 467 kvm->arch.kvm_ops = kvm_ops; 468 r = kvmppc_core_init_vm(kvm); 469 if (r) 470 module_put(kvm_ops->owner); 471 return r; 472 err_out: 473 return -EINVAL; 474 } 475 476 void kvm_arch_destroy_vm(struct kvm *kvm) 477 { 478 #ifdef CONFIG_KVM_XICS 479 /* 480 * We call kick_all_cpus_sync() to ensure that all 481 * CPUs have executed any pending IPIs before we 482 * continue and free VCPUs structures below. 483 */ 484 if (is_kvmppc_hv_enabled(kvm)) 485 kick_all_cpus_sync(); 486 #endif 487 488 kvm_destroy_vcpus(kvm); 489 490 mutex_lock(&kvm->lock); 491 492 kvmppc_core_destroy_vm(kvm); 493 494 mutex_unlock(&kvm->lock); 495 496 /* drop the module reference */ 497 module_put(kvm->arch.kvm_ops->owner); 498 } 499 500 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 501 { 502 int r; 503 /* Assume we're using HV mode when the HV module is loaded */ 504 int hv_enabled = kvmppc_hv_ops ? 1 : 0; 505 506 if (kvm) { 507 /* 508 * Hooray - we know which VM type we're running on. Depend on 509 * that rather than the guess above. 510 */ 511 hv_enabled = is_kvmppc_hv_enabled(kvm); 512 } 513 514 switch (ext) { 515 #ifdef CONFIG_BOOKE 516 case KVM_CAP_PPC_BOOKE_SREGS: 517 case KVM_CAP_PPC_BOOKE_WATCHDOG: 518 case KVM_CAP_PPC_EPR: 519 #else 520 case KVM_CAP_PPC_SEGSTATE: 521 case KVM_CAP_PPC_HIOR: 522 case KVM_CAP_PPC_PAPR: 523 #endif 524 case KVM_CAP_PPC_UNSET_IRQ: 525 case KVM_CAP_PPC_IRQ_LEVEL: 526 case KVM_CAP_ENABLE_CAP: 527 case KVM_CAP_ONE_REG: 528 case KVM_CAP_IOEVENTFD: 529 case KVM_CAP_DEVICE_CTRL: 530 case KVM_CAP_IMMEDIATE_EXIT: 531 case KVM_CAP_SET_GUEST_DEBUG: 532 r = 1; 533 break; 534 case KVM_CAP_PPC_GUEST_DEBUG_SSTEP: 535 case KVM_CAP_PPC_PAIRED_SINGLES: 536 case KVM_CAP_PPC_OSI: 537 case KVM_CAP_PPC_GET_PVINFO: 538 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC) 539 case KVM_CAP_SW_TLB: 540 #endif 541 /* We support this only for PR */ 542 r = !hv_enabled; 543 break; 544 #ifdef CONFIG_KVM_MPIC 545 case KVM_CAP_IRQ_MPIC: 546 r = 1; 547 break; 548 #endif 549 550 #ifdef CONFIG_PPC_BOOK3S_64 551 case KVM_CAP_SPAPR_TCE: 552 case KVM_CAP_SPAPR_TCE_64: 553 r = 1; 554 break; 555 case KVM_CAP_SPAPR_TCE_VFIO: 556 r = !!cpu_has_feature(CPU_FTR_HVMODE); 557 break; 558 case KVM_CAP_PPC_RTAS: 559 case KVM_CAP_PPC_FIXUP_HCALL: 560 case KVM_CAP_PPC_ENABLE_HCALL: 561 #ifdef CONFIG_KVM_XICS 562 case KVM_CAP_IRQ_XICS: 563 #endif 564 case KVM_CAP_PPC_GET_CPU_CHAR: 565 r = 1; 566 break; 567 #ifdef CONFIG_KVM_XIVE 568 case KVM_CAP_PPC_IRQ_XIVE: 569 /* 570 * We need XIVE to be enabled on the platform (implies 571 * a POWER9 processor) and the PowerNV platform, as 572 * nested is not yet supported. 573 */ 574 r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE) && 575 kvmppc_xive_native_supported(); 576 break; 577 #endif 578 579 case KVM_CAP_PPC_ALLOC_HTAB: 580 r = hv_enabled; 581 break; 582 #endif /* CONFIG_PPC_BOOK3S_64 */ 583 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 584 case KVM_CAP_PPC_SMT: 585 r = 0; 586 if (kvm) { 587 if (kvm->arch.emul_smt_mode > 1) 588 r = kvm->arch.emul_smt_mode; 589 else 590 r = kvm->arch.smt_mode; 591 } else if (hv_enabled) { 592 if (cpu_has_feature(CPU_FTR_ARCH_300)) 593 r = 1; 594 else 595 r = threads_per_subcore; 596 } 597 break; 598 case KVM_CAP_PPC_SMT_POSSIBLE: 599 r = 1; 600 if (hv_enabled) { 601 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 602 r = ((threads_per_subcore << 1) - 1); 603 else 604 /* P9 can emulate dbells, so allow any mode */ 605 r = 8 | 4 | 2 | 1; 606 } 607 break; 608 case KVM_CAP_PPC_RMA: 609 r = 0; 610 break; 611 case KVM_CAP_PPC_HWRNG: 612 r = kvmppc_hwrng_present(); 613 break; 614 case KVM_CAP_PPC_MMU_RADIX: 615 r = !!(hv_enabled && radix_enabled()); 616 break; 617 case KVM_CAP_PPC_MMU_HASH_V3: 618 r = !!(hv_enabled && kvmppc_hv_ops->hash_v3_possible && 619 kvmppc_hv_ops->hash_v3_possible()); 620 break; 621 case KVM_CAP_PPC_NESTED_HV: 622 r = !!(hv_enabled && kvmppc_hv_ops->enable_nested && 623 !kvmppc_hv_ops->enable_nested(NULL)); 624 break; 625 #endif 626 case KVM_CAP_SYNC_MMU: 627 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 628 r = hv_enabled; 629 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER) 630 r = 1; 631 #else 632 r = 0; 633 #endif 634 break; 635 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 636 case KVM_CAP_PPC_HTAB_FD: 637 r = hv_enabled; 638 break; 639 #endif 640 case KVM_CAP_NR_VCPUS: 641 /* 642 * Recommending a number of CPUs is somewhat arbitrary; we 643 * return the number of present CPUs for -HV (since a host 644 * will have secondary threads "offline"), and for other KVM 645 * implementations just count online CPUs. 646 */ 647 if (hv_enabled) 648 r = min_t(unsigned int, num_present_cpus(), KVM_MAX_VCPUS); 649 else 650 r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS); 651 break; 652 case KVM_CAP_MAX_VCPUS: 653 r = KVM_MAX_VCPUS; 654 break; 655 case KVM_CAP_MAX_VCPU_ID: 656 r = KVM_MAX_VCPU_IDS; 657 break; 658 #ifdef CONFIG_PPC_BOOK3S_64 659 case KVM_CAP_PPC_GET_SMMU_INFO: 660 r = 1; 661 break; 662 case KVM_CAP_SPAPR_MULTITCE: 663 r = 1; 664 break; 665 case KVM_CAP_SPAPR_RESIZE_HPT: 666 r = !!hv_enabled; 667 break; 668 #endif 669 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 670 case KVM_CAP_PPC_FWNMI: 671 r = hv_enabled; 672 break; 673 #endif 674 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 675 case KVM_CAP_PPC_HTM: 676 r = !!(cur_cpu_spec->cpu_user_features2 & PPC_FEATURE2_HTM) || 677 (hv_enabled && cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)); 678 break; 679 #endif 680 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE) 681 case KVM_CAP_PPC_SECURE_GUEST: 682 r = hv_enabled && kvmppc_hv_ops->enable_svm && 683 !kvmppc_hv_ops->enable_svm(NULL); 684 break; 685 case KVM_CAP_PPC_DAWR1: 686 r = !!(hv_enabled && kvmppc_hv_ops->enable_dawr1 && 687 !kvmppc_hv_ops->enable_dawr1(NULL)); 688 break; 689 case KVM_CAP_PPC_RPT_INVALIDATE: 690 r = 1; 691 break; 692 #endif 693 case KVM_CAP_PPC_AIL_MODE_3: 694 r = 0; 695 /* 696 * KVM PR, POWER7, and some POWER9s don't support AIL=3 mode. 697 * The POWER9s can support it if the guest runs in hash mode, 698 * but QEMU doesn't necessarily query the capability in time. 699 */ 700 if (hv_enabled) { 701 if (kvmhv_on_pseries()) { 702 if (pseries_reloc_on_exception()) 703 r = 1; 704 } else if (cpu_has_feature(CPU_FTR_ARCH_207S) && 705 !cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG)) { 706 r = 1; 707 } 708 } 709 break; 710 default: 711 r = 0; 712 break; 713 } 714 return r; 715 716 } 717 718 long kvm_arch_dev_ioctl(struct file *filp, 719 unsigned int ioctl, unsigned long arg) 720 { 721 return -EINVAL; 722 } 723 724 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) 725 { 726 kvmppc_core_free_memslot(kvm, slot); 727 } 728 729 int kvm_arch_prepare_memory_region(struct kvm *kvm, 730 const struct kvm_memory_slot *old, 731 struct kvm_memory_slot *new, 732 enum kvm_mr_change change) 733 { 734 return