arch/powerpc/kvm/book3s_pr.c

   1 /*
   2  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   3  *
   4  * Authors:
   5  *    Alexander Graf <agraf@suse.de>
   6  *    Kevin Wolf <mail@kevin-wolf.de>
   7  *    Paul Mackerras <paulus@samba.org>
   8  *
   9  * Description:
  10  * Functions relating to running KVM on Book 3S processors where
  11  * we don't have access to hypervisor mode, and we run the guest
  12  * in problem state (user mode).
  13  *
  14  * This file is derived from arch/powerpc/kvm/44x.c,
  15  * by Hollis Blanchard <hollisb@us.ibm.com>.
  16  *
  17  * This program is free software; you can redistribute it and/or modify
  18  * it under the terms of the GNU General Public License, version 2, as
  19  * published by the Free Software Foundation.
  20  */
  21
  22 #include <linux/kvm_host.h>
  23 #include <linux/export.h>
  24 #include <linux/err.h>
  25 #include <linux/slab.h>
  26
  27 #include <asm/reg.h>
  28 #include <asm/cputable.h>
  29 #include <asm/cacheflush.h>
  30 #include <asm/tlbflush.h>
  31 #include <asm/uaccess.h>
  32 #include <asm/io.h>
  33 #include <asm/kvm_ppc.h>
  34 #include <asm/kvm_book3s.h>
  35 #include <asm/mmu_context.h>
  36 #include <asm/switch_to.h>
  37 #include <linux/gfp.h>
  38 #include <linux/sched.h>
  39 #include <linux/vmalloc.h>
  40 #include <linux/highmem.h>
  41
  42 #include "trace.h"
  43
  44 /* #define EXIT_DEBUG */
  45 /* #define DEBUG_EXT */
  46
  47 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  48                              ulong msr);
  49
  50 /* Some compatibility defines */
  51 #ifdef CONFIG_PPC_BOOK3S_32
  52 #define MSR_USER32 MSR_USER
  53 #define MSR_USER64 MSR_USER
  54 #define HW_PAGE_SIZE PAGE_SIZE
  55 #endif
  56
  57 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  58 {
  59 #ifdef CONFIG_PPC_BOOK3S_64
  60         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  61         memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
  62         memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
  63                sizeof(get_paca()->shadow_vcpu));
  64         svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
  65         svcpu_put(svcpu);
  66 #endif
  67         vcpu->cpu = smp_processor_id();
  68 #ifdef CONFIG_PPC_BOOK3S_32
  69         current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
  70 #endif
  71 }
  72
  73 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  74 {
  75 #ifdef CONFIG_PPC_BOOK3S_64
  76         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  77         memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
  78         memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  79                sizeof(get_paca()->shadow_vcpu));
  80         to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
  81         svcpu_put(svcpu);
  82 #endif
  83
  84         kvmppc_giveup_ext(vcpu, MSR_FP);
  85         kvmppc_giveup_ext(vcpu, MSR_VEC);
  86         kvmppc_giveup_ext(vcpu, MSR_VSX);
  87         vcpu->cpu = -1;
  88 }
  89
  90 int kvmppc_core_check_requests(struct kvm_vcpu *vcpu)
  91 {
  92         int r = 1; /* Indicate we want to get back into the guest */
  93
  94         /* We misuse TLB_FLUSH to indicate that we want to clear
  95            all shadow cache entries */
  96         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
  97                 kvmppc_mmu_pte_flush(vcpu, 0, 0);
  98
  99         return r;
 100 }
 101
 102 /************* MMU Notifiers *************/
 103
 104 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
 105 {
 106         trace_kvm_unmap_hva(hva);
 107
 108         /*
 109          * Flush all shadow tlb entries everywhere. This is slow, but
 110          * we are 100% sure that we catch the to be unmapped page
 111          */
 112         kvm_flush_remote_tlbs(kvm);
 113
 114         return 0;
 115 }
 116
 117 int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
 118 {
 119         /* kvm_unmap_hva flushes everything anyways */
 120         kvm_unmap_hva(kvm, start);
 121
 122         return 0;
 123 }
 124
 125 int kvm_age_hva(struct kvm *kvm, unsigned long hva)
 126 {
 127         /* XXX could be more clever ;) */
 128         return 0;
 129 }
 130
 131 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
 132 {
 133         /* XXX could be more clever ;) */
 134         return 0;
 135 }
 136
 137 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
 138 {
 139         /* The page will get remapped properly on its next fault */
 140         kvm_unmap_hva(kvm, hva);
 141 }
 142
 143 /*****************************************/
 144
 145 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 146 {
 147         ulong smsr = vcpu->arch.shared->msr;
 148
 149         /* Guest MSR values */
 150         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE;
 151         /* Process MSR values */
 152         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 153         /* External providers the guest reserved */
 154         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
 155         /* 64-bit Process MSR values */
 156 #ifdef CONFIG_PPC_BOOK3S_64
 157         smsr |= MSR_ISF | MSR_HV;
 158 #endif
 159         vcpu->arch.shadow_msr = smsr;
 160 }
 161
 162 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 163 {
 164         ulong old_msr = vcpu->arch.shared->msr;
 165
 166 #ifdef EXIT_DEBUG
 167         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 168 #endif
 169
 170         msr &= to_book3s(vcpu)->msr_mask;
