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 start,
 299                           unsigned long end)
 300 {
 301         /* XXX could be more clever ;) */
 302         return 0;
 303 }
 304
 305 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
 306 {
 307         /* XXX could be more clever ;) */
 308         return 0;
 309 }
 310
 311 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
 312 {
 313         /* The page will get remapped properly on its next fault */
 314         do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
 315 }
 316
 317 /*****************************************/
 318
 319 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 320 {
 321         ulong guest_msr = kvmppc_get_msr(vcpu);
 322         ulong smsr = guest_msr;
 323
 324         /* Guest MSR values */
 325         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE;
 326         /* Process MSR values */
 327         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 328         /* External providers the guest reserved */
 329         smsr |= (guest_msr & vcpu->arch.guest_owned_ext);
 330         /* 64-bit Process MSR values */
 331 #ifdef CONFIG_PPC_BOOK3S_64
 332         smsr |= MSR_ISF | MSR_HV;
 333 #endif
 334         vcpu->arch.shadow_msr = smsr;
 335 }
 336
 337 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
 338 {
 339         ulong old_msr = kvmppc_get_msr(vcpu);
 340
 341 #ifdef EXIT_DEBUG
 342         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 343 #endif
 344
 345         msr &= to_book3s(vcpu)->msr_mask;
 346         kvmppc_set_msr_fast(vcpu, msr);
 347         kvmppc_recalc_shadow_msr(vcpu);
 348
 349         if (msr & MSR_POW) {
 350                 if (!vcpu->arch.pending_exceptions) {
 351                         kvm_vcpu_block(vcpu);
 352                         clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 353                         vcpu->stat.halt_wakeup++;
 354
 355                         /* Unset POW bit after we woke up */
 356                         msr &= ~MSR_POW;
 357                         kvmppc_set_msr_fast(vcpu, msr);
 358                 }
 359         }
 360
 361         if (kvmppc_is_split_real(vcpu))
 362                 kvmppc_fixup_split_real(vcpu);
 363         else
 364                 kvmppc_unfixup_split_real(vcpu);
 365
 366         if ((kvmppc_get_msr(vcpu) & (MSR_PR|MSR_IR|MSR_DR)) !=
 367                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 368                 kvmppc_mmu_flush_segments(vcpu);
 369                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 370
 371                 /* Preload magic page segment when in kernel mode */
 372                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 373                         struct kvm_vcpu_arch *a = &vcpu->arch;
 374
 375                         if (msr & MSR_DR)
 376                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 377                         else
 378                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 379                 }
 380         }
 381
 382         /*
 383          * When switching from 32 to 64-bit, we may have a stale 32-bit
 384          * magic page around, we need to flush it. Typically 32-bit magic
 385          * page will be instanciated when calling into RTAS. Note: We
 386          * assume that such transition only happens while in kernel mode,
 387          * ie, we never transition from user 32-bit to kernel 64-bit with
 388          * a 32-bit magic page around.
 389          */
 390         if (vcpu->arch.magic_page_pa &&
 391             !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
 392                 /* going from RTAS to normal kernel code */
 393                 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
 394                                      ~0xFFFUL);
 395         }
 396
 397         /* Preload FPU if it's enabled */
 398         if (kvmppc_get_msr(vcpu) & MSR_FP)
 399                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 400 }
 401
 402 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
 403 {
 404         u32 host_pvr;
 405
 406         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 407         vcpu->arch.pvr = pvr;
 408 #ifdef CONFIG_PPC_BOOK3S_64
 409         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 410                 kvmppc_mmu_book3s_64_init(vcpu);
 411                 if (!to_book3s(vcpu)->hior_explicit)
 412                         to_book3s(vcpu)->hior = 0xfff00000;
 413                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 414                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 415         } else
 416 #endif
 417         {
 418                 kvmppc_mmu_book3s_32_init(vcpu);
 419                 if (!to_book3s(vcpu)->hior_explicit)
 420                         to_book3s(vcpu)->hior = 0;
 421                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 422                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 423         }
 424
 425         kvmppc_sanity_check(vcpu);
 426
 427         /* If we are in hypervisor level on 970, we can tell the CPU to
 428          * treat DCBZ as 32 bytes store */
 429         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 430         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 431             !strcmp(cur_cpu_spec->platform, "ppc970"))
 432                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 433
 434         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 435            really needs them in a VM on Cell and force disable them. */
 436         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 437                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 438
 439         /*
 440          * If they're asking for POWER6 or later, set the flag
 441          * indicating that we can do multiple large page sizes
 442          * and 1TB segments.
 443          * Also set the flag that indicates that tlbie has the large
 444          * page bit in the RB operand instead of the instruction.
