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 <asm/firmware.h>
  38 #include <asm/hvcall.h>
  39 #include <linux/gfp.h>
  40 #include <linux/sched.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/highmem.h>
  43 #include <linux/module.h>
  44 #include <linux/miscdevice.h>
  45
  46 #include "book3s.h"
  47
  48 #define CREATE_TRACE_POINTS
  49 #include "trace_pr.h"
  50
  51 /* #define EXIT_DEBUG */
  52 /* #define DEBUG_EXT */
  53
  54 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  55                              ulong msr);
  56 static void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac);
  57
  58 /* Some compatibility defines */
  59 #ifdef CONFIG_PPC_BOOK3S_32
  60 #define MSR_USER32 MSR_USER
  61 #define MSR_USER64 MSR_USER
  62 #define HW_PAGE_SIZE PAGE_SIZE
  63 #endif
  64
  65 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
  66 {
  67 #ifdef CONFIG_PPC_BOOK3S_64
  68         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  69         memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
  70         svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
  71         svcpu->in_use = 0;
  72         svcpu_put(svcpu);
  73 #endif
  74         vcpu->cpu = smp_processor_id();
  75 #ifdef CONFIG_PPC_BOOK3S_32
  76         current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
  77 #endif
  78 }
  79
  80 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
  81 {
  82 #ifdef CONFIG_PPC_BOOK3S_64
  83         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  84         if (svcpu->in_use) {
  85                 kvmppc_copy_from_svcpu(vcpu, svcpu);
  86         }
  87         memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
  88         to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
  89         svcpu_put(svcpu);
  90 #endif
  91
  92         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
  93         kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
  94         vcpu->cpu = -1;
  95 }
  96
  97 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
  98 void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu *svcpu,
  99                           struct kvm_vcpu *vcpu)
 100 {
 101         svcpu->gpr[0] = vcpu->arch.gpr[0];
 102         svcpu->gpr[1] = vcpu->arch.gpr[1];
 103         svcpu->gpr[2] = vcpu->arch.gpr[2];
 104         svcpu->gpr[3] = vcpu->arch.gpr[3];
 105         svcpu->gpr[4] = vcpu->arch.gpr[4];
 106         svcpu->gpr[5] = vcpu->arch.gpr[5];
 107         svcpu->gpr[6] = vcpu->arch.gpr[6];
 108         svcpu->gpr[7] = vcpu->arch.gpr[7];
 109         svcpu->gpr[8] = vcpu->arch.gpr[8];
 110         svcpu->gpr[9] = vcpu->arch.gpr[9];
 111         svcpu->gpr[10] = vcpu->arch.gpr[10];
 112         svcpu->gpr[11] = vcpu->arch.gpr[11];
 113         svcpu->gpr[12] = vcpu->arch.gpr[12];
 114         svcpu->gpr[13] = vcpu->arch.gpr[13];
 115         svcpu->cr  = vcpu->arch.cr;
 116         svcpu->xer = vcpu->arch.xer;
 117         svcpu->ctr = vcpu->arch.ctr;
 118         svcpu->lr  = vcpu->arch.lr;
 119         svcpu->pc  = vcpu->arch.pc;
 120 #ifdef CONFIG_PPC_BOOK3S_64
 121         svcpu->shadow_fscr = vcpu->arch.shadow_fscr;
 122 #endif
 123         svcpu->in_use = true;
 124 }
 125
 126 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
 127 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu,
 128                             struct kvmppc_book3s_shadow_vcpu *svcpu)
 129 {
 130         /*
 131          * vcpu_put would just call us again because in_use hasn't
 132          * been updated yet.
 133          */
 134         preempt_disable();
 135
 136         /*
 137          * Maybe we were already preempted and synced the svcpu from
 138          * our preempt notifiers. Don't bother touching this svcpu then.
 139          */
 140         if (!svcpu->in_use)
 141                 goto out;
 142
 143         vcpu->arch.gpr[0] = svcpu->gpr[0];
 144         vcpu->arch.gpr[1] = svcpu->gpr[1];
 145         vcpu->arch.gpr[2] = svcpu->gpr[2];
 146         vcpu->arch.gpr[3] = svcpu->gpr[3];
 147         vcpu->arch.gpr[4] = svcpu->gpr[4];
 148         vcpu->arch.gpr[5] = svcpu->gpr[5];
 149         vcpu->arch.gpr[6] = svcpu->gpr[6];
 150         vcpu->arch.gpr[7] = svcpu->gpr[7];
 151         vcpu->arch.gpr[8] = svcpu->gpr[8];
 152         vcpu->arch.gpr[9] = svcpu->gpr[9];
 153         vcpu->arch.gpr[10] = svcpu->gpr[10];
 154         vcpu->arch.gpr[11] = svcpu->gpr[11];
 155         vcpu->arch.gpr[12] = svcpu->gpr[12];
 156         vcpu->arch.gpr[13] = svcpu->gpr[13];
 157         vcpu->arch.cr  = svcpu->cr;
 158         vcpu->arch.xer = svcpu->xer;
 159         vcpu->arch.ctr = svcpu->ctr;
 160         vcpu->arch.lr  = svcpu->lr;
 161         vcpu->arch.pc  = svcpu->pc;
 162         vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
 163         vcpu->arch.fault_dar   = svcpu->fault_dar;
 164         vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
 165         vcpu->arch.last_inst   = svcpu->last_inst;
 166 #ifdef CONFIG_PPC_BOOK3S_64
 167         vcpu->arch.shadow_fscr = svcpu->shadow_fscr;
 168 #endif
 169         svcpu->in_use = false;
 170
 171 out:
 172         preempt_enable();
 173 }
 174
 175 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
 176 {
 177         int r = 1; /* Indicate we want to get back into the guest */
 178
 179         /* We misuse TLB_FLUSH to indicate that we want to clear
 180            all shadow cache entries */
 181         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
 182                 kvmppc_mmu_pte_flush(vcpu, 0, 0);
 183
 184         return r;
 185 }
 186
 187 /************* MMU Notifiers *************/
 188 static void do_kvm_unmap_hva(struct kvm *kvm, unsigned long start,
 189                              unsigned long end)
 190 {
 191         long i;
 192         struct kvm_vcpu *vcpu;
 193         struct kvm_memslots *slots;
 194         struct kvm_memory_slot *memslot;
 195
 196         slots = kvm_memslots(kvm);
 197         kvm_for_each_memslot(memslot, slots) {
 198                 unsigned long hva_start, hva_end;
 199                 gfn_t gfn, gfn_end;
 200
 201                 hva_start = max(start, memslot->userspace_addr);
 202                 hva_end = min(end, memslot->userspace_addr +
 203                                         (memslot->npages << PAGE_SHIFT));
 204                 if (hva_start >= hva_end)
 205                         continue;
 206                 /*
 207                  * {gfn(page) | page intersects with [hva_start, hva_end)} =
 208                  * {gfn, gfn+1, ..., gfn_end-1}.
