1 /* 2 * This program is free software; you can redistribute it and/or modify 3 * it under the terms of the GNU General Public License, version 2, as 4 * published by the Free Software Foundation. 5 * 6 * This program is distributed in the hope that it will be useful, 7 * but WITHOUT ANY WARRANTY; without even the implied warranty of 8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 9 * GNU General Public License for more details. 10 * 11 * You should have received a copy of the GNU General Public License 12 * along with this program; if not, write to the Free Software 13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. 14 * 15 * Copyright IBM Corp. 2007 16 * 17 * Authors: Hollis Blanchard <hollisb@us.ibm.com> 18 * Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com> 19 */ 20 21 #include <linux/errno.h> 22 #include <linux/err.h> 23 #include <linux/kvm_host.h> 24 #include <linux/vmalloc.h> 25 #include <linux/hrtimer.h> 26 #include <linux/sched/signal.h> 27 #include <linux/fs.h> 28 #include <linux/slab.h> 29 #include <linux/file.h> 30 #include <linux/module.h> 31 #include <linux/irqbypass.h> 32 #include <linux/kvm_irqfd.h> 33 #include <asm/cputable.h> 34 #include <linux/uaccess.h> 35 #include <asm/kvm_ppc.h> 36 #include <asm/tlbflush.h> 37 #include <asm/cputhreads.h> 38 #include <asm/irqflags.h> 39 #include <asm/iommu.h> 40 #include <asm/switch_to.h> 41 #include <asm/xive.h> 42 43 #include "timing.h" 44 #include "irq.h" 45 #include "../mm/mmu_decl.h" 46 47 #define CREATE_TRACE_POINTS 48 #include "trace.h" 49 50 struct kvmppc_ops *kvmppc_hv_ops; 51 EXPORT_SYMBOL_GPL(kvmppc_hv_ops); 52 struct kvmppc_ops *kvmppc_pr_ops; 53 EXPORT_SYMBOL_GPL(kvmppc_pr_ops); 54 55 56 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) 57 { 58 return !!(v->arch.pending_exceptions) || kvm_request_pending(v); 59 } 60 61 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 62 { 63 return false; 64 } 65 66 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 67 { 68 return 1; 69 } 70 71 /* 72 * Common checks before entering the guest world. Call with interrupts 73 * disabled. 74 * 75 * returns: 76 * 77 * == 1 if we're ready to go into guest state 78 * <= 0 if we need to go back to the host with return value 79 */ 80 int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu) 81 { 82 int r; 83 84 WARN_ON(irqs_disabled()); 85 hard_irq_disable(); 86 87 while (true) { 88 if (need_resched()) { 89 local_irq_enable(); 90 cond_resched(); 91 hard_irq_disable(); 92 continue; 93 } 94 95 if (signal_pending(current)) { 96 kvmppc_account_exit(vcpu, SIGNAL_EXITS); 97 vcpu->run->exit_reason = KVM_EXIT_INTR; 98 r = -EINTR; 99 break; 100 } 101 102 vcpu->mode = IN_GUEST_MODE; 103 104 /* 105 * Reading vcpu->requests must happen after setting vcpu->mode, 106 * so we don't miss a request because the requester sees 107 * OUTSIDE_GUEST_MODE and assumes we'll be checking requests 108 * before next entering the guest (and thus doesn't IPI). 109 * This also orders the write to mode from any reads 110 * to the page tables done while the VCPU is running. 111 * Please see the comment in kvm_flush_remote_tlbs. 112 */ 113 smp_mb(); 114 115 if (kvm_request_pending(vcpu)) { 116 /* Make sure we process requests preemptable */ 117 local_irq_enable(); 118 trace_kvm_check_requests(vcpu); 119 r = kvmppc_core_check_requests(vcpu); 120 hard_irq_disable(); 121 if (r > 0) 122 continue; 123 break; 124 } 125 126 if (kvmppc_core_prepare_to_enter(vcpu)) { 127 /* interrupts got enabled in between, so we 128 are back at square 1 */ 129 continue; 130 } 131 132 guest_enter_irqoff(); 133 return 1; 134 } 135 136 /* return to host */ 137 local_irq_enable(); 138 return r; 139 } 140 EXPORT_SYMBOL_GPL(kvmppc_prepare_to_enter); 141 142 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE) 143 static void kvmppc_swab_shared(struct kvm_vcpu *vcpu) 144 { 145 struct kvm_vcpu_arch_shared *shared = vcpu->arch.shared; 146 int i; 147 148 shared->sprg0 = swab64(shared->sprg0); 149 shared->sprg1 = swab64(shared->sprg1); 150 shared->sprg2 = swab64(shared->sprg2); 151 shared->sprg3 = swab64(shared->sprg3); 152 shared->srr0 = swab64(shared->srr0); 153 shared->srr1 = swab64(shared->srr1); 154 shared->dar = swab64(shared->dar); 155 shared->msr = swab64(shared->msr); 156 shared->dsisr = swab32(shared->dsisr); 157 shared->int_pending = swab32(shared->int_pending); 158 for (i = 0; i < ARRAY_SIZE(shared->sr); i++) 159 shared->sr[i] = swab32(shared->sr[i]); 160 } 161 #endif 162 163 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu) 164 { 165 int nr = kvmppc_get_gpr(vcpu, 11); 166 int r; 167 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3); 168 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4); 169 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5); 170 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6); 171 unsigned long r2 = 0; 172 173 if (!(kvmppc_get_msr(vcpu) & MSR_SF)) { 174 /* 32 bit mode */ 175 param1 &= 0xffffffff; 176 param2 &= 0xffffffff; 177 param3 &= 0xffffffff; 178 param4 &= 0xffffffff; 179 } 180 181 switch (nr) { 182 case KVM_HCALL_TOKEN(KVM_HC_PPC_MAP_MAGIC_PAGE): 183 { 184 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_KVM_BOOK3S_PR_POSSIBLE) 185 /* Book3S can be little endian, find it out here */ 186 int shared_big_endian = true; 187 if (vcpu->arch.intr_msr & MSR_LE) 188 shared_big_endian = false; 189 if (shared_big_endian != vcpu->arch.shared_big_endian) 190 kvmppc_swab_shared(vcpu); 191 vcpu->arch.shared_big_endian = shared_big_endian; 192 #endif 193 194 if (!(param2 & MAGIC_PAGE_FLAG_NOT_MAPPED_NX)) { 195 /* 196 * Older versions of the Linux magic page code had 197 * a bug where they would map their trampoline code 198 * NX. If that's the case, remove !PR NX capability. 