kvmppc_core_prepare_memory_region(kvm, old, new, change); 735 } 736 737 void kvm_arch_commit_memory_region(struct kvm *kvm, 738 struct kvm_memory_slot *old, 739 const struct kvm_memory_slot *new, 740 enum kvm_mr_change change) 741 { 742 kvmppc_core_commit_memory_region(kvm, old, new, change); 743 } 744 745 void kvm_arch_flush_shadow_memslot(struct kvm *kvm, 746 struct kvm_memory_slot *slot) 747 { 748 kvmppc_core_flush_memslot(kvm, slot); 749 } 750 751 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) 752 { 753 return 0; 754 } 755 756 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer) 757 { 758 struct kvm_vcpu *vcpu; 759 760 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer); 761 kvmppc_decrementer_func(vcpu); 762 763 return HRTIMER_NORESTART; 764 } 765 766 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) 767 { 768 int err; 769 770 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS); 771 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup; 772 773 #ifdef CONFIG_KVM_EXIT_TIMING 774 mutex_init(&vcpu->arch.exit_timing_lock); 775 #endif 776 err = kvmppc_subarch_vcpu_init(vcpu); 777 if (err) 778 return err; 779 780 err = kvmppc_core_vcpu_create(vcpu); 781 if (err) 782 goto out_vcpu_uninit; 783 784 rcuwait_init(&vcpu->arch.wait); 785 vcpu->arch.waitp = &vcpu->arch.wait; 786 return 0; 787 788 out_vcpu_uninit: 789 kvmppc_subarch_vcpu_uninit(vcpu); 790 return err; 791 } 792 793 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 794 { 795 } 796 797 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 798 { 799 /* Make sure we're not using the vcpu anymore */ 800 hrtimer_cancel(&vcpu->arch.dec_timer); 801 802 switch (vcpu->arch.irq_type) { 803 case KVMPPC_IRQ_MPIC: 804 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu); 805 break; 806 case KVMPPC_IRQ_XICS: 807 if (xics_on_xive()) 808 kvmppc_xive_cleanup_vcpu(vcpu); 809 else 810 kvmppc_xics_free_icp(vcpu); 811 break; 812 case KVMPPC_IRQ_XIVE: 813 kvmppc_xive_native_cleanup_vcpu(vcpu); 814 break; 815 } 816 817 kvmppc_core_vcpu_free(vcpu); 818 819 kvmppc_subarch_vcpu_uninit(vcpu); 820 } 821 822 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) 823 { 824 return kvmppc_core_pending_dec(vcpu); 825 } 826 827 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 828 { 829 #ifdef CONFIG_BOOKE 830 /* 831 * vrsave (formerly usprg0) isn't used by Linux, but may 832 * be used by the guest. 833 * 834 * On non-booke this is associated with Altivec and 835 * is handled by code in book3s.c. 836 */ 837 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave); 838 #endif 839 kvmppc_core_vcpu_load(vcpu, cpu); 840 } 841 842 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 843 { 844 kvmppc_core_vcpu_put(vcpu); 845 #ifdef CONFIG_BOOKE 846 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE); 847 #endif 848 } 849 850 /* 851 * irq_bypass_add_producer and irq_bypass_del_producer are only 852 * useful if the architecture supports PCI passthrough. 853 * irq_bypass_stop and irq_bypass_start are not needed and so 854 * kvm_ops are not defined for them. 855 */ 856 bool kvm_arch_has_irq_bypass(void) 857 { 858 return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) || 859 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer)); 860 } 861 862 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, 863 struct irq_bypass_producer *prod) 864 { 865 struct kvm_kernel_irqfd *irqfd = 866 container_of(cons, struct kvm_kernel_irqfd, consumer); 867 struct kvm *kvm = irqfd->kvm; 868 869 if (kvm->arch.kvm_ops->irq_bypass_add_producer) 870 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod); 871 872 return 0; 873 } 874 875 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, 876 struct irq_bypass_producer *prod) 877 { 878 struct kvm_kernel_irqfd *irqfd = 879 container_of(cons, struct kvm_kernel_irqfd, consumer); 880 struct kvm *kvm = irqfd->kvm; 881 882 if (kvm->arch.kvm_ops->irq_bypass_del_producer) 883 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod); 884 } 885 886 #ifdef CONFIG_VSX 887 static inline int kvmppc_get_vsr_dword_offset(int index) 888 { 889 int offset; 890 891 if ((index != 0) && (index != 1)) 892 return -1; 893 894 #ifdef __BIG_ENDIAN 895 offset = index; 896 #else 897 offset = 1 - index; 898 #endif 899 900 return offset; 901 } 902 903 static inline int kvmppc_get_vsr_word_offset(int index) 904 { 905 int offset; 906 907 if ((index > 3) || (index < 0)) 908 return -1; 909 910 #ifdef __BIG_ENDIAN 911 offset = index; 912 #else 913 offset = 3 - index; 914 #endif 915 return offset; 916 } 917 918 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu, 919 u64 gpr) 920 { 921 union kvmppc_one_reg val; 922 int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset); 923 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 924 925 if (offset == -1) 926 return; 927 928 if (index >= 32) { 929 val.vval = VCPU_VSX_VR(vcpu, index - 32); 930 val.vsxval[offset] = gpr; 931 VCPU_VSX_VR(vcpu, index - 32) = val.vval; 932 } else { 933 VCPU_VSX_FPR(vcpu, index, offset) = gpr; 934 } 935 } 936 937 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu, 938 u64 gpr) 939 { 940 union kvmppc_one_reg val; 941 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 942 943 if (index >= 32) { 944 val.vval = VCPU_VSX_VR(vcpu, index - 32); 945 val.vsxval[0] = gpr; 946 val.vsxval[1] = gpr; 947 VCPU_VSX_VR(vcpu, index - 32) = val.vval; 948 } else { 949 VCPU_VSX_FPR(vcpu, index, 0) = gpr; 950 VCPU_VSX_FPR(vcpu, index, 1) = gpr; 951 } 952 } 953 954 static inline void kvmppc_set_vsr_word_dump(struct kvm_vcpu *vcpu, 955 u32 gpr) 956 { 957 union kvmppc_one_reg val; 958 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 959 960 if (index >= 32) { 961 val.vsx32val[0] = gpr; 962 val.vsx32val[1] = gpr; 963 val.vsx32val[2] = gpr; 964 val.vsx32val[3] = gpr; 965 VCPU_VSX_VR(vcpu, index - 32) = val.vval; 966 } else { 967 val.vsx32val[0] = gpr; 968 val.vsx32val[1] = gpr; 969 VCPU_VSX_FPR(vcpu, index, 0) = val.vsxval[0]; 970 VCPU_VSX_FPR(vcpu, index, 1) = val.vsxval[0]; 971 } 972 } 973 974 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu, 975 u32 gpr32) 976 { 977 union kvmppc_one_reg val; 978 int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset); 979 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 980 int dword_offset, word_offset; 981 982 if (offset == -1) 983 return; 984 985 if (index >= 32) { 986 val.vval = VCPU_VSX_VR(vcpu, index - 32); 987 val.vsx32val[offset] = gpr32; 988 VCPU_VSX_VR(vcpu, index - 32) = val.vval; 989 } else { 990 dword_offset = offset / 2; 991 word_offset = offset % 2; 992 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset); 993 val.vsx32val[word_offset] = gpr32; 994 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0]; 995 } 996 } 997 #endif /* CONFIG_VSX */ 998 999 #ifdef CONFIG_ALTIVEC 1000 static inline int kvmppc_get_vmx_offset_generic(struct kvm_vcpu *vcpu, 1001 int index, int element_size) 