 209                  */
 210                 gfn = hva_to_gfn_memslot(hva_start, memslot);
 211                 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
 212                 kvm_for_each_vcpu(i, vcpu, kvm)
 213                         kvmppc_mmu_pte_pflush(vcpu, gfn << PAGE_SHIFT,
 214                                               gfn_end << PAGE_SHIFT);
 215         }
 216 }
 217
 218 static int kvm_unmap_hva_pr(struct kvm *kvm, unsigned long hva)
 219 {
 220         trace_kvm_unmap_hva(hva);
 221
 222         do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
 223
 224         return 0;
 225 }
 226
 227 static int kvm_unmap_hva_range_pr(struct kvm *kvm, unsigned long start,
 228                                   unsigned long end)
 229 {
 230         do_kvm_unmap_hva(kvm, start, end);
 231
 232         return 0;
 233 }
 234
 235 static int kvm_age_hva_pr(struct kvm *kvm, unsigned long hva)
 236 {
 237         /* XXX could be more clever ;) */
 238         return 0;
 239 }
 240
 241 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
 242 {
 243         /* XXX could be more clever ;) */
 244         return 0;
 245 }
 246
 247 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
 248 {
 249         /* The page will get remapped properly on its next fault */
 250         do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
 251 }
 252
 253 /*****************************************/
 254
 255 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 256 {
 257         ulong guest_msr = kvmppc_get_msr(vcpu);
 258         ulong smsr = guest_msr;
 259
 260         /* Guest MSR values */
 261         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE;
 262         /* Process MSR values */
 263         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 264         /* External providers the guest reserved */
 265         smsr |= (guest_msr & vcpu->arch.guest_owned_ext);
 266         /* 64-bit Process MSR values */
 267 #ifdef CONFIG_PPC_BOOK3S_64
 268         smsr |= MSR_ISF | MSR_HV;
 269 #endif
 270         vcpu->arch.shadow_msr = smsr;
 271 }
 272
 273 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
 274 {
 275         ulong old_msr = kvmppc_get_msr(vcpu);
 276
 277 #ifdef EXIT_DEBUG
 278         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 279 #endif
 280
 281         msr &= to_book3s(vcpu)->msr_mask;
 282         kvmppc_set_msr_fast(vcpu, msr);
 283         kvmppc_recalc_shadow_msr(vcpu);
 284
 285         if (msr & MSR_POW) {
 286                 if (!vcpu->arch.pending_exceptions) {
 287                         kvm_vcpu_block(vcpu);
 288                         clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 289                         vcpu->stat.halt_wakeup++;
 290
 291                         /* Unset POW bit after we woke up */
 292                         msr &= ~MSR_POW;
 293                         kvmppc_set_msr_fast(vcpu, msr);
 294                 }
 295         }
 296
 297         if ((kvmppc_get_msr(vcpu) & (MSR_PR|MSR_IR|MSR_DR)) !=
 298                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 299                 kvmppc_mmu_flush_segments(vcpu);
 300                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 301
 302                 /* Preload magic page segment when in kernel mode */
 303                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 304                         struct kvm_vcpu_arch *a = &vcpu->arch;
 305
 306                         if (msr & MSR_DR)
 307                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 308                         else
 309                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 310                 }
 311         }
 312
 313         /*
 314          * When switching from 32 to 64-bit, we may have a stale 32-bit
 315          * magic page around, we need to flush it. Typically 32-bit magic
 316          * page will be instanciated when calling into RTAS. Note: We
 317          * assume that such transition only happens while in kernel mode,
 318          * ie, we never transition from user 32-bit to kernel 64-bit with
 319          * a 32-bit magic page around.
 320          */
 321         if (vcpu->arch.magic_page_pa &&
 322             !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
 323                 /* going from RTAS to normal kernel code */
 324                 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
 325                                      ~0xFFFUL);
 326         }
 327
 328         /* Preload FPU if it's enabled */
 329         if (kvmppc_get_msr(vcpu) & MSR_FP)
 330                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 331 }
 332
 333 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
 334 {
 335         u32 host_pvr;
 336
 337         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 338         vcpu->arch.pvr = pvr;
 339 #ifdef CONFIG_PPC_BOOK3S_64
 340         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 341                 kvmppc_mmu_book3s_64_init(vcpu);
 342                 if (!to_book3s(vcpu)->hior_explicit)
 343                         to_book3s(vcpu)->hior = 0xfff00000;
 344                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 345                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 346         } else
 347 #endif
 348         {
 349                 kvmppc_mmu_book3s_32_init(vcpu);
 350                 if (!to_book3s(vcpu)->hior_explicit)
 351                         to_book3s(vcpu)->hior = 0;
 352                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 353                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 354         }
 355
 356         kvmppc_sanity_check(vcpu);
 357
 358         /* If we are in hypervisor level on 970, we can tell the CPU to
 359          * treat DCBZ as 32 bytes store */
 360         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 361         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 362             !strcmp(cur_cpu_spec->platform, "ppc970"))
 363                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 364
 365         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 366            really needs them in a VM on Cell and force disable them. */
 367         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 368                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 369
 370         /*
 371          * If they're asking for POWER6 or later, set the flag
 372          * indicating that we can do multiple large page sizes
 373          * and 1TB segments.
 374          * Also set the flag that indicates that tlbie has the large
 375          * page bit in the RB operand instead of the instruction.
 376          */
 377         switch (PVR_VER(pvr)) {
 378         case PVR_POWER6:
 379         case PVR_POWER7:
 380         case PVR_POWER7p:
 381         case PVR_POWER8:
 382                 vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
 383                         BOOK3S_HFLAG_NEW_TLBIE;
 384                 break;
 385         }
 386
 387 #ifdef CONFIG_PPC_BOOK3S_32
 388         /* 32 bit Book3S always has 32 byte dcbz */
 389         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 390 #endif
 391
 392         /* On some CPUs we can execute paired single operations natively */
 393         asm ( "mfpvr %0" : "=r"(host_pvr));
 394         switch (host_pvr) {
 395         case 0x00080200:        /* lonestar 2.0 */
 396         case 0x00088202:        /* lonestar 2.2 */
 397         case 0x70000100:        /* gekko 1.0 */
 398         case 0x00080100:        /* gekko 2.0 */
 399         case 0x00083203:        /* gekko 2.3a */
 400         case 0x00083213:        /* gekko 2.3b */
 401         case 0x00083204:        /* gekko 2.4 */
 402         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 403         case 0x00087200:        /* broadway */
 404                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 405                 /* Enable HID2.PSE - in case we need it later */
 406                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 407         }
 408 }
 409
 410 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 411  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 412  * emulate 32 bytes dcbz length.