199 */ 200 vcpu->arch.disable_kernel_nx = true; 201 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 202 } 203 204 vcpu->arch.magic_page_pa = param1 & ~0xfffULL; 205 vcpu->arch.magic_page_ea = param2 & ~0xfffULL; 206 207 #ifdef CONFIG_PPC_64K_PAGES 208 /* 209 * Make sure our 4k magic page is in the same window of a 64k 210 * page within the guest and within the host's page. 211 */ 212 if ((vcpu->arch.magic_page_pa & 0xf000) != 213 ((ulong)vcpu->arch.shared & 0xf000)) { 214 void *old_shared = vcpu->arch.shared; 215 ulong shared = (ulong)vcpu->arch.shared; 216 void *new_shared; 217 218 shared &= PAGE_MASK; 219 shared |= vcpu->arch.magic_page_pa & 0xf000; 220 new_shared = (void*)shared; 221 memcpy(new_shared, old_shared, 0x1000); 222 vcpu->arch.shared = new_shared; 223 } 224 #endif 225 226 r2 = KVM_MAGIC_FEAT_SR | KVM_MAGIC_FEAT_MAS0_TO_SPRG7; 227 228 r = EV_SUCCESS; 229 break; 230 } 231 case KVM_HCALL_TOKEN(KVM_HC_FEATURES): 232 r = EV_SUCCESS; 233 #if defined(CONFIG_PPC_BOOK3S) || defined(CONFIG_KVM_E500V2) 234 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE); 235 #endif 236 237 /* Second return value is in r4 */ 238 break; 239 case EV_HCALL_TOKEN(EV_IDLE): 240 r = EV_SUCCESS; 241 kvm_vcpu_block(vcpu); 242 kvm_clear_request(KVM_REQ_UNHALT, vcpu); 243 break; 244 default: 245 r = EV_UNIMPLEMENTED; 246 break; 247 } 248 249 kvmppc_set_gpr(vcpu, 4, r2); 250 251 return r; 252 } 253 EXPORT_SYMBOL_GPL(kvmppc_kvm_pv); 254 255 int kvmppc_sanity_check(struct kvm_vcpu *vcpu) 256 { 257 int r = false; 258 259 /* We have to know what CPU to virtualize */ 260 if (!vcpu->arch.pvr) 261 goto out; 262 263 /* PAPR only works with book3s_64 */ 264 if ((vcpu->arch.cpu_type != KVM_CPU_3S_64) && vcpu->arch.papr_enabled) 265 goto out; 266 267 /* HV KVM can only do PAPR mode for now */ 268 if (!vcpu->arch.papr_enabled && is_kvmppc_hv_enabled(vcpu->kvm)) 269 goto out; 270 271 #ifdef CONFIG_KVM_BOOKE_HV 272 if (!cpu_has_feature(CPU_FTR_EMB_HV)) 273 goto out; 274 #endif 275 276 r = true; 277 278 out: 279 vcpu->arch.sane = r; 280 return r ? 0 : -EINVAL; 281 } 282 EXPORT_SYMBOL_GPL(kvmppc_sanity_check); 283 284 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu) 285 { 286 enum emulation_result er; 287 int r; 288 289 er = kvmppc_emulate_loadstore(vcpu); 290 switch (er) { 291 case EMULATE_DONE: 292 /* Future optimization: only reload non-volatiles if they were 293 * actually modified. */ 294 r = RESUME_GUEST_NV; 295 break; 296 case EMULATE_AGAIN: 297 r = RESUME_GUEST; 298 break; 299 case EMULATE_DO_MMIO: 300 run->exit_reason = KVM_EXIT_MMIO; 301 /* We must reload nonvolatiles because "update" load/store 302 * instructions modify register state. */ 303 /* Future optimization: only reload non-volatiles if they were 304 * actually modified. */ 305 r = RESUME_HOST_NV; 306 break; 307 case EMULATE_FAIL: 308 { 309 u32 last_inst; 310 311 kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst); 312 /* XXX Deliver Program interrupt to guest. */ 313 pr_emerg("%s: emulation failed (%08x)\n", __func__, last_inst); 314 r = RESUME_HOST; 315 break; 316 } 317 default: 318 WARN_ON(1); 319 r = RESUME_GUEST; 320 } 321 322 return r; 323 } 324 EXPORT_SYMBOL_GPL(kvmppc_emulate_mmio); 325 326 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr, 327 bool data) 328 { 329 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK; 330 struct kvmppc_pte pte; 331 int r; 332 333 vcpu->stat.st++; 334 335 r = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST, 336 XLATE_WRITE, &pte); 337 if (r < 0) 338 return r; 339 340 *eaddr = pte.raddr; 341 342 if (!pte.may_write) 343 return -EPERM; 344 345 /* Magic page override */ 346 if (kvmppc_supports_magic_page(vcpu) && mp_pa && 347 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) && 348 !(kvmppc_get_msr(vcpu) & MSR_PR)) { 349 void *magic = vcpu->arch.shared; 350 magic += pte.eaddr & 0xfff; 351 memcpy(magic, ptr, size); 352 return EMULATE_DONE; 353 } 354 355 if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size)) 356 return EMULATE_DO_MMIO; 357 358 return EMULATE_DONE; 359 } 360 EXPORT_SYMBOL_GPL(kvmppc_st); 361 362 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr, 363 bool data) 364 { 365 ulong mp_pa = vcpu->arch.magic_page_pa & KVM_PAM & PAGE_MASK; 366 struct kvmppc_pte pte; 367 int rc; 368 369 vcpu->stat.ld++; 370 371 rc = kvmppc_xlate(vcpu, *eaddr, data ? XLATE_DATA : XLATE_INST, 372 XLATE_READ, &pte); 373 if (rc) 374 return rc; 375 376 *eaddr = pte.raddr; 377 378 if (!pte.may_read) 379 return -EPERM; 380 381 if (!data && !pte.may_execute) 382 return -ENOEXEC; 383 384 /* Magic page override */ 385 if (kvmppc_supports_magic_page(vcpu) && mp_pa && 386 ((pte.raddr & KVM_PAM & PAGE_MASK) == mp_pa) && 387 !(kvmppc_get_msr(vcpu) & MSR_PR)) { 388 void *magic = vcpu->arch.shared; 389 magic += pte.eaddr & 0xfff; 390 memcpy(ptr, magic, size); 391 return EMULATE_DONE; 392 } 393 394 if (kvm_read_guest(vcpu->kvm, pte.raddr, ptr, size)) 395 return EMULATE_DO_MMIO; 396 397 return EMULATE_DONE; 398 } 399 EXPORT_SYMBOL_GPL(kvmppc_ld); 400 401 int kvm_arch_hardware_enable(void) 402 { 403 return 0; 404 } 405 406 int kvm_arch_hardware_setup(void) 407 { 408 return 0; 409 } 410 411 void kvm_arch_check_processor_compat(void *rtn) 412 { 413 *(int *)rtn = kvmppc_core_check_processor_compat(); 414 } 415 416 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 417 { 418 struct kvmppc_ops *kvm_ops = NULL; 419 /* 420 * if we have both HV and PR enabled, default is HV 421 */ 422 if (type == 0) { 423 if (kvmppc_hv_ops) 424 kvm_ops = kvmppc_hv_ops; 425 else 426 kvm_ops = kvmppc_pr_ops; 427 if (!kvm_ops) 428 goto err_out; 429 } else if (type == KVM_VM_PPC_HV) { 430 if (!kvmppc_hv_ops) 431 goto err_out; 432 kvm_ops = kvmppc_hv_ops; 433 } else if (type == KVM_VM_PPC_PR) { 434 if (!kvmppc_pr_ops) 435 goto err_out; 436 kvm_ops = kvmppc_pr_ops; 437 } else 438 goto err_out; 439 440 if (kvm_ops->owner && !try_module_get(kvm_ops->owner)) 441 return -ENOENT; 442 443 kvm->arch.kvm_ops = kvm_ops; 444 return kvmppc_core_init_vm(kvm); 445 err_out: 446 return -EINVAL; 447 } 448 449 bool kvm_arch_has_vcpu_debugfs(void) 450 { 451 return false; 452 } 453 454 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu) 455 { 456 return 0; 457 } 458 459 void kvm_arch_destroy_vm(struct kvm *kvm) 460 { 461 unsigned int i; 462 struct kvm_vcpu *vcpu; 463 464 #ifdef CONFIG_KVM_XICS 465 /* 466 * We call kick_all_cpus_sync() to ensure that all 467 * CPUs have executed any pending IPIs before we 468 * continue and free VCPUs structures below. 