 445          */
 446         switch (PVR_VER(pvr)) {
 447         case PVR_POWER6:
 448         case PVR_POWER7:
 449         case PVR_POWER7p:
 450         case PVR_POWER8:
 451                 vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
 452                         BOOK3S_HFLAG_NEW_TLBIE;
 453                 break;
 454         }
 455
 456 #ifdef CONFIG_PPC_BOOK3S_32
 457         /* 32 bit Book3S always has 32 byte dcbz */
 458         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 459 #endif
 460
 461         /* On some CPUs we can execute paired single operations natively */
 462         asm ( "mfpvr %0" : "=r"(host_pvr));
 463         switch (host_pvr) {
 464         case 0x00080200:        /* lonestar 2.0 */
 465         case 0x00088202:        /* lonestar 2.2 */
 466         case 0x70000100:        /* gekko 1.0 */
 467         case 0x00080100:        /* gekko 2.0 */
 468         case 0x00083203:        /* gekko 2.3a */
 469         case 0x00083213:        /* gekko 2.3b */
 470         case 0x00083204:        /* gekko 2.4 */
 471         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 472         case 0x00087200:        /* broadway */
 473                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 474                 /* Enable HID2.PSE - in case we need it later */
 475                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 476         }
 477 }
 478
 479 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 480  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 481  * emulate 32 bytes dcbz length.
 482  *
 483  * The Book3s_64 inventors also realized this case and implemented a special bit
 484  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 485  *
 486  * My approach here is to patch the dcbz instruction on executing pages.
 487  */
 488 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 489 {
 490         struct page *hpage;
 491         u64 hpage_offset;
 492         u32 *page;
 493         int i;
 494
 495         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 496         if (is_error_page(hpage))
 497                 return;
 498
 499         hpage_offset = pte->raddr & ~PAGE_MASK;
 500         hpage_offset &= ~0xFFFULL;
 501         hpage_offset /= 4;
 502
 503         get_page(hpage);
 504         page = kmap_atomic(hpage);
 505
 506         /* patch dcbz into reserved instruction, so we trap */
 507         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 508                 if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
 509                         page[i] &= cpu_to_be32(0xfffffff7);
 510
 511         kunmap_atomic(page);
 512         put_page(hpage);
 513 }
 514
 515 static bool kvmppc_visible_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
 516 {
 517         ulong mp_pa = vcpu->arch.magic_page_pa;
 518
 519         if (!(kvmppc_get_msr(vcpu) & MSR_SF))
 520                 mp_pa = (uint32_t)mp_pa;
 521
 522         gpa &= ~0xFFFULL;
 523         if (unlikely(mp_pa) && unlikely((mp_pa & KVM_PAM) == (gpa & KVM_PAM))) {
 524                 return true;
 525         }
 526
 527         return kvm_is_visible_gfn(vcpu->kvm, gpa >> PAGE_SHIFT);
 528 }
 529
 530 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 531                             ulong eaddr, int vec)
 532 {
 533         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 534         bool iswrite = false;
 535         int r = RESUME_GUEST;
 536         int relocated;
 537         int page_found = 0;
 538         struct kvmppc_pte pte;
 539         bool is_mmio = false;
 540         bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
 541         bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
 542         u64 vsid;
 543
 544         relocated = data ? dr : ir;
 545         if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
 546                 iswrite = true;
 547
 548         /* Resolve real address if translation turned on */
 549         if (relocated) {
 550                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
 551         } else {
 552                 pte.may_execute = true;
 553                 pte.may_read = true;
 554                 pte.may_write = true;
 555                 pte.raddr = eaddr & KVM_PAM;
 556                 pte.eaddr = eaddr;
 557                 pte.vpage = eaddr >> 12;
 558                 pte.page_size = MMU_PAGE_64K;
 559         }
 560
 561         switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
 562         case 0:
 563                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 564                 break;
 565         case MSR_DR:
 566                 if (!data &&
 567                     (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
 568                     ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
 569                         pte.raddr &= ~SPLIT_HACK_MASK;
 570                 /* fall through */
 571         case MSR_IR:
 572                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 573
 574                 if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
 575                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 576                 else
 577                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 578                 pte.vpage |= vsid;
 579
 580                 if (vsid == -1)
 581                         page_found = -EINVAL;
 582                 break;
 583         }
 584
 585         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 586            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 587                 /*
 588                  * If we do the dcbz hack, we have to NX on every execution,
 589                  * so we can patch the executing code. This renders our guest
 590                  * NX-less.
 591                  */
 592                 pte.may_execute = !data;
 593         }
 594
 595         if (page_found == -ENOENT) {
 596                 /* Page not found in guest PTE entries */
 597                 u64 ssrr1 = vcpu->arch.shadow_srr1;
 598                 u64 msr = kvmppc_get_msr(vcpu);
 599                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 600                 kvmppc_set_dsisr(vcpu, vcpu->arch.fault_dsisr);
 601                 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
 602                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 603         } else if (page_found == -EPERM) {
 604                 /* Storage protection */
 605                 u32 dsisr = vcpu->arch.fault_dsisr;
 606                 u64 ssrr1 = vcpu->arch.shadow_srr1;
 607                 u64 msr = kvmppc_get_msr(vcpu);
 608                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 609                 dsisr = (dsisr & ~DSISR_NOHPTE) | DSISR_PROTFAULT;
 610                 kvmppc_set_dsisr(vcpu, dsisr);
 611                 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
 612                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 613         } else if (page_found == -EINVAL) {
 614                 /* Page not found in guest SLB */
 615                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 616                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 617         } else if (!is_mmio &&
 618                    kvmppc_visible_gpa(vcpu, pte.raddr)) {
 619                 if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
 620                         /*
 621                          * There is already a host HPTE there, presumably
 622                          * a read-only one for a page the guest thinks
 623                          * is writable, so get rid of it first.