 171         vcpu->arch.shared->msr = msr;
 172         kvmppc_recalc_shadow_msr(vcpu);
 173
 174         if (msr & MSR_POW) {
 175                 if (!vcpu->arch.pending_exceptions) {
 176                         kvm_vcpu_block(vcpu);
 177                         clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 178                         vcpu->stat.halt_wakeup++;
 179
 180                         /* Unset POW bit after we woke up */
 181                         msr &= ~MSR_POW;
 182                         vcpu->arch.shared->msr = msr;
 183                 }
 184         }
 185
 186         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 187                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 188                 kvmppc_mmu_flush_segments(vcpu);
 189                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 190
 191                 /* Preload magic page segment when in kernel mode */
 192                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 193                         struct kvm_vcpu_arch *a = &vcpu->arch;
 194
 195                         if (msr & MSR_DR)
 196                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 197                         else
 198                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 199                 }
 200         }
 201
 202         /*
 203          * When switching from 32 to 64-bit, we may have a stale 32-bit
 204          * magic page around, we need to flush it. Typically 32-bit magic
 205          * page will be instanciated when calling into RTAS. Note: We
 206          * assume that such transition only happens while in kernel mode,
 207          * ie, we never transition from user 32-bit to kernel 64-bit with
 208          * a 32-bit magic page around.
 209          */
 210         if (vcpu->arch.magic_page_pa &&
 211             !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
 212                 /* going from RTAS to normal kernel code */
 213                 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
 214                                      ~0xFFFUL);
 215         }
 216
 217         /* Preload FPU if it's enabled */
 218         if (vcpu->arch.shared->msr & MSR_FP)
 219                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 220 }
 221
 222 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 223 {
 224         u32 host_pvr;
 225
 226         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 227         vcpu->arch.pvr = pvr;
 228 #ifdef CONFIG_PPC_BOOK3S_64
 229         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 230                 kvmppc_mmu_book3s_64_init(vcpu);
 231                 if (!to_book3s(vcpu)->hior_explicit)
 232                         to_book3s(vcpu)->hior = 0xfff00000;
 233                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 234                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 235         } else
 236 #endif
 237         {
 238                 kvmppc_mmu_book3s_32_init(vcpu);
 239                 if (!to_book3s(vcpu)->hior_explicit)
 240                         to_book3s(vcpu)->hior = 0;
 241                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 242                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 243         }
 244
 245         kvmppc_sanity_check(vcpu);
 246
 247         /* If we are in hypervisor level on 970, we can tell the CPU to
 248          * treat DCBZ as 32 bytes store */
 249         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 250         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 251             !strcmp(cur_cpu_spec->platform, "ppc970"))
 252                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 253
 254         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 255            really needs them in a VM on Cell and force disable them. */
 256         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 257                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 258
 259 #ifdef CONFIG_PPC_BOOK3S_32
 260         /* 32 bit Book3S always has 32 byte dcbz */
 261         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 262 #endif
 263
 264         /* On some CPUs we can execute paired single operations natively */
 265         asm ( "mfpvr %0" : "=r"(host_pvr));
 266         switch (host_pvr) {
 267         case 0x00080200:        /* lonestar 2.0 */
 268         case 0x00088202:        /* lonestar 2.2 */
 269         case 0x70000100:        /* gekko 1.0 */
 270         case 0x00080100:        /* gekko 2.0 */
 271         case 0x00083203:        /* gekko 2.3a */
 272         case 0x00083213:        /* gekko 2.3b */
 273         case 0x00083204:        /* gekko 2.4 */
 274         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 275         case 0x00087200:        /* broadway */
 276                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 277                 /* Enable HID2.PSE - in case we need it later */
 278                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 279         }
 280 }
 281
 282 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 283  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 284  * emulate 32 bytes dcbz length.
 285  *
 286  * The Book3s_64 inventors also realized this case and implemented a special bit
 287  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 288  *
 289  * My approach here is to patch the dcbz instruction on executing pages.