469 */ 470 if (is_kvmppc_hv_enabled(kvm)) 471 kick_all_cpus_sync(); 472 #endif 473 474 kvm_for_each_vcpu(i, vcpu, kvm) 475 kvm_arch_vcpu_free(vcpu); 476 477 mutex_lock(&kvm->lock); 478 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) 479 kvm->vcpus[i] = NULL; 480 481 atomic_set(&kvm->online_vcpus, 0); 482 483 kvmppc_core_destroy_vm(kvm); 484 485 mutex_unlock(&kvm->lock); 486 487 /* drop the module reference */ 488 module_put(kvm->arch.kvm_ops->owner); 489 } 490 491 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 492 { 493 int r; 494 /* Assume we're using HV mode when the HV module is loaded */ 495 int hv_enabled = kvmppc_hv_ops ? 1 : 0; 496 497 if (kvm) { 498 /* 499 * Hooray - we know which VM type we're running on. Depend on 500 * that rather than the guess above. 501 */ 502 hv_enabled = is_kvmppc_hv_enabled(kvm); 503 } 504 505 switch (ext) { 506 #ifdef CONFIG_BOOKE 507 case KVM_CAP_PPC_BOOKE_SREGS: 508 case KVM_CAP_PPC_BOOKE_WATCHDOG: 509 case KVM_CAP_PPC_EPR: 510 #else 511 case KVM_CAP_PPC_SEGSTATE: 512 case KVM_CAP_PPC_HIOR: 513 case KVM_CAP_PPC_PAPR: 514 #endif 515 case KVM_CAP_PPC_UNSET_IRQ: 516 case KVM_CAP_PPC_IRQ_LEVEL: 517 case KVM_CAP_ENABLE_CAP: 518 case KVM_CAP_ENABLE_CAP_VM: 519 case KVM_CAP_ONE_REG: 520 case KVM_CAP_IOEVENTFD: 521 case KVM_CAP_DEVICE_CTRL: 522 case KVM_CAP_IMMEDIATE_EXIT: 523 r = 1; 524 break; 525 case KVM_CAP_PPC_PAIRED_SINGLES: 526 case KVM_CAP_PPC_OSI: 527 case KVM_CAP_PPC_GET_PVINFO: 528 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC) 529 case KVM_CAP_SW_TLB: 530 #endif 531 /* We support this only for PR */ 532 r = !hv_enabled; 533 break; 534 #ifdef CONFIG_KVM_MPIC 535 case KVM_CAP_IRQ_MPIC: 536 r = 1; 537 break; 538 #endif 539 540 #ifdef CONFIG_PPC_BOOK3S_64 541 case KVM_CAP_SPAPR_TCE: 542 case KVM_CAP_SPAPR_TCE_64: 543 /* fallthrough */ 544 case KVM_CAP_SPAPR_TCE_VFIO: 545 case KVM_CAP_PPC_RTAS: 546 case KVM_CAP_PPC_FIXUP_HCALL: 547 case KVM_CAP_PPC_ENABLE_HCALL: 548 #ifdef CONFIG_KVM_XICS 549 case KVM_CAP_IRQ_XICS: 550 #endif 551 r = 1; 552 break; 553 554 case KVM_CAP_PPC_ALLOC_HTAB: 555 r = hv_enabled; 556 break; 557 #endif /* CONFIG_PPC_BOOK3S_64 */ 558 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 559 case KVM_CAP_PPC_SMT: 560 r = 0; 561 if (kvm) { 562 if (kvm->arch.emul_smt_mode > 1) 563 r = kvm->arch.emul_smt_mode; 564 else 565 r = kvm->arch.smt_mode; 566 } else if (hv_enabled) { 567 if (cpu_has_feature(CPU_FTR_ARCH_300)) 568 r = 1; 569 else 570 r = threads_per_subcore; 571 } 572 break; 573 case KVM_CAP_PPC_SMT_POSSIBLE: 574 r = 1; 575 if (hv_enabled) { 576 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 577 r = ((threads_per_subcore << 1) - 1); 578 else 579 /* P9 can emulate dbells, so allow any mode */ 580 r = 8 | 4 | 2 | 1; 581 } 582 break; 583 case KVM_CAP_PPC_RMA: 584 r = 0; 585 break; 586 case KVM_CAP_PPC_HWRNG: 587 r = kvmppc_hwrng_present(); 588 break; 589 case KVM_CAP_PPC_MMU_RADIX: 590 r = !!(hv_enabled && radix_enabled()); 591 break; 592 case KVM_CAP_PPC_MMU_HASH_V3: 593 r = !!(hv_enabled && !radix_enabled() && 594 cpu_has_feature(CPU_FTR_ARCH_300)); 595 break; 596 #endif 597 case KVM_CAP_SYNC_MMU: 598 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 599 r = hv_enabled; 600 #elif defined(KVM_ARCH_WANT_MMU_NOTIFIER) 601 r = 1; 602 #else 603 r = 0; 604 #endif 605 break; 606 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 607 case KVM_CAP_PPC_HTAB_FD: 608 r = hv_enabled; 609 break; 610 #endif 611 case KVM_CAP_NR_VCPUS: 612 /* 613 * Recommending a number of CPUs is somewhat arbitrary; we 614 * return the number of present CPUs for -HV (since a host 615 * will have secondary threads "offline"), and for other KVM 616 * implementations just count online CPUs. 617 */ 618 if (hv_enabled) 619 r = num_present_cpus(); 620 else 621 r = num_online_cpus(); 622 break; 623 case KVM_CAP_NR_MEMSLOTS: 624 r = KVM_USER_MEM_SLOTS; 625 break; 626 case KVM_CAP_MAX_VCPUS: 627 r = KVM_MAX_VCPUS; 628 break; 629 #ifdef CONFIG_PPC_BOOK3S_64 630 case KVM_CAP_PPC_GET_SMMU_INFO: 631 r = 1; 632 break; 633 case KVM_CAP_SPAPR_MULTITCE: 634 r = 1; 635 break; 636 case KVM_CAP_SPAPR_RESIZE_HPT: 637 /* Disable this on POWER9 until code handles new HPTE format */ 638 r = !!hv_enabled && !cpu_has_feature(CPU_FTR_ARCH_300); 639 break; 640 #endif 641 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 642 case KVM_CAP_PPC_FWNMI: 643 r = hv_enabled; 644 break; 645 #endif 646 case KVM_CAP_PPC_HTM: 647 r = cpu_has_feature(CPU_FTR_TM_COMP) && 648 is_kvmppc_hv_enabled(kvm); 649 break; 650 default: 651 r = 0; 652 break; 653 } 654 return r; 655 656 } 657 658 long kvm_arch_dev_ioctl(struct file *filp, 659 unsigned int ioctl, unsigned long arg) 660 { 661 return -EINVAL; 662 } 663 664 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, 665 struct kvm_memory_slot *dont) 666 { 667 kvmppc_core_free_memslot(kvm, free, dont); 668 } 669 670 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, 671 unsigned long npages) 672 { 673 return kvmppc_core_create_memslot(kvm, slot, npages); 674 } 675 676 int kvm_arch_prepare_memory_region(struct kvm *kvm, 677 struct kvm_memory_slot *memslot, 678 const struct kvm_userspace_memory_region *mem, 679 enum kvm_mr_change change) 680 { 681 return kvmppc_core_prepare_memory_region(kvm, memslot, mem); 682 } 683 684 void kvm_arch_commit_memory_region(struct kvm *kvm, 685 const struct kvm_userspace_memory_region *mem, 686 const struct kvm_memory_slot *old, 687 const struct kvm_memory_slot *new, 688 enum kvm_mr_change change) 689 { 690 kvmppc_core_commit_memory_region(kvm, mem, old, new); 691 } 692 693 void kvm_arch_flush_shadow_memslot(struct kvm *kvm, 694 struct kvm_memory_slot *slot) 695 { 696 kvmppc_core_flush_memslot(kvm, slot); 697 } 698 699 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) 700 { 701 struct kvm_vcpu *vcpu; 702 vcpu = kvmppc_core_vcpu_create(kvm, id); 703 if (!IS_ERR(vcpu)) { 704 vcpu->arch.wqp = &vcpu->wq; 705 kvmppc_create_vcpu_debugfs(vcpu, id); 706 } 707 return vcpu; 708 } 709 710 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 711 { 712 } 713 714 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) 715 { 716 /* Make sure we're not using the vcpu anymore */ 717 hrtimer_cancel(&vcpu->arch.dec_timer); 718 719 kvmppc_remove_vcpu_debugfs(vcpu); 720 721 switch (vcpu->arch.irq_type) { 