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 bool kvmppc_is_split_real(struct kvm_vcpu *vcpu)
  66 {
  67         ulong msr = kvmppc_get_msr(vcpu);
  68         return (msr & (MSR_IR|MSR_DR)) == MSR_DR;
  69 }
  70
  71 static void kvmppc_fixup_split_real(struct kvm_vcpu *vcpu)
  72 {
  73         ulong msr = kvmppc_get_msr(vcpu);
  74         ulong pc = kvmppc_get_pc(vcpu);
  75
  76         /* We are in DR only split real mode */
  77         if ((msr & (MSR_IR|MSR_DR)) != MSR_DR)
  78                 return;
  79
  80         /* We have not fixed up the guest already */
  81         if (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK)
  82                 return;
  83
  84         /* The code is in fixupable address space */
  85         if (pc & SPLIT_HACK_MASK)
  86                 return;
  87
  88         vcpu->arch.hflags |= BOOK3S_HFLAG_SPLIT_HACK;
  89         kvmppc_set_pc(vcpu, pc | SPLIT_HACK_OFFS);
  90 }
  91
  92 void kvmppc_unfixup_split_real(struct kvm_vcpu *vcpu);
  93
  94 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
  95 {
  96 #ifdef CONFIG_PPC_BOOK3S_64
  97         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  98         memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
  99         svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
 100         svcpu->in_use = 0;
 101         svcpu_put(svcpu);
 102 #endif
 103
 104         /* Disable AIL if supported */
 105         if (cpu_has_feature(CPU_FTR_HVMODE) &&
 106             cpu_has_feature(CPU_FTR_ARCH_207S))
 107                 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) & ~LPCR_AIL);
 108
 109         vcpu->cpu = smp_processor_id();
 110 #ifdef CONFIG_PPC_BOOK3S_32
 111         current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
 112 #endif
 113
 114         if (kvmppc_is_split_real(vcpu))
 115                 kvmppc_fixup_split_real(vcpu);
 116 }
 117
 118 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
 119 {
 120 #ifdef CONFIG_PPC_BOOK3S_64
 121         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 122         if (svcpu->in_use) {
 123                 kvmppc_copy_from_svcpu(vcpu, svcpu);
 124         }
 125         memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
 126         to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
 127         svcpu_put(svcpu);
 128 #endif
 129
 130         if (kvmppc_is_split_real(vcpu))
 131                 kvmppc_unfixup_split_real(vcpu);
 132
 133         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
 134         kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
 135
 136         /* Enable AIL if supported */
 137         if (cpu_has_feature(CPU_FTR_HVMODE) &&
 138             cpu_has_feature(CPU_FTR_ARCH_207S))
 139                 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_AIL_3);
 140
 141         vcpu->cpu = -1;
 142 }
 143
 144 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
 145 void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu *svcpu,
 146                           struct kvm_vcpu *vcpu)
 147 {
 148         svcpu->gpr[0] = vcpu->arch.gpr[0];
 149         svcpu->gpr[1] = vcpu->arch.gpr[1];
 150         svcpu->gpr[2] = vcpu->arch.gpr[2];
 151         svcpu->gpr[3] = vcpu->arch.gpr[3];
 152         svcpu->gpr[4] = vcpu->arch.gpr[4];
 153         svcpu->gpr[5] = vcpu->arch.gpr[5];
 154         svcpu->gpr[6] = vcpu->arch.gpr[6];
 155         svcpu->gpr[7] = vcpu->arch.gpr[7];
 156         svcpu->gpr[8] = vcpu->arch.gpr[8];
 157         svcpu->gpr[9] = vcpu->arch.gpr[9];
 158         svcpu->gpr[10] = vcpu->arch.gpr[10];
 159         svcpu->gpr[11] = vcpu->arch.gpr[11];
 160         svcpu->gpr[12] = vcpu->arch.gpr[12];
 161         svcpu->gpr[13] = vcpu->arch.gpr[13];
 162         svcpu->cr  = vcpu->arch.cr;
 163         svcpu->xer = vcpu->arch.xer;
 164         svcpu->ctr = vcpu->arch.ctr;
 165         svcpu->lr  = vcpu->arch.lr;
 166         svcpu->pc  = vcpu->arch.pc;
 167 #ifdef CONFIG_PPC_BOOK3S_64
 168         svcpu->shadow_fscr = vcpu->arch.shadow_fscr;
 169 #endif
 170         /*
 171          * Now also save the current time base value. We use this
 172          * to find the guest purr and spurr value.
 173          */
 174         vcpu->arch.entry_tb = get_tb();
 175         vcpu->arch.entry_vtb = get_vtb();
 176         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 177                 vcpu->arch.entry_ic = mfspr(SPRN_IC);
 178         svcpu->in_use = true;
 179 }
 180
 181 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
 182 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu,
 183                             struct kvmppc_book3s_shadow_vcpu *svcpu)
 184 {
 185         /*
 186          * vcpu_put would just call us again because in_use hasn't
 187          * been updated yet.
 188          */
 189         preempt_disable();
 190
 191         /*
 192          * Maybe we were already preempted and synced the svcpu from
 193          * our preempt notifiers. Don't bother touching this svcpu then.
 194          */
 195         if (!svcpu->in_use)
 196                 goto out;
 197
 198         vcpu->arch.gpr[0] = svcpu->gpr[0];
 199         vcpu->arch.gpr[1] = svcpu->gpr[1];
 200         vcpu->arch.gpr[2] = svcpu->gpr[2];
 201         vcpu->arch.gpr[3] = svcpu->gpr[3];
 202         vcpu->arch.gpr[4] = svcpu->gpr[4];
 203         vcpu->arch.gpr[5] = svcpu->gpr[5];
 204         vcpu->arch.gpr[6] = svcpu->gpr[6];
 205         vcpu->arch.gpr[7] = svcpu->gpr[7];
 206         vcpu->arch.gpr[8] = svcpu->gpr[8];
 207         vcpu->arch.gpr[9] = svcpu->gpr[9];
 208         vcpu->arch.gpr[10] = svcpu->gpr[10];
 209         vcpu->arch.gpr[11] = svcpu->gpr[11];
 210         vcpu->arch.gpr[12] = svcpu->gpr[12];
 211         vcpu->arch.gpr[13] = svcpu->gpr[13];
 212         vcpu->arch.cr  = svcpu->cr;
 213         vcpu->arch.xer = svcpu->xer;
 214         vcpu->arch.ctr = svcpu->ctr;
 215         vcpu->arch.lr  = svcpu->lr;
 216         vcpu->arch.pc  = svcpu->pc;
 217         vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
 218         vcpu->arch.fault_dar   = svcpu->fault_dar;
 219         vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
 220         vcpu->arch.last_inst   = svcpu->last_inst;
 221 #ifdef CONFIG_PPC_BOOK3S_64
 222         vcpu->arch.shadow_fscr = svcpu->shadow_fscr;
 223 #endif
 224         /*
 225          * Update purr and spurr using time base on exit.