 624                          */
 625                         kvmppc_mmu_unmap_page(vcpu, &pte);
 626                 }
 627                 /* The guest's PTE is not mapped yet. Map on the host */
 628                 kvmppc_mmu_map_page(vcpu, &pte, iswrite);
 629                 if (data)
 630                         vcpu->stat.sp_storage++;
 631                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 632                          (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 633                         kvmppc_patch_dcbz(vcpu, &pte);
 634         } else {
 635                 /* MMIO */
 636                 vcpu->stat.mmio_exits++;
 637                 vcpu->arch.paddr_accessed = pte.raddr;
 638                 vcpu->arch.vaddr_accessed = pte.eaddr;
 639                 r = kvmppc_emulate_mmio(run, vcpu);
 640                 if ( r == RESUME_HOST_NV )
 641                         r = RESUME_HOST;
 642         }
 643
 644         return r;
 645 }
 646
 647 /* Give up external provider (FPU, Altivec, VSX) */
 648 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 649 {
 650         struct thread_struct *t = &current->thread;
 651
 652         /*
 653          * VSX instructions can access FP and vector registers, so if
 654          * we are giving up VSX, make sure we give up FP and VMX as well.
 655          */
 656         if (msr & MSR_VSX)
 657                 msr |= MSR_FP | MSR_VEC;
 658
 659         msr &= vcpu->arch.guest_owned_ext;
 660         if (!msr)
 661                 return;
 662
 663 #ifdef DEBUG_EXT
 664         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 665 #endif
 666
 667         if (msr & MSR_FP) {
 668                 /*
 669                  * Note that on CPUs with VSX, giveup_fpu stores
 670                  * both the traditional FP registers and the added VSX
 671                  * registers into thread.fp_state.fpr[].
 672                  */
 673                 if (t->regs->msr & MSR_FP)
 674                         giveup_fpu(current);
 675                 t->fp_save_area = NULL;
 676         }
 677
 678 #ifdef CONFIG_ALTIVEC
 679         if (msr & MSR_VEC) {
 680                 if (current->thread.regs->msr & MSR_VEC)
 681                         giveup_altivec(current);
 682                 t->vr_save_area = NULL;
 683         }
 684 #endif
 685
 686         vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
 687         kvmppc_recalc_shadow_msr(vcpu);
 688 }
 689
 690 /* Give up facility (TAR / EBB / DSCR) */
 691 static void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
 692 {
 693 #ifdef CONFIG_PPC_BOOK3S_64
 694         if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
 695                 /* Facility not available to the guest, ignore giveup request*/
 696                 return;
 697         }
 698
 699         switch (fac) {
 700         case FSCR_TAR_LG:
 701                 vcpu->arch.tar = mfspr(SPRN_TAR);
 702                 mtspr(SPRN_TAR, current->thread.tar);
 703                 vcpu->arch.shadow_fscr &= ~FSCR_TAR;
 704                 break;
 705         }
 706 #endif
 707 }
 708
 709 /* Handle external providers (FPU, Altivec, VSX) */
 710 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 711                              ulong msr)
 712 {
 713         struct thread_struct *t = &current->thread;
 714
 715         /* When we have paired singles, we emulate in software */
 716         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 717                 return RESUME_GUEST;
 718
 719         if (!(kvmppc_get_msr(vcpu) & msr)) {
 720                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 721                 return RESUME_GUEST;
 722         }
 723
 724         if (msr == MSR_VSX) {
 725                 /* No VSX?  Give an illegal instruction interrupt */
 726 #ifdef CONFIG_VSX
 727                 if (!cpu_has_feature(CPU_FTR_VSX))
 728 #endif
 729                 {
 730                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 731                         return RESUME_GUEST;
 732                 }
 733
 734                 /*
 735                  * We have to load up all the FP and VMX registers before
 736                  * we can let the guest use VSX instructions.
 737                  */
 738                 msr = MSR_FP | MSR_VEC | MSR_VSX;
 739         }
 740
 741         /* See if we already own all the ext(s) needed */
 742         msr &= ~vcpu->arch.guest_owned_ext;
 743         if (!msr)
 744                 return RESUME_GUEST;
 745
 746 #ifdef DEBUG_EXT
 747         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 748 #endif
 749
 750         if (msr & MSR_FP) {
 751                 preempt_disable();
 752                 enable_kernel_fp();
 753                 load_fp_state(&vcpu->arch.fp);
 754                 disable_kernel_fp();
 755                 t->fp_save_area = &vcpu->arch.fp;
 756                 preempt_enable();
 757         }
 758
 759         if (msr & MSR_VEC) {
 760 #ifdef CONFIG_ALTIVEC
 761                 preempt_disable();
 762                 enable_kernel_altivec();
 763                 load_vr_state(&vcpu->arch.vr);
 764                 disable_kernel_altivec();
 765                 t->vr_save_area = &vcpu->arch.vr;
 766                 preempt_enable();
 767 #endif
 768         }
 769
 770         t->regs->msr |= msr;
 771         vcpu->arch.guest_owned_ext |= msr;
 772         kvmppc_recalc_shadow_msr(vcpu);
 773
 774         return RESUME_GUEST;
 775 }
 776
 777 /*
 778  * Kernel code using FP or VMX could have flushed guest state to
 779  * the thread_struct; if so, get it back now.