722 case KVMPPC_IRQ_MPIC: 723 kvmppc_mpic_disconnect_vcpu(vcpu->arch.mpic, vcpu); 724 break; 725 case KVMPPC_IRQ_XICS: 726 if (xive_enabled()) 727 kvmppc_xive_cleanup_vcpu(vcpu); 728 else 729 kvmppc_xics_free_icp(vcpu); 730 break; 731 } 732 733 kvmppc_core_vcpu_free(vcpu); 734 } 735 736 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 737 { 738 kvm_arch_vcpu_free(vcpu); 739 } 740 741 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) 742 { 743 return kvmppc_core_pending_dec(vcpu); 744 } 745 746 static enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer) 747 { 748 struct kvm_vcpu *vcpu; 749 750 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer); 751 kvmppc_decrementer_func(vcpu); 752 753 return HRTIMER_NORESTART; 754 } 755 756 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) 757 { 758 int ret; 759 760 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS); 761 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup; 762 vcpu->arch.dec_expires = ~(u64)0; 763 764 #ifdef CONFIG_KVM_EXIT_TIMING 765 mutex_init(&vcpu->arch.exit_timing_lock); 766 #endif 767 ret = kvmppc_subarch_vcpu_init(vcpu); 768 return ret; 769 } 770 771 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) 772 { 773 kvmppc_mmu_destroy(vcpu); 774 kvmppc_subarch_vcpu_uninit(vcpu); 775 } 776 777 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 778 { 779 #ifdef CONFIG_BOOKE 780 /* 781 * vrsave (formerly usprg0) isn't used by Linux, but may 782 * be used by the guest. 783 * 784 * On non-booke this is associated with Altivec and 785 * is handled by code in book3s.c. 786 */ 787 mtspr(SPRN_VRSAVE, vcpu->arch.vrsave); 788 #endif 789 kvmppc_core_vcpu_load(vcpu, cpu); 790 } 791 792 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 793 { 794 kvmppc_core_vcpu_put(vcpu); 795 #ifdef CONFIG_BOOKE 796 vcpu->arch.vrsave = mfspr(SPRN_VRSAVE); 797 #endif 798 } 799 800 /* 801 * irq_bypass_add_producer and irq_bypass_del_producer are only 802 * useful if the architecture supports PCI passthrough. 803 * irq_bypass_stop and irq_bypass_start are not needed and so 804 * kvm_ops are not defined for them. 805 */ 806 bool kvm_arch_has_irq_bypass(void) 807 { 808 return ((kvmppc_hv_ops && kvmppc_hv_ops->irq_bypass_add_producer) || 809 (kvmppc_pr_ops && kvmppc_pr_ops->irq_bypass_add_producer)); 810 } 811 812 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, 813 struct irq_bypass_producer *prod) 814 { 815 struct kvm_kernel_irqfd *irqfd = 816 container_of(cons, struct kvm_kernel_irqfd, consumer); 817 struct kvm *kvm = irqfd->kvm; 818 819 if (kvm->arch.kvm_ops->irq_bypass_add_producer) 820 return kvm->arch.kvm_ops->irq_bypass_add_producer(cons, prod); 821 822 return 0; 823 } 824 825 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, 826 struct irq_bypass_producer *prod) 827 { 828 struct kvm_kernel_irqfd *irqfd = 829 container_of(cons, struct kvm_kernel_irqfd, consumer); 830 struct kvm *kvm = irqfd->kvm; 831 832 if (kvm->arch.kvm_ops->irq_bypass_del_producer) 833 kvm->arch.kvm_ops->irq_bypass_del_producer(cons, prod); 834 } 835 836 #ifdef CONFIG_VSX 837 static inline int kvmppc_get_vsr_dword_offset(int index) 838 { 839 int offset; 840 841 if ((index != 0) && (index != 1)) 842 return -1; 843 844 #ifdef __BIG_ENDIAN 845 offset = index; 846 #else 847 offset = 1 - index; 848 #endif 849 850 return offset; 851 } 852 853 static inline int kvmppc_get_vsr_word_offset(int index) 854 { 855 int offset; 856 857 if ((index > 3) || (index < 0)) 858 return -1; 859 860 #ifdef __BIG_ENDIAN 861 offset = index; 862 #else 863 offset = 3 - index; 864 #endif 865 return offset; 866 } 867 868 static inline void kvmppc_set_vsr_dword(struct kvm_vcpu *vcpu, 869 u64 gpr) 870 { 871 union kvmppc_one_reg val; 872 int offset = kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset); 873 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 874 875 if (offset == -1) 876 return; 877 878 if (vcpu->arch.mmio_vsx_tx_sx_enabled) { 879 val.vval = VCPU_VSX_VR(vcpu, index); 880 val.vsxval[offset] = gpr; 881 VCPU_VSX_VR(vcpu, index) = val.vval; 882 } else { 883 VCPU_VSX_FPR(vcpu, index, offset) = gpr; 884 } 885 } 886 887 static inline void kvmppc_set_vsr_dword_dump(struct kvm_vcpu *vcpu, 888 u64 gpr) 889 { 890 union kvmppc_one_reg val; 891 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 892 893 if (vcpu->arch.mmio_vsx_tx_sx_enabled) { 894 val.vval = VCPU_VSX_VR(vcpu, index); 895 val.vsxval[0] = gpr; 896 val.vsxval[1] = gpr; 897 VCPU_VSX_VR(vcpu, index) = val.vval; 898 } else { 899 VCPU_VSX_FPR(vcpu, index, 0) = gpr; 900 VCPU_VSX_FPR(vcpu, index, 1) = gpr; 901 } 902 } 903 904 static inline void kvmppc_set_vsr_word(struct kvm_vcpu *vcpu, 905 u32 gpr32) 906 { 907 union kvmppc_one_reg val; 908 int offset = kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset); 909 int index = vcpu->arch.io_gpr & KVM_MMIO_REG_MASK; 910 int dword_offset, word_offset; 911 912 if (offset == -1) 913 return; 914 915 if (vcpu->arch.mmio_vsx_tx_sx_enabled) { 916 val.vval = VCPU_VSX_VR(vcpu, index); 917 val.vsx32val[offset] = gpr32; 918 VCPU_VSX_VR(vcpu, index) = val.vval; 919 } else { 920 dword_offset = offset / 2; 921 word_offset = offset % 2; 922 val.vsxval[0] = VCPU_VSX_FPR(vcpu, index, dword_offset); 923 val.vsx32val[word_offset] = gpr32; 924 VCPU_VSX_FPR(vcpu, index, dword_offset) = val.vsxval[0]; 925 } 926 } 927 #endif /* CONFIG_VSX */ 928 929 #ifdef CONFIG_PPC_FPU 930 static inline u64 sp_to_dp(u32 fprs) 931 { 932 u64 fprd; 933 934 preempt_disable(); 935 enable_kernel_fp(); 936 asm ("lfs%U1%X1 0,%1; stfd%U0%X0 0,%0" : "=m" (fprd) : "m" (fprs) 937 : "fr0"); 938 preempt_enable(); 939 return fprd; 940 } 941 942 static inline u32 dp_to_sp(u64 fprd) 943 { 944 u32 fprs; 945 946 preempt_disable(); 947 enable_kernel_fp(); 948 asm ("lfd%U1%X1 0,%1; stfs%U0%X0 0,%0" : "=m" (fprs) : "m" (fprd) 949 : "fr0"); 950 preempt_enable(); 951 return fprs; 952 } 953 954 #else 955 #define sp_to_dp(x) (x) 956 #define dp_to_sp(x) (x) 957 #endif /* CONFIG_PPC_FPU */ 958 959 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu, 960 struct kvm_run *run) 961 { 962 u64 uninitialized_var(gpr); 963 964 if (run->mmio.len > sizeof(gpr)) { 965 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len); 