1002 { 1003 int offset; 1004 int elts = sizeof(vector128)/element_size; 1005 1006 if ((index < 0) || (index >= elts)) 1007 return -1; 1008 1009 if (kvmppc_need_byteswap(vcpu)) 1010 offset = elts - index - 1; 1011 else 1012 offset = index; 1013 1014 return offset; 1015 } 1016 1017 static inline int kvmppc_get_vmx_dword_offset(struct kvm_vcpu *vcpu, 1018 int index) 1019 { 1020 return kvmppc_get_vmx_offset_generic(vcpu, index, 8); 1021 } 1022 1023 static inline int kvmppc_get_vmx_word_offset(struct kvm_vcpu *vcpu, 1024 int index) 1025 { 1026 return kvmppc_get_vmx_offset_generic(vcpu, index, 4); 1027 } 1028 1029 static inline int kvmppc_get_vmx_hword_offset(struct kvm_vcpu *vcpu, 1030 int index) 1031 { 1032 return kvmppc_get_vmx_offset_generic(vcpu, index, 2); 1033 } 1034 1035 static inline int kvmppc_get_vmx_byte_offset(struct kvm_vcpu *vcpu, 1036 int index) 1037 { 1038 return kvmppc_get_vmx_offset_generic(vcpu, index, 1); 1039 } 1040 1041 1042 static inline void kvmppc_set_vmx_dword(struct kvm_vcpu *vcpu, 1043 u64 gpr) 1044 { 1045 union kvmppc_one_reg val; 1046 int offset = kvmppc_get_vmx_dword_offset(vcpu, 1047 vcpu->arch.mmio_vmx_offset); 1048 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 1049 1050 if (offset == -1) 1051 return; 1052 1053 val.vval = VCPU_VSX_VR(vcpu, index); 1054 val.vsxval[offset] = gpr; 1055 VCPU_VSX_VR(vcpu, index) = val.vval; 1056 } 1057 1058 static inline void kvmppc_set_vmx_word(struct kvm_vcpu *vcpu, 1059 u32 gpr32) 1060 { 1061 union kvmppc_one_reg val; 1062 int offset = kvmppc_get_vmx_word_offset(vcpu, 1063 vcpu->arch.mmio_vmx_offset); 1064 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 1065 1066 if (offset == -1) 1067 return; 1068 1069 val.vval = VCPU_VSX_VR(vcpu, index); 1070 val.vsx32val[offset] = gpr32; 1071 VCPU_VSX_VR(vcpu, index) = val.vval; 1072 } 1073 1074 static inline void kvmppc_set_vmx_hword(struct kvm_vcpu *vcpu, 1075 u16 gpr16) 1076 { 1077 union kvmppc_one_reg val; 1078 int offset = kvmppc_get_vmx_hword_offset(vcpu, 1079 vcpu->arch.mmio_vmx_offset); 1080 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 1081 1082 if (offset == -1) 1083 return; 1084 1085 val.vval = VCPU_VSX_VR(vcpu, index); 1086 val.vsx16val[offset] = gpr16; 1087 VCPU_VSX_VR(vcpu, index) = val.vval; 1088 } 1089 1090 static inline void kvmppc_set_vmx_byte(struct kvm_vcpu *vcpu, 1091 u8 gpr8) 1092 { 1093 union kvmppc_one_reg val; 1094 int offset = kvmppc_get_vmx_byte_offset(vcpu, 1095 vcpu->arch.mmio_vmx_offset); 1096 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 1097 1098 if (offset == -1) 1099 return; 1100 1101 val.vval = VCPU_VSX_VR(vcpu, index); 1102 val.vsx8val[offset] = gpr8; 1103 VCPU_VSX_VR(vcpu, index) = val.vval; 1104 } 1105 #endif /* CONFIG_ALTIVEC */ 1106 1107 #ifdef CONFIG_PPC_FPU 1108 static inline u64 sp_to_dp(u32 fprs) 1109 { 1110 u64 fprd; 1111 1112 preempt_disable(); 1113 enable_kernel_fp(); 1114 asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m<>" (fprd) : "m<>" (fprs) 1115 : "fr0"); 1116 preempt_enable(); 1117 return fprd; 1118 } 1119 1120 static inline u32 dp_to_sp(u64 fprd) 1121 { 1122 u32 fprs; 1123 1124 preempt_disable(); 1125 enable_kernel_fp(); 1126 asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m<>" (fprs) : "m<>" (fprd) 1127 : "fr0"); 1128 preempt_enable(); 1129 return fprs; 1130 } 1131 1132 #else 1133 #define sp_to_dp(x) (x) 1134 #define dp_to_sp(x) (x) 1135 #endif /* CONFIG_PPC_FPU */ 1136 1137 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu) 1138 { 1139 struct kvm_run *run = vcpu->run; 1140 u64 gpr; 1141 1142 if (run->mmio.len > sizeof(gpr)) 1143 return; 1144 1145 if (!vcpu->arch.mmio_host_swabbed) { 1146 switch (run->mmio.len) { 1147 case 8: gpr = *(u64 *)run->mmio.data; break; 1148 case 4: gpr = *(u32 *)run->mmio.data; break; 1149 case 2: gpr = *(u16 *)run->mmio.data; break; 1150 case 1: gpr = *(u8 *)run->mmio.data; break; 1151 } 1152 } else { 1153 switch (run->mmio.len) { 1154 case 8: gpr = swab64(*(u64 *)run->mmio.data); break; 1155 case 4: gpr = swab32(*(u32 *)run->mmio.data); break; 1156 case 2: gpr = swab16(*(u16 *)run->mmio.data); break; 1157 case 1: gpr = *(u8 *)run->mmio.data; break; 1158 } 1159 } 1160 1161 /* conversion between single and double precision */ 1162 if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4)) 1163 gpr = sp_to_dp(gpr); 1164 1165 if (vcpu->arch.mmio_sign_extend) { 1166 switch (run->mmio.len) { 1167 #ifdef CONFIG_PPC64 1168 case 4: 1169 gpr = (s64)(s32)gpr; 1170 break; 1171 #endif 1172 case 2: 1173 gpr = (s64)(s16)gpr; 1174 break; 1175 case 1: 1176 gpr = (s64)(s8)gpr; 1177 break; 1178 } 1179 } 1180 1181 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) { 1182 case KVM_MMIO_REG_GPR: 1183 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr); 1184 break; 1185 case KVM_MMIO_REG_FPR: 1186 if (vcpu->kvm->arch.kvm_ops->giveup_ext) 1187 vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_FP); 1188 1189 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr; 1190 break; 1191 #ifdef CONFIG_PPC_BOOK3S 1192 case KVM_MMIO_REG_QPR: 1193 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr; 1194 break; 1195 case KVM_MMIO_REG_FQPR: 1196 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr; 1197 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr; 1198 break; 1199 #endif 1200 #ifdef CONFIG_VSX 1201 case KVM_MMIO_REG_VSX: 1202 if (vcpu->kvm->arch.kvm_ops->giveup_ext) 1203 vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VSX); 1204 1205 if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_DWORD) 1206 kvmppc_set_vsr_dword(vcpu, gpr); 1207 else if (vcpu->arch.mmio_copy_type == KVMPPC_VSX_COPY_WORD) 1208 kvmppc_set_vsr_word(vcpu, gpr); 1209 else if (vcpu->arch.mmio_copy_type == 1210 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP) 1211 kvmppc_set_vsr_dword_dump(vcpu, gpr); 1212 else if (vcpu->arch.mmio_copy_type == 1213 KVMPPC_VSX_COPY_WORD_LOAD_DUMP) 1214 kvmppc_set_vsr_word_dump(vcpu, gpr); 1215 break; 1216 #endif 1217 #ifdef CONFIG_ALTIVEC 1218 case KVM_MMIO_REG_VMX: 1219 if (vcpu->kvm->arch.kvm_ops->giveup_ext) 1220 vcpu->kvm->arch.kvm_ops->giveup_ext(vcpu, MSR_VEC); 1221 1222 if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_DWORD) 1223 kvmppc_set_vmx_dword(vcpu, gpr); 1224 else if (vcpu->arch.mmio_copy_type == KVMPPC_VMX_COPY_WORD) 1225 kvmppc_set_vmx_word(vcpu, gpr); 1226 else if (vcpu->arch.mmio_copy_type == 1227 KVMPPC_VMX_COPY_HWORD) 1228 kvmppc_set_vmx_hword(vcpu, gpr); 1229 else if (vcpu->arch.mmio_copy_type == 1230 KVMPPC_VMX_COPY_BYTE) 1231 kvmppc_set_vmx_byte(vcpu, gpr); 1232 break; 1233 #endif 1234 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 1235 case KVM_MMIO_REG_NESTED_GPR: 1236 if (kvmppc_need_byteswap(vcpu)) 1237 gpr = swab64(gpr); 1238 kvm_vcpu_write_guest(vcpu, vcpu->arch.nested_io_gpr, &gpr, 1239 sizeof(gpr)); 1240 break; 1241 #endif 1242 default: 1243 BUG(); 1244 } 1245 } 1246 1247 static int __kvmppc_handle_load(struct kvm_vcpu *vcpu, 1248 unsigned int rt, unsigned int bytes, 1249 int is_default_endian, int sign_extend) 1250 { 1251 struct kvm_run *run = vcpu->run; 1252 int idx, ret; 1253 bool host_swabbed; 1254 1255 /* Pity C doesn't have a logical XOR operator */ 1256 if (kvmppc_need_byteswap(vcpu)) { 1257 host_swabbed = is_default_endian; 1258 } else { 1259 host_swabbed = !is_default_endian; 1260 } 1261 1262 if (bytes > sizeof(run->mmio.data)) 1263 return EMULATE_FAIL; 1264 1265 run->mmio.phys_addr = vcpu->arch.paddr_accessed; 1266 run->mmio.len = bytes; 1267 run->mmio.is_write = 0; 1268 1269 vcpu->arch.io_gpr = rt; 1270 vcpu->arch.mmio_host_swabbed = host_swabbed; 1271 vcpu->mmio_needed = 1; 1272 vcpu->mmio_is_write = 0; 1273 vcpu->arch.mmio_sign_extend = sign_extend; 1274 1275 idx = srcu_read_lock(&vcpu->kvm->srcu); 1276 1277 ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr, 1278 bytes, &run->mmio.data); 1279 1280 srcu_read_unlock(&vcpu->kvm->srcu, idx); 1281 1282 if (!ret) { 1283 kvmppc_complete_mmio_load(vcpu); 1284 vcpu->mmio_needed = 0; 1285 return EMULATE_DONE; 1286 } 1287 1288 return EMULATE_DO_MMIO; 1289 } 1290 1291 int kvmppc_handle_load(struct kvm_vcpu *vcpu, 1292 unsigned int rt, unsigned int bytes, 1293 int is_default_endian) 1294 { 1295 return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 0); 1296 } 1297 EXPORT_SYMBOL_GPL(kvmppc_handle_load); 1298 1299 /* Same as above, but sign extends */ 1300 int kvmppc_handle_loads(struct kvm_vcpu *vcpu, 1301 unsigned int rt, unsigned int bytes, 1302 int is_default_endian) 1303 { 1304 return __kvmppc_handle_load(vcpu, rt, bytes, is_default_endian, 1); 1305 } 1306 1307 #ifdef CONFIG_VSX 1308 int kvmppc_handle_vsx_load(struct kvm_vcpu *vcpu, 1309 unsigned int rt, unsigned int bytes, 1310 int is_default_endian, int mmio_sign_extend) 1311 { 1312 enum emulation_result emulated = EMULATE_DONE; 1313 1314 /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */ 1315 if (vcpu->arch.mmio_vsx_copy_nums > 4) 1316 return EMULATE_FAIL; 1317 1318 while (vcpu->arch.mmio_vsx_copy_nums) { 1319 emulated = __kvmppc_handle_load(vcpu, rt, bytes, 1320 is_default_endian, mmio_sign_extend); 1321 1322 if (emulated != EMULATE_DONE) 1323 break; 1324 1325 vcpu->arch.paddr_accessed += vcpu->run->mmio.len; 1326 1327 vcpu->arch.mmio_vsx_copy_nums--; 1328 vcpu->arch.mmio_vsx_offset++; 1329 } 1330 return emulated; 1331 } 1332 #endif /* CONFIG_VSX */ 1333 1334 int kvmppc_handle_store(struct kvm_vcpu *vcpu, 1335 u64 val, unsigned int bytes, int is_default_endian) 1336 { 1337 struct kvm_run *run = vcpu->run; 1338 void *data = run->mmio.data; 1339 int idx, ret; 1340 bool host_swabbed; 1341 1342 /* Pity C doesn't have a logical XOR operator */ 1343 if (kvmppc_need_byteswap(vcpu)) { 1344 host_swabbed = is_default_endian; 1345 } else { 1346 host_swabbed = !is_default_endian; 1347 } 1348 1349 if (bytes > sizeof(run->mmio.data)) 1350 return EMULATE_FAIL; 1351 1352 run->mmio.phys_addr = vcpu->arch.paddr_accessed; 1353 run->mmio.len = bytes; 1354 run->mmio.is_write = 1; 1355 vcpu->mmio_needed = 1; 1356 vcpu->mmio_is_write = 1; 1357 1358 if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4)) 1359 val = dp_to_sp(val); 1360 1361 /* Store the value at the lowest bytes in 'data'. */ 1362 if (!host_swabbed) { 1363 switch (bytes) { 1364 case 8: *(u64 *)data = val; break; 1365 case 4: *(u32 *)data = val; break; 1366 case 2: *(u16 *)data = val; break; 1367 case 1: *(u8 *)data = val; break; 1368 } 1369 } else { 1370 switch (bytes) { 1371 case 8: *(u64 *)data = swab64(val); break; 1372 case 4: *(u32 *)data = swab32(val); break; 1373 case 2: *(u16 *)data = swab16(val); break; 1374 case 1: *(u8 *)data = val; break; 1375 } 1376 } 1377 1378 idx = srcu_read_lock(&vcpu->kvm->srcu); 1379 1380 ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr, 1381 bytes, &run->mmio.data); 1382 1383 srcu_read_unlock(&vcpu->kvm->srcu, idx); 1384 1385 if (!ret) { 1386 vcpu->mmio_needed = 0; 1387 return EMULATE_DONE; 1388 } 1389 1390 return EMULATE_DO_MMIO; 1391 } 1392 EXPORT_SYMBOL_GPL(kvmppc_handle_store); 1393 1394 #ifdef CONFIG_VSX 1395 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val) 1396 { 1397 u32 dword_offset, word_offset; 1398 union kvmppc_one_reg reg; 1399 int vsx_offset = 0; 1400 int copy_type = vcpu->arch.mmio_copy_type; 1401 int result = 0; 1402 1403 switch (copy_type) { 1404 case KVMPPC_VSX_COPY_DWORD: 1405 vsx_offset = 1406 kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset); 1407 1408 if (vsx_offset == -1) { 1409 result = -1; 1410 break; 1411 } 1412 1413 if (rs < 32) { 1414 *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset); 1415 } else { 1416 reg.vval = VCPU_VSX_VR(vcpu, rs - 32); 1417 *val = reg.vsxval[vsx_offset]; 1418 } 1419 break; 1420 1421 case KVMPPC_VSX_COPY_WORD: 1422 vsx_offset = 1423 kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset); 1424 1425 if (vsx_offset == -1) { 1426 result = -1; 1427 break; 1428 } 1429 1430 if (rs < 32) { 1431 dword_offset = vsx_offset / 2; 1432 word_offset = vsx_offset % 2; 1433 reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset); 1434 *val = reg.vsx32val[word_offset]; 1435 } else { 1436 reg.vval = VCPU_VSX_VR(vcpu, rs - 32); 1437 *val = reg.vsx32val[vsx_offset]; 1438 } 1439 break; 1440 1441 default: 1442 result = -1; 1443 break; 1444 } 1445 1446 return result; 1447 } 1448 1449 int kvmppc_handle_vsx_store(struct kvm_vcpu *vcpu, 1450 int rs, unsigned int bytes, int is_default_endian) 1451 { 1452 u64 val; 1453 enum emulation_result emulated = EMULATE_DONE; 1454 1455 vcpu->arch.io_gpr = rs; 1456 1457 /* Currently, mmio_vsx_copy_nums only allowed to be 4 or less */ 1458 if (vcpu->arch.mmio_vsx_copy_nums > 4) 1459 return EMULATE_FAIL; 1460 1461 while (vcpu->arch.mmio_vsx_copy_nums) { 1462 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1) 1463 return EMULATE_FAIL; 1464 1465 emulated = kvmppc_handle_store(vcpu, 1466 val, bytes, is_default_endian); 1467 1468 if (emulated != EMULATE_DONE) 1469 break; 1470 1471 vcpu->arch.paddr_accessed += vcpu->run->mmio.len; 1472 1473 vcpu->arch.mmio_vsx_copy_nums--; 1474 vcpu->arch.mmio_vsx_offset++; 1475 } 1476 1477 return emulated; 1478 } 1479 1480 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu) 1481 { 1482 struct kvm_run *run = vcpu->run; 1483 enum emulation_result emulated = EMULATE_FAIL; 1484 int r; 1485 1486 vcpu->arch.paddr_accessed += run->mmio.len; 1487 1488 if (!vcpu->mmio_is_write) { 1489 emulated = kvmppc_handle_vsx_load(vcpu, vcpu->arch.io_gpr, 1490 run->mmio.len, 1, vcpu->arch.mmio_sign_extend); 1491 } else { 1492 emulated = kvmppc_handle_vsx_store(vcpu, 1493 vcpu->arch.io_gpr, run->mmio.len, 1); 1494 } 1495 1496 switch (emulated) { 1497 case EMULATE_DO_MMIO: 1498 run->exit_reason = KVM_EXIT_MMIO; 1499 r = RESUME_HOST; 1500 break; 1501 case EMULATE_FAIL: 1502 pr_info("KVM: MMIO emulation failed (VSX repeat)\n"); 1503 