 780  */
 781 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
 782 {
 783         unsigned long lost_ext;
 784
 785         lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
 786         if (!lost_ext)
 787                 return;
 788
 789         if (lost_ext & MSR_FP) {
 790                 preempt_disable();
 791                 enable_kernel_fp();
 792                 load_fp_state(&vcpu->arch.fp);
 793                 disable_kernel_fp();
 794                 preempt_enable();
 795         }
 796 #ifdef CONFIG_ALTIVEC
 797         if (lost_ext & MSR_VEC) {
 798                 preempt_disable();
 799                 enable_kernel_altivec();
 800                 load_vr_state(&vcpu->arch.vr);
 801                 disable_kernel_altivec();
 802                 preempt_enable();
 803         }
 804 #endif
 805         current->thread.regs->msr |= lost_ext;
 806 }
 807
 808 #ifdef CONFIG_PPC_BOOK3S_64
 809
 810 static void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
 811 {
 812         /* Inject the Interrupt Cause field and trigger a guest interrupt */
 813         vcpu->arch.fscr &= ~(0xffULL << 56);
 814         vcpu->arch.fscr |= (fac << 56);
 815         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
 816 }
 817
 818 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
 819 {
 820         enum emulation_result er = EMULATE_FAIL;
 821
 822         if (!(kvmppc_get_msr(vcpu) & MSR_PR))
 823                 er = kvmppc_emulate_instruction(vcpu->run, vcpu);
 824
 825         if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
 826                 /* Couldn't emulate, trigger interrupt in guest */
 827                 kvmppc_trigger_fac_interrupt(vcpu, fac);
 828         }
 829 }
 830
 831 /* Enable facilities (TAR, EBB, DSCR) for the guest */
 832 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
 833 {
 834         bool guest_fac_enabled;
 835         BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
 836
 837         /*
 838          * Not every facility is enabled by FSCR bits, check whether the
 839          * guest has this facility enabled at all.
 840          */
 841         switch (fac) {
 842         case FSCR_TAR_LG:
 843         case FSCR_EBB_LG:
 844                 guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
 845                 break;
 846         case FSCR_TM_LG:
 847                 guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
 848                 break;
 849         default:
 850                 guest_fac_enabled = false;
 851                 break;
 852         }
 853
 854         if (!guest_fac_enabled) {
 855                 /* Facility not enabled by the guest */
 856                 kvmppc_trigger_fac_interrupt(vcpu, fac);
 857                 return RESUME_GUEST;
 858         }
 859
 860         switch (fac) {
 861         case FSCR_TAR_LG:
 862                 /* TAR switching isn't lazy in Linux yet */
 863                 current->thread.tar = mfspr(SPRN_TAR);
 864                 mtspr(SPRN_TAR, vcpu->arch.tar);
 865                 vcpu->arch.shadow_fscr |= FSCR_TAR;
 866                 break;
 867         default:
 868                 kvmppc_emulate_fac(vcpu, fac);
 869                 break;
 870         }
 871
 872         return RESUME_GUEST;
 873 }
 874
 875 void kvmppc_set_fscr(struct kvm_vcpu *vcpu, u64 fscr)
 876 {
 877         if ((vcpu->arch.fscr & FSCR_TAR) && !(fscr & FSCR_TAR)) {
 878                 /* TAR got dropped, drop it in shadow too */
 879                 kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
 880         }
 881         vcpu->arch.fscr = fscr;
 882 }
 883 #endif
 884
 885 static void kvmppc_setup_debug(struct kvm_vcpu *vcpu)
 886 {
 887         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
 888                 u64 msr = kvmppc_get_msr(vcpu);
 889
 890                 kvmppc_set_msr(vcpu, msr | MSR_SE);
 891         }
 892 }
 893
 894 static void kvmppc_clear_debug(struct kvm_vcpu *vcpu)
 895 {
 896         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
 897                 u64 msr = kvmppc_get_msr(vcpu);
 898
 899                 kvmppc_set_msr(vcpu, msr & ~MSR_SE);
 900         }
 901 }
 902
 903 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
 904                           unsigned int exit_nr)
 905 {
 906         int r = RESUME_HOST;
 907         int s;
 908
 909         vcpu->stat.sum_exits++;
 910
 911         run->exit_reason = KVM_EXIT_UNKNOWN;
 912         run->ready_for_interrupt_injection = 1;
 913
 914         /* We get here with MSR.EE=1 */
 915
 916         trace_kvm_exit(exit_nr, vcpu);
 917         kvm_guest_exit();
 918
 919         switch (exit_nr) {
 920         case BOOK3S_INTERRUPT_INST_STORAGE:
 921         {
 922                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 923                 vcpu->stat.pf_instruc++;
 924
 925                 if (kvmppc_is_split_real(vcpu))
 926                         kvmppc_fixup_split_real(vcpu);
 927
 928 #ifdef CONFIG_PPC_BOOK3S_32
 929                 /* We set segments as unused segments when invalidating them. So
 930                  * treat the respective fault as segment fault. */
 931                 {
 932                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 933                         u32 sr;
 934
 935                         svcpu = svcpu_get(vcpu);
 936                         sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
 937                         svcpu_put(svcpu);
 938                         if (sr == SR_INVALID) {
 939                                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 940                                 r = RESUME_GUEST;
 941                                 break;
 942                         }
 943                 }
 944 #endif
 945
 946                 /* only care about PTEG not found errors, but leave NX alone */
 947                 if (shadow_srr1 & 0x40000000) {
 948                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
 949                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 950                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 951                         vcpu->stat.sp_instruc++;
 952                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 953                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 954                         /*
 955                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 956                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 957                          *     that no guest that needs the dcbz hack does NX.