 413  *
 414  * The Book3s_64 inventors also realized this case and implemented a special bit
 415  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 416  *
 417  * My approach here is to patch the dcbz instruction on executing pages.
 418  */
 419 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 420 {
 421         struct page *hpage;
 422         u64 hpage_offset;
 423         u32 *page;
 424         int i;
 425
 426         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 427         if (is_error_page(hpage))
 428                 return;
 429
 430         hpage_offset = pte->raddr & ~PAGE_MASK;
 431         hpage_offset &= ~0xFFFULL;
 432         hpage_offset /= 4;
 433
 434         get_page(hpage);
 435         page = kmap_atomic(hpage);
 436
 437         /* patch dcbz into reserved instruction, so we trap */
 438         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 439                 if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
 440                         page[i] &= cpu_to_be32(0xfffffff7);
 441
 442         kunmap_atomic(page);
 443         put_page(hpage);
 444 }
 445
 446 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 447 {
 448         ulong mp_pa = vcpu->arch.magic_page_pa;
 449
 450         if (!(kvmppc_get_msr(vcpu) & MSR_SF))
 451                 mp_pa = (uint32_t)mp_pa;
 452
 453         if (unlikely(mp_pa) &&
 454             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 455                 return 1;
 456         }
 457
 458         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 459 }
 460
 461 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 462                             ulong eaddr, int vec)
 463 {
 464         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 465         bool iswrite = false;
 466         int r = RESUME_GUEST;
 467         int relocated;
 468         int page_found = 0;
 469         struct kvmppc_pte pte;
 470         bool is_mmio = false;
 471         bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
 472         bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
 473         u64 vsid;
 474
 475         relocated = data ? dr : ir;
 476         if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
 477                 iswrite = true;
 478
 479         /* Resolve real address if translation turned on */
 480         if (relocated) {
 481                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
 482         } else {
 483                 pte.may_execute = true;
 484                 pte.may_read = true;
 485                 pte.may_write = true;
 486                 pte.raddr = eaddr & KVM_PAM;
 487                 pte.eaddr = eaddr;
 488                 pte.vpage = eaddr >> 12;
 489                 pte.page_size = MMU_PAGE_64K;
 490         }
 491
 492         switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
 493         case 0:
 494                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 495                 break;
 496         case MSR_DR:
 497         case MSR_IR:
 498                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 499
 500                 if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
 501                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 502                 else
 503                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 504                 pte.vpage |= vsid;
 505
 506                 if (vsid == -1)
 507                         page_found = -EINVAL;
 508                 break;
 509         }
 510
 511         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 512            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 513                 /*
 514                  * If we do the dcbz hack, we have to NX on every execution,
 515                  * so we can patch the executing code. This renders our guest
 516                  * NX-less.
 517                  */
 518                 pte.may_execute = !data;
 519         }
 520
 521         if (page_found == -ENOENT) {
 522                 /* Page not found in guest PTE entries */
 523                 u64 ssrr1 = vcpu->arch.shadow_srr1;
 524                 u64 msr = kvmppc_get_msr(vcpu);
 525                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 526                 kvmppc_set_dsisr(vcpu, vcpu->arch.fault_dsisr);
 527                 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
 528                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 529         } else if (page_found == -EPERM) {
 530                 /* Storage protection */
 531                 u32 dsisr = vcpu->arch.fault_dsisr;
 532                 u64 ssrr1 = vcpu->arch.shadow_srr1;
 533                 u64 msr = kvmppc_get_msr(vcpu);
 534                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 535                 dsisr = (dsisr & ~DSISR_NOHPTE) | DSISR_PROTFAULT;
 536                 kvmppc_set_dsisr(vcpu, dsisr);
 537                 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
 538                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 539         } else if (page_found == -EINVAL) {
 540                 /* Page not found in guest SLB */
 541                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 542                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 543         } else if (!is_mmio &&
 544                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 545                 if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
 546                         /*
 547                          * There is already a host HPTE there, presumably
 548                          * a read-only one for a page the guest thinks
 549                          * is writable, so get rid of it first.
 550                          */
 551                         kvmppc_mmu_unmap_page(vcpu, &pte);
 552                 }
 553                 /* The guest's PTE is not mapped yet. Map on the host */
 554                 kvmppc_mmu_map_page(vcpu, &pte, iswrite);
 555                 if (data)
 556                         vcpu->stat.sp_storage++;
 557                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 558                          (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 559                         kvmppc_patch_dcbz(vcpu, &pte);
 560         } else {
 561                 /* MMIO */
 562                 vcpu->stat.mmio_exits++;
 563                 vcpu->arch.paddr_accessed = pte.raddr;
 564                 vcpu->arch.vaddr_accessed = pte.eaddr;
 565                 r = kvmppc_emulate_mmio(run, vcpu);
 566                 if ( r == RESUME_HOST_NV )
 567                         r = RESUME_HOST;
 568         }
 569
 570         return r;
 571 }
 572
 573 static inline int get_fpr_index(int i)
 574 {
 575         return i * TS_FPRWIDTH;
 576 }
 577
 578 /* Give up external provider (FPU, Altivec, VSX) */
 579 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 580 {
 581         struct thread_struct *t = &current->thread;
 582
 583         /*
 584          * VSX instructions can access FP and vector registers, so if
 585          * we are giving up VSX, make sure we give up FP and VMX as well.
 586          */
 587         if (msr & MSR_VSX)
 588                 msr |= MSR_FP | MSR_VEC;
 589
 590         msr &= vcpu->arch.guest_owned_ext;
 591         if (!msr)
 592                 return;
 593
 594 #ifdef DEBUG_EXT
 595         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 596 #endif
 597
 598         if (msr & MSR_FP) {
 599                 /*
 600                  * Note that on CPUs with VSX, giveup_fpu stores
 601                  * both the traditional FP registers and the added VSX
 602                  * registers into thread.fp_state.fpr[].