 226          */
 227         vcpu->arch.purr += get_tb() - vcpu->arch.entry_tb;
 228         vcpu->arch.spurr += get_tb() - vcpu->arch.entry_tb;
 229         vcpu->arch.vtb += get_vtb() - vcpu->arch.entry_vtb;
 230         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 231                 vcpu->arch.ic += mfspr(SPRN_IC) - vcpu->arch.entry_ic;
 232         svcpu->in_use = false;
 233
 234 out:
 235         preempt_enable();
 236 }
 237
 238 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
 239 {
 240         int r = 1; /* Indicate we want to get back into the guest */
 241
 242         /* We misuse TLB_FLUSH to indicate that we want to clear
 243            all shadow cache entries */
 244         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
 245                 kvmppc_mmu_pte_flush(vcpu, 0, 0);
 246
 247         return r;
 248 }
 249
 250 /************* MMU Notifiers *************/
 251 static void do_kvm_unmap_hva(struct kvm *kvm, unsigned long start,
 252                              unsigned long end)
 253 {
 254         long i;
 255         struct kvm_vcpu *vcpu;
 256         struct kvm_memslots *slots;
 257         struct kvm_memory_slot *memslot;
 258
 259         slots = kvm_memslots(kvm);
 260         kvm_for_each_memslot(memslot, slots) {
 261                 unsigned long hva_start, hva_end;
 262                 gfn_t gfn, gfn_end;
 263
 264                 hva_start = max(start, memslot->userspace_addr);
 265                 hva_end = min(end, memslot->userspace_addr +
 266                                         (memslot->npages << PAGE_SHIFT));
 267                 if (hva_start >= hva_end)
 268                         continue;
 269                 /*
 270                  * {gfn(page) | page intersects with [hva_start, hva_end)} =
 271                  * {gfn, gfn+1, ..., gfn_end-1}.
 272                  */
 273                 gfn = hva_to_gfn_memslot(hva_start, memslot);
 274                 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
 275                 kvm_for_each_vcpu(i, vcpu, kvm)
 276                         kvmppc_mmu_pte_pflush(vcpu, gfn << PAGE_SHIFT,
 277                                               gfn_end << PAGE_SHIFT);
 278         }
 279 }
 280
 281 static int kvm_unmap_hva_pr(struct kvm *kvm, unsigned long hva)
 282 {
 283         trace_kvm_unmap_hva(hva);
 284
 285         do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
 286
 287         return 0;
 288 }
 289
 290 static int kvm_unmap_hva_range_pr(struct kvm *kvm, unsigned long start,
 291                                   unsigned long end)
 292 {
 293         do_kvm_unmap_hva(kvm, start, end);
 294
 295         return 0;
 296 }
 297
 298 static int kvm_age_hva_pr(struct kvm *kvm, unsigned long hva)
 299 {
 300         /* XXX could be more clever ;) */
 301         return 0;
 302 }
 303
 304 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
 305 {
 306         /* XXX could be more clever ;) */
 307         return 0;
 308 }
 309
 310 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
 311 {
 312         /* The page will get remapped properly on its next fault */
 313         do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
 314 }
 315
 316 /*****************************************/
 317
 318 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 319 {
 320         ulong guest_msr = kvmppc_get_msr(vcpu);
 321         ulong smsr = guest_msr;
 322
 323         /* Guest MSR values */
 324         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE;
 325         /* Process MSR values */
 326         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 327         /* External providers the guest reserved */
 328         smsr |= (guest_msr & vcpu->arch.guest_owned_ext);
 329         /* 64-bit Process MSR values */
 330 #ifdef CONFIG_PPC_BOOK3S_64
 331         smsr |= MSR_ISF | MSR_HV;
 332 #endif
 333         vcpu->arch.shadow_msr = smsr;
 334 }
 335
 336 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
 337 {
 338         ulong old_msr = kvmppc_get_msr(vcpu);
 339
 340 #ifdef EXIT_DEBUG
 341         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 342 #endif
 343
 344         msr &= to_book3s(vcpu)->msr_mask;
 345         kvmppc_set_msr_fast(vcpu, msr);
 346         kvmppc_recalc_shadow_msr(vcpu);
 347
 348         if (msr & MSR_POW) {
 349                 if (!vcpu->arch.pending_exceptions) {
 350                         kvm_vcpu_block(vcpu);
 351                         clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 352                         vcpu->stat.halt_wakeup++;
 353
 354                         /* Unset POW bit after we woke up */
 355                         msr &= ~MSR_POW;
 356                         kvmppc_set_msr_fast(vcpu, msr);
 357                 }
 358         }
 359
 360         if (kvmppc_is_split_real(vcpu))
 361                 kvmppc_fixup_split_real(vcpu);
 362         else
 363                 kvmppc_unfixup_split_real(vcpu);
 364
 365         if ((kvmppc_get_msr(vcpu) & (MSR_PR|MSR_IR|MSR_DR)) !=
 366                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 367                 kvmppc_mmu_flush_segments(vcpu);
 368                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 369
 370                 /* Preload magic page segment when in kernel mode */
 371                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 372                         struct kvm_vcpu_arch *a = &vcpu->arch;
 373
 374                         if (msr & MSR_DR)
 375                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 376                         else
 377                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 378                 }
 379         }
 380
 381         /*
 382          * When switching from 32 to 64-bit, we may have a stale 32-bit
 383          * magic page around, we need to flush it. Typically 32-bit magic
 384          * page will be instanciated when calling into RTAS. Note: We
 385          * assume that such transition only happens while in kernel mode,
 386          * ie, we never transition from user 32-bit to kernel 64-bit with
 387          * a 32-bit magic page around.
 388          */
 389         if (vcpu->arch.magic_page_pa &&
 390             !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
 391                 /* going from RTAS to normal kernel code */
 392                 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
 393                                      ~0xFFFUL);
 394         }
 395
 396         /* Preload FPU if it's enabled */
 397         if (kvmppc_get_msr(vcpu) & MSR_FP)
 398                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 399 }
 400
 401 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
 402 {
 403         u32 host_pvr;
 404
 405         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 406         vcpu->arch.pvr = pvr;
 407 #ifdef CONFIG_PPC_BOOK3S_64
 408         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 409                 kvmppc_mmu_book3s_64_init(vcpu);
 410                 if (!to_book3s(vcpu)->hior_explicit)
 411                         to_book3s(vcpu)->hior = 0xfff00000;
 412                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 413                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 414         } else
 415 #endif
 416         {
 417                 kvmppc_mmu_book3s_32_init(vcpu);
 418                 if (!to_book3s(vcpu)->hior_explicit)
 419                         to_book3s(vcpu)->hior = 0;
 420                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 421                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 422         }
 423
 424         kvmppc_sanity_check(vcpu);
 425
 426         /* If we are in hypervisor level on 970, we can tell the CPU to
 427          * treat DCBZ as 32 bytes store */
 428         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 429         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 430             !strcmp(cur_cpu_spec->platform, "ppc970"))
 431                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 432
 433         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 434            really needs them in a VM on Cell and force disable them. */
 435         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 436                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 437
 438         /*
 439          * If they're asking for POWER6 or later, set the flag
 440          * indicating that we can do multiple large page sizes
 441          * and 1TB segments.