 290  */
 291 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 292 {
 293         struct page *hpage;
 294         u64 hpage_offset;
 295         u32 *page;
 296         int i;
 297
 298         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 299         if (is_error_page(hpage))
 300                 return;
 301
 302         hpage_offset = pte->raddr & ~PAGE_MASK;
 303         hpage_offset &= ~0xFFFULL;
 304         hpage_offset /= 4;
 305
 306         get_page(hpage);
 307         page = kmap_atomic(hpage);
 308
 309         /* patch dcbz into reserved instruction, so we trap */
 310         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 311                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 312                         page[i] &= 0xfffffff7;
 313
 314         kunmap_atomic(page);
 315         put_page(hpage);
 316 }
 317
 318 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 319 {
 320         ulong mp_pa = vcpu->arch.magic_page_pa;
 321
 322         if (!(vcpu->arch.shared->msr & MSR_SF))
 323                 mp_pa = (uint32_t)mp_pa;
 324
 325         if (unlikely(mp_pa) &&
 326             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 327                 return 1;
 328         }
 329
 330         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 331 }
 332
 333 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 334                             ulong eaddr, int vec)
 335 {
 336         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 337         int r = RESUME_GUEST;
 338         int relocated;
 339         int page_found = 0;
 340         struct kvmppc_pte pte;
 341         bool is_mmio = false;
 342         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 343         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 344         u64 vsid;
 345
 346         relocated = data ? dr : ir;
 347
 348         /* Resolve real address if translation turned on */
 349         if (relocated) {
 350                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 351         } else {
 352                 pte.may_execute = true;
 353                 pte.may_read = true;
 354                 pte.may_write = true;
 355                 pte.raddr = eaddr & KVM_PAM;
 356                 pte.eaddr = eaddr;
 357                 pte.vpage = eaddr >> 12;
 358         }
 359
 360         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 361         case 0:
 362                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 363                 break;
 364         case MSR_DR:
 365         case MSR_IR:
 366                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 367
 368                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 369                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 370                 else
 371                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 372                 pte.vpage |= vsid;
 373
 374                 if (vsid == -1)
 375                         page_found = -EINVAL;
 376                 break;
 377         }
 378
 379         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 380            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 381                 /*
 382                  * If we do the dcbz hack, we have to NX on every execution,
 383                  * so we can patch the executing code. This renders our guest
 384                  * NX-less.
 385                  */
 386                 pte.may_execute = !data;
 387         }
 388
 389         if (page_found == -ENOENT) {
 390                 /* Page not found in guest PTE entries */
 391                 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 392                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 393                 vcpu->arch.shared->dsisr = svcpu->fault_dsisr;
 394                 vcpu->arch.shared->msr |=
 395                         (svcpu->shadow_srr1 & 0x00000000f8000000ULL);
 396                 svcpu_put(svcpu);
 397                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 398         } else if (page_found == -EPERM) {
 399                 /* Storage protection */
 400                 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 401                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 402                 vcpu->arch.shared->dsisr = svcpu->fault_dsisr & ~DSISR_NOHPTE;
 403                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 404                 vcpu->arch.shared->msr |=
 405                         svcpu->shadow_srr1 & 0x00000000f8000000ULL;
 406                 svcpu_put(svcpu);
 407                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 408         } else if (page_found == -EINVAL) {
 409                 /* Page not found in guest SLB */
 410                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 411                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 412         } else if (!is_mmio &&
 413                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 414                 /* The guest's PTE is not mapped yet. Map on the host */
 415                 kvmppc_mmu_map_page(vcpu, &pte);
 416                 if (data)
 417                         vcpu->stat.sp_storage++;
 418                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 419                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 420                         kvmppc_patch_dcbz(vcpu, &pte);
 421         } else {
 422                 /* MMIO */
 423                 vcpu->stat.mmio_exits++;
 424                 vcpu->arch.paddr_accessed = pte.raddr;
 425                 vcpu->arch.vaddr_accessed = pte.eaddr;
 426                 r = kvmppc_emulate_mmio(run, vcpu);
 427                 if ( r == RESUME_HOST_NV )
 428                         r = RESUME_HOST;
 429         }
 430
 431         return r;
 432 }
 433
 434 static inline int get_fpr_index(int i)
 435 {
 436 #ifdef CONFIG_VSX
 437         i *= 2;
 438 #endif
 439         return i;
 440 }
 441
 442 /* Give up external provider (FPU, Altivec, VSX) */
 443 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 444 {
 445         struct thread_struct *t = &current->thread;
 446         u64 *vcpu_fpr = vcpu->arch.fpr;
 447 #ifdef CONFIG_VSX
 448         u64 *vcpu_vsx = vcpu->arch.vsr;
 449 #endif
 450         u64 *thread_fpr = (u64*)t->fpr;
 451         int i;
 452
 453         if (!(vcpu->arch.guest_owned_ext & msr))
 454                 return;
 455
 456 #ifdef DEBUG_EXT