 603                  */
 604                 if (t->regs->msr & MSR_FP)
 605                         giveup_fpu(current);
 606                 t->fp_save_area = NULL;
 607         }
 608
 609 #ifdef CONFIG_ALTIVEC
 610         if (msr & MSR_VEC) {
 611                 if (current->thread.regs->msr & MSR_VEC)
 612                         giveup_altivec(current);
 613                 t->vr_save_area = NULL;
 614         }
 615 #endif
 616
 617         vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
 618         kvmppc_recalc_shadow_msr(vcpu);
 619 }
 620
 621 /* Give up facility (TAR / EBB / DSCR) */
 622 static void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
 623 {
 624 #ifdef CONFIG_PPC_BOOK3S_64
 625         if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
 626                 /* Facility not available to the guest, ignore giveup request*/
 627                 return;
 628         }
 629
 630         switch (fac) {
 631         case FSCR_TAR_LG:
 632                 vcpu->arch.tar = mfspr(SPRN_TAR);
 633                 mtspr(SPRN_TAR, current->thread.tar);
 634                 vcpu->arch.shadow_fscr &= ~FSCR_TAR;
 635                 break;
 636         }
 637 #endif
 638 }
 639
 640 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 641 {
 642         ulong srr0 = kvmppc_get_pc(vcpu);
 643         u32 last_inst = kvmppc_get_last_inst(vcpu);
 644         int ret;
 645
 646         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 647         if (ret == -ENOENT) {
 648                 ulong msr = kvmppc_get_msr(vcpu);
 649
 650                 msr = kvmppc_set_field(msr, 33, 33, 1);
 651                 msr = kvmppc_set_field(msr, 34, 36, 0);
 652                 msr = kvmppc_set_field(msr, 42, 47, 0);
 653                 kvmppc_set_msr_fast(vcpu, msr);
 654                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 655                 return EMULATE_AGAIN;
 656         }
 657
 658         return EMULATE_DONE;
 659 }
 660
 661 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 662 {
 663
 664         /* Need to do paired single emulation? */
 665         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 666                 return EMULATE_DONE;
 667
 668         /* Read out the instruction */
 669         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 670                 /* Need to emulate */
 671                 return EMULATE_FAIL;
 672
 673         return EMULATE_AGAIN;
 674 }
 675
 676 /* Handle external providers (FPU, Altivec, VSX) */
 677 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 678                              ulong msr)
 679 {
 680         struct thread_struct *t = &current->thread;
 681
 682         /* When we have paired singles, we emulate in software */
 683         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 684                 return RESUME_GUEST;
 685
 686         if (!(kvmppc_get_msr(vcpu) & msr)) {
 687                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 688                 return RESUME_GUEST;
 689         }
 690
 691         if (msr == MSR_VSX) {
 692                 /* No VSX?  Give an illegal instruction interrupt */
 693 #ifdef CONFIG_VSX
 694                 if (!cpu_has_feature(CPU_FTR_VSX))
 695 #endif
 696                 {
 697                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 698                         return RESUME_GUEST;
 699                 }
 700
 701                 /*
 702                  * We have to load up all the FP and VMX registers before
 703                  * we can let the guest use VSX instructions.
 704                  */
 705                 msr = MSR_FP | MSR_VEC | MSR_VSX;
 706         }
 707
 708         /* See if we already own all the ext(s) needed */
 709         msr &= ~vcpu->arch.guest_owned_ext;
 710         if (!msr)
 711                 return RESUME_GUEST;
 712
 713 #ifdef DEBUG_EXT
 714         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 715 #endif
 716
 717         if (msr & MSR_FP) {
 718                 preempt_disable();
 719                 enable_kernel_fp();
 720                 load_fp_state(&vcpu->arch.fp);
 721                 t->fp_save_area = &vcpu->arch.fp;
 722                 preempt_enable();
 723         }
 724
 725         if (msr & MSR_VEC) {
 726 #ifdef CONFIG_ALTIVEC
 727                 preempt_disable();
 728                 enable_kernel_altivec();
 729                 load_vr_state(&vcpu->arch.vr);
 730                 t->vr_save_area = &vcpu->arch.vr;
 731                 preempt_enable();
 732 #endif
 733         }
 734
 735         t->regs->msr |= msr;
 736         vcpu->arch.guest_owned_ext |= msr;
 737         kvmppc_recalc_shadow_msr(vcpu);
 738
 739         return RESUME_GUEST;
 740 }
 741
 742 /*
 743  * Kernel code using FP or VMX could have flushed guest state to
 744  * the thread_struct; if so, get it back now.
 745  */
 746 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
 747 {
 748         unsigned long lost_ext;
 749
 750         lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
 751         if (!lost_ext)
 752                 return;
 753
 754         if (lost_ext & MSR_FP) {
 755                 preempt_disable();
 756                 enable_kernel_fp();
 757                 load_fp_state(&vcpu->arch.fp);
 758                 preempt_enable();
 759         }
 760 #ifdef CONFIG_ALTIVEC
 761         if (lost_ext & MSR_VEC) {
 762                 preempt_disable();
 763                 enable_kernel_altivec();
 764                 load_vr_state(&vcpu->arch.vr);
 765                 preempt_enable();
 766         }
 767 #endif
 768         current->thread.regs->msr |= lost_ext;
 769 }
 770
 771 #ifdef CONFIG_PPC_BOOK3S_64
 772
 773 static void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
 774 {
 775         /* Inject the Interrupt Cause field and trigger a guest interrupt */
 776         vcpu->arch.fscr &= ~(0xffULL << 56);
 777         vcpu->arch.fscr |= (fac << 56);
 778         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
 779 }
 780
 781 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
 782 {
 783         enum emulation_result er = EMULATE_FAIL;
 784
 785         if (!(kvmppc_get_msr(vcpu) & MSR_PR))
 786                 er = kvmppc_emulate_instruction(vcpu->run, vcpu);
 787
 788         if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
 789                 /* Couldn't emulate, trigger interrupt in guest */
 790                 kvmppc_trigger_fac_interrupt(vcpu, fac);
 791         }
 792 }
 793
 794 /* Enable facilities (TAR, EBB, DSCR) for the guest */
 795 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
 796 {
 797         BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
 798
 799         if (!(vcpu->arch.fscr & (1ULL << fac))) {
 800                 /* Facility not enabled by the guest */
 801                 kvmppc_trigger_fac_interrupt(vcpu, fac);
 802                 return RESUME_GUEST;
 803         }
 804
 805         switch (fac) {
 806         case FSCR_TAR_LG:
 807                 /* TAR switching isn't lazy in Linux yet */
 808                 current->thread.tar = mfspr(SPRN_TAR);
 809                 mtspr(SPRN_TAR, vcpu->arch.tar);
 810                 vcpu->arch.shadow_fscr |= FSCR_TAR;
 811                 break;
 812         default:
 813                 kvmppc_emulate_fac(vcpu, fac);
 814                 break;
 815         }
 816
 817         return RESUME_GUEST;
 818 }
 819 #endif
 820
 821 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
 822                           unsigned int exit_nr)
 823 {
 824         int r = RESUME_HOST;
 825         int s;
 826
 827         vcpu->stat.sum_exits++;
 828
 829         run->exit_reason = KVM_EXIT_UNKNOWN;
 830         run->ready_for_interrupt_injection = 1;
 831
 832         /* We get here with MSR.EE=1 */
 833
 834         trace_kvm_exit(exit_nr, vcpu);
 835         kvm_guest_exit();
 836
 837         switch (exit_nr) {
 838         case BOOK3S_INTERRUPT_INST_STORAGE:
 839         {
 840                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 841                 vcpu->stat.pf_instruc++;
 842
 843 #ifdef CONFIG_PPC_BOOK3S_32
 844                 /* We set segments as unused segments when invalidating them. So
 845                  * treat the respective fault as segment fault. */
 846                 {
 847                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 848                         u32 sr;
 849
 850                         svcpu = svcpu_get(vcpu);
 851                         sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
 852                         svcpu_put(svcpu);
 853                         if (sr == SR_INVALID) {
 854                                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 855                                 r = RESUME_GUEST;
 856                                 break;
 857                         }
 858                 }
 859 #endif
 860
 861                 /* only care about PTEG not found errors, but leave NX alone */
 862                 if (shadow_srr1 & 0x40000000) {
 863                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
 864                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 865                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 866                         vcpu->stat.sp_instruc++;
 867                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 868                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 869                         /*
 870                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 871                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 872                          *     that no guest that needs the dcbz hack does NX.