966 return; 967 } 968 969 if (!vcpu->arch.mmio_host_swabbed) { 970 switch (run->mmio.len) { 971 case 8: gpr = *(u64 *)run->mmio.data; break; 972 case 4: gpr = *(u32 *)run->mmio.data; break; 973 case 2: gpr = *(u16 *)run->mmio.data; break; 974 case 1: gpr = *(u8 *)run->mmio.data; break; 975 } 976 } else { 977 switch (run->mmio.len) { 978 case 8: gpr = swab64(*(u64 *)run->mmio.data); break; 979 case 4: gpr = swab32(*(u32 *)run->mmio.data); break; 980 case 2: gpr = swab16(*(u16 *)run->mmio.data); break; 981 case 1: gpr = *(u8 *)run->mmio.data; break; 982 } 983 } 984 985 /* conversion between single and double precision */ 986 if ((vcpu->arch.mmio_sp64_extend) && (run->mmio.len == 4)) 987 gpr = sp_to_dp(gpr); 988 989 if (vcpu->arch.mmio_sign_extend) { 990 switch (run->mmio.len) { 991 #ifdef CONFIG_PPC64 992 case 4: 993 gpr = (s64)(s32)gpr; 994 break; 995 #endif 996 case 2: 997 gpr = (s64)(s16)gpr; 998 break; 999 case 1: 1000 gpr = (s64)(s8)gpr; 1001 break; 1002 } 1003 } 1004 1005 switch (vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) { 1006 case KVM_MMIO_REG_GPR: 1007 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr); 1008 break; 1009 case KVM_MMIO_REG_FPR: 1010 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr; 1011 break; 1012 #ifdef CONFIG_PPC_BOOK3S 1013 case KVM_MMIO_REG_QPR: 1014 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr; 1015 break; 1016 case KVM_MMIO_REG_FQPR: 1017 VCPU_FPR(vcpu, vcpu->arch.io_gpr & KVM_MMIO_REG_MASK) = gpr; 1018 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_MMIO_REG_MASK] = gpr; 1019 break; 1020 #endif 1021 #ifdef CONFIG_VSX 1022 case KVM_MMIO_REG_VSX: 1023 if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_DWORD) 1024 kvmppc_set_vsr_dword(vcpu, gpr); 1025 else if (vcpu->arch.mmio_vsx_copy_type == KVMPPC_VSX_COPY_WORD) 1026 kvmppc_set_vsr_word(vcpu, gpr); 1027 else if (vcpu->arch.mmio_vsx_copy_type == 1028 KVMPPC_VSX_COPY_DWORD_LOAD_DUMP) 1029 kvmppc_set_vsr_dword_dump(vcpu, gpr); 1030 break; 1031 #endif 1032 default: 1033 BUG(); 1034 } 1035 } 1036 1037 static int __kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu, 1038 unsigned int rt, unsigned int bytes, 1039 int is_default_endian, int sign_extend) 1040 { 1041 int idx, ret; 1042 bool host_swabbed; 1043 1044 /* Pity C doesn't have a logical XOR operator */ 1045 if (kvmppc_need_byteswap(vcpu)) { 1046 host_swabbed = is_default_endian; 1047 } else { 1048 host_swabbed = !is_default_endian; 1049 } 1050 1051 if (bytes > sizeof(run->mmio.data)) { 1052 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__, 1053 run->mmio.len); 1054 } 1055 1056 run->mmio.phys_addr = vcpu->arch.paddr_accessed; 1057 run->mmio.len = bytes; 1058 run->mmio.is_write = 0; 1059 1060 vcpu->arch.io_gpr = rt; 1061 vcpu->arch.mmio_host_swabbed = host_swabbed; 1062 vcpu->mmio_needed = 1; 1063 vcpu->mmio_is_write = 0; 1064 vcpu->arch.mmio_sign_extend = sign_extend; 1065 1066 idx = srcu_read_lock(&vcpu->kvm->srcu); 1067 1068 ret = kvm_io_bus_read(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr, 1069 bytes, &run->mmio.data); 1070 1071 srcu_read_unlock(&vcpu->kvm->srcu, idx); 1072 1073 if (!ret) { 1074 kvmppc_complete_mmio_load(vcpu, run); 1075 vcpu->mmio_needed = 0; 1076 return EMULATE_DONE; 1077 } 1078 1079 return EMULATE_DO_MMIO; 1080 } 1081 1082 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu, 1083 unsigned int rt, unsigned int bytes, 1084 int is_default_endian) 1085 { 1086 return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 0); 1087 } 1088 EXPORT_SYMBOL_GPL(kvmppc_handle_load); 1089 1090 /* Same as above, but sign extends */ 1091 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu, 1092 unsigned int rt, unsigned int bytes, 1093 int is_default_endian) 1094 { 1095 return __kvmppc_handle_load(run, vcpu, rt, bytes, is_default_endian, 1); 1096 } 1097 1098 #ifdef CONFIG_VSX 1099 int kvmppc_handle_vsx_load(struct kvm_run *run, struct kvm_vcpu *vcpu, 1100 unsigned int rt, unsigned int bytes, 1101 int is_default_endian, int mmio_sign_extend) 1102 { 1103 enum emulation_result emulated = EMULATE_DONE; 1104 1105 /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */ 1106 if ( (vcpu->arch.mmio_vsx_copy_nums > 4) || 1107 (vcpu->arch.mmio_vsx_copy_nums < 0) ) { 1108 return EMULATE_FAIL; 1109 } 1110 1111 while (vcpu->arch.mmio_vsx_copy_nums) { 1112 emulated = __kvmppc_handle_load(run, vcpu, rt, bytes, 1113 is_default_endian, mmio_sign_extend); 1114 1115 if (emulated != EMULATE_DONE) 1116 break; 1117 1118 vcpu->arch.paddr_accessed += run->mmio.len; 1119 1120 vcpu->arch.mmio_vsx_copy_nums--; 1121 vcpu->arch.mmio_vsx_offset++; 1122 } 1123 return emulated; 1124 } 1125 #endif /* CONFIG_VSX */ 1126 1127 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu, 1128 u64 val, unsigned int bytes, int is_default_endian) 1129 { 1130 void *data = run->mmio.data; 1131 int idx, ret; 1132 bool host_swabbed; 1133 1134 /* Pity C doesn't have a logical XOR operator */ 1135 if (kvmppc_need_byteswap(vcpu)) { 1136 host_swabbed = is_default_endian; 1137 } else { 1138 host_swabbed = !is_default_endian; 1139 } 1140 1141 if (bytes > sizeof(run->mmio.data)) { 1142 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__, 1143 run->mmio.len); 1144 } 1145 1146 run->mmio.phys_addr = vcpu->arch.paddr_accessed; 1147 run->mmio.len = bytes; 1148 run->mmio.is_write = 1; 1149 vcpu->mmio_needed = 1; 1150 vcpu->mmio_is_write = 1; 1151 1152 if ((vcpu->arch.mmio_sp64_extend) && (bytes == 4)) 1153 val = dp_to_sp(val); 1154 1155 /* Store the value at the lowest bytes in 'data'. */ 1156 if (!host_swabbed) { 1157 switch (bytes) { 1158 case 8: *(u64 *)data = val; break; 1159 case 4: *(u32 *)data = val; break; 1160 case 2: *(u16 *)data = val; break; 1161 case 1: *(u8 *)data = val; break; 1162 } 1163 } else { 1164 switch (bytes) { 1165 case 8: *(u64 *)data = swab64(val); break; 1166 case 4: *(u32 *)data = swab32(val); break; 1167 case 2: *(u16 *)data = swab16(val); break; 1168 case 1: *(u8 *)data = val; break; 1169 } 1170 } 1171 1172 idx = srcu_read_lock(&vcpu->kvm->srcu); 1173 1174 ret = kvm_io_bus_write(vcpu, KVM_MMIO_BUS, run->mmio.phys_addr, 1175 bytes, &run->mmio.data); 1176 1177 srcu_read_unlock(&vcpu->kvm->srcu, idx); 