 958                          */
 959                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 960                         r = RESUME_GUEST;
 961                 } else {
 962                         u64 msr = kvmppc_get_msr(vcpu);
 963                         msr |= shadow_srr1 & 0x58000000;
 964                         kvmppc_set_msr_fast(vcpu, msr);
 965                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 966                         r = RESUME_GUEST;
 967                 }
 968                 break;
 969         }
 970         case BOOK3S_INTERRUPT_DATA_STORAGE:
 971         {
 972                 ulong dar = kvmppc_get_fault_dar(vcpu);
 973                 u32 fault_dsisr = vcpu->arch.fault_dsisr;
 974                 vcpu->stat.pf_storage++;
 975
 976 #ifdef CONFIG_PPC_BOOK3S_32
 977                 /* We set segments as unused segments when invalidating them. So
 978                  * treat the respective fault as segment fault. */
 979                 {
 980                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 981                         u32 sr;
 982
 983                         svcpu = svcpu_get(vcpu);
 984                         sr = svcpu->sr[dar >> SID_SHIFT];
 985                         svcpu_put(svcpu);
 986                         if (sr == SR_INVALID) {
 987                                 kvmppc_mmu_map_segment(vcpu, dar);
 988                                 r = RESUME_GUEST;
 989                                 break;
 990                         }
 991                 }
 992 #endif
 993
 994                 /*
 995                  * We need to handle missing shadow PTEs, and
 996                  * protection faults due to us mapping a page read-only
 997                  * when the guest thinks it is writable.
 998                  */
 999                 if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
1000                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
1001                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
1002                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1003                 } else {
1004                         kvmppc_set_dar(vcpu, dar);
1005                         kvmppc_set_dsisr(vcpu, fault_dsisr);
1006                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1007                         r = RESUME_GUEST;
1008                 }
1009                 break;
1010         }
1011         case BOOK3S_INTERRUPT_DATA_SEGMENT:
1012                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
1013                         kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
1014                         kvmppc_book3s_queue_irqprio(vcpu,
1015                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
1016                 }
1017                 r = RESUME_GUEST;
1018                 break;
1019         case BOOK3S_INTERRUPT_INST_SEGMENT:
1020                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1021                         kvmppc_book3s_queue_irqprio(vcpu,
1022                                 BOOK3S_INTERRUPT_INST_SEGMENT);
1023                 }
1024                 r = RESUME_GUEST;
1025                 break;
1026         /* We're good on these - the host merely wanted to get our attention */
1027         case BOOK3S_INTERRUPT_DECREMENTER:
1028         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1029         case BOOK3S_INTERRUPT_DOORBELL:
1030         case BOOK3S_INTERRUPT_H_DOORBELL:
1031                 vcpu->stat.dec_exits++;
1032                 r = RESUME_GUEST;
1033                 break;
1034         case BOOK3S_INTERRUPT_EXTERNAL:
1035         case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
1036         case BOOK3S_INTERRUPT_EXTERNAL_HV:
1037                 vcpu->stat.ext_intr_exits++;
1038                 r = RESUME_GUEST;
1039                 break;
1040         case BOOK3S_INTERRUPT_PERFMON:
1041                 r = RESUME_GUEST;
1042                 break;
1043         case BOOK3S_INTERRUPT_PROGRAM:
1044         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1045         {
1046                 enum emulation_result er;
1047                 ulong flags;
1048                 u32 last_inst;
1049                 int emul;
1050
1051 program_interrupt:
1052                 flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
1053
1054                 emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1055                 if (emul != EMULATE_DONE) {
1056                         r = RESUME_GUEST;
1057                         break;
1058                 }
1059
1060                 if (kvmppc_get_msr(vcpu) & MSR_PR) {
1061 #ifdef EXIT_DEBUG
1062                         pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
1063                                 kvmppc_get_pc(vcpu), last_inst);
1064 #endif
1065                         if ((last_inst & 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), last_inst);
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         {
1102                 u32 last_sc;
1103                 int emul;
1104
1105                 /* Get last sc for papr */
1106                 if (vcpu->arch.papr_enabled) {
1107                         /* The sc instuction points SRR0 to the next inst */
1108                         emul = kvmppc_get_last_inst(vcpu, INST_SC, &last_sc);
1109                         if (emul != EMULATE_DONE) {
1110                                 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) - 4);
1111                                 r = RESUME_GUEST;
1112                                 break;
1113                         }
1114                 }
1115
1116                 if (vcpu->arch.papr_enabled &&
1117                     (last_sc == 0x44000022) &&
1118                     !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1119                         /* SC 1 papr hypercalls */
1120                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
1121                         int i;
1122
1123 #ifdef CONFIG_PPC_BOOK3S_64