run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 1504 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; 1505 r = RESUME_HOST; 1506 break; 1507 default: 1508 r = RESUME_GUEST; 1509 break; 1510 } 1511 return r; 1512 } 1513 #endif /* CONFIG_VSX */ 1514 1515 #ifdef CONFIG_ALTIVEC 1516 int kvmppc_handle_vmx_load(struct kvm_vcpu *vcpu, 1517 unsigned int rt, unsigned int bytes, int is_default_endian) 1518 { 1519 enum emulation_result emulated = EMULATE_DONE; 1520 1521 if (vcpu->arch.mmio_vmx_copy_nums > 2) 1522 return EMULATE_FAIL; 1523 1524 while (vcpu->arch.mmio_vmx_copy_nums) { 1525 emulated = __kvmppc_handle_load(vcpu, rt, bytes, 1526 is_default_endian, 0); 1527 1528 if (emulated != EMULATE_DONE) 1529 break; 1530 1531 vcpu->arch.paddr_accessed += vcpu->run->mmio.len; 1532 vcpu->arch.mmio_vmx_copy_nums--; 1533 vcpu->arch.mmio_vmx_offset++; 1534 } 1535 1536 return emulated; 1537 } 1538 1539 static int kvmppc_get_vmx_dword(struct kvm_vcpu *vcpu, int index, u64 *val) 1540 { 1541 union kvmppc_one_reg reg; 1542 int vmx_offset = 0; 1543 int result = 0; 1544 1545 vmx_offset = 1546 kvmppc_get_vmx_dword_offset(vcpu, vcpu->arch.mmio_vmx_offset); 1547 1548 if (vmx_offset == -1) 1549 return -1; 1550 1551 reg.vval = VCPU_VSX_VR(vcpu, index); 1552 *val = reg.vsxval[vmx_offset]; 1553 1554 return result; 1555 } 1556 1557 static int kvmppc_get_vmx_word(struct kvm_vcpu *vcpu, int index, u64 *val) 1558 { 1559 union kvmppc_one_reg reg; 1560 int vmx_offset = 0; 1561 int result = 0; 1562 1563 vmx_offset = 1564 kvmppc_get_vmx_word_offset(vcpu, vcpu->arch.mmio_vmx_offset); 1565 1566 if (vmx_offset == -1) 1567 return -1; 1568 1569 reg.vval = VCPU_VSX_VR(vcpu, index); 1570 *val = reg.vsx32val[vmx_offset]; 1571 1572 return result; 1573 } 1574 1575 static int kvmppc_get_vmx_hword(struct kvm_vcpu *vcpu, int index, u64 *val) 1576 { 1577 union kvmppc_one_reg reg; 1578 int vmx_offset = 0; 1579 int result = 0; 1580 1581 vmx_offset = 1582 kvmppc_get_vmx_hword_offset(vcpu, vcpu->arch.mmio_vmx_offset); 1583 1584 if (vmx_offset == -1) 1585 return -1; 1586 1587 reg.vval = VCPU_VSX_VR(vcpu, index); 1588 *val = reg.vsx16val[vmx_offset]; 1589 1590 return result; 1591 } 1592 1593 static int kvmppc_get_vmx_byte(struct kvm_vcpu *vcpu, int index, u64 *val) 1594 { 1595 union kvmppc_one_reg reg; 1596 int vmx_offset = 0; 1597 int result = 0; 1598 1599 vmx_offset = 1600 kvmppc_get_vmx_byte_offset(vcpu, vcpu->arch.mmio_vmx_offset); 1601 1602 if (vmx_offset == -1) 1603 return -1; 1604 1605 reg.vval = VCPU_VSX_VR(vcpu, index); 1606 *val = reg.vsx8val[vmx_offset]; 1607 1608 return result; 1609 } 1610 1611 int kvmppc_handle_vmx_store(struct kvm_vcpu *vcpu, 1612 unsigned int rs, unsigned int bytes, int is_default_endian) 1613 { 1614 u64 val = 0; 1615 unsigned int index = rs & KVM_MMIO_REG_MASK; 1616 enum emulation_result emulated = EMULATE_DONE; 1617 1618 if (vcpu->arch.mmio_vmx_copy_nums > 2) 1619 return EMULATE_FAIL; 1620 1621 vcpu->arch.io_gpr = rs; 1622 1623 while (vcpu->arch.mmio_vmx_copy_nums) { 1624 switch (vcpu->arch.mmio_copy_type) { 1625 case KVMPPC_VMX_COPY_DWORD: 1626 if (kvmppc_get_vmx_dword(vcpu, index, &val) == -1) 1627 return EMULATE_FAIL; 1628 1629 break; 1630 case KVMPPC_VMX_COPY_WORD: 1631 if (kvmppc_get_vmx_word(vcpu, index, &val) == -1) 1632 return EMULATE_FAIL; 1633 break; 1634 case KVMPPC_VMX_COPY_HWORD: 1635 if (kvmppc_get_vmx_hword(vcpu, index, &val) == -1) 1636 return EMULATE_FAIL; 1637 break; 1638 case KVMPPC_VMX_COPY_BYTE: 1639 if (kvmppc_get_vmx_byte(vcpu, index, &val) == -1) 1640 return EMULATE_FAIL; 1641 break; 1642 default: 1643 return EMULATE_FAIL; 1644 } 1645 1646 emulated = kvmppc_handle_store(vcpu, val, bytes, 1647 is_default_endian); 1648 if (emulated != EMULATE_DONE) 1649 break; 1650 1651 vcpu->arch.paddr_accessed += vcpu->run->mmio.len; 1652 vcpu->arch.mmio_vmx_copy_nums--; 1653 vcpu->arch.mmio_vmx_offset++; 1654 } 1655 1656 return emulated; 1657 } 1658 1659 static int kvmppc_emulate_mmio_vmx_loadstore(struct kvm_vcpu *vcpu) 1660 { 1661 struct kvm_run *run = vcpu->run; 1662 enum emulation_result emulated = EMULATE_FAIL; 1663 int r; 1664 1665 vcpu->arch.paddr_accessed += run->mmio.len; 1666 1667 if (!vcpu->mmio_is_write) { 1668 emulated = kvmppc_handle_vmx_load(vcpu, 1669 vcpu->arch.io_gpr, run->mmio.len, 1); 1670 } else { 1671 emulated = kvmppc_handle_vmx_store(vcpu, 1672 vcpu->arch.io_gpr, run->mmio.len, 1); 1673 } 1674 1675 switch (emulated) { 1676 case EMULATE_DO_MMIO: 1677 run->exit_reason = KVM_EXIT_MMIO; 1678 r = RESUME_HOST; 1679 break; 1680 case EMULATE_FAIL: 1681 pr_info("KVM: MMIO emulation failed (VMX repeat)\n"); 1682 run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 1683 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; 1684 r = RESUME_HOST; 1685 break; 1686 default: 1687 r = RESUME_GUEST; 1688 break; 1689 } 1690 return r; 1691 } 1692 #endif /* CONFIG_ALTIVEC */ 1693 1694 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg) 1695 { 1696 int r = 0; 1697 union kvmppc_one_reg val; 1698 int size; 1699 1700 size = one_reg_size(reg->id); 1701 if (size > sizeof(val)) 1702 return -EINVAL; 1703 1704 r = kvmppc_get_one_reg(vcpu, reg->id, &val); 1705 if (r == -EINVAL) { 1706 r = 0; 1707 switch (reg->id) { 1708 #ifdef CONFIG_ALTIVEC 1709 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31: 1710 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1711 r = -ENXIO; 1712 break; 1713 } 1714 val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0]; 1715 break; 1716 case KVM_REG_PPC_VSCR: 1717 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1718 r = -ENXIO; 1719 break; 1720 } 1721 val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]); 1722 break; 1723 case KVM_REG_PPC_VRSAVE: 1724 val = get_reg_val(reg->id, vcpu->arch.vrsave); 1725 break; 1726 #endif /* CONFIG_ALTIVEC */ 1727 default: 1728 r = -EINVAL; 1729 break; 1730 } 1731 } 1732 1733 if (r) 1734 return r; 1735 1736 if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size)) 1737 r = -EFAULT; 1738 1739 return r; 1740 } 1741 1742 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg) 1743 { 1744 int r; 1745 union kvmppc_one_reg val; 1746 int size; 1747 1748 size = one_reg_size(reg->id); 1749 if (size > sizeof(val)) 1750 return -EINVAL; 1751 1752 if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size)) 1753 return -EFAULT; 1754 1755 r = kvmppc_set_one_reg(vcpu, reg->id, &val); 1756 if (r == -EINVAL) { 1757 r = 0; 1758 switch (reg->id) { 1759 #ifdef CONFIG_ALTIVEC 1760 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31: 1761 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1762 r = -ENXIO; 1763 break; 1764 } 1765 vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval; 1766 break; 1767 case KVM_REG_PPC_VSCR: 1768 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1769 r = -ENXIO; 1770 break; 1771 } 1772 vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val); 1773 break; 1774 case KVM_REG_PPC_VRSAVE: 1775 