 442          * Also set the flag that indicates that tlbie has the large
 443          * page bit in the RB operand instead of the instruction.
 444          */
 445         switch (PVR_VER(pvr)) {
 446         case PVR_POWER6:
 447         case PVR_POWER7:
 448         case PVR_POWER7p:
 449         case PVR_POWER8:
 450                 vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
 451                         BOOK3S_HFLAG_NEW_TLBIE;
 452                 break;
 453         }
 454
 455 #ifdef CONFIG_PPC_BOOK3S_32
 456         /* 32 bit Book3S always has 32 byte dcbz */
 457         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 458 #endif
 459
 460         /* On some CPUs we can execute paired single operations natively */
 461         asm ( "mfpvr %0" : "=r"(host_pvr));
 462         switch (host_pvr) {
 463         case 0x00080200:        /* lonestar 2.0 */
 464         case 0x00088202:        /* lonestar 2.2 */
 465         case 0x70000100:        /* gekko 1.0 */
 466         case 0x00080100:        /* gekko 2.0 */
 467         case 0x00083203:        /* gekko 2.3a */
 468         case 0x00083213:        /* gekko 2.3b */
 469         case 0x00083204:        /* gekko 2.4 */
 470         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 471         case 0x00087200:        /* broadway */
 472                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 473                 /* Enable HID2.PSE - in case we need it later */
 474                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 475         }
 476 }
 477
 478 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 479  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 480  * emulate 32 bytes dcbz length.
 481  *
 482  * The Book3s_64 inventors also realized this case and implemented a special bit
 483  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 484  *
 485  * My approach here is to patch the dcbz instruction on executing pages.
 486  */
 487 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 488 {
 489         struct page *hpage;
 490         u64 hpage_offset;
 491         u32 *page;
 492         int i;
 493
 494         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 495         if (is_error_page(hpage))
 496                 return;
 497
 498         hpage_offset = pte->raddr & ~PAGE_MASK;
 499         hpage_offset &= ~0xFFFULL;
 500         hpage_offset /= 4;
 501
 502         get_page(hpage);
 503         page = kmap_atomic(hpage);
 504
 505         /* patch dcbz into reserved instruction, so we trap */
 506         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 507                 if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
 508                         page[i] &= cpu_to_be32(0xfffffff7);
 509
 510         kunmap_atomic(page);
 511         put_page(hpage);
 512 }
 513
 514 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 515 {
 516         ulong mp_pa = vcpu->arch.magic_page_pa;
 517
 518         if (!(kvmppc_get_msr(vcpu) & MSR_SF))
 519                 mp_pa = (uint32_t)mp_pa;
 520
 521         if (unlikely(mp_pa) &&
 522             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 523                 return 1;
 524         }
 525
 526         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 527 }
 528
 529 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 530                             ulong eaddr, int vec)
 531 {
 532         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 533         bool iswrite = false;
 534         int r = RESUME_GUEST;
 535         int relocated;
 536         int page_found = 0;
 537         struct kvmppc_pte pte;
 538         bool is_mmio = false;
 539         bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
 540         bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
 541         u64 vsid;
 542
 543         relocated = data ? dr : ir;
 544         if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
 545                 iswrite = true;
 546
 547         /* Resolve real address if translation turned on */
 548         if (relocated) {
 549                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
 550         } else {
 551                 pte.may_execute = true;
 552                 pte.may_read = true;
 553                 pte.may_write = true;
 554                 pte.raddr = eaddr & KVM_PAM;
 555                 pte.eaddr = eaddr;
 556                 pte.vpage = eaddr >> 12;
 557                 pte.page_size = MMU_PAGE_64K;
 558         }
 559
 560         switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
 561         case 0:
 562                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 563                 break;
 564         case MSR_DR:
 565                 if (!data &&
 566                     (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
 567                     ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
 568                         pte.raddr &= ~SPLIT_HACK_MASK;
 569                 /* fall through */
 570         case MSR_IR:
 571                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 572
 573                 if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
 574                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 575                 else
 576                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 577                 pte.vpage |= vsid;
 578
 579                 if (vsid == -1)
 580                         page_found = -EINVAL;
 581                 break;
 582         }
 583
 584         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 585            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 586                 /*
 587                  * If we do the dcbz hack, we have to NX on every execution,
 588                  * so we can patch the executing code. This renders our guest
 589                  * NX-less.
 590                  */
 591                 pte.may_execute = !data;
 592         }
 593
 594         if (page_found == -ENOENT) {
 595                 /* Page not found in guest PTE entries */
 596                 u64 ssrr1 = vcpu->arch.shadow_srr1;
 597                 u64 msr = kvmppc_get_msr(vcpu);
 598                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 599                 kvmppc_set_dsisr(vcpu, vcpu->arch.fault_dsisr);
 600                 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
 601                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 602         } else if (page_found == -EPERM) {
 603                 /* Storage protection */
 604                 u32 dsisr = vcpu->arch.fault_dsisr;
 605                 u64 ssrr1 = vcpu->arch.shadow_srr1;
 606                 u64 msr = kvmppc_get_msr(vcpu);
 607                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 608                 dsisr = (dsisr & ~DSISR_NOHPTE) | DSISR_PROTFAULT;
 609                 kvmppc_set_dsisr(vcpu, dsisr);
 610                 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
 611                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 612         } else if (page_found == -EINVAL) {
 613                 /* Page not found in guest SLB */
 614                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 615                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 616         } else if (!is_mmio &&
 617                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 618                 if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
 619                         /*
 620                          * There is already a host HPTE there, presumably
 621                          * a read-only one for a page the guest thinks
 622                          * is writable, so get rid of it first.
 623                          */
 624                         kvmppc_mmu_unmap_page(vcpu, &pte);
 625                 }
 626                 /* The guest's PTE is not mapped yet. Map on the host */
 627                 kvmppc_mmu_map_page(vcpu, &pte, iswrite);
 628                 if (data)
 629                         vcpu->stat.sp_storage++;
 630                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 631                          (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 632                         kvmppc_patch_dcbz(vcpu, &pte);
 633         } else {
 634                 /* MMIO */
 635                 vcpu->stat.mmio_exits++;
 636                 vcpu->arch.paddr_accessed = pte.raddr;
 637                 vcpu->arch.vaddr_accessed = pte.eaddr;
 638                 r = kvmppc_emulate_mmio(run, vcpu);
 639                 if ( r == RESUME_HOST_NV )
 640                         r = RESUME_HOST;
 641         }
 642
 643         return r;
 644 }
 645
 646 static inline int get_fpr_index(int i)
 647 {
 648         return i * TS_FPRWIDTH;
 649 }
 650
 651 /* Give up external provider (FPU, Altivec, VSX) */
 652 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 653 {
 654         struct thread_struct *t = &current->thread;
 655
 656         /*
 657          * VSX instructions can access FP and vector registers, so if
 658          * we are giving up VSX, make sure we give up FP and VMX as well.