1124                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1125                                 r = RESUME_GUEST;
1126                                 break;
1127                         }
1128 #endif
1129
1130                         run->papr_hcall.nr = cmd;
1131                         for (i = 0; i < 9; ++i) {
1132                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1133                                 run->papr_hcall.args[i] = gpr;
1134                         }
1135                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
1136                         vcpu->arch.hcall_needed = 1;
1137                         r = RESUME_HOST;
1138                 } else if (vcpu->arch.osi_enabled &&
1139                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1140                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1141                         /* MOL hypercalls */
1142                         u64 *gprs = run->osi.gprs;
1143                         int i;
1144
1145                         run->exit_reason = KVM_EXIT_OSI;
1146                         for (i = 0; i < 32; i++)
1147                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
1148                         vcpu->arch.osi_needed = 1;
1149                         r = RESUME_HOST_NV;
1150                 } else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1151                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1152                         /* KVM PV hypercalls */
1153                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1154                         r = RESUME_GUEST;
1155                 } else {
1156                         /* Guest syscalls */
1157                         vcpu->stat.syscall_exits++;
1158                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1159                         r = RESUME_GUEST;
1160                 }
1161                 break;
1162         }
1163         case BOOK3S_INTERRUPT_FP_UNAVAIL:
1164         case BOOK3S_INTERRUPT_ALTIVEC:
1165         case BOOK3S_INTERRUPT_VSX:
1166         {
1167                 int ext_msr = 0;
1168                 int emul;
1169                 u32 last_inst;
1170
1171                 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) {
1172                         /* Do paired single instruction emulation */
1173                         emul = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1174                                                     &last_inst);
1175                         if (emul == EMULATE_DONE)
1176                                 goto program_interrupt;
1177                         else
1178                                 r = RESUME_GUEST;
1179
1180                         break;
1181                 }
1182
1183                 /* Enable external provider */
1184                 switch (exit_nr) {
1185                 case BOOK3S_INTERRUPT_FP_UNAVAIL:
1186                         ext_msr = MSR_FP;
1187                         break;
1188
1189                 case BOOK3S_INTERRUPT_ALTIVEC:
1190                         ext_msr = MSR_VEC;
1191                         break;
1192
1193                 case BOOK3S_INTERRUPT_VSX:
1194                         ext_msr = MSR_VSX;
1195                         break;
1196                 }
1197
1198                 r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1199                 break;
1200         }
1201         case BOOK3S_INTERRUPT_ALIGNMENT:
1202         {
1203                 u32 last_inst;
1204                 int emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1205
1206                 if (emul == EMULATE_DONE) {
1207                         u32 dsisr;
1208                         u64 dar;
1209
1210                         dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1211                         dar = kvmppc_alignment_dar(vcpu, last_inst);
1212
1213                         kvmppc_set_dsisr(vcpu, dsisr);
1214                         kvmppc_set_dar(vcpu, dar);
1215
1216                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1217                 }
1218                 r = RESUME_GUEST;
1219                 break;
1220         }
1221 #ifdef CONFIG_PPC_BOOK3S_64
1222         case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1223                 kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1224                 r = RESUME_GUEST;
1225                 break;
1226 #endif
1227         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1228                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1229                 r = RESUME_GUEST;
1230                 break;
1231         case BOOK3S_INTERRUPT_TRACE:
1232                 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1233                         run->exit_reason = KVM_EXIT_DEBUG;
1234                         r = RESUME_HOST;
1235                 } else {
1236                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1237                         r = RESUME_GUEST;
1238                 }
1239                 break;
1240         default:
1241         {
1242                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1243                 /* Ugh - bork here! What did we get? */
1244                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1245                         exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1246                 r = RESUME_HOST;
1247                 BUG();
1248                 break;
1249         }
1250         }
1251
1252         if (!(r & RESUME_HOST)) {
1253                 /* To avoid clobbering exit_reason, only check for signals if
1254                  * we aren't already exiting to userspace for some other
1255                  * reason. */
1256
1257                 /*
1258                  * Interrupts could be timers for the guest which we have to
1259                  * inject again, so let's postpone them until we're in the guest
1260                  * and if we really did time things so badly, then we just exit
1261                  * again due to a host external interrupt.