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1776 r = -ENXIO; 1777 break; 1778 } 1779 vcpu->arch.vrsave = set_reg_val(reg->id, val); 1780 break; 1781 #endif /* CONFIG_ALTIVEC */ 1782 default: 1783 r = -EINVAL; 1784 break; 1785 } 1786 } 1787 1788 return r; 1789 } 1790 1791 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) 1792 { 1793 struct kvm_run *run = vcpu->run; 1794 int r; 1795 1796 vcpu_load(vcpu); 1797 1798 if (vcpu->mmio_needed) { 1799 vcpu->mmio_needed = 0; 1800 if (!vcpu->mmio_is_write) 1801 kvmppc_complete_mmio_load(vcpu); 1802 #ifdef CONFIG_VSX 1803 if (vcpu->arch.mmio_vsx_copy_nums > 0) { 1804 vcpu->arch.mmio_vsx_copy_nums--; 1805 vcpu->arch.mmio_vsx_offset++; 1806 } 1807 1808 if (vcpu->arch.mmio_vsx_copy_nums > 0) { 1809 r = kvmppc_emulate_mmio_vsx_loadstore(vcpu); 1810 if (r == RESUME_HOST) { 1811 vcpu->mmio_needed = 1; 1812 goto out; 1813 } 1814 } 1815 #endif 1816 #ifdef CONFIG_ALTIVEC 1817 if (vcpu->arch.mmio_vmx_copy_nums > 0) { 1818 vcpu->arch.mmio_vmx_copy_nums--; 1819 vcpu->arch.mmio_vmx_offset++; 1820 } 1821 1822 if (vcpu->arch.mmio_vmx_copy_nums > 0) { 1823 r = kvmppc_emulate_mmio_vmx_loadstore(vcpu); 1824 if (r == RESUME_HOST) { 1825 vcpu->mmio_needed = 1; 1826 goto out; 1827 } 1828 } 1829 #endif 1830 } else if (vcpu->arch.osi_needed) { 1831 u64 *gprs = run->osi.gprs; 1832 int i; 1833 1834 for (i = 0; i < 32; i++) 1835 kvmppc_set_gpr(vcpu, i, gprs[i]); 1836 vcpu->arch.osi_needed = 0; 1837 } else if (vcpu->arch.hcall_needed) { 1838 int i; 1839 1840 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret); 1841 for (i = 0; i < 9; ++i) 1842 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]); 1843 vcpu->arch.hcall_needed = 0; 1844 #ifdef CONFIG_BOOKE 1845 } else if (vcpu->arch.epr_needed) { 1846 kvmppc_set_epr(vcpu, run->epr.epr); 1847 vcpu->arch.epr_needed = 0; 1848 #endif 1849 } 1850 1851 kvm_sigset_activate(vcpu); 1852 1853 if (run->immediate_exit) 1854 r = -EINTR; 1855 else 1856 r = kvmppc_vcpu_run(vcpu); 1857 1858 kvm_sigset_deactivate(vcpu); 1859 1860 #ifdef CONFIG_ALTIVEC 1861 out: 1862 #endif 1863 1864 /* 1865 * We're already returning to userspace, don't pass the 1866 * RESUME_HOST flags along. 1867 */ 1868 if (r > 0) 1869 r = 0; 1870 1871 vcpu_put(vcpu); 1872 return r; 1873 } 1874 1875 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq) 1876 { 1877 if (irq->irq == KVM_INTERRUPT_UNSET) { 1878 kvmppc_core_dequeue_external(vcpu); 1879 return 0; 1880 } 1881 1882 kvmppc_core_queue_external(vcpu, irq); 1883 1884 kvm_vcpu_kick(vcpu); 1885 1886 return 0; 1887 } 1888 1889 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, 1890 struct kvm_enable_cap *cap) 1891 { 1892 int r; 1893 1894 if (cap->flags) 1895 return -EINVAL; 1896 1897 switch (cap->cap) { 1898 case KVM_CAP_PPC_OSI: 1899 r = 0; 1900 vcpu->arch.osi_enabled = true; 1901 break; 1902 case KVM_CAP_PPC_PAPR: 1903 r = 0; 1904 vcpu->arch.papr_enabled = true; 1905 break; 1906 case KVM_CAP_PPC_EPR: 1907 r = 0; 1908 if (cap->args[0]) 1909 vcpu->arch.epr_flags |= KVMPPC_EPR_USER; 1910 else 1911 vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER; 1912 break; 1913 #ifdef CONFIG_BOOKE 1914 case KVM_CAP_PPC_BOOKE_WATCHDOG: 1915 r = 0; 1916 vcpu->arch.watchdog_enabled = true; 1917 break; 1918 #endif 1919 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC) 1920 case KVM_CAP_SW_TLB: { 1921 struct kvm_config_tlb cfg; 1922 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0]; 1923 1924 r = -EFAULT; 1925 if (copy_from_user(&cfg, user_ptr, sizeof(cfg))) 1926 break; 1927 1928 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg); 1929 break; 1930 } 1931 #endif 1932 #ifdef CONFIG_KVM_MPIC 1933 case KVM_CAP_IRQ_MPIC: { 1934 struct fd f; 1935 struct kvm_device *dev; 1936 1937 r = -EBADF; 1938 f = fdget(cap->args[0]); 1939 if (!f.file) 1940 break; 1941 1942 r = -EPERM; 1943 dev = kvm_device_from_filp(f.file); 1944 if (dev) 1945 r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]); 1946 1947 fdput(f); 1948 break; 1949 } 1950 #endif 1951 #ifdef CONFIG_KVM_XICS 1952 case KVM_CAP_IRQ_XICS: { 1953 struct fd f; 1954 struct kvm_device *dev; 1955 1956 r = -EBADF; 1957 f = fdget(cap->args[0]); 1958 if (!f.file) 1959 break; 1960 1961 r = -EPERM; 1962 dev = kvm_device_from_filp(f.file); 1963 if (dev) { 1964 if (xics_on_xive()) 1965 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]); 1966 else 1967 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]); 1968 } 1969 1970 fdput(f); 1971 break; 1972 } 1973 #endif /* CONFIG_KVM_XICS */ 1974 #ifdef CONFIG_KVM_XIVE 1975 case KVM_CAP_PPC_IRQ_XIVE: { 1976 struct fd f; 1977 struct kvm_device *dev; 1978 1979 r = -EBADF; 1980 f = fdget(cap->args[0]); 1981 if (!f.file) 1982 break; 1983 1984 r = -ENXIO; 1985 if (!xive_enabled()) 1986 break; 1987 1988 r = -EPERM; 1989 dev = kvm_device_from_filp(f.file); 1990 if (dev) 1991 r = kvmppc_xive_native_connect_vcpu(dev, vcpu, 1992 cap->args[1]); 1993 1994 fdput(f); 1995 break; 1996 } 1997 #endif /* CONFIG_KVM_XIVE */ 1998 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 1999 case KVM_CAP_PPC_FWNMI: 2000 r = -EINVAL; 2001 if (!is_kvmppc_hv_enabled(vcpu->kvm)) 2002 break; 2003 r = 0; 2004 vcpu->kvm->arch.fwnmi_enabled = true; 2005 break; 2006 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */ 2007 default: 2008 r = -EINVAL; 2009 break; 2010 } 2011 2012 if (!r) 2013 r = kvmppc_sanity_check(vcpu); 2014 2015 return r; 2016 } 2017 2018 bool kvm_arch_intc_initialized(struct kvm *kvm) 2019 { 2020 #ifdef CONFIG_KVM_MPIC 2021 if (kvm->arch.mpic) 2022 return true; 2023 #endif 2024 #ifdef CONFIG_KVM_XICS 2025 if (kvm->arch.xics || kvm->arch.xive) 2026 return true; 2027 #endif 2028 return false; 2029 } 2030 2031 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 2032 struct kvm_mp_state *mp_state) 2033 { 2034 return -EINVAL; 2035 } 2036 2037 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 2038 struct kvm_mp_state *mp_state) 2039 { 2040 return -EINVAL; 2041 } 2042 2043 long kvm_arch_vcpu_async_ioctl(struct file *filp, 2044 unsigned int ioctl, unsigned long arg) 2045 { 2046 struct kvm_vcpu *vcpu = filp->private_data; 2047 void __user *argp = (void __user *)arg; 2048 2049 if (ioctl == KVM_INTERRUPT) { 2050 struct kvm_interrupt irq; 2051 if (copy_from_user(&irq, argp, sizeof(irq))) 2052 return -EFAULT; 2053 return kvm_vcpu_ioctl_interrupt(vcpu, &irq); 2054 } 2055 return -ENOIOCTLCMD; 2056 } 2057 2058 long kvm_arch_vcpu_ioctl(struct file *filp, 2059 unsigned int ioctl, unsigned long arg) 2060 { 2061 struct kvm_vcpu *vcpu = filp->private_data; 2062 void __user *argp = (void __user *)arg; 2063 long r; 2064 2065 switch (ioctl) { 2066 case KVM_ENABLE_CAP: 2067 { 2068 struct kvm_enable_cap cap; 2069 r = -EFAULT; 2070 if (copy_from_user(&cap, argp, sizeof(cap))) 2071 goto out; 2072 vcpu_load(vcpu); 2073 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); 2074 vcpu_put(vcpu); 2075 break; 2076 } 2077 2078 