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