 873                          */
 874                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 875                         r = RESUME_GUEST;
 876                 } else {
 877                         u64 msr = kvmppc_get_msr(vcpu);
 878                         msr |= shadow_srr1 & 0x58000000;
 879                         kvmppc_set_msr_fast(vcpu, msr);
 880                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 881                         r = RESUME_GUEST;
 882                 }
 883                 break;
 884         }
 885         case BOOK3S_INTERRUPT_DATA_STORAGE:
 886         {
 887                 ulong dar = kvmppc_get_fault_dar(vcpu);
 888                 u32 fault_dsisr = vcpu->arch.fault_dsisr;
 889                 vcpu->stat.pf_storage++;
 890
 891 #ifdef CONFIG_PPC_BOOK3S_32
 892                 /* We set segments as unused segments when invalidating them. So
 893                  * treat the respective fault as segment fault. */
 894                 {
 895                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 896                         u32 sr;
 897
 898                         svcpu = svcpu_get(vcpu);
 899                         sr = svcpu->sr[dar >> SID_SHIFT];
 900                         svcpu_put(svcpu);
 901                         if (sr == SR_INVALID) {
 902                                 kvmppc_mmu_map_segment(vcpu, dar);
 903                                 r = RESUME_GUEST;
 904                                 break;
 905                         }
 906                 }
 907 #endif
 908
 909                 /*
 910                  * We need to handle missing shadow PTEs, and
 911                  * protection faults due to us mapping a page read-only
 912                  * when the guest thinks it is writable.
 913                  */
 914                 if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
 915                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
 916                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 917                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 918                 } else {
 919                         kvmppc_set_dar(vcpu, dar);
 920                         kvmppc_set_dsisr(vcpu, fault_dsisr);
 921                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 922                         r = RESUME_GUEST;
 923                 }
 924                 break;
 925         }
 926         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 927                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 928                         kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 929                         kvmppc_book3s_queue_irqprio(vcpu,
 930                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 931                 }
 932                 r = RESUME_GUEST;
 933                 break;
 934         case BOOK3S_INTERRUPT_INST_SEGMENT:
 935                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 936                         kvmppc_book3s_queue_irqprio(vcpu,
 937                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 938                 }
 939                 r = RESUME_GUEST;
 940                 break;
 941         /* We're good on these - the host merely wanted to get our attention */
 942         case BOOK3S_INTERRUPT_DECREMENTER:
 943         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 944         case BOOK3S_INTERRUPT_DOORBELL:
 945                 vcpu->stat.dec_exits++;
 946                 r = RESUME_GUEST;
 947                 break;
 948         case BOOK3S_INTERRUPT_EXTERNAL:
 949         case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
 950         case BOOK3S_INTERRUPT_EXTERNAL_HV:
 951                 vcpu->stat.ext_intr_exits++;
 952                 r = RESUME_GUEST;
 953                 break;
 954         case BOOK3S_INTERRUPT_PERFMON:
 955                 r = RESUME_GUEST;
 956                 break;
 957         case BOOK3S_INTERRUPT_PROGRAM:
 958         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 959         {
 960                 enum emulation_result er;
 961                 ulong flags;
 962
 963 program_interrupt:
 964                 flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
 965
 966                 if (kvmppc_get_msr(vcpu) & MSR_PR) {
 967 #ifdef EXIT_DEBUG
 968                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 969 #endif
 970                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 971                             (INS_DCBZ & 0xfffffff7)) {
 972                                 kvmppc_core_queue_program(vcpu, flags);
 973                                 r = RESUME_GUEST;
 974                                 break;
 975                         }
 976                 }
 977
 978                 vcpu->stat.emulated_inst_exits++;
 979                 er = kvmppc_emulate_instruction(run, vcpu);
 980                 switch (er) {
 981                 case EMULATE_DONE:
 982                         r = RESUME_GUEST_NV;
 983                         break;
 984                 case EMULATE_AGAIN:
 985                         r = RESUME_GUEST;
 986                         break;
 987                 case EMULATE_FAIL:
 988                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 989                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 990                         kvmppc_core_queue_program(vcpu, flags);
 991                         r = RESUME_GUEST;
 992                         break;
 993                 case EMULATE_DO_MMIO:
 994                         run->exit_reason = KVM_EXIT_MMIO;
 995                         r = RESUME_HOST_NV;
 996                         break;
 997                 case EMULATE_EXIT_USER:
 998                         r = RESUME_HOST_NV;
 999                         break;
1000                 default:
1001                         BUG();
1002                 }
1003                 break;
1004         }
1005         case BOOK3S_INTERRUPT_SYSCALL:
1006                 if (vcpu->arch.papr_enabled &&
1007                     (kvmppc_get_last_sc(vcpu) == 0x44000022) &&
1008                     !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1009                         /* SC 1 papr hypercalls */
1010                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
1011                         int i;
1012
1013 #ifdef CONFIG_PPC_BOOK3S_64
1014                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1015                                 r = RESUME_GUEST;
1016                                 break;
1017                         }
1018 #endif
1019
1020                         run->papr_hcall.nr = cmd;
1021                         for (i = 0; i < 9; ++i) {
1022                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1023                                 run->papr_hcall.args[i] = gpr;
1024                         }
1025                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
1026                         vcpu->arch.hcall_needed = 1;
1027                         r = RESUME_HOST;
1028                 } else if (vcpu->arch.osi_enabled &&
1029                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1030                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1031                         /* MOL hypercalls */
1032                         u64 *gprs = run->osi.gprs;
1033                         int i;
1034
1035                         run->exit_reason = KVM_EXIT_OSI;
1036                         for (i = 0; i < 32; i++)
1037                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
1038                         vcpu->arch.osi_needed = 1;
1039                         r = RESUME_HOST_NV;
1040                 } else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1041                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1042                         /* KVM PV hypercalls */
1043                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1044                         r = RESUME_GUEST;
1045                 } else {
1046                         /* Guest syscalls */
1047                         vcpu->stat.syscall_exits++;
1048                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1049                         r = RESUME_GUEST;
1050                 }
1051                 break;
1052         case BOOK3S_INTERRUPT_FP_UNAVAIL:
1053         case BOOK3S_INTERRUPT_ALTIVEC:
1054         case BOOK3S_INTERRUPT_VSX:
1055         {
1056                 int ext_msr = 0;
1057
1058                 switch (exit_nr) {
1059                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
1060                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
1061                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
1062                 }
1063
1064                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
1065                 case EMULATE_DONE:
1066                         /* everything ok - let's enable the ext */
1067                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1068                         break;
1069                 case EMULATE_FAIL:
1070                         /* we need to emulate this instruction */
1071                         goto program_interrupt;
1072                         break;
1073                 default:
1074                         /* nothing to worry about - go again */
1075                         break;
1076                 }
1077                 break;
1078         }
1079         case BOOK3S_INTERRUPT_ALIGNMENT:
1080                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
1081                         u32 last_inst = kvmppc_get_last_inst(vcpu);
1082                         u32 dsisr;
1083                         u64 dar;
1084
1085                         dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1086                         dar = kvmppc_alignment_dar(vcpu, last_inst);
1087
1088                         kvmppc_set_dsisr(vcpu, dsisr);
1089                         kvmppc_set_dar(vcpu, dar);
1090
1091                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1092                 }
1093                 r = RESUME_GUEST;
1094                 break;
1095 #ifdef CONFIG_PPC_BOOK3S_64
1096         case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1097                 kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1098                 r = RESUME_GUEST;
1099                 break;
1100 #endif
1101         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1102         case BOOK3S_INTERRUPT_TRACE:
1103                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1104                 r = RESUME_GUEST;
1105                 break;
1106         default:
1107         {
1108                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1109                 /* Ugh - bork here! What did we get? */
1110                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1111                         exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1112                 r = RESUME_HOST;
1113                 BUG();
1114                 break;
1115         }
1116         }
1117
1118         if (!(r & RESUME_HOST)) {
1119                 /* To avoid clobbering exit_reason, only check for signals if
1120                  * we aren't already exiting to userspace for some other
1121                  * reason. */
1122
1123                 /*
1124                  * Interrupts could be timers for the guest which we have to
1125                  * inject again, so let's postpone them until we're in the guest
1126                  * and if we really did time things so badly, then we just exit
1127                  * again due to a host external interrupt.
1128                  */
1129                 s = kvmppc_prepare_to_enter(vcpu);
1130                 if (s <= 0)
1131                         r = s;
1132                 else {
1133                         /* interrupts now hard-disabled */
1134                         kvmppc_fix_ee_before_entry();
1135                 }
1136
1137                 kvmppc_handle_lost_ext(vcpu);
1138         }
1139
1140         trace_kvm_book3s_reenter(r, vcpu);
1141
1142         return r;
1143 }
1144
1145 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1146                                             struct kvm_sregs *sregs)
1147 {
1148         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1149         int i;
1150
1151         sregs->pvr = vcpu->arch.pvr;
1152
1153         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1154         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1155                 for (i = 0; i < 64; i++) {
1156                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1157                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1158                 }
1159         } else {
1160                 for (i = 0; i < 16; i++)
1161                         sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1162
1163                 for (i = 0; i < 8; i++) {
1164                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1165                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1166                 }
1167         }
1168
1169         return 0;
1170 }
1171
1172 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1173                                             struct kvm_sregs *sregs)
1174 {
1175         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1176         int i;
1177
1178         kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1179
1180         vcpu3s->sdr1 = sregs->u.s.sdr1;
1181         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1182                 for (i = 0; i < 64; i++) {
1183                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1184                                                     sregs->u.s.ppc64.slb[i].slbe);
1185                 }
1186         } else {
1187                 for (i = 0; i < 16; i++) {
1188                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1189                 }
1190                 for (i = 0; i < 8; i++) {
1191                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1192                                        (u32)sregs->u.s.ppc32.ibat[i]);
1193                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1194                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1195                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1196                                        (u32)sregs->u.s.ppc32.dbat[i]);
1197                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1198                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1199                 }
1200         }
1201
1202         /* Flush the MMU after messing with the segments */
1203         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1204
1205         return 0;
1206 }
1207
1208 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1209                                  union kvmppc_one_reg *val)
1210 {
1211         int r = 0;
1212
1213         switch (id) {
1214         case KVM_REG_PPC_HIOR:
1215                 *val = get_reg_val(id, to_book3s(vcpu)->hior);
1216                 break;
1217         case KVM_REG_PPC_LPCR:
1218                 /*
1219                  * We are only interested in the LPCR_ILE bit
1220                  */
1221                 if (vcpu->arch.intr_msr & MSR_LE)
1222                         *val = get_reg_val(id, LPCR_ILE);
1223                 else
1224                         *val = get_reg_val(id, 0);
1225                 break;
1226         default:
1227                 r = -EINVAL;
1228                 break;
1229         }
1230
1231         return r;
1232 }
1233
1234 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1235 {
1236         if (new_lpcr & LPCR_ILE)
1237                 vcpu->arch.intr_msr |= MSR_LE;
1238         else
1239                 vcpu->arch.intr_msr &= ~MSR_LE;
1240 }
1241
1242 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1243                                  union kvmppc_one_reg *val)
1244 {
1245         int r = 0;
1246
1247         switch (id) {
1248         case KVM_REG_PPC_HIOR:
1249                 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1250                 to_book3s(vcpu)->hior_explicit = true;
1251                 break;
1252         case KVM_REG_PPC_LPCR:
1253                 kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1254                 break;
1255         default:
1256                 r = -EINVAL;
1257                 break;
1258         }
1259
1260         return r;
1261 }
1262
1263 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1264                                                    unsigned int id)
1265 {
1266         struct kvmppc_vcpu_book3s *vcpu_book3s;
1267         struct kvm_vcpu *vcpu;
1268         int err = -ENOMEM;
1269         unsigned long p;
1270
1271         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1272         if (!vcpu)
1273                 goto out;
1274
1275         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1276         if (!vcpu_book3s)
1277                 goto free_vcpu;
1278         vcpu->arch.book3s = vcpu_book3s;
1279
1280 #ifdef CONFIG_KVM_BOOK3S_32
1281         vcpu->arch.shadow_vcpu =
1282                 kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1283         if (!vcpu->arch.shadow_vcpu)
1284                 goto free_vcpu3s;
1285 #endif
1286
1287         err = kvm_vcpu_init(vcpu, kvm, id);
1288         if (err)
1289                 goto free_shadow_vcpu;
1290
1291         err = -ENOMEM;
1292         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1293         if (!p)
1294                 goto uninit_vcpu;
1295         /* the real shared page fills the last 4k of our page */
1296         vcpu->arch.shared = (void *)(p + PAGE_SIZE - 4096);
1297 #ifdef CONFIG_PPC_BOOK3S_64
1298         /* Always start the shared struct in native endian mode */
1299 #ifdef __BIG_ENDIAN__
1300         vcpu->arch.shared_big_endian = true;
1301 #else
1302         vcpu->arch.shared_big_endian = false;
1303 #endif
1304
1305         /*
1306          * Default to the same as the host if we're on sufficiently
1307          * recent machine that we have 1TB segments;
1308          * otherwise default to PPC970FX.