 659          */
 660         if (msr & MSR_VSX)
 661                 msr |= MSR_FP | MSR_VEC;
 662
 663         msr &= vcpu->arch.guest_owned_ext;
 664         if (!msr)
 665                 return;
 666
 667 #ifdef DEBUG_EXT
 668         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 669 #endif
 670
 671         if (msr & MSR_FP) {
 672                 /*
 673                  * Note that on CPUs with VSX, giveup_fpu stores
 674                  * both the traditional FP registers and the added VSX
 675                  * registers into thread.fp_state.fpr[].
 676                  */
 677                 if (t->regs->msr & MSR_FP)
 678                         giveup_fpu(current);
 679                 t->fp_save_area = NULL;
 680         }
 681
 682 #ifdef CONFIG_ALTIVEC
 683         if (msr & MSR_VEC) {
 684                 if (current->thread.regs->msr & MSR_VEC)
 685                         giveup_altivec(current);
 686                 t->vr_save_area = NULL;
 687         }
 688 #endif
 689
 690         vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
 691         kvmppc_recalc_shadow_msr(vcpu);
 692 }
 693
 694 /* Give up facility (TAR / EBB / DSCR) */
 695 static void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
 696 {
 697 #ifdef CONFIG_PPC_BOOK3S_64
 698         if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
 699                 /* Facility not available to the guest, ignore giveup request*/
 700                 return;
 701         }
 702
 703         switch (fac) {
 704         case FSCR_TAR_LG:
 705                 vcpu->arch.tar = mfspr(SPRN_TAR);
 706                 mtspr(SPRN_TAR, current->thread.tar);
 707                 vcpu->arch.shadow_fscr &= ~FSCR_TAR;
 708                 break;
 709         }
 710 #endif
 711 }
 712
 713 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 714 {
 715         ulong srr0 = kvmppc_get_pc(vcpu);
 716         u32 last_inst = kvmppc_get_last_inst(vcpu);
 717         int ret;
 718
 719         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 720         if (ret == -ENOENT) {
 721                 ulong msr = kvmppc_get_msr(vcpu);
 722
 723                 msr = kvmppc_set_field(msr, 33, 33, 1);
 724                 msr = kvmppc_set_field(msr, 34, 36, 0);
 725                 msr = kvmppc_set_field(msr, 42, 47, 0);
 726                 kvmppc_set_msr_fast(vcpu, msr);
 727                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 728                 return EMULATE_AGAIN;
 729         }
 730
 731         return EMULATE_DONE;
 732 }
 733
 734 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 735 {
 736
 737         /* Need to do paired single emulation? */
 738         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 739                 return EMULATE_DONE;
 740
 741         /* Read out the instruction */
 742         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 743                 /* Need to emulate */
 744                 return EMULATE_FAIL;
 745
 746         return EMULATE_AGAIN;
 747 }
 748
 749 /* Handle external providers (FPU, Altivec, VSX) */
 750 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 751                              ulong msr)
 752 {
 753         struct thread_struct *t = &current->thread;
 754
 755         /* When we have paired singles, we emulate in software */
 756         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 757                 return RESUME_GUEST;
 758
 759         if (!(kvmppc_get_msr(vcpu) & msr)) {
 760                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 761                 return RESUME_GUEST;
 762         }
 763
 764         if (msr == MSR_VSX) {
 765                 /* No VSX?  Give an illegal instruction interrupt */
 766 #ifdef CONFIG_VSX
 767                 if (!cpu_has_feature(CPU_FTR_VSX))
 768 #endif
 769                 {
 770                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 771                         return RESUME_GUEST;
 772                 }
 773
 774                 /*
 775                  * We have to load up all the FP and VMX registers before
 776                  * we can let the guest use VSX instructions.
 777                  */
 778                 msr = MSR_FP | MSR_VEC | MSR_VSX;
 779         }
 780
 781         /* See if we already own all the ext(s) needed */
 782         msr &= ~vcpu->arch.guest_owned_ext;
 783         if (!msr)
 784                 return RESUME_GUEST;
 785
 786 #ifdef DEBUG_EXT
 787         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 788 #endif
 789
 790         if (msr & MSR_FP) {
 791                 preempt_disable();
 792                 enable_kernel_fp();
 793                 load_fp_state(&vcpu->arch.fp);
 794                 t->fp_save_area = &vcpu->arch.fp;
 795                 preempt_enable();
 796         }
 797
 798         if (msr & MSR_VEC) {
 799 #ifdef CONFIG_ALTIVEC
 800                 preempt_disable();
 801                 enable_kernel_altivec();
 802                 load_vr_state(&vcpu->arch.vr);
 803                 t->vr_save_area = &vcpu->arch.vr;
 804                 preempt_enable();
 805 #endif
 806         }
 807
 808         t->regs->msr |= msr;
 809         vcpu->arch.guest_owned_ext |= msr;
 810         kvmppc_recalc_shadow_msr(vcpu);
 811
 812         return RESUME_GUEST;
 813 }
 814
 815 /*
 816  * Kernel code using FP or VMX could have flushed guest state to
 817  * the thread_struct; if so, get it back now.
 818  */
 819 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
 820 {
 821         unsigned long lost_ext;
 822
 823         lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
 824         if (!lost_ext)
 825                 return;
 826
 827         if (lost_ext & MSR_FP) {
 828                 preempt_disable();
 829                 enable_kernel_fp();
 830                 load_fp_state(&vcpu->arch.fp);
 831                 preempt_enable();
 832         }
 833 #ifdef CONFIG_ALTIVEC
 834         if (lost_ext & MSR_VEC) {
 835                 preempt_disable();
 836                 enable_kernel_altivec();
 837                 load_vr_state(&vcpu->arch.vr);
 838                 preempt_enable();
 839         }
 840 #endif
 841         current->thread.regs->msr |= lost_ext;
 842 }
 843
 844 #ifdef CONFIG_PPC_BOOK3S_64
 845
 846 static void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
 847 {
 848         /* Inject the Interrupt Cause field and trigger a guest interrupt */
 849         vcpu->arch.fscr &= ~(0xffULL << 56);
 850         vcpu->arch.fscr |= (fac << 56);
 851         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
 852 }
 853
 854 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
 855 {
 856         enum emulation_result er = EMULATE_FAIL;
 857
 858         if (!(kvmppc_get_msr(vcpu) & MSR_PR))
 859                 er = kvmppc_emulate_instruction(vcpu->run, vcpu);
 860
 861         if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
 862                 /* Couldn't emulate, trigger interrupt in guest */
 863                 kvmppc_trigger_fac_interrupt(vcpu, fac);
 864         }
 865 }
 866
 867 /* Enable facilities (TAR, EBB, DSCR) for the guest */
 868 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
 869 {
 870         bool guest_fac_enabled;
 871         BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
 872
 873         /*
 874          * Not every facility is enabled by FSCR bits, check whether the
 875          * guest has this facility enabled at all.