case KVM_SET_ONE_REG: 2079 case KVM_GET_ONE_REG: 2080 { 2081 struct kvm_one_reg reg; 2082 r = -EFAULT; 2083 if (copy_from_user(®, argp, sizeof(reg))) 2084 goto out; 2085 if (ioctl == KVM_SET_ONE_REG) 2086 r = kvm_vcpu_ioctl_set_one_reg(vcpu, ®); 2087 else 2088 r = kvm_vcpu_ioctl_get_one_reg(vcpu, ®); 2089 break; 2090 } 2091 2092 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC) 2093 case KVM_DIRTY_TLB: { 2094 struct kvm_dirty_tlb dirty; 2095 r = -EFAULT; 2096 if (copy_from_user(&dirty, argp, sizeof(dirty))) 2097 goto out; 2098 vcpu_load(vcpu); 2099 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty); 2100 vcpu_put(vcpu); 2101 break; 2102 } 2103 #endif 2104 default: 2105 r = -EINVAL; 2106 } 2107 2108 out: 2109 return r; 2110 } 2111 2112 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 2113 { 2114 return VM_FAULT_SIGBUS; 2115 } 2116 2117 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo) 2118 { 2119 u32 inst_nop = 0x60000000; 2120 #ifdef CONFIG_KVM_BOOKE_HV 2121 u32 inst_sc1 = 0x44000022; 2122 pvinfo->hcall[0] = cpu_to_be32(inst_sc1); 2123 pvinfo->hcall[1] = cpu_to_be32(inst_nop); 2124 pvinfo->hcall[2] = cpu_to_be32(inst_nop); 2125 pvinfo->hcall[3] = cpu_to_be32(inst_nop); 2126 #else 2127 u32 inst_lis = 0x3c000000; 2128 u32 inst_ori = 0x60000000; 2129 u32 inst_sc = 0x44000002; 2130 u32 inst_imm_mask = 0xffff; 2131 2132 /* 2133 * The hypercall to get into KVM from within guest context is as 2134 * follows: 2135 * 2136 * lis r0, r0, KVM_SC_MAGIC_R0@h 2137 * ori r0, KVM_SC_MAGIC_R0@l 2138 * sc 2139 * nop 2140 */ 2141 pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask)); 2142 pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask)); 2143 pvinfo->hcall[2] = cpu_to_be32(inst_sc); 2144 pvinfo->hcall[3] = cpu_to_be32(inst_nop); 2145 #endif 2146 2147 pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE; 2148 2149 return 0; 2150 } 2151 2152 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm) 2153 { 2154 int ret = 0; 2155 2156 #ifdef CONFIG_KVM_MPIC 2157 ret = ret || (kvm->arch.mpic != NULL); 2158 #endif 2159 #ifdef CONFIG_KVM_XICS 2160 ret = ret || (kvm->arch.xics != NULL); 2161 ret = ret || (kvm->arch.xive != NULL); 2162 #endif 2163 smp_rmb(); 2164 return ret; 2165 } 2166 2167 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event, 2168 bool line_status) 2169 { 2170 if (!kvm_arch_irqchip_in_kernel(kvm)) 2171 return -ENXIO; 2172 2173 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, 2174 irq_event->irq, irq_event->level, 2175 line_status); 2176 return 0; 2177 } 2178 2179 2180 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, 2181 struct kvm_enable_cap *cap) 2182 { 2183 int r; 2184 2185 if (cap->flags) 2186 return -EINVAL; 2187 2188 switch (cap->cap) { 2189 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER 2190 case KVM_CAP_PPC_ENABLE_HCALL: { 2191 unsigned long hcall = cap->args[0]; 2192 2193 r = -EINVAL; 2194 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) || 2195 cap->args[1] > 1) 2196 break; 2197 if (!kvmppc_book3s_hcall_implemented(kvm, hcall)) 2198 break; 2199 if (cap->args[1]) 2200 set_bit(hcall / 4, kvm->arch.enabled_hcalls); 2201 else 2202 clear_bit(hcall / 4, kvm->arch.enabled_hcalls); 2203 r = 0; 2204 break; 2205 } 2206 case KVM_CAP_PPC_SMT: { 2207 unsigned long mode = cap->args[0]; 2208 unsigned long flags = cap->args[1]; 2209 2210 r = -EINVAL; 2211 if (kvm->arch.kvm_ops->set_smt_mode) 2212 r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags); 2213 break; 2214 } 2215 2216 case KVM_CAP_PPC_NESTED_HV: 2217 r = -EINVAL; 2218 if (!is_kvmppc_hv_enabled(kvm) || 2219 !kvm->arch.kvm_ops->enable_nested) 2220 break; 2221 r = kvm->arch.kvm_ops->enable_nested(kvm); 2222 break; 2223 #endif 2224 #if defined(CONFIG_KVM_BOOK3S_HV_POSSIBLE) 2225 case KVM_CAP_PPC_SECURE_GUEST: 2226 r = -EINVAL; 2227 if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_svm) 2228 break; 2229 r = kvm->arch.kvm_ops->enable_svm(kvm); 2230 break; 2231 case KVM_CAP_PPC_DAWR1: 2232 r = -EINVAL; 2233 if (!is_kvmppc_hv_enabled(kvm) || !kvm->arch.kvm_ops->enable_dawr1) 2234 break; 2235 r = kvm->arch.kvm_ops->enable_dawr1(kvm); 2236 break; 2237 #endif 2238 default: 2239 r = -EINVAL; 2240 break; 2241 } 2242 2243 return r; 2244 } 2245 2246 #ifdef CONFIG_PPC_BOOK3S_64 2247 /* 2248 * These functions check whether the underlying hardware is safe 2249 * against attacks based on observing the effects of speculatively 2250 * executed instructions, and whether it supplies instructions for 2251 * use in workarounds. The information comes from firmware, either 2252 * via the device tree on powernv platforms or from an hcall on 2253 * pseries platforms. 2254 */ 2255 #ifdef CONFIG_PPC_PSERIES 2256 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp) 2257 { 2258 struct h_cpu_char_result c; 2259 unsigned long rc; 2260 2261 if (!machine_is(pseries)) 2262 return -ENOTTY; 2263 2264 rc = plpar_get_cpu_characteristics(&c); 2265 if (rc == H_SUCCESS) { 2266 cp->character = c.character; 2267 cp->behaviour = c.behaviour; 2268 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 | 2269 KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED | 2270 KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 | 2271 KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 | 2272 KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV | 2273 KVM_PPC_CPU_CHAR_BR_HINT_HONOURED | 2274 KVM_PPC_CPU_CHAR_MTTRIG_THR_RECONF | 2275 KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS | 2276 KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST; 2277 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY | 2278 KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR | 2279 KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR | 2280 KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE; 2281 } 2282 return 0; 2283 } 2284 #else 2285 static int pseries_get_cpu_char(struct kvm_ppc_cpu_char *cp) 2286 { 2287 return -ENOTTY; 2288 } 2289 #endif 2290 2291 static inline bool have_fw_feat(struct device_node *fw_features, 2292 const char *state, const char *name) 2293 { 2294 struct device_node *np; 2295 bool r = false; 2296 2297 np = of_get_child_by_name(fw_features, name); 2298 if (np) { 2299 r = of_property_read_bool(np, state); 2300 of_node_put(np); 2301 } 2302 return r; 2303 } 2304 2305 static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp) 2306 { 2307 struct device_node *np, *fw_features; 2308 int r; 2309 2310 memset(cp, 0, sizeof(*cp)); 2311 r = pseries_get_cpu_char(cp); 2312 if (r != -ENOTTY) 2313 return r; 2314 2315 np = of_find_node_by_name(NULL, "ibm,opal"); 2316 if (np) { 2317 fw_features = of_get_child_by_name(np, "fw-features"); 2318 of_node_put(np); 2319 if (!fw_features) 2320 return 0; 2321 if (have_fw_feat(fw_features, "enabled", 2322 "inst-spec-barrier-ori31,31,0")) 2323 cp->character |= KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31; 2324 if (have_fw_feat(fw_features, "enabled", 2325 "fw-bcctrl-serialized")) 2326 cp->character |= KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED; 2327 if (have_fw_feat(fw_features, "enabled", 2328 "inst-l1d-flush-ori30,30,0")) 2329 cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30; 2330 if (have_fw_feat(fw_features, "enabled", 2331 "inst-l1d-flush-trig2")) 2332 cp->character |= KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2; 2333 if (have_fw_feat(fw_features, "enabled", 2334 "fw-l1d-thread-split")) 2335 cp->character |= KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV; 2336 if (have_fw_feat(fw_features, "enabled", 2337 "fw-count-cache-disabled")) 2338 cp->character |= KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS; 2339 if (have_fw_feat(fw_features, "enabled", 2340 "fw-count-cache-flush-bcctr2,0,0")) 2341 cp->character |= KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST; 2342 cp->character_mask = KVM_PPC_CPU_CHAR_SPEC_BAR_ORI31 | 2343 KVM_PPC_CPU_CHAR_BCCTRL_SERIALISED | 2344 KVM_PPC_CPU_CHAR_L1D_FLUSH_ORI30 | 2345 KVM_PPC_CPU_CHAR_L1D_FLUSH_TRIG2 | 2346 KVM_PPC_CPU_CHAR_L1D_THREAD_PRIV | 2347 KVM_PPC_CPU_CHAR_COUNT_CACHE_DIS | 2348 KVM_PPC_CPU_CHAR_BCCTR_FLUSH_ASSIST; 2349 2350 if (have_fw_feat(fw_features, "enabled", 2351 "speculation-policy-favor-security")) 2352 cp->behaviour |= KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY; 2353 if (!have_fw_feat(fw_features, "disabled", 2354 "needs-l1d-flush-msr-pr-0-to-1")) 2355 cp->behaviour |= KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR; 2356 if (!have_fw_feat(fw_features, "disabled", 2357 "needs-spec-barrier-for-bound-checks")) 2358 cp->behaviour |= KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR; 2359 if (have_fw_feat(fw_features, "enabled", 2360 "needs-count-cache-flush-on-context-switch")) 2361 cp->behaviour |= KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE; 2362 cp->behaviour_mask = KVM_PPC_CPU_BEHAV_FAVOUR_SECURITY | 2363 KVM_PPC_CPU_BEHAV_L1D_FLUSH_PR | 2364 KVM_PPC_CPU_BEHAV_BNDS_CHK_SPEC_BAR | 2365 KVM_PPC_CPU_BEHAV_FLUSH_COUNT_CACHE; 2366 2367 of_node_put(fw_features); 2368 } 2369 2370 return 0; 2371 } 2372 #endif 2373 2374 long kvm_arch_vm_ioctl(struct file *filp, 2375 unsigned int ioctl, unsigned long arg) 2376 { 2377 struct kvm *kvm __maybe_unused = filp->private_data; 2378 void __user *argp = (void __user *)arg; 2379 long r; 2380 2381 switch (ioctl) { 2382 case KVM_PPC_GET_PVINFO: { 2383 struct kvm_ppc_pvinfo pvinfo; 2384 memset(&pvinfo, 0, sizeof(pvinfo)); 2385 r = kvm_vm_ioctl_get_pvinfo(&pvinfo); 2386 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) { 2387 r = -EFAULT; 2388 goto out; 2389 } 2390 2391 break; 2392 } 2393 #ifdef CONFIG_SPAPR_TCE_IOMMU 2394 case KVM_CREATE_SPAPR_TCE_64: { 2395 struct kvm_create_spapr_tce_64 create_tce_64; 2396 2397 r = -EFAULT; 2398 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64))) 2399 goto out; 2400 if (create_tce_64.flags) { 2401 r = -EINVAL; 2402 goto out; 2403 } 2404 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64); 2405 goto out; 2406 } 2407 case KVM_CREATE_SPAPR_TCE: { 2408 struct kvm_create_spapr_tce create_tce; 2409 struct kvm_create_spapr_tce_64 create_tce_64; 2410 2411 r = -EFAULT; 2412 if (copy_from_user(&create_tce, argp, sizeof(create_tce))) 2413 goto out; 2414 2415 create_tce_64.liobn = create_tce.liobn; 2416 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K; 2417 create_tce_64.offset = 0; 2418 create_tce_64.size = create_tce.window_size >> 2419 IOMMU_PAGE_SHIFT_4K; 2420 create_tce_64.flags = 0; 2421 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64); 2422 goto out; 2423 } 2424 #endif 2425 #ifdef CONFIG_PPC_BOOK3S_64 2426 case KVM_PPC_GET_SMMU_INFO: { 2427 struct kvm_ppc_smmu_info info; 2428 struct kvm *kvm = filp->private_data; 2429 2430 memset(&info, 0, sizeof(info)); 2431 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info); 2432 if (r >= 0 && copy_to_user(argp, &info, sizeof(info))) 2433 r = -EFAULT; 2434 break; 2435 } 2436 case KVM_PPC_RTAS_DEFINE_TOKEN: { 2437 struct kvm *kvm = filp->private_data; 2438 2439 r = kvm_vm_ioctl_rtas_define_token(kvm, argp); 2440 break; 2441 } 2442 case KVM_PPC_CONFIGURE_V3_MMU: { 2443 struct kvm *kvm = filp->private_data; 2444 struct kvm_ppc_mmuv3_cfg cfg; 2445 2446 r = -EINVAL; 2447 if (!kvm->arch.kvm_ops->configure_mmu) 2448 goto out; 2449 r = -EFAULT; 2450 if (copy_from_user(&cfg, argp, sizeof(cfg))) 2451 goto out; 2452 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg); 2453 break; 2454 } 2455 case KVM_PPC_GET_RMMU_INFO: { 2456 struct kvm *kvm = filp->private_data; 2457 struct kvm_ppc_rmmu_info info; 2458 2459 r = -EINVAL; 2460 if (!kvm->arch.kvm_ops->get_rmmu_info) 2461 goto out; 2462 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info); 2463 if (r >= 0 && copy_to_user(argp, &info, sizeof(info))) 2464 r = -EFAULT; 2465 break; 2466 } 2467 case KVM_PPC_GET_CPU_CHAR: { 2468 struct kvm_ppc_cpu_char cpuchar; 2469 2470 r = kvmppc_get_cpu_char(&cpuchar); 2471 if (r >= 0 && copy_to_user(argp, &cpuchar, sizeof(cpuchar))) 2472 r = -EFAULT; 2473 break; 2474 } 2475 case KVM_PPC_SVM_OFF: { 2476 struct kvm *kvm = filp->private_data; 2477 2478 r = 0; 2479 if (!kvm->arch.kvm_ops->svm_off) 2480 goto out; 2481 2482 r = kvm->arch.kvm_ops->svm_off(kvm); 2483 break; 2484 } 2485 default: { 2486 struct kvm *kvm = filp->private_data; 2487 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg); 2488 } 2489 #else /* CONFIG_PPC_BOOK3S_64 */ 2490 default: 2491 r = -ENOTTY; 2492 #endif 2493 } 2494 out: 2495 return r; 2496 } 2497 2498 static DEFINE_IDA(lpid_inuse); 2499 static unsigned long nr_lpids; 2500 2501 long kvmppc_alloc_lpid(void) 2502 { 2503 int lpid; 2504 2505 /* The host LPID must always be 0 (allocation starts at 1) */ 2506 lpid = ida_alloc_range(&lpid_inuse, 1, nr_lpids - 1, GFP_KERNEL); 2507 if (lpid < 0) { 2508 if (lpid == -ENOMEM) 2509 pr_err("%s: Out of memory\n", __func__); 2510 else 2511 pr_err("%s: No LPIDs free\n", __func__); 2512 return -ENOMEM; 2513 } 2514 2515 return lpid; 2516 } 2517 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid); 2518 2519 void kvmppc_free_lpid(long lpid) 2520 { 2521 ida_free(&lpid_inuse, lpid); 2522 } 2523 EXPORT_SYMBOL_GPL(kvmppc_free_lpid); 2524 2525 /* nr_lpids_param includes the host LPID */ 2526 void kvmppc_init_lpid(unsigned long nr_lpids_param) 2527 { 2528 nr_lpids = nr_lpids_param; 2529 } 2530 EXPORT_SYMBOL_GPL(kvmppc_init_lpid); 2531 2532 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr); 2533 2534 void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry) 2535 { 2536 if (vcpu->kvm->arch.kvm_ops->create_vcpu_debugfs) 2537 vcpu->kvm->arch.kvm_ops->create_vcpu_debugfs(vcpu, debugfs_dentry); 2538 } 2539 2540 int kvm_arch_create_vm_debugfs(struct kvm *kvm) 2541 { 2542 if (kvm->arch.kvm_ops->create_vm_debugfs) 2543 kvm->arch.kvm_ops->create_vm_debugfs(kvm); 2544 return 0; 2545 } 2546