1178 1179 if (!ret) { 1180 vcpu->mmio_needed = 0; 1181 return EMULATE_DONE; 1182 } 1183 1184 return EMULATE_DO_MMIO; 1185 } 1186 EXPORT_SYMBOL_GPL(kvmppc_handle_store); 1187 1188 #ifdef CONFIG_VSX 1189 static inline int kvmppc_get_vsr_data(struct kvm_vcpu *vcpu, int rs, u64 *val) 1190 { 1191 u32 dword_offset, word_offset; 1192 union kvmppc_one_reg reg; 1193 int vsx_offset = 0; 1194 int copy_type = vcpu->arch.mmio_vsx_copy_type; 1195 int result = 0; 1196 1197 switch (copy_type) { 1198 case KVMPPC_VSX_COPY_DWORD: 1199 vsx_offset = 1200 kvmppc_get_vsr_dword_offset(vcpu->arch.mmio_vsx_offset); 1201 1202 if (vsx_offset == -1) { 1203 result = -1; 1204 break; 1205 } 1206 1207 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) { 1208 *val = VCPU_VSX_FPR(vcpu, rs, vsx_offset); 1209 } else { 1210 reg.vval = VCPU_VSX_VR(vcpu, rs); 1211 *val = reg.vsxval[vsx_offset]; 1212 } 1213 break; 1214 1215 case KVMPPC_VSX_COPY_WORD: 1216 vsx_offset = 1217 kvmppc_get_vsr_word_offset(vcpu->arch.mmio_vsx_offset); 1218 1219 if (vsx_offset == -1) { 1220 result = -1; 1221 break; 1222 } 1223 1224 if (!vcpu->arch.mmio_vsx_tx_sx_enabled) { 1225 dword_offset = vsx_offset / 2; 1226 word_offset = vsx_offset % 2; 1227 reg.vsxval[0] = VCPU_VSX_FPR(vcpu, rs, dword_offset); 1228 *val = reg.vsx32val[word_offset]; 1229 } else { 1230 reg.vval = VCPU_VSX_VR(vcpu, rs); 1231 *val = reg.vsx32val[vsx_offset]; 1232 } 1233 break; 1234 1235 default: 1236 result = -1; 1237 break; 1238 } 1239 1240 return result; 1241 } 1242 1243 int kvmppc_handle_vsx_store(struct kvm_run *run, struct kvm_vcpu *vcpu, 1244 int rs, unsigned int bytes, int is_default_endian) 1245 { 1246 u64 val; 1247 enum emulation_result emulated = EMULATE_DONE; 1248 1249 vcpu->arch.io_gpr = rs; 1250 1251 /* Currently, mmio_vsx_copy_nums only allowed to be less than 4 */ 1252 if ( (vcpu->arch.mmio_vsx_copy_nums > 4) || 1253 (vcpu->arch.mmio_vsx_copy_nums < 0) ) { 1254 return EMULATE_FAIL; 1255 } 1256 1257 while (vcpu->arch.mmio_vsx_copy_nums) { 1258 if (kvmppc_get_vsr_data(vcpu, rs, &val) == -1) 1259 return EMULATE_FAIL; 1260 1261 emulated = kvmppc_handle_store(run, vcpu, 1262 val, bytes, is_default_endian); 1263 1264 if (emulated != EMULATE_DONE) 1265 break; 1266 1267 vcpu->arch.paddr_accessed += run->mmio.len; 1268 1269 vcpu->arch.mmio_vsx_copy_nums--; 1270 vcpu->arch.mmio_vsx_offset++; 1271 } 1272 1273 return emulated; 1274 } 1275 1276 static int kvmppc_emulate_mmio_vsx_loadstore(struct kvm_vcpu *vcpu, 1277 struct kvm_run *run) 1278 { 1279 enum emulation_result emulated = EMULATE_FAIL; 1280 int r; 1281 1282 vcpu->arch.paddr_accessed += run->mmio.len; 1283 1284 if (!vcpu->mmio_is_write) { 1285 emulated = kvmppc_handle_vsx_load(run, vcpu, vcpu->arch.io_gpr, 1286 run->mmio.len, 1, vcpu->arch.mmio_sign_extend); 1287 } else { 1288 emulated = kvmppc_handle_vsx_store(run, vcpu, 1289 vcpu->arch.io_gpr, run->mmio.len, 1); 1290 } 1291 1292 switch (emulated) { 1293 case EMULATE_DO_MMIO: 1294 run->exit_reason = KVM_EXIT_MMIO; 1295 r = RESUME_HOST; 1296 break; 1297 case EMULATE_FAIL: 1298 pr_info("KVM: MMIO emulation failed (VSX repeat)\n"); 1299 run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 1300 run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; 1301 r = RESUME_HOST; 1302 break; 1303 default: 1304 r = RESUME_GUEST; 1305 break; 1306 } 1307 return r; 1308 } 1309 #endif /* CONFIG_VSX */ 1310 1311 int kvm_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg) 1312 { 1313 int r = 0; 1314 union kvmppc_one_reg val; 1315 int size; 1316 1317 size = one_reg_size(reg->id); 1318 if (size > sizeof(val)) 1319 return -EINVAL; 1320 1321 r = kvmppc_get_one_reg(vcpu, reg->id, &val); 1322 if (r == -EINVAL) { 1323 r = 0; 1324 switch (reg->id) { 1325 #ifdef CONFIG_ALTIVEC 1326 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31: 1327 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1328 r = -ENXIO; 1329 break; 1330 } 1331 val.vval = vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0]; 1332 break; 1333 case KVM_REG_PPC_VSCR: 1334 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1335 r = -ENXIO; 1336 break; 1337 } 1338 val = get_reg_val(reg->id, vcpu->arch.vr.vscr.u[3]); 1339 break; 1340 case KVM_REG_PPC_VRSAVE: 1341 val = get_reg_val(reg->id, vcpu->arch.vrsave); 1342 break; 1343 #endif /* CONFIG_ALTIVEC */ 1344 default: 1345 r = -EINVAL; 1346 break; 1347 } 1348 } 1349 1350 if (r) 1351 return r; 1352 1353 if (copy_to_user((char __user *)(unsigned long)reg->addr, &val, size)) 1354 r = -EFAULT; 1355 1356 return r; 1357 } 1358 1359 int kvm_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, struct kvm_one_reg *reg) 1360 { 1361 int r; 1362 union kvmppc_one_reg val; 1363 int size; 1364 1365 size = one_reg_size(reg->id); 1366 if (size > sizeof(val)) 1367 return -EINVAL; 1368 1369 if (copy_from_user(&val, (char __user *)(unsigned long)reg->addr, size)) 1370 return -EFAULT; 1371 1372 r = kvmppc_set_one_reg(vcpu, reg->id, &val); 1373 if (r == -EINVAL) { 1374 r = 0; 1375 switch (reg->id) { 1376 #ifdef CONFIG_ALTIVEC 1377 case KVM_REG_PPC_VR0 ... KVM_REG_PPC_VR31: 1378 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1379 r = -ENXIO; 1380 break; 1381 } 1382 vcpu->arch.vr.vr[reg->id - KVM_REG_PPC_VR0] = val.vval; 1383 break; 1384 case KVM_REG_PPC_VSCR: 1385 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1386 r = -ENXIO; 1387 break; 1388 } 1389 vcpu->arch.vr.vscr.u[3] = set_reg_val(reg->id, val); 1390 break; 1391 case KVM_REG_PPC_VRSAVE: 1392 if (!cpu_has_feature(CPU_FTR_ALTIVEC)) { 1393 r = -ENXIO; 1394 break; 1395 } 1396 vcpu->arch.vrsave = set_reg_val(reg->id, val); 1397 break; 1398 #endif /* CONFIG_ALTIVEC */ 1399 default: 1400 r = -EINVAL; 1401 break; 1402 } 1403 } 1404 1405 return r; 1406 } 1407 1408 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) 1409 { 1410 int r; 1411 sigset_t sigsaved; 1412 1413 if (vcpu->mmio_needed) { 1414 vcpu->mmio_needed = 0; 1415 if (!vcpu->mmio_is_write) 1416 kvmppc_complete_mmio_load(vcpu, run); 1417 #ifdef CONFIG_VSX 1418 if (vcpu->arch.mmio_vsx_copy_nums > 0) { 1419 vcpu->arch.mmio_vsx_copy_nums--; 1420 vcpu->arch.mmio_vsx_offset++; 1421 } 1422 1423 if (vcpu->arch.mmio_vsx_copy_nums > 0) { 1424 