 876          */
 877         switch (fac) {
 878         case FSCR_TAR_LG:
 879         case FSCR_EBB_LG:
 880                 guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
 881                 break;
 882         case FSCR_TM_LG:
 883                 guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
 884                 break;
 885         default:
 886                 guest_fac_enabled = false;
 887                 break;
 888         }
 889
 890         if (!guest_fac_enabled) {
 891                 /* Facility not enabled by the guest */
 892                 kvmppc_trigger_fac_interrupt(vcpu, fac);
 893                 return RESUME_GUEST;
 894         }
 895
 896         switch (fac) {
 897         case FSCR_TAR_LG:
 898                 /* TAR switching isn't lazy in Linux yet */
 899                 current->thread.tar = mfspr(SPRN_TAR);
 900                 mtspr(SPRN_TAR, vcpu->arch.tar);
 901                 vcpu->arch.shadow_fscr |= FSCR_TAR;
 902                 break;
 903         default:
 904                 kvmppc_emulate_fac(vcpu, fac);
 905                 break;
 906         }
 907
 908         return RESUME_GUEST;
 909 }
 910 #endif
 911
 912 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
 913                           unsigned int exit_nr)
 914 {
 915         int r = RESUME_HOST;
 916         int s;
 917
 918         vcpu->stat.sum_exits++;
 919
 920         run->exit_reason = KVM_EXIT_UNKNOWN;
 921         run->ready_for_interrupt_injection = 1;
 922
 923         /* We get here with MSR.EE=1 */
 924
 925         trace_kvm_exit(exit_nr, vcpu);
 926         kvm_guest_exit();
 927
 928         switch (exit_nr) {
 929         case BOOK3S_INTERRUPT_INST_STORAGE:
 930         {
 931                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 932                 vcpu->stat.pf_instruc++;
 933
 934                 if (kvmppc_is_split_real(vcpu))
 935                         kvmppc_fixup_split_real(vcpu);
 936
 937 #ifdef CONFIG_PPC_BOOK3S_32
 938                 /* We set segments as unused segments when invalidating them. So
 939                  * treat the respective fault as segment fault. */
 940                 {
 941                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 942                         u32 sr;
 943
 944                         svcpu = svcpu_get(vcpu);
 945                         sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
 946                         svcpu_put(svcpu);
 947                         if (sr == SR_INVALID) {
 948                                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 949                                 r = RESUME_GUEST;
 950                                 break;
 951                         }
 952                 }
 953 #endif
 954
 955                 /* only care about PTEG not found errors, but leave NX alone */
 956                 if (shadow_srr1 & 0x40000000) {
 957                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
 958                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 959                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 960                         vcpu->stat.sp_instruc++;
 961                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 962                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 963                         /*
 964                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 965                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 966                          *     that no guest that needs the dcbz hack does NX.
 967                          */
 968                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 969                         r = RESUME_GUEST;
 970                 } else {
 971                         u64 msr = kvmppc_get_msr(vcpu);
 972                         msr |= shadow_srr1 & 0x58000000;
 973                         kvmppc_set_msr_fast(vcpu, msr);
 974                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 975                         r = RESUME_GUEST;
 976                 }
 977                 break;
 978         }
 979         case BOOK3S_INTERRUPT_DATA_STORAGE:
 980         {
 981                 ulong dar = kvmppc_get_fault_dar(vcpu);
 982                 u32 fault_dsisr = vcpu->arch.fault_dsisr;
 983                 vcpu->stat.pf_storage++;
 984
 985 #ifdef CONFIG_PPC_BOOK3S_32
 986                 /* We set segments as unused segments when invalidating them. So
 987                  * treat the respective fault as segment fault. */
 988                 {
 989                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 990                         u32 sr;
 991
 992                         svcpu = svcpu_get(vcpu);
 993                         sr = svcpu->sr[dar >> SID_SHIFT];
 994                         svcpu_put(svcpu);
 995                         if (sr == SR_INVALID) {
 996                                 kvmppc_mmu_map_segment(vcpu, dar);
 997                                 r = RESUME_GUEST;
 998                                 break;
 999                         }
1000                 }
1001 #endif
1002
1003                 /*
1004                  * We need to handle missing shadow PTEs, and
1005                  * protection faults due to us mapping a page read-only
1006                  * when the guest thinks it is writable.
1007                  */
1008                 if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
1009                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
1010                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
1011                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1012                 } else {
1013                         kvmppc_set_dar(vcpu, dar);
1014                         kvmppc_set_dsisr(vcpu, fault_dsisr);
1015                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1016                         r = RESUME_GUEST;
1017                 }
1018                 break;
1019         }
1020         case BOOK3S_INTERRUPT_DATA_SEGMENT:
1021                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
1022                         kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
1023                         kvmppc_book3s_queue_irqprio(vcpu,
1024                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
1025                 }
1026                 r = RESUME_GUEST;
1027                 break;
1028         case BOOK3S_INTERRUPT_INST_SEGMENT:
1029                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1030                         kvmppc_book3s_queue_irqprio(vcpu,
1031                                 BOOK3S_INTERRUPT_INST_SEGMENT);
1032                 }
1033                 r = RESUME_GUEST;
1034                 break;
1035         /* We're good on these - the host merely wanted to get our attention */
1036         case BOOK3S_INTERRUPT_DECREMENTER:
1037         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1038         case BOOK3S_INTERRUPT_DOORBELL:
1039         case BOOK3S_INTERRUPT_H_DOORBELL:
1040                 vcpu->stat.dec_exits++;
1041                 r = RESUME_GUEST;
1042                 break;
1043         case BOOK3S_INTERRUPT_EXTERNAL:
1044         case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
1045         case BOOK3S_INTERRUPT_EXTERNAL_HV:
1046                 vcpu->stat.ext_intr_exits++;
1047                 r = RESUME_GUEST;
1048                 break;
1049         case BOOK3S_INTERRUPT_PERFMON:
1050                 r = RESUME_GUEST;
1051                 break;
1052         case BOOK3S_INTERRUPT_PROGRAM:
1053         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1054         {
1055                 enum emulation_result er;
1056                 ulong flags;
1057
1058 program_interrupt:
1059                 flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
1060
1061                 if (kvmppc_get_msr(vcpu) & MSR_PR) {
1062 #ifdef EXIT_DEBUG
1063                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
1064 #endif
1065                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
1066                             (INS_DCBZ & 0xfffffff7)) {
1067                                 kvmppc_core_queue_program(vcpu, flags);
1068                                 r = RESUME_GUEST;
1069                                 break;
1070                         }
1071                 }
1072
1073                 vcpu->stat.emulated_inst_exits++;
1074                 er = kvmppc_emulate_instruction(run, vcpu);
1075                 switch (er) {