 457         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 458 #endif
 459
 460         switch (msr) {
 461         case MSR_FP:
 462                 giveup_fpu(current);
 463                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 464                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 465
 466                 vcpu->arch.fpscr = t->fpscr.val;
 467                 break;
 468         case MSR_VEC:
 469 #ifdef CONFIG_ALTIVEC
 470                 giveup_altivec(current);
 471                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 472                 vcpu->arch.vscr = t->vscr;
 473 #endif
 474                 break;
 475         case MSR_VSX:
 476 #ifdef CONFIG_VSX
 477                 __giveup_vsx(current);
 478                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 479                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 480 #endif
 481                 break;
 482         default:
 483                 BUG();
 484         }
 485
 486         vcpu->arch.guest_owned_ext &= ~msr;
 487         current->thread.regs->msr &= ~msr;
 488         kvmppc_recalc_shadow_msr(vcpu);
 489 }
 490
 491 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 492 {
 493         ulong srr0 = kvmppc_get_pc(vcpu);
 494         u32 last_inst = kvmppc_get_last_inst(vcpu);
 495         int ret;
 496
 497         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 498         if (ret == -ENOENT) {
 499                 ulong msr = vcpu->arch.shared->msr;
 500
 501                 msr = kvmppc_set_field(msr, 33, 33, 1);
 502                 msr = kvmppc_set_field(msr, 34, 36, 0);
 503                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 504                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 505                 return EMULATE_AGAIN;
 506         }
 507
 508         return EMULATE_DONE;
 509 }
 510
 511 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 512 {
 513
 514         /* Need to do paired single emulation? */
 515         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 516                 return EMULATE_DONE;
 517
 518         /* Read out the instruction */
 519         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 520                 /* Need to emulate */
 521                 return EMULATE_FAIL;
 522
 523         return EMULATE_AGAIN;
 524 }
 525
 526 /* Handle external providers (FPU, Altivec, VSX) */
 527 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 528                              ulong msr)
 529 {
 530         struct thread_struct *t = &current->thread;
 531         u64 *vcpu_fpr = vcpu->arch.fpr;
 532 #ifdef CONFIG_VSX
 533         u64 *vcpu_vsx = vcpu->arch.vsr;
 534 #endif
 535         u64 *thread_fpr = (u64*)t->fpr;
 536         int i;
 537
 538         /* When we have paired singles, we emulate in software */
 539         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 540                 return RESUME_GUEST;
 541
 542         if (!(vcpu->arch.shared->msr & msr)) {
 543                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 544                 return RESUME_GUEST;
 545         }
 546
 547         /* We already own the ext */
 548         if (vcpu->arch.guest_owned_ext & msr) {
 549                 return RESUME_GUEST;
 550         }
 551
 552 #ifdef DEBUG_EXT
 553         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 554 #endif
 555
 556         current->thread.regs->msr |= msr;
 557
 558         switch (msr) {
 559         case MSR_FP:
 560                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 561                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 562
 563                 t->fpscr.val = vcpu->arch.fpscr;
 564                 t->fpexc_mode = 0;
 565                 kvmppc_load_up_fpu();
 566                 break;
 567         case MSR_VEC:
 568 #ifdef CONFIG_ALTIVEC
 569                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 570                 t->vscr = vcpu->arch.vscr;
 571                 t->vrsave = -1;
 572                 kvmppc_load_up_altivec();
 573 #endif
 574                 break;
 575         case MSR_VSX:
 576 #ifdef CONFIG_VSX
 577                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 578                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 579                 kvmppc_load_up_vsx();
 580 #endif
 581                 break;
 582         default:
 583                 BUG();
 584         }
 585
 586         vcpu->arch.guest_owned_ext |= msr;
 587
 588         kvmppc_recalc_shadow_msr(vcpu);
 589
 590         return RESUME_GUEST;
 591 }
 592
 593 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 594                        unsigned int exit_nr)
 595 {
 596         int r = RESUME_HOST;
 597         int s;
 598
 599         vcpu->stat.sum_exits++;
 600
 601         run->exit_reason = KVM_EXIT_UNKNOWN;
 602         run->ready_for_interrupt_injection = 1;
 603
 604         /* We get here with MSR.EE=1 */
 605
 606         trace_kvm_exit(exit_nr, vcpu);
 607         kvm_guest_exit();
 608
 609         switch (exit_nr) {
 610         case BOOK3S_INTERRUPT_INST_STORAGE:
 611         {
 612                 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 613                 ulong shadow_srr1 = svcpu->shadow_srr1;
 614                 vcpu->stat.pf_instruc++;
 615
 616 #ifdef CONFIG_PPC_BOOK3S_32
 617                 /* We set segments as unused segments when invalidating them. So
 618                  * treat the respective fault as segment fault. */
 619                 if (svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT] == SR_INVALID) {
 620                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 621                         r = RESUME_GUEST;
 622                         svcpu_put(svcpu);
 623                         break;
 624                 }
 625 #endif
 626                 svcpu_put(svcpu);
 627
 628                 /* only care about PTEG not found errors, but leave NX alone */
 629                 if (shadow_srr1 & 0x40000000) {
 630                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 631                         vcpu->stat.sp_instruc++;
 632                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 633                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 634                         /*
 635                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 636                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 637                          *     that no guest that needs the dcbz hack does NX.