1262                  */
1263                 s = kvmppc_prepare_to_enter(vcpu);
1264                 if (s <= 0)
1265                         r = s;
1266                 else {
1267                         /* interrupts now hard-disabled */
1268                         kvmppc_fix_ee_before_entry();
1269                 }
1270
1271                 kvmppc_handle_lost_ext(vcpu);
1272         }
1273
1274         trace_kvm_book3s_reenter(r, vcpu);
1275
1276         return r;
1277 }
1278
1279 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1280                                             struct kvm_sregs *sregs)
1281 {
1282         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1283         int i;
1284
1285         sregs->pvr = vcpu->arch.pvr;
1286
1287         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1288         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1289                 for (i = 0; i < 64; i++) {
1290                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1291                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1292                 }
1293         } else {
1294                 for (i = 0; i < 16; i++)
1295                         sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1296
1297                 for (i = 0; i < 8; i++) {
1298                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1299                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1300                 }
1301         }
1302
1303         return 0;
1304 }
1305
1306 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1307                                             struct kvm_sregs *sregs)
1308 {
1309         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1310         int i;
1311
1312         kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1313
1314         vcpu3s->sdr1 = sregs->u.s.sdr1;
1315         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1316                 for (i = 0; i < 64; i++) {
1317                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1318                                                     sregs->u.s.ppc64.slb[i].slbe);
1319                 }
1320         } else {
1321                 for (i = 0; i < 16; i++) {
1322                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1323                 }
1324                 for (i = 0; i < 8; i++) {
1325                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1326                                        (u32)sregs->u.s.ppc32.ibat[i]);
1327                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1328                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1329                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1330                                        (u32)sregs->u.s.ppc32.dbat[i]);
1331                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1332                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1333                 }
1334         }
1335
1336         /* Flush the MMU after messing with the segments */
1337         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1338
1339         return 0;
1340 }
1341
1342 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1343                                  union kvmppc_one_reg *val)
1344 {
1345         int r = 0;
1346
1347         switch (id) {
1348         case KVM_REG_PPC_DEBUG_INST:
1349                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1350                 break;
1351         case KVM_REG_PPC_HIOR:
1352                 *val = get_reg_val(id, to_book3s(vcpu)->hior);
1353                 break;
1354         case KVM_REG_PPC_LPCR:
1355         case KVM_REG_PPC_LPCR_64:
1356                 /*
1357                  * We are only interested in the LPCR_ILE bit
1358                  */
1359                 if (vcpu->arch.intr_msr & MSR_LE)
1360                         *val = get_reg_val(id, LPCR_ILE);
1361                 else
1362                         *val = get_reg_val(id, 0);
1363                 break;
1364         default:
1365                 r = -EINVAL;
1366                 break;
1367         }
1368
1369         return r;
1370 }
1371
1372 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1373 {
1374         if (new_lpcr & LPCR_ILE)
1375                 vcpu->arch.intr_msr |= MSR_LE;
1376         else
1377                 vcpu->arch.intr_msr &= ~MSR_LE;
1378 }
1379
1380 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1381                                  union kvmppc_one_reg *val)
1382 {
1383         int r = 0;
1384
1385         switch (id) {
1386         case KVM_REG_PPC_HIOR:
1387                 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1388                 to_book3s(vcpu)->hior_explicit = true;
1389                 break;
1390         case KVM_REG_PPC_LPCR:
1391         case KVM_REG_PPC_LPCR_64:
1392                 kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1393                 break;
1394         default:
1395                 r = -EINVAL;
1396                 break;
1397         }
1398
1399         return r;
1400 }
1401
1402 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1403                                                    unsigned int id)
1404 {
1405         struct kvmppc_vcpu_book3s *vcpu_book3s;
1406         struct kvm_vcpu *vcpu;
1407         int err = -ENOMEM;
1408         unsigned long p;
1409
1410         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1411         if (!vcpu)
1412                 goto out;
1413
1414         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1415         if (!vcpu_book3s)
1416                 goto free_vcpu;
1417         vcpu->arch.book3s = vcpu_book3s;
1418
1419 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1420         vcpu->arch.shadow_vcpu =
1421                 kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1422         if (!vcpu->arch.shadow_vcpu)
1423                 goto free_vcpu3s;
1424 #endif
1425
1426         err = kvm_vcpu_init(vcpu, kvm, id);
1427         if (err)
1428                 goto free_shadow_vcpu;
1429
1430         err = -ENOMEM;
1431         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1432         if (!p)
1433                 goto uninit_vcpu;
1434         vcpu->arch.shared = (void *)p;
1435 #ifdef CONFIG_PPC_BOOK3S_64
1436         /* Always start the shared struct in native endian mode */
1437 #ifdef __BIG_ENDIAN__
1438         vcpu->arch.shared_big_endian = true;
1439 #else
1440         vcpu->arch.shared_big_endian = false;
1441 #endif
1442
1443         /*
1444          * Default to the same as the host if we're on sufficiently
1445          * recent machine that we have 1TB segments;
1446          * otherwise default to PPC970FX.
1447          */
1448         vcpu->arch.pvr = 0x3C0301;
1449         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1450                 vcpu->arch.pvr = mfspr(SPRN_PVR);
1451         vcpu->arch.intr_msr = MSR_SF;
1452 #else
1453         /* default to book3s_32 (750) */
1454         vcpu->arch.pvr = 0x84202;
1455 #endif
1456         kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1457         vcpu->arch.slb_nr = 64;
1458
1459         vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1460
1461         err = kvmppc_mmu_init(vcpu);
1462         if (err < 0)
1463                 goto uninit_vcpu;
1464
1465         return vcpu;
1466
1467 uninit_vcpu:
1468         kvm_vcpu_uninit(vcpu);
1469 free_shadow_vcpu:
1470 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1471         kfree(vcpu->arch.shadow_vcpu);
1472 free_vcpu3s:
1473 #endif
1474         vfree(vcpu_book3s);
1475 free_vcpu:
1476         kmem_cache_free(kvm_vcpu_cache, vcpu);
1477 out:
1478         return ERR_PTR(err);
1479 }
1480
1481 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1482 {
1483         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1484
1485         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1486         kvm_vcpu_uninit(vcpu);
1487 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1488         kfree(vcpu->arch.shadow_vcpu);
1489 #endif
1490         vfree(vcpu_book3s);
1491         kmem_cache_free(kvm_vcpu_cache, vcpu);
1492 }
1493
1494 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1495 {
1496         int ret;
1497 #ifdef CONFIG_ALTIVEC
1498         unsigned long uninitialized_var(vrsave);
1499 #endif
1500
1501         /* Check if we can run the vcpu at all */
1502         if (!vcpu->arch.sane) {
1503                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1504                 ret = -EINVAL;
1505                 goto out;
1506         }
1507
1508         kvmppc_setup_debug(vcpu);
1509
1510         /*
1511          * Interrupts could be timers for the guest which we have to inject
1512          * again, so let's postpone them until we're in the guest and if we
1513          * really did time things so badly, then we just exit again due to
1514          * a host external interrupt.