1076                 case EMULATE_DONE:
1077                         r = RESUME_GUEST_NV;
1078                         break;
1079                 case EMULATE_AGAIN:
1080                         r = RESUME_GUEST;
1081                         break;
1082                 case EMULATE_FAIL:
1083                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
1084                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
1085                         kvmppc_core_queue_program(vcpu, flags);
1086                         r = RESUME_GUEST;
1087                         break;
1088                 case EMULATE_DO_MMIO:
1089                         run->exit_reason = KVM_EXIT_MMIO;
1090                         r = RESUME_HOST_NV;
1091                         break;
1092                 case EMULATE_EXIT_USER:
1093                         r = RESUME_HOST_NV;
1094                         break;
1095                 default:
1096                         BUG();
1097                 }
1098                 break;
1099         }
1100         case BOOK3S_INTERRUPT_SYSCALL:
1101                 if (vcpu->arch.papr_enabled &&
1102                     (kvmppc_get_last_sc(vcpu) == 0x44000022) &&
1103                     !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1104                         /* SC 1 papr hypercalls */
1105                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
1106                         int i;
1107
1108 #ifdef CONFIG_PPC_BOOK3S_64
1109                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1110                                 r = RESUME_GUEST;
1111                                 break;
1112                         }
1113 #endif
1114
1115                         run->papr_hcall.nr = cmd;
1116                         for (i = 0; i < 9; ++i) {
1117                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1118                                 run->papr_hcall.args[i] = gpr;
1119                         }
1120                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
1121                         vcpu->arch.hcall_needed = 1;
1122                         r = RESUME_HOST;
1123                 } else if (vcpu->arch.osi_enabled &&
1124                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1125                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1126                         /* MOL hypercalls */
1127                         u64 *gprs = run->osi.gprs;
1128                         int i;
1129
1130                         run->exit_reason = KVM_EXIT_OSI;
1131                         for (i = 0; i < 32; i++)
1132                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
1133                         vcpu->arch.osi_needed = 1;
1134                         r = RESUME_HOST_NV;
1135                 } else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1136                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1137                         /* KVM PV hypercalls */
1138                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1139                         r = RESUME_GUEST;
1140                 } else {
1141                         /* Guest syscalls */
1142                         vcpu->stat.syscall_exits++;
1143                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1144                         r = RESUME_GUEST;
1145                 }
1146                 break;
1147         case BOOK3S_INTERRUPT_FP_UNAVAIL:
1148         case BOOK3S_INTERRUPT_ALTIVEC:
1149         case BOOK3S_INTERRUPT_VSX:
1150         {
1151                 int ext_msr = 0;
1152
1153                 switch (exit_nr) {
1154                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
1155                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
1156                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
1157                 }
1158
1159                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
1160                 case EMULATE_DONE:
1161                         /* everything ok - let's enable the ext */
1162                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1163                         break;
1164                 case EMULATE_FAIL:
1165                         /* we need to emulate this instruction */
1166                         goto program_interrupt;
1167                         break;
1168                 default:
1169                         /* nothing to worry about - go again */
1170                         break;
1171                 }
1172                 break;
1173         }
1174         case BOOK3S_INTERRUPT_ALIGNMENT:
1175                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
1176                         u32 last_inst = kvmppc_get_last_inst(vcpu);
1177                         u32 dsisr;
1178                         u64 dar;
1179
1180                         dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1181                         dar = kvmppc_alignment_dar(vcpu, last_inst);
1182
1183                         kvmppc_set_dsisr(vcpu, dsisr);
1184                         kvmppc_set_dar(vcpu, dar);
1185
1186                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1187                 }
1188                 r = RESUME_GUEST;
1189                 break;
1190 #ifdef CONFIG_PPC_BOOK3S_64
1191         case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1192                 kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1193                 r = RESUME_GUEST;
1194                 break;
1195 #endif
1196         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1197         case BOOK3S_INTERRUPT_TRACE:
1198                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1199                 r = RESUME_GUEST;
1200                 break;
1201         default:
1202         {
1203                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1204                 /* Ugh - bork here! What did we get? */
1205                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1206                         exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1207                 r = RESUME_HOST;
1208                 BUG();
1209                 break;
1210         }
1211         }
1212
1213         if (!(r & RESUME_HOST)) {
1214                 /* To avoid clobbering exit_reason, only check for signals if
1215                  * we aren't already exiting to userspace for some other
1216                  * reason. */
1217
1218                 /*
1219                  * Interrupts could be timers for the guest which we have to
1220                  * inject again, so let's postpone them until we're in the guest
1221                  * and if we really did time things so badly, then we just exit
1222                  * again due to a host external interrupt.
1223                  */
1224                 s = kvmppc_prepare_to_enter(vcpu);
1225                 if (s <= 0)
1226                         r = s;
1227                 else {
1228                         /* interrupts now hard-disabled */
1229                         kvmppc_fix_ee_before_entry();
1230                 }
1231
1232                 kvmppc_handle_lost_ext(vcpu);
1233         }
1234
1235         trace_kvm_book3s_reenter(r, vcpu);
1236
1237         return r;
1238 }
1239
1240 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1241                                             struct kvm_sregs *sregs)
1242 {
1243         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1244         int i;
1245
1246         sregs->pvr = vcpu->arch.pvr;
1247
1248         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1249         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1250                 for (i = 0; i < 64; i++) {
1251                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1252                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1253                 }
1254         } else {
1255                 for (i = 0; i < 16; i++)
1256                         sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1257
1258                 for (i = 0; i < 8; i++) {
1259                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1260                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1261                 }
1262         }
1263
1264         return 0;
1265 }
1266
1267 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1268                                             struct kvm_sregs *sregs)
1269 {
1270         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1271         int i;
1272