 638                          */
 639                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 640                         r = RESUME_GUEST;
 641                 } else {
 642                         vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
 643                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 644                         r = RESUME_GUEST;
 645                 }
 646                 break;
 647         }
 648         case BOOK3S_INTERRUPT_DATA_STORAGE:
 649         {
 650                 ulong dar = kvmppc_get_fault_dar(vcpu);
 651                 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 652                 u32 fault_dsisr = svcpu->fault_dsisr;
 653                 vcpu->stat.pf_storage++;
 654
 655 #ifdef CONFIG_PPC_BOOK3S_32
 656                 /* We set segments as unused segments when invalidating them. So
 657                  * treat the respective fault as segment fault. */
 658                 if ((svcpu->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 659                         kvmppc_mmu_map_segment(vcpu, dar);
 660                         r = RESUME_GUEST;
 661                         svcpu_put(svcpu);
 662                         break;
 663                 }
 664 #endif
 665                 svcpu_put(svcpu);
 666
 667                 /* The only case we need to handle is missing shadow PTEs */
 668                 if (fault_dsisr & DSISR_NOHPTE) {
 669                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 670                 } else {
 671                         vcpu->arch.shared->dar = dar;
 672                         vcpu->arch.shared->dsisr = fault_dsisr;
 673                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 674                         r = RESUME_GUEST;
 675                 }
 676                 break;
 677         }
 678         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 679                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 680                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 681                         kvmppc_book3s_queue_irqprio(vcpu,
 682                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 683                 }
 684                 r = RESUME_GUEST;
 685                 break;
 686         case BOOK3S_INTERRUPT_INST_SEGMENT:
 687                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 688                         kvmppc_book3s_queue_irqprio(vcpu,
 689                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 690                 }
 691                 r = RESUME_GUEST;
 692                 break;
 693         /* We're good on these - the host merely wanted to get our attention */
 694         case BOOK3S_INTERRUPT_DECREMENTER:
 695         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 696                 vcpu->stat.dec_exits++;
 697                 r = RESUME_GUEST;
 698                 break;
 699         case BOOK3S_INTERRUPT_EXTERNAL:
 700         case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
 701         case BOOK3S_INTERRUPT_EXTERNAL_HV:
 702                 vcpu->stat.ext_intr_exits++;
 703                 r = RESUME_GUEST;
 704                 break;
 705         case BOOK3S_INTERRUPT_PERFMON:
 706                 r = RESUME_GUEST;
 707                 break;
 708         case BOOK3S_INTERRUPT_PROGRAM:
 709         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 710         {
 711                 enum emulation_result er;
 712                 struct kvmppc_book3s_shadow_vcpu *svcpu;
 713                 ulong flags;
 714
 715 program_interrupt:
 716                 svcpu = svcpu_get(vcpu);
 717                 flags = svcpu->shadow_srr1 & 0x1f0000ull;
 718                 svcpu_put(svcpu);
 719
 720                 if (vcpu->arch.shared->msr & MSR_PR) {
 721 #ifdef EXIT_DEBUG
 722                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 723 #endif
 724                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 725                             (INS_DCBZ & 0xfffffff7)) {
 726                                 kvmppc_core_queue_program(vcpu, flags);
 727                                 r = RESUME_GUEST;
 728                                 break;
 729                         }
 730                 }
 731
 732                 vcpu->stat.emulated_inst_exits++;
 733                 er = kvmppc_emulate_instruction(run, vcpu);
 734                 switch (er) {
 735                 case EMULATE_DONE:
 736                         r = RESUME_GUEST_NV;
 737                         break;
 738                 case EMULATE_AGAIN:
 739                         r = RESUME_GUEST;
 740                         break;
 741                 case EMULATE_FAIL:
 742                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 743                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 744                         kvmppc_core_queue_program(vcpu, flags);
 745                         r = RESUME_GUEST;
 746                         break;
 747                 case EMULATE_DO_MMIO:
 748                         run->exit_reason = KVM_EXIT_MMIO;
 749                         r = RESUME_HOST_NV;
 750                         break;
 751                 default:
 752                         BUG();
 753                 }
 754                 break;
 755         }
 756         case BOOK3S_INTERRUPT_SYSCALL:
 757                 if (vcpu->arch.papr_enabled &&
 758                     (kvmppc_get_last_inst(vcpu) == 0x44000022) &&
 759                     !(vcpu->arch.shared->msr & MSR_PR)) {
 760                         /* SC 1 papr hypercalls */
 761                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
 762                         int i;
 763
 764 #ifdef CONFIG_KVM_BOOK3S_64_PR
 765                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
 766                                 r = RESUME_GUEST;
 767                                 break;
 768                         }
 769 #endif
 770
 771                         run->papr_hcall.nr = cmd;
 772                         for (i = 0; i < 9; ++i) {
 773                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
 774                                 run->papr_hcall.args[i] = gpr;
 775                         }
 776                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
 777                         vcpu->arch.hcall_needed = 1;
 778                         r = RESUME_HOST;
 779                 } else if (vcpu->arch.osi_enabled &&