1309          */
1310         vcpu->arch.pvr = 0x3C0301;
1311         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1312                 vcpu->arch.pvr = mfspr(SPRN_PVR);
1313         vcpu->arch.intr_msr = MSR_SF;
1314 #else
1315         /* default to book3s_32 (750) */
1316         vcpu->arch.pvr = 0x84202;
1317 #endif
1318         kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1319         vcpu->arch.slb_nr = 64;
1320
1321         vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1322
1323         err = kvmppc_mmu_init(vcpu);
1324         if (err < 0)
1325                 goto uninit_vcpu;
1326
1327         return vcpu;
1328
1329 uninit_vcpu:
1330         kvm_vcpu_uninit(vcpu);
1331 free_shadow_vcpu:
1332 #ifdef CONFIG_KVM_BOOK3S_32
1333         kfree(vcpu->arch.shadow_vcpu);
1334 free_vcpu3s:
1335 #endif
1336         vfree(vcpu_book3s);
1337 free_vcpu:
1338         kmem_cache_free(kvm_vcpu_cache, vcpu);
1339 out:
1340         return ERR_PTR(err);
1341 }
1342
1343 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1344 {
1345         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1346
1347         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1348         kvm_vcpu_uninit(vcpu);
1349 #ifdef CONFIG_KVM_BOOK3S_32
1350         kfree(vcpu->arch.shadow_vcpu);
1351 #endif
1352         vfree(vcpu_book3s);
1353         kmem_cache_free(kvm_vcpu_cache, vcpu);
1354 }
1355
1356 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1357 {
1358         int ret;
1359 #ifdef CONFIG_ALTIVEC
1360         unsigned long uninitialized_var(vrsave);
1361 #endif
1362
1363         /* Check if we can run the vcpu at all */
1364         if (!vcpu->arch.sane) {
1365                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1366                 ret = -EINVAL;
1367                 goto out;
1368         }
1369
1370         /*
1371          * Interrupts could be timers for the guest which we have to inject
1372          * again, so let's postpone them until we're in the guest and if we
1373          * really did time things so badly, then we just exit again due to
1374          * a host external interrupt.
1375          */
1376         ret = kvmppc_prepare_to_enter(vcpu);
1377         if (ret <= 0)
1378                 goto out;
1379         /* interrupts now hard-disabled */
1380
1381         /* Save FPU state in thread_struct */
1382         if (current->thread.regs->msr & MSR_FP)
1383                 giveup_fpu(current);
1384
1385 #ifdef CONFIG_ALTIVEC
1386         /* Save Altivec state in thread_struct */
1387         if (current->thread.regs->msr & MSR_VEC)
1388                 giveup_altivec(current);
1389 #endif
1390
1391 #ifdef CONFIG_VSX
1392         /* Save VSX state in thread_struct */
1393         if (current->thread.regs->msr & MSR_VSX)
1394                 __giveup_vsx(current);
1395 #endif
1396
1397         /* Preload FPU if it's enabled */
1398         if (kvmppc_get_msr(vcpu) & MSR_FP)
1399                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1400
1401         kvmppc_fix_ee_before_entry();
1402
1403         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1404
1405         /* No need for kvm_guest_exit. It's done in handle_exit.
1406            We also get here with interrupts enabled. */
1407
1408         /* Make sure we save the guest FPU/Altivec/VSX state */
1409         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1410
1411         /* Make sure we save the guest TAR/EBB/DSCR state */
1412         kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1413
1414 out:
1415         vcpu->mode = OUTSIDE_GUEST_MODE;
1416         return ret;
1417 }
1418
1419 /*
1420  * Get (and clear) the dirty memory log for a memory slot.
1421  */
1422 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1423                                          struct kvm_dirty_log *log)
1424 {
1425         struct kvm_memory_slot *memslot;
1426         struct kvm_vcpu *vcpu;
1427         ulong ga, ga_end;
1428         int is_dirty = 0;
1429         int r;
1430         unsigned long n;
1431
1432         mutex_lock(&kvm->slots_lock);
1433
1434         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1435         if (r)
1436                 goto out;
1437
1438         /* If nothing is dirty, don't bother messing with page tables. */
1439         if (is_dirty) {
1440                 memslot = id_to_memslot(kvm->memslots, log->slot);
1441
1442                 ga = memslot->base_gfn << PAGE_SHIFT;
1443                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1444
1445                 kvm_for_each_vcpu(n, vcpu, kvm)
1446                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1447
1448                 n = kvm_dirty_bitmap_bytes(memslot);
1449                 memset(memslot->dirty_bitmap, 0, n);
1450         }
1451
1452         r = 0;
1453 out:
1454         mutex_unlock(&kvm->slots_lock);
1455         return r;
1456 }
1457
1458 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1459                                          struct kvm_memory_slot *memslot)
1460 {
1461         return;
1462 }
1463
1464 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1465                                         struct kvm_memory_slot *memslot,
1466                                         struct kvm_userspace_memory_region *mem)
1467 {
1468         return 0;
1469 }
1470
1471 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1472                                 struct kvm_userspace_memory_region *mem,
1473                                 const struct kvm_memory_slot *old)
1474 {
1475         return;
1476 }
1477
1478 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1479                                         struct kvm_memory_slot *dont)
1480 {
1481         return;
1482 }
1483
1484 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1485                                          unsigned long npages)
1486 {
1487         return 0;
1488 }
1489
1490
1491 #ifdef CONFIG_PPC64
1492 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1493                                          struct kvm_ppc_smmu_info *info)
1494 {
1495         long int i;
1496         struct kvm_vcpu *vcpu;
1497
1498         info->flags = 0;
1499
1500         /* SLB is always 64 entries */
1501         info->slb_size = 64;
1502
1503         /* Standard 4k base page size segment */
1504         info->sps[0].page_shift = 12;
1505         info->sps[0].slb_enc = 0;
1506         info->sps[0].enc[0].page_shift = 12;
1507         info->sps[0].enc[0].pte_enc = 0;
1508
1509         /*
1510          * 64k large page size.