r = kvmppc_emulate_mmio_vsx_loadstore(vcpu, run); 1425 if (r == RESUME_HOST) { 1426 vcpu->mmio_needed = 1; 1427 return r; 1428 } 1429 } 1430 #endif 1431 } else if (vcpu->arch.osi_needed) { 1432 u64 *gprs = run->osi.gprs; 1433 int i; 1434 1435 for (i = 0; i < 32; i++) 1436 kvmppc_set_gpr(vcpu, i, gprs[i]); 1437 vcpu->arch.osi_needed = 0; 1438 } else if (vcpu->arch.hcall_needed) { 1439 int i; 1440 1441 kvmppc_set_gpr(vcpu, 3, run->papr_hcall.ret); 1442 for (i = 0; i < 9; ++i) 1443 kvmppc_set_gpr(vcpu, 4 + i, run->papr_hcall.args[i]); 1444 vcpu->arch.hcall_needed = 0; 1445 #ifdef CONFIG_BOOKE 1446 } else if (vcpu->arch.epr_needed) { 1447 kvmppc_set_epr(vcpu, run->epr.epr); 1448 vcpu->arch.epr_needed = 0; 1449 #endif 1450 } 1451 1452 if (vcpu->sigset_active) 1453 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); 1454 1455 if (run->immediate_exit) 1456 r = -EINTR; 1457 else 1458 r = kvmppc_vcpu_run(run, vcpu); 1459 1460 if (vcpu->sigset_active) 1461 sigprocmask(SIG_SETMASK, &sigsaved, NULL); 1462 1463 return r; 1464 } 1465 1466 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq) 1467 { 1468 if (irq->irq == KVM_INTERRUPT_UNSET) { 1469 kvmppc_core_dequeue_external(vcpu); 1470 return 0; 1471 } 1472 1473 kvmppc_core_queue_external(vcpu, irq); 1474 1475 kvm_vcpu_kick(vcpu); 1476 1477 return 0; 1478 } 1479 1480 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, 1481 struct kvm_enable_cap *cap) 1482 { 1483 int r; 1484 1485 if (cap->flags) 1486 return -EINVAL; 1487 1488 switch (cap->cap) { 1489 case KVM_CAP_PPC_OSI: 1490 r = 0; 1491 vcpu->arch.osi_enabled = true; 1492 break; 1493 case KVM_CAP_PPC_PAPR: 1494 r = 0; 1495 vcpu->arch.papr_enabled = true; 1496 break; 1497 case KVM_CAP_PPC_EPR: 1498 r = 0; 1499 if (cap->args[0]) 1500 vcpu->arch.epr_flags |= KVMPPC_EPR_USER; 1501 else 1502 vcpu->arch.epr_flags &= ~KVMPPC_EPR_USER; 1503 break; 1504 #ifdef CONFIG_BOOKE 1505 case KVM_CAP_PPC_BOOKE_WATCHDOG: 1506 r = 0; 1507 vcpu->arch.watchdog_enabled = true; 1508 break; 1509 #endif 1510 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC) 1511 case KVM_CAP_SW_TLB: { 1512 struct kvm_config_tlb cfg; 1513 void __user *user_ptr = (void __user *)(uintptr_t)cap->args[0]; 1514 1515 r = -EFAULT; 1516 if (copy_from_user(&cfg, user_ptr, sizeof(cfg))) 1517 break; 1518 1519 r = kvm_vcpu_ioctl_config_tlb(vcpu, &cfg); 1520 break; 1521 } 1522 #endif 1523 #ifdef CONFIG_KVM_MPIC 1524 case KVM_CAP_IRQ_MPIC: { 1525 struct fd f; 1526 struct kvm_device *dev; 1527 1528 r = -EBADF; 1529 f = fdget(cap->args[0]); 1530 if (!f.file) 1531 break; 1532 1533 r = -EPERM; 1534 dev = kvm_device_from_filp(f.file); 1535 if (dev) 1536 r = kvmppc_mpic_connect_vcpu(dev, vcpu, cap->args[1]); 1537 1538 fdput(f); 1539 break; 1540 } 1541 #endif 1542 #ifdef CONFIG_KVM_XICS 1543 case KVM_CAP_IRQ_XICS: { 1544 struct fd f; 1545 struct kvm_device *dev; 1546 1547 r = -EBADF; 1548 f = fdget(cap->args[0]); 1549 if (!f.file) 1550 break; 1551 1552 r = -EPERM; 1553 dev = kvm_device_from_filp(f.file); 1554 if (dev) { 1555 if (xive_enabled()) 1556 r = kvmppc_xive_connect_vcpu(dev, vcpu, cap->args[1]); 1557 else 1558 r = kvmppc_xics_connect_vcpu(dev, vcpu, cap->args[1]); 1559 } 1560 1561 fdput(f); 1562 break; 1563 } 1564 #endif /* CONFIG_KVM_XICS */ 1565 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE 1566 case KVM_CAP_PPC_FWNMI: 1567 r = -EINVAL; 1568 if (!is_kvmppc_hv_enabled(vcpu->kvm)) 1569 break; 1570 r = 0; 1571 vcpu->kvm->arch.fwnmi_enabled = true; 1572 break; 1573 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */ 1574 default: 1575 r = -EINVAL; 1576 break; 1577 } 1578 1579 if (!r) 1580 r = kvmppc_sanity_check(vcpu); 1581 1582 return r; 1583 } 1584 1585 bool kvm_arch_intc_initialized(struct kvm *kvm) 1586 { 1587 #ifdef CONFIG_KVM_MPIC 1588 if (kvm->arch.mpic) 1589 return true; 1590 #endif 1591 #ifdef CONFIG_KVM_XICS 1592 if (kvm->arch.xics || kvm->arch.xive) 1593 return true; 1594 #endif 1595 return false; 1596 } 1597 1598 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 1599 struct kvm_mp_state *mp_state) 1600 { 1601 return -EINVAL; 1602 } 1603 1604 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 1605 struct kvm_mp_state *mp_state) 1606 { 1607 return -EINVAL; 1608 } 1609 1610 long kvm_arch_vcpu_ioctl(struct file *filp, 1611 unsigned int ioctl, unsigned long arg) 1612 { 1613 struct kvm_vcpu *vcpu = filp->private_data; 1614 void __user *argp = (void __user *)arg; 1615 long r; 1616 1617 switch (ioctl) { 1618 case KVM_INTERRUPT: { 1619 struct kvm_interrupt irq; 1620 r = -EFAULT; 1621 if (copy_from_user(&irq, argp, sizeof(irq))) 1622 goto out; 1623 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq); 1624 goto out; 1625 } 1626 1627 case KVM_ENABLE_CAP: 1628 { 1629 struct kvm_enable_cap cap; 1630 r = -EFAULT; 1631 if (copy_from_user(&cap, argp, sizeof(cap))) 1632 goto out; 1633 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); 1634 break; 1635 } 1636 1637 case KVM_SET_ONE_REG: 1638 case KVM_GET_ONE_REG: 1639 { 1640 struct kvm_one_reg reg; 1641 r = -EFAULT; 1642 if (copy_from_user(®, argp, sizeof(reg))) 1643 goto out; 1644 if (ioctl == KVM_SET_ONE_REG) 1645 r = kvm_vcpu_ioctl_set_one_reg(vcpu, ®); 1646 else 1647 r = kvm_vcpu_ioctl_get_one_reg(vcpu, ®); 1648 break; 1649 } 1650 1651 #if defined(CONFIG_KVM_E500V2) || defined(CONFIG_KVM_E500MC) 1652 case KVM_DIRTY_TLB: { 1653 struct kvm_dirty_tlb dirty; 1654 r = -EFAULT; 1655 if (copy_from_user(&dirty, argp, sizeof(dirty))) 1656 goto out; 1657 r = kvm_vcpu_ioctl_dirty_tlb(vcpu, &dirty); 1658 break; 1659 } 1660 #endif 1661 default: 1662 r = -EINVAL; 1663 } 1664 1665 out: 1666 return r; 1667 } 1668 1669 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 1670 { 1671 return VM_FAULT_SIGBUS; 1672 } 1673 1674 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo) 1675 { 1676 u32 inst_nop = 0x60000000; 1677 #ifdef CONFIG_KVM_BOOKE_HV 1678 u32 inst_sc1 = 0x44000022; 1679 pvinfo->hcall[0] = cpu_to_be32(inst_sc1); 1680 pvinfo->hcall[1] = cpu_to_be32(inst_nop); 1681 pvinfo->hcall[2] = cpu_to_be32(inst_nop); 1682 pvinfo->hcall[3] = cpu_to_be32(inst_nop); 1683 #else 1684 u32 inst_lis = 0x3c000000; 1685 u32 inst_ori = 0x60000000; 1686 u32 inst_sc = 0x44000002; 1687 