 780                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 781                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 782                         /* MOL hypercalls */
 783                         u64 *gprs = run->osi.gprs;
 784                         int i;
 785
 786                         run->exit_reason = KVM_EXIT_OSI;
 787                         for (i = 0; i < 32; i++)
 788                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 789                         vcpu->arch.osi_needed = 1;
 790                         r = RESUME_HOST_NV;
 791                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 792                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 793                         /* KVM PV hypercalls */
 794                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 795                         r = RESUME_GUEST;
 796                 } else {
 797                         /* Guest syscalls */
 798                         vcpu->stat.syscall_exits++;
 799                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 800                         r = RESUME_GUEST;
 801                 }
 802                 break;
 803         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 804         case BOOK3S_INTERRUPT_ALTIVEC:
 805         case BOOK3S_INTERRUPT_VSX:
 806         {
 807                 int ext_msr = 0;
 808
 809                 switch (exit_nr) {
 810                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 811                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 812                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 813                 }
 814
 815                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
 816                 case EMULATE_DONE:
 817                         /* everything ok - let's enable the ext */
 818                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
 819                         break;
 820                 case EMULATE_FAIL:
 821                         /* we need to emulate this instruction */
 822                         goto program_interrupt;
 823                         break;
 824                 default:
 825                         /* nothing to worry about - go again */
 826                         break;
 827                 }
 828                 break;
 829         }
 830         case BOOK3S_INTERRUPT_ALIGNMENT:
 831                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
 832                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
 833                                 kvmppc_get_last_inst(vcpu));
 834                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
 835                                 kvmppc_get_last_inst(vcpu));
 836                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 837                 }
 838                 r = RESUME_GUEST;
 839                 break;
 840         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 841         case BOOK3S_INTERRUPT_TRACE:
 842                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 843                 r = RESUME_GUEST;
 844                 break;
 845         default:
 846         {
 847                 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 848                 ulong shadow_srr1 = svcpu->shadow_srr1;
 849                 svcpu_put(svcpu);
 850                 /* Ugh - bork here! What did we get? */
 851                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
 852                         exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
 853                 r = RESUME_HOST;
 854                 BUG();
 855                 break;
 856         }
 857         }
 858
 859         if (!(r & RESUME_HOST)) {
 860                 /* To avoid clobbering exit_reason, only check for signals if
 861                  * we aren't already exiting to userspace for some other
 862                  * reason. */
 863
 864                 /*
 865                  * Interrupts could be timers for the guest which we have to
 866                  * inject again, so let's postpone them until we're in the guest
 867                  * and if we really did time things so badly, then we just exit
 868                  * again due to a host external interrupt.
 869                  */
 870                 local_irq_disable();
 871                 s = kvmppc_prepare_to_enter(vcpu);
 872                 if (s <= 0) {
 873                         local_irq_enable();
 874                         r = s;
 875                 } else {
 876                         kvmppc_lazy_ee_enable();
 877                 }
 878         }
 879
 880         trace_kvm_book3s_reenter(r, vcpu);
 881
 882         return r;
 883 }
 884
 885 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 886                                   struct kvm_sregs *sregs)
 887 {
 888         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 889         int i;
 890
 891         sregs->pvr = vcpu->arch.pvr;
 892
 893         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 894         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 895                 for (i = 0; i < 64; i++) {
 896                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
 897                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 898                 }
 899         } else {
 900                 for (i = 0; i < 16; i++)
 901                         sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
 902
 903                 for (i = 0; i < 8; i++) {
 904                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 905                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 906                 }
 907         }
 908
 909         return 0;
 910 }
 911
 912 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 913                                   struct kvm_sregs *sregs)
 914 {
 915         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 916         int i;
 917
 918         kvmppc_set_pvr(vcpu, sregs->pvr);
 919
 920         vcpu3s->sdr1 = sregs->u.s.sdr1;
 921         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 922                 for (i = 0; i < 64; i++) {
 923                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 924                                                     sregs->u.s.ppc64.slb[i].slbe);
 925                 }
 926         } else {
 927                 for (i = 0; i < 16; i++) {
 928                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 929                 }