1273         kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1274
1275         vcpu3s->sdr1 = sregs->u.s.sdr1;
1276         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1277                 for (i = 0; i < 64; i++) {
1278                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1279                                                     sregs->u.s.ppc64.slb[i].slbe);
1280                 }
1281         } else {
1282                 for (i = 0; i < 16; i++) {
1283                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1284                 }
1285                 for (i = 0; i < 8; i++) {
1286                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1287                                        (u32)sregs->u.s.ppc32.ibat[i]);
1288                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1289                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1290                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1291                                        (u32)sregs->u.s.ppc32.dbat[i]);
1292                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1293                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1294                 }
1295         }
1296
1297         /* Flush the MMU after messing with the segments */
1298         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1299
1300         return 0;
1301 }
1302
1303 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1304                                  union kvmppc_one_reg *val)
1305 {
1306         int r = 0;
1307
1308         switch (id) {
1309         case KVM_REG_PPC_HIOR:
1310                 *val = get_reg_val(id, to_book3s(vcpu)->hior);
1311                 break;
1312         case KVM_REG_PPC_LPCR:
1313                 /*
1314                  * We are only interested in the LPCR_ILE bit
1315                  */
1316                 if (vcpu->arch.intr_msr & MSR_LE)
1317                         *val = get_reg_val(id, LPCR_ILE);
1318                 else
1319                         *val = get_reg_val(id, 0);
1320                 break;
1321         default:
1322                 r = -EINVAL;
1323                 break;
1324         }
1325
1326         return r;
1327 }
1328
1329 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1330 {
1331         if (new_lpcr & LPCR_ILE)
1332                 vcpu->arch.intr_msr |= MSR_LE;
1333         else
1334                 vcpu->arch.intr_msr &= ~MSR_LE;
1335 }
1336
1337 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1338                                  union kvmppc_one_reg *val)
1339 {
1340         int r = 0;
1341
1342         switch (id) {
1343         case KVM_REG_PPC_HIOR:
1344                 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1345                 to_book3s(vcpu)->hior_explicit = true;
1346                 break;
1347         case KVM_REG_PPC_LPCR:
1348                 kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1349                 break;
1350         default:
1351                 r = -EINVAL;
1352                 break;
1353         }
1354
1355         return r;
1356 }
1357
1358 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1359                                                    unsigned int id)
1360 {
1361         struct kvmppc_vcpu_book3s *vcpu_book3s;
1362         struct kvm_vcpu *vcpu;
1363         int err = -ENOMEM;
1364         unsigned long p;
1365
1366         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1367         if (!vcpu)
1368                 goto out;
1369
1370         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1371         if (!vcpu_book3s)
1372                 goto free_vcpu;
1373         vcpu->arch.book3s = vcpu_book3s;
1374
1375 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1376         vcpu->arch.shadow_vcpu =
1377                 kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1378         if (!vcpu->arch.shadow_vcpu)
1379                 goto free_vcpu3s;
1380 #endif
1381
1382         err = kvm_vcpu_init(vcpu, kvm, id);
1383         if (err)
1384                 goto free_shadow_vcpu;
1385
1386         err = -ENOMEM;
1387         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1388         if (!p)
1389                 goto uninit_vcpu;
1390         /* the real shared page fills the last 4k of our page */
1391         vcpu->arch.shared = (void *)(p + PAGE_SIZE - 4096);
1392 #ifdef CONFIG_PPC_BOOK3S_64
1393         /* Always start the shared struct in native endian mode */
1394 #ifdef __BIG_ENDIAN__
1395         vcpu->arch.shared_big_endian = true;
1396 #else
1397         vcpu->arch.shared_big_endian = false;
1398 #endif
1399
1400         /*
1401          * Default to the same as the host if we're on sufficiently
1402          * recent machine that we have 1TB segments;
1403          * otherwise default to PPC970FX.
1404          */
1405         vcpu->arch.pvr = 0x3C0301;
1406         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1407                 vcpu->arch.pvr = mfspr(SPRN_PVR);
1408         vcpu->arch.intr_msr = MSR_SF;
1409 #else
1410         /* default to book3s_32 (750) */
1411         vcpu->arch.pvr = 0x84202;
1412 #endif
1413         kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1414         vcpu->arch.slb_nr = 64;
1415
1416         vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1417
1418         err = kvmppc_mmu_init(vcpu);
1419         if (err < 0)
1420                 goto uninit_vcpu;
1421
1422         return vcpu;
1423
1424 uninit_vcpu:
1425         kvm_vcpu_uninit(vcpu);
1426 free_shadow_vcpu:
1427 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1428         kfree(vcpu->arch.shadow_vcpu);
1429 free_vcpu3s:
1430 #endif
1431         vfree(vcpu_book3s);
1432 free_vcpu:
1433         kmem_cache_free(kvm_vcpu_cache, vcpu);
1434 out:
1435         return ERR_PTR(err);
1436 }
1437
1438 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1439 {
1440         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1441
1442         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1443         kvm_vcpu_uninit(vcpu);
1444 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1445         kfree(vcpu->arch.shadow_vcpu);
1446 #endif
1447         vfree(vcpu_book3s);
1448         kmem_cache_free(kvm_vcpu_cache, vcpu);
1449 }
1450
1451 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1452 {
1453         int ret;
1454 #ifdef CONFIG_ALTIVEC
1455         unsigned long uninitialized_var(vrsave);
1456 #endif
1457
1458         /* Check if we can run the vcpu at all */
1459         if (!vcpu->arch.sane) {
1460                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1461                 ret = -EINVAL;
1462                 goto out;
1463         }
1464
1465         /*
1466          * Interrupts could be timers for the guest which we have to inject
1467          * again, so let's postpone them until we're in the guest and if we
1468          * really did time things so badly, then we just exit again due to
1469          * a host external interrupt.
1470          */
1471         ret = kvmppc_prepare_to_enter(vcpu);
1472         if (ret <= 0)
1473                 goto out;
1474         /* interrupts now hard-disabled */
1475
1476         /* Save FPU state in thread_struct */
1477         if (current->thread.regs->msr & MSR_FP)
1478                 giveup_fpu(current);
1479
1480 #ifdef CONFIG_ALTIVEC
1481         /* Save Altivec state in thread_struct */
1482         if (current->thread.regs->msr & MSR_VEC)
1483                 giveup_altivec(current);
1484 #endif
1485
1486 #ifdef CONFIG_VSX
1487         /* Save VSX state in thread_struct */
1488         if (current->thread.regs->msr & MSR_VSX)
1489                 __giveup_vsx(current);
1490 #endif
1491
1492         /* Preload FPU if it's enabled */
1493         if (kvmppc_get_msr(vcpu) & MSR_FP)
1494                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1495
1496         kvmppc_fix_ee_before_entry();
1497
1498         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1499
1500         /* No need for kvm_guest_exit. It's done in handle_exit.
1501            We also get here with interrupts enabled. */
1502
1503         /* Make sure we save the guest FPU/Altivec/VSX state */
1504         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1505
1506         /* Make sure we save the guest TAR/EBB/DSCR state */
1507         kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1508
1509 out:
1510         vcpu->mode = OUTSIDE_GUEST_MODE;
1511         return ret;
1512 }
1513
1514 /*
1515  * Get (and clear) the dirty memory log for a memory slot.