1511          * We only want to put this in if the CPUs we're emulating
1512          * support it, but unfortunately we don't have a vcpu easily
1513          * to hand here to test.  Just pick the first vcpu, and if
1514          * that doesn't exist yet, report the minimum capability,
1515          * i.e., no 64k pages.
1516          * 1T segment support goes along with 64k pages.
1517          */
1518         i = 1;
1519         vcpu = kvm_get_vcpu(kvm, 0);
1520         if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1521                 info->flags = KVM_PPC_1T_SEGMENTS;
1522                 info->sps[i].page_shift = 16;
1523                 info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1524                 info->sps[i].enc[0].page_shift = 16;
1525                 info->sps[i].enc[0].pte_enc = 1;
1526                 ++i;
1527         }
1528
1529         /* Standard 16M large page size segment */
1530         info->sps[i].page_shift = 24;
1531         info->sps[i].slb_enc = SLB_VSID_L;
1532         info->sps[i].enc[0].page_shift = 24;
1533         info->sps[i].enc[0].pte_enc = 0;
1534
1535         return 0;
1536 }
1537 #else
1538 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1539                                          struct kvm_ppc_smmu_info *info)
1540 {
1541         /* We should not get called */
1542         BUG();
1543 }
1544 #endif /* CONFIG_PPC64 */
1545
1546 static unsigned int kvm_global_user_count = 0;
1547 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1548
1549 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1550 {
1551         mutex_init(&kvm->arch.hpt_mutex);
1552
1553         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1554                 spin_lock(&kvm_global_user_count_lock);
1555                 if (++kvm_global_user_count == 1)
1556                         pSeries_disable_reloc_on_exc();
1557                 spin_unlock(&kvm_global_user_count_lock);
1558         }
1559         return 0;
1560 }
1561
1562 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1563 {
1564 #ifdef CONFIG_PPC64
1565         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1566 #endif
1567
1568         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1569                 spin_lock(&kvm_global_user_count_lock);
1570                 BUG_ON(kvm_global_user_count == 0);
1571                 if (--kvm_global_user_count == 0)
1572                         pSeries_enable_reloc_on_exc();
1573                 spin_unlock(&kvm_global_user_count_lock);
1574         }
1575 }
1576
1577 static int kvmppc_core_check_processor_compat_pr(void)
1578 {
1579         /* we are always compatible */
1580         return 0;
1581 }
1582
1583 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1584                                  unsigned int ioctl, unsigned long arg)
1585 {
1586         return -ENOTTY;
1587 }
1588
1589 static struct kvmppc_ops kvm_ops_pr = {
1590         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1591         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1592         .get_one_reg = kvmppc_get_one_reg_pr,
1593         .set_one_reg = kvmppc_set_one_reg_pr,
1594         .vcpu_load   = kvmppc_core_vcpu_load_pr,
1595         .vcpu_put    = kvmppc_core_vcpu_put_pr,
1596         .set_msr     = kvmppc_set_msr_pr,
1597         .vcpu_run    = kvmppc_vcpu_run_pr,
1598         .vcpu_create = kvmppc_core_vcpu_create_pr,
1599         .vcpu_free   = kvmppc_core_vcpu_free_pr,
1600         .check_requests = kvmppc_core_check_requests_pr,
1601         .get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1602         .flush_memslot = kvmppc_core_flush_memslot_pr,
1603         .prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1604         .commit_memory_region = kvmppc_core_commit_memory_region_pr,
1605         .unmap_hva = kvm_unmap_hva_pr,
1606         .unmap_hva_range = kvm_unmap_hva_range_pr,
1607         .age_hva  = kvm_age_hva_pr,
1608         .test_age_hva = kvm_test_age_hva_pr,
1609         .set_spte_hva = kvm_set_spte_hva_pr,
1610         .mmu_destroy  = kvmppc_mmu_destroy_pr,
1611         .free_memslot = kvmppc_core_free_memslot_pr,
1612         .create_memslot = kvmppc_core_create_memslot_pr,
1613         .init_vm = kvmppc_core_init_vm_pr,
1614         .destroy_vm = kvmppc_core_destroy_vm_pr,
1615         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1616         .emulate_op = kvmppc_core_emulate_op_pr,
1617         .emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1618         .emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1619         .fast_vcpu_kick = kvm_vcpu_kick,
1620         .arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1621 };
1622
1623
1624 int kvmppc_book3s_init_pr(void)
1625 {
1626         int r;
1627
1628         r = kvmppc_core_check_processor_compat_pr();
1629         if (r < 0)
1630                 return r;
1631
1632         kvm_ops_pr.owner = THIS_MODULE;
1633         kvmppc_pr_ops = &kvm_ops_pr;
1634
1635         r = kvmppc_mmu_hpte_sysinit();
1636         return r;
1637 }
1638
1639 void kvmppc_book3s_exit_pr(void)
1640 {
1641         kvmppc_pr_ops = NULL;
1642         kvmppc_mmu_hpte_sysexit();
1643 }
1644
1645 /*
1646  * We only support separate modules for book3s 64
1647  */
1648 #ifdef CONFIG_PPC_BOOK3S_64
1649
1650 module_init(kvmppc_book3s_init_pr);
1651 module_exit(kvmppc_book3s_exit_pr);
1652
1653 MODULE_LICENSE("GPL");
1654 MODULE_ALIAS_MISCDEV(KVM_MINOR);
1655 MODULE_ALIAS("devname:kvm");
1656 #endif