 930                 for (i = 0; i < 8; i++) {
 931                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 932                                        (u32)sregs->u.s.ppc32.ibat[i]);
 933                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 934                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 935                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 936                                        (u32)sregs->u.s.ppc32.dbat[i]);
 937                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 938                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 939                 }
 940         }
 941
 942         /* Flush the MMU after messing with the segments */
 943         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 944
 945         return 0;
 946 }
 947
 948 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
 949 {
 950         int r = 0;
 951
 952         switch (id) {
 953         case KVM_REG_PPC_HIOR:
 954                 *val = get_reg_val(id, to_book3s(vcpu)->hior);
 955                 break;
 956         default:
 957                 r = -EINVAL;
 958                 break;
 959         }
 960
 961         return r;
 962 }
 963
 964 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
 965 {
 966         int r = 0;
 967
 968         switch (id) {
 969         case KVM_REG_PPC_HIOR:
 970                 to_book3s(vcpu)->hior = set_reg_val(id, *val);
 971                 to_book3s(vcpu)->hior_explicit = true;
 972                 break;
 973         default:
 974                 r = -EINVAL;
 975                 break;
 976         }
 977
 978         return r;
 979 }
 980
 981 int kvmppc_core_check_processor_compat(void)
 982 {
 983         return 0;
 984 }
 985
 986 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 987 {
 988         struct kvmppc_vcpu_book3s *vcpu_book3s;
 989         struct kvm_vcpu *vcpu;
 990         int err = -ENOMEM;
 991         unsigned long p;
 992
 993         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
 994         if (!vcpu_book3s)
 995                 goto out;
 996
 997         vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
 998                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
 999         if (!vcpu_book3s->shadow_vcpu)
1000                 goto free_vcpu;
1001
1002         vcpu = &vcpu_book3s->vcpu;
1003         err = kvm_vcpu_init(vcpu, kvm, id);
1004         if (err)
1005                 goto free_shadow_vcpu;
1006
1007         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1008         /* the real shared page fills the last 4k of our page */
1009         vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
1010         if (!p)
1011                 goto uninit_vcpu;
1012
1013 #ifdef CONFIG_PPC_BOOK3S_64
1014         /* default to book3s_64 (970fx) */
1015         vcpu->arch.pvr = 0x3C0301;
1016 #else
1017         /* default to book3s_32 (750) */
1018         vcpu->arch.pvr = 0x84202;
1019 #endif
1020         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1021         vcpu->arch.slb_nr = 64;
1022
1023         vcpu->arch.shadow_msr = MSR_USER64;
1024
1025         err = kvmppc_mmu_init(vcpu);
1026         if (err < 0)
1027                 goto uninit_vcpu;
1028
1029         return vcpu;
1030
1031 uninit_vcpu:
1032         kvm_vcpu_uninit(vcpu);
1033 free_shadow_vcpu:
1034         kfree(vcpu_book3s->shadow_vcpu);
1035 free_vcpu:
1036         vfree(vcpu_book3s);
1037 out:
1038         return ERR_PTR(err);
1039 }
1040
1041 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1042 {
1043         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1044
1045         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1046         kvm_vcpu_uninit(vcpu);
1047         kfree(vcpu_book3s->shadow_vcpu);
1048         vfree(vcpu_book3s);
1049 }
1050
1051 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1052 {
1053         int ret;
1054         double fpr[32][TS_FPRWIDTH];
1055         unsigned int fpscr;
1056         int fpexc_mode;
1057 #ifdef CONFIG_ALTIVEC
1058         vector128 vr[32];
1059         vector128 vscr;
1060         unsigned long uninitialized_var(vrsave);
1061         int used_vr;
1062 #endif
1063 #ifdef CONFIG_VSX
1064         int used_vsr;
1065 #endif
1066         ulong ext_msr;
1067
1068         /* Check if we can run the vcpu at all */
1069         if (!vcpu->arch.sane) {
1070                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1071                 ret = -EINVAL;
1072                 goto out;
1073         }
1074
1075         /*
1076          * Interrupts could be timers for the guest which we have to inject
1077          * again, so let's postpone them until we're in the guest and if we
1078          * really did time things so badly, then we just exit again due to
1079          * a host external interrupt.
1080          */
1081         local_irq_disable();
1082         ret = kvmppc_prepare_to_enter(vcpu);
1083         if (ret <= 0) {
1084                 local_irq_enable();
1085                 goto out;
1086         }
1087
1088         /* Save FPU state in stack */
1089         if (current->thread.regs->msr & MSR_FP)
1090                 giveup_fpu(current);
1091         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1092         fpscr = current->thread.fpscr.val;
1093         fpexc_mode = current->thread.fpexc_mode;
1094
1095 #ifdef CONFIG_ALTIVEC
1096         /* Save Altivec state in stack */
1097         used_vr = current->thread.used_vr;
1098         if (used_vr) {
1099                 if (current->thread.regs->msr & MSR_VEC)
1100                         giveup_altivec(current);
1101                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1102                 vscr = current->thread.vscr;
1103                 vrsave = current->thread.vrsave;
1104         }
1105 #endif
1106
1107 #ifdef CONFIG_VSX
1108         /* Save VSX state in stack */
1109         used_vsr = current->thread.used_vsr;
1110         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1111                         __giveup_vsx(current);
1112 #endif
1113
1114         /* Remember the MSR with disabled extensions */
1115         ext_msr = current->thread.regs->msr;
1116
1117         /* Preload FPU if it's enabled */
1118         if (vcpu->arch.shared->msr & MSR_FP)
1119                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1120
1121         kvmppc_lazy_ee_enable();
1122
1123         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1124
1125         /* No need for kvm_guest_exit. It's done in handle_exit.