1516  */
1517 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1518                                          struct kvm_dirty_log *log)
1519 {
1520         struct kvm_memory_slot *memslot;
1521         struct kvm_vcpu *vcpu;
1522         ulong ga, ga_end;
1523         int is_dirty = 0;
1524         int r;
1525         unsigned long n;
1526
1527         mutex_lock(&kvm->slots_lock);
1528
1529         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1530         if (r)
1531                 goto out;
1532
1533         /* If nothing is dirty, don't bother messing with page tables. */
1534         if (is_dirty) {
1535                 memslot = id_to_memslot(kvm->memslots, log->slot);
1536
1537                 ga = memslot->base_gfn << PAGE_SHIFT;
1538                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1539
1540                 kvm_for_each_vcpu(n, vcpu, kvm)
1541                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1542
1543                 n = kvm_dirty_bitmap_bytes(memslot);
1544                 memset(memslot->dirty_bitmap, 0, n);
1545         }
1546
1547         r = 0;
1548 out:
1549         mutex_unlock(&kvm->slots_lock);
1550         return r;
1551 }
1552
1553 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1554                                          struct kvm_memory_slot *memslot)
1555 {
1556         return;
1557 }
1558
1559 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1560                                         struct kvm_memory_slot *memslot,
1561                                         struct kvm_userspace_memory_region *mem)
1562 {
1563         return 0;
1564 }
1565
1566 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1567                                 struct kvm_userspace_memory_region *mem,
1568                                 const struct kvm_memory_slot *old)
1569 {
1570         return;
1571 }
1572
1573 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1574                                         struct kvm_memory_slot *dont)
1575 {
1576         return;
1577 }
1578
1579 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1580                                          unsigned long npages)
1581 {
1582         return 0;
1583 }
1584
1585
1586 #ifdef CONFIG_PPC64
1587 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1588                                          struct kvm_ppc_smmu_info *info)
1589 {
1590         long int i;
1591         struct kvm_vcpu *vcpu;
1592
1593         info->flags = 0;
1594
1595         /* SLB is always 64 entries */
1596         info->slb_size = 64;
1597
1598         /* Standard 4k base page size segment */
1599         info->sps[0].page_shift = 12;
1600         info->sps[0].slb_enc = 0;
1601         info->sps[0].enc[0].page_shift = 12;
1602         info->sps[0].enc[0].pte_enc = 0;
1603
1604         /*
1605          * 64k large page size.
1606          * We only want to put this in if the CPUs we're emulating
1607          * support it, but unfortunately we don't have a vcpu easily
1608          * to hand here to test.  Just pick the first vcpu, and if
1609          * that doesn't exist yet, report the minimum capability,
1610          * i.e., no 64k pages.
1611          * 1T segment support goes along with 64k pages.
1612          */
1613         i = 1;
1614         vcpu = kvm_get_vcpu(kvm, 0);
1615         if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1616                 info->flags = KVM_PPC_1T_SEGMENTS;
1617                 info->sps[i].page_shift = 16;
1618                 info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1619                 info->sps[i].enc[0].page_shift = 16;
1620                 info->sps[i].enc[0].pte_enc = 1;
1621                 ++i;
1622         }
1623
1624         /* Standard 16M large page size segment */
1625         info->sps[i].page_shift = 24;
1626         info->sps[i].slb_enc = SLB_VSID_L;
1627         info->sps[i].enc[0].page_shift = 24;
1628         info->sps[i].enc[0].pte_enc = 0;
1629
1630         return 0;
1631 }
1632 #else
1633 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1634                                          struct kvm_ppc_smmu_info *info)
1635 {
1636         /* We should not get called */
1637         BUG();
1638 }
1639 #endif /* CONFIG_PPC64 */
1640
1641 static unsigned int kvm_global_user_count = 0;
1642 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1643
1644 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1645 {
1646         mutex_init(&kvm->arch.hpt_mutex);
1647
1648 #ifdef CONFIG_PPC_BOOK3S_64
1649         /* Start out with the default set of hcalls enabled */
1650         kvmppc_pr_init_default_hcalls(kvm);
1651 #endif
1652
1653         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1654                 spin_lock(&kvm_global_user_count_lock);
1655                 if (++kvm_global_user_count == 1)
1656                         pSeries_disable_reloc_on_exc();
1657                 spin_unlock(&kvm_global_user_count_lock);
1658         }
1659         return 0;
1660 }
1661
1662 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1663 {
1664 #ifdef CONFIG_PPC64
1665         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1666 #endif
1667
1668         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1669                 spin_lock(&kvm_global_user_count_lock);
1670                 BUG_ON(kvm_global_user_count == 0);
1671                 if (--kvm_global_user_count == 0)
1672                         pSeries_enable_reloc_on_exc();
1673                 spin_unlock(&kvm_global_user_count_lock);
1674         }
1675 }
1676
1677 static int kvmppc_core_check_processor_compat_pr(void)
1678 {
1679         /* we are always compatible */
1680         return 0;
1681 }
1682
1683 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1684                                  unsigned int ioctl, unsigned long arg)
1685 {
1686         return -ENOTTY;
1687 }
1688
1689 static struct kvmppc_ops kvm_ops_pr = {
1690         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1691         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1692         .get_one_reg = kvmppc_get_one_reg_pr,
1693         .set_one_reg = kvmppc_set_one_reg_pr,
1694         .vcpu_load   = kvmppc_core_vcpu_load_pr,
1695         .vcpu_put    = kvmppc_core_vcpu_put_pr,
1696         .set_msr     = kvmppc_set_msr_pr,
1697         .vcpu_run    = kvmppc_vcpu_run_pr,
1698         .vcpu_create = kvmppc_core_vcpu_create_pr,
1699         .vcpu_free   = kvmppc_core_vcpu_free_pr,
1700         .check_requests = kvmppc_core_check_requests_pr,
1701         .get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1702         .flush_memslot = kvmppc_core_flush_memslot_pr,
1703         .prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1704         .commit_memory_region = kvmppc_core_commit_memory_region_pr,
1705         .unmap_hva = kvm_unmap_hva_pr,
1706         .unmap_hva_range = kvm_unmap_hva_range_pr,
1707         .age_hva  = kvm_age_hva_pr,
1708         .test_age_hva = kvm_test_age_hva_pr,
1709         .set_spte_hva = kvm_set_spte_hva_pr,
1710         .mmu_destroy  = kvmppc_mmu_destroy_pr,
1711         .free_memslot = kvmppc_core_free_memslot_pr,
1712         .create_memslot = kvmppc_core_create_memslot_pr,
1713         .init_vm = kvmppc_core_init_vm_pr,
1714         .destroy_vm = kvmppc_core_destroy_vm_pr,
1715         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1716         .emulate_op = kvmppc_core_emulate_op_pr,
1717         .emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1718         .emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1719         .fast_vcpu_kick = kvm_vcpu_kick,
1720         .arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1721 #ifdef CONFIG_PPC_BOOK3S_64
1722         .hcall_implemented = kvmppc_hcall_impl_pr,
1723 #endif
1724 };
1725
1726
1727 int kvmppc_book3s_init_pr(void)
1728 {
1729         int r;
1730
1731         r = kvmppc_core_check_processor_compat_pr();
1732         if (r < 0)
1733                 return r;
1734
1735         kvm_ops_pr.owner = THIS_MODULE;
1736         kvmppc_pr_ops = &kvm_ops_pr;
1737
1738         r = kvmppc_mmu_hpte_sysinit();
1739         return r;
1740 }
1741
1742 void kvmppc_book3s_exit_pr(void)
1743 {
1744         kvmppc_pr_ops = NULL;
1745         kvmppc_mmu_hpte_sysexit();
1746 }
1747
1748 /*
1749  * We only support separate modules for book3s 64
1750  */
1751 #ifdef CONFIG_PPC_BOOK3S_64
1752
1753 module_init(kvmppc_book3s_init_pr);
1754 module_exit(kvmppc_book3s_exit_pr);
1755
1756 MODULE_LICENSE("GPL");
1757 MODULE_ALIAS_MISCDEV(KVM_MINOR);
1758 MODULE_ALIAS("devname:kvm");
1759 #endif