u32 inst_imm_mask = 0xffff; 1688 1689 /* 1690 * The hypercall to get into KVM from within guest context is as 1691 * follows: 1692 * 1693 * lis r0, r0, KVM_SC_MAGIC_R0@h 1694 * ori r0, KVM_SC_MAGIC_R0@l 1695 * sc 1696 * nop 1697 */ 1698 pvinfo->hcall[0] = cpu_to_be32(inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask)); 1699 pvinfo->hcall[1] = cpu_to_be32(inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask)); 1700 pvinfo->hcall[2] = cpu_to_be32(inst_sc); 1701 pvinfo->hcall[3] = cpu_to_be32(inst_nop); 1702 #endif 1703 1704 pvinfo->flags = KVM_PPC_PVINFO_FLAGS_EV_IDLE; 1705 1706 return 0; 1707 } 1708 1709 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event, 1710 bool line_status) 1711 { 1712 if (!irqchip_in_kernel(kvm)) 1713 return -ENXIO; 1714 1715 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, 1716 irq_event->irq, irq_event->level, 1717 line_status); 1718 return 0; 1719 } 1720 1721 1722 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, 1723 struct kvm_enable_cap *cap) 1724 { 1725 int r; 1726 1727 if (cap->flags) 1728 return -EINVAL; 1729 1730 switch (cap->cap) { 1731 #ifdef CONFIG_KVM_BOOK3S_64_HANDLER 1732 case KVM_CAP_PPC_ENABLE_HCALL: { 1733 unsigned long hcall = cap->args[0]; 1734 1735 r = -EINVAL; 1736 if (hcall > MAX_HCALL_OPCODE || (hcall & 3) || 1737 cap->args[1] > 1) 1738 break; 1739 if (!kvmppc_book3s_hcall_implemented(kvm, hcall)) 1740 break; 1741 if (cap->args[1]) 1742 set_bit(hcall / 4, kvm->arch.enabled_hcalls); 1743 else 1744 clear_bit(hcall / 4, kvm->arch.enabled_hcalls); 1745 r = 0; 1746 break; 1747 } 1748 case KVM_CAP_PPC_SMT: { 1749 unsigned long mode = cap->args[0]; 1750 unsigned long flags = cap->args[1]; 1751 1752 r = -EINVAL; 1753 if (kvm->arch.kvm_ops->set_smt_mode) 1754 r = kvm->arch.kvm_ops->set_smt_mode(kvm, mode, flags); 1755 break; 1756 } 1757 #endif 1758 default: 1759 r = -EINVAL; 1760 break; 1761 } 1762 1763 return r; 1764 } 1765 1766 long kvm_arch_vm_ioctl(struct file *filp, 1767 unsigned int ioctl, unsigned long arg) 1768 { 1769 struct kvm *kvm __maybe_unused = filp->private_data; 1770 void __user *argp = (void __user *)arg; 1771 long r; 1772 1773 switch (ioctl) { 1774 case KVM_PPC_GET_PVINFO: { 1775 struct kvm_ppc_pvinfo pvinfo; 1776 memset(&pvinfo, 0, sizeof(pvinfo)); 1777 r = kvm_vm_ioctl_get_pvinfo(&pvinfo); 1778 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) { 1779 r = -EFAULT; 1780 goto out; 1781 } 1782 1783 break; 1784 } 1785 case KVM_ENABLE_CAP: 1786 { 1787 struct kvm_enable_cap cap; 1788 r = -EFAULT; 1789 if (copy_from_user(&cap, argp, sizeof(cap))) 1790 goto out; 1791 r = kvm_vm_ioctl_enable_cap(kvm, &cap); 1792 break; 1793 } 1794 #ifdef CONFIG_SPAPR_TCE_IOMMU 1795 case KVM_CREATE_SPAPR_TCE_64: { 1796 struct kvm_create_spapr_tce_64 create_tce_64; 1797 1798 r = -EFAULT; 1799 if (copy_from_user(&create_tce_64, argp, sizeof(create_tce_64))) 1800 goto out; 1801 if (create_tce_64.flags) { 1802 r = -EINVAL; 1803 goto out; 1804 } 1805 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64); 1806 goto out; 1807 } 1808 case KVM_CREATE_SPAPR_TCE: { 1809 struct kvm_create_spapr_tce create_tce; 1810 struct kvm_create_spapr_tce_64 create_tce_64; 1811 1812 r = -EFAULT; 1813 if (copy_from_user(&create_tce, argp, sizeof(create_tce))) 1814 goto out; 1815 1816 create_tce_64.liobn = create_tce.liobn; 1817 create_tce_64.page_shift = IOMMU_PAGE_SHIFT_4K; 1818 create_tce_64.offset = 0; 1819 create_tce_64.size = create_tce.window_size >> 1820 IOMMU_PAGE_SHIFT_4K; 1821 create_tce_64.flags = 0; 1822 r = kvm_vm_ioctl_create_spapr_tce(kvm, &create_tce_64); 1823 goto out; 1824 } 1825 #endif 1826 #ifdef CONFIG_PPC_BOOK3S_64 1827 case KVM_PPC_GET_SMMU_INFO: { 1828 struct kvm_ppc_smmu_info info; 1829 struct kvm *kvm = filp->private_data; 1830 1831 memset(&info, 0, sizeof(info)); 1832 r = kvm->arch.kvm_ops->get_smmu_info(kvm, &info); 1833 if (r >= 0 && copy_to_user(argp, &info, sizeof(info))) 1834 r = -EFAULT; 1835 break; 1836 } 1837 case KVM_PPC_RTAS_DEFINE_TOKEN: { 1838 struct kvm *kvm = filp->private_data; 1839 1840 r = kvm_vm_ioctl_rtas_define_token(kvm, argp); 1841 break; 1842 } 1843 case KVM_PPC_CONFIGURE_V3_MMU: { 1844 struct kvm *kvm = filp->private_data; 1845 struct kvm_ppc_mmuv3_cfg cfg; 1846 1847 r = -EINVAL; 1848 if (!kvm->arch.kvm_ops->configure_mmu) 1849 goto out; 1850 r = -EFAULT; 1851 if (copy_from_user(&cfg, argp, sizeof(cfg))) 1852 goto out; 1853 r = kvm->arch.kvm_ops->configure_mmu(kvm, &cfg); 1854 break; 1855 } 1856 case KVM_PPC_GET_RMMU_INFO: { 1857 struct kvm *kvm = filp->private_data; 1858 struct kvm_ppc_rmmu_info info; 1859 1860 r = -EINVAL; 1861 if (!kvm->arch.kvm_ops->get_rmmu_info) 1862 goto out; 1863 r = kvm->arch.kvm_ops->get_rmmu_info(kvm, &info); 1864 if (r >= 0 && copy_to_user(argp, &info, sizeof(info))) 1865 r = -EFAULT; 1866 break; 1867 } 1868 default: { 1869 struct kvm *kvm = filp->private_data; 1870 r = kvm->arch.kvm_ops->arch_vm_ioctl(filp, ioctl, arg); 1871 } 1872 #else /* CONFIG_PPC_BOOK3S_64 */ 1873 default: 1874 r = -ENOTTY; 1875 #endif 1876 } 1877 out: 1878 return r; 1879 } 1880 1881 static unsigned long lpid_inuse[BITS_TO_LONGS(KVMPPC_NR_LPIDS)]; 1882 static unsigned long nr_lpids; 1883 1884 long kvmppc_alloc_lpid(void) 1885 { 1886 long lpid; 1887 1888 do { 1889 lpid = find_first_zero_bit(lpid_inuse, KVMPPC_NR_LPIDS); 1890 if (lpid >= nr_lpids) { 1891 pr_err("%s: No LPIDs free\n", __func__); 1892 return -ENOMEM; 1893 } 1894 } while (test_and_set_bit(lpid, lpid_inuse)); 1895 1896 return lpid; 1897 } 1898 EXPORT_SYMBOL_GPL(kvmppc_alloc_lpid); 1899 1900 void kvmppc_claim_lpid(long lpid) 1901 { 1902 set_bit(lpid, lpid_inuse); 1903 } 1904 EXPORT_SYMBOL_GPL(kvmppc_claim_lpid); 1905 1906 void kvmppc_free_lpid(long lpid) 1907 { 1908 clear_bit(lpid, lpid_inuse); 1909 } 1910 EXPORT_SYMBOL_GPL(kvmppc_free_lpid); 1911 1912 void kvmppc_init_lpid(unsigned long nr_lpids_param) 1913 { 1914 nr_lpids = min_t(unsigned long, KVMPPC_NR_LPIDS, nr_lpids_param); 1915 memset(lpid_inuse, 0, sizeof(lpid_inuse)); 1916 } 1917 EXPORT_SYMBOL_GPL(kvmppc_init_lpid); 1918 1919 int kvm_arch_init(void *opaque) 1920 { 1921 return 0; 1922 } 1923 1924 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ppc_instr); 1925