1126            We also get here with interrupts enabled. */
1127
1128         current->thread.regs->msr = ext_msr;
1129
1130         /* Make sure we save the guest FPU/Altivec/VSX state */
1131         kvmppc_giveup_ext(vcpu, MSR_FP);
1132         kvmppc_giveup_ext(vcpu, MSR_VEC);
1133         kvmppc_giveup_ext(vcpu, MSR_VSX);
1134
1135         /* Restore FPU state from stack */
1136         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1137         current->thread.fpscr.val = fpscr;
1138         current->thread.fpexc_mode = fpexc_mode;
1139
1140 #ifdef CONFIG_ALTIVEC
1141         /* Restore Altivec state from stack */
1142         if (used_vr && current->thread.used_vr) {
1143                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1144                 current->thread.vscr = vscr;
1145                 current->thread.vrsave = vrsave;
1146         }
1147         current->thread.used_vr = used_vr;
1148 #endif
1149
1150 #ifdef CONFIG_VSX
1151         current->thread.used_vsr = used_vsr;
1152 #endif
1153
1154 out:
1155         vcpu->mode = OUTSIDE_GUEST_MODE;
1156         return ret;
1157 }
1158
1159 /*
1160  * Get (and clear) the dirty memory log for a memory slot.
1161  */
1162 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1163                                       struct kvm_dirty_log *log)
1164 {
1165         struct kvm_memory_slot *memslot;
1166         struct kvm_vcpu *vcpu;
1167         ulong ga, ga_end;
1168         int is_dirty = 0;
1169         int r;
1170         unsigned long n;
1171
1172         mutex_lock(&kvm->slots_lock);
1173
1174         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1175         if (r)
1176                 goto out;
1177
1178         /* If nothing is dirty, don't bother messing with page tables. */
1179         if (is_dirty) {
1180                 memslot = id_to_memslot(kvm->memslots, log->slot);
1181
1182                 ga = memslot->base_gfn << PAGE_SHIFT;
1183                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1184
1185                 kvm_for_each_vcpu(n, vcpu, kvm)
1186                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1187
1188                 n = kvm_dirty_bitmap_bytes(memslot);
1189                 memset(memslot->dirty_bitmap, 0, n);
1190         }
1191
1192         r = 0;
1193 out:
1194         mutex_unlock(&kvm->slots_lock);
1195         return r;
1196 }
1197
1198 #ifdef CONFIG_PPC64
1199 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1200 {
1201         /* No flags */
1202         info->flags = 0;
1203
1204         /* SLB is always 64 entries */
1205         info->slb_size = 64;
1206
1207         /* Standard 4k base page size segment */
1208         info->sps[0].page_shift = 12;
1209         info->sps[0].slb_enc = 0;
1210         info->sps[0].enc[0].page_shift = 12;
1211         info->sps[0].enc[0].pte_enc = 0;
1212
1213         /* Standard 16M large page size segment */
1214         info->sps[1].page_shift = 24;
1215         info->sps[1].slb_enc = SLB_VSID_L;
1216         info->sps[1].enc[0].page_shift = 24;
1217         info->sps[1].enc[0].pte_enc = 0;
1218
1219         return 0;
1220 }
1221 #endif /* CONFIG_PPC64 */
1222
1223 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1224                               struct kvm_memory_slot *dont)
1225 {
1226 }
1227
1228 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1229                                unsigned long npages)
1230 {
1231         return 0;
1232 }
1233
1234 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1235                                       struct kvm_memory_slot *memslot,
1236                                       struct kvm_userspace_memory_region *mem)
1237 {
1238         return 0;
1239 }
1240
1241 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1242                                 struct kvm_userspace_memory_region *mem,
1243                                 struct kvm_memory_slot old)
1244 {
1245 }
1246
1247 void kvmppc_core_flush_memslot(struct kvm *kvm, struct kvm_memory_slot *memslot)
1248 {
1249 }
1250
1251 int kvmppc_core_init_vm(struct kvm *kvm)
1252 {
1253 #ifdef CONFIG_PPC64
1254         INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1255 #endif
1256
1257         return 0;
1258 }
1259
1260 void kvmppc_core_destroy_vm(struct kvm *kvm)
1261 {
1262 #ifdef CONFIG_PPC64
1263         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1264 #endif
1265 }
1266
1267 static int kvmppc_book3s_init(void)
1268 {
1269         int r;
1270
1271         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1272                      THIS_MODULE);
1273
1274         if (r)
1275                 return r;
1276
1277         r = kvmppc_mmu_hpte_sysinit();
1278
1279         return r;
1280 }
1281
1282 static void kvmppc_book3s_exit(void)
1283 {
1284         kvmppc_mmu_hpte_sysexit();
1285         kvm_exit();
1286 }
1287
1288 module_init(kvmppc_book3s_init);
1289 module_exit(kvmppc_book3s_exit);