1515          */
1516         ret = kvmppc_prepare_to_enter(vcpu);
1517         if (ret <= 0)
1518                 goto out;
1519         /* interrupts now hard-disabled */
1520
1521         /* Save FPU, Altivec and VSX state */
1522         giveup_all(current);
1523
1524         /* Preload FPU if it's enabled */
1525         if (kvmppc_get_msr(vcpu) & MSR_FP)
1526                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1527
1528         kvmppc_fix_ee_before_entry();
1529
1530         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1531
1532         kvmppc_clear_debug(vcpu);
1533
1534         /* No need for kvm_guest_exit. It's done in handle_exit.
1535            We also get here with interrupts enabled. */
1536
1537         /* Make sure we save the guest FPU/Altivec/VSX state */
1538         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1539
1540         /* Make sure we save the guest TAR/EBB/DSCR state */
1541         kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1542
1543 out:
1544         vcpu->mode = OUTSIDE_GUEST_MODE;
1545         return ret;
1546 }
1547
1548 /*
1549  * Get (and clear) the dirty memory log for a memory slot.
1550  */
1551 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1552                                          struct kvm_dirty_log *log)
1553 {
1554         struct kvm_memslots *slots;
1555         struct kvm_memory_slot *memslot;
1556         struct kvm_vcpu *vcpu;
1557         ulong ga, ga_end;
1558         int is_dirty = 0;
1559         int r;
1560         unsigned long n;
1561
1562         mutex_lock(&kvm->slots_lock);
1563
1564         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1565         if (r)
1566                 goto out;
1567
1568         /* If nothing is dirty, don't bother messing with page tables. */
1569         if (is_dirty) {
1570                 slots = kvm_memslots(kvm);
1571                 memslot = id_to_memslot(slots, log->slot);
1572
1573                 ga = memslot->base_gfn << PAGE_SHIFT;
1574                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1575
1576                 kvm_for_each_vcpu(n, vcpu, kvm)
1577                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1578
1579                 n = kvm_dirty_bitmap_bytes(memslot);
1580                 memset(memslot->dirty_bitmap, 0, n);
1581         }
1582
1583         r = 0;
1584 out:
1585         mutex_unlock(&kvm->slots_lock);
1586         return r;
1587 }
1588
1589 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1590                                          struct kvm_memory_slot *memslot)
1591 {
1592         return;
1593 }
1594
1595 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1596                                         struct kvm_memory_slot *memslot,
1597                                         const struct kvm_userspace_memory_region *mem)
1598 {
1599         return 0;
1600 }
1601
1602 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1603                                 const struct kvm_userspace_memory_region *mem,
1604                                 const struct kvm_memory_slot *old,
1605                                 const struct kvm_memory_slot *new)
1606 {
1607         return;
1608 }
1609
1610 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1611                                         struct kvm_memory_slot *dont)
1612 {
1613         return;
1614 }
1615
1616 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1617                                          unsigned long npages)
1618 {
1619         return 0;
1620 }
1621
1622
1623 #ifdef CONFIG_PPC64
1624 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1625                                          struct kvm_ppc_smmu_info *info)
1626 {
1627         long int i;
1628         struct kvm_vcpu *vcpu;
1629
1630         info->flags = 0;
1631
1632         /* SLB is always 64 entries */
1633         info->slb_size = 64;
1634
1635         /* Standard 4k base page size segment */
1636         info->sps[0].page_shift = 12;
1637         info->sps[0].slb_enc = 0;
1638         info->sps[0].enc[0].page_shift = 12;
1639         info->sps[0].enc[0].pte_enc = 0;
1640
1641         /*
1642          * 64k large page size.
1643          * We only want to put this in if the CPUs we're emulating
1644          * support it, but unfortunately we don't have a vcpu easily
1645          * to hand here to test.  Just pick the first vcpu, and if
1646          * that doesn't exist yet, report the minimum capability,
1647          * i.e., no 64k pages.
1648          * 1T segment support goes along with 64k pages.
1649          */
1650         i = 1;
1651         vcpu = kvm_get_vcpu(kvm, 0);
1652         if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1653                 info->flags = KVM_PPC_1T_SEGMENTS;
1654                 info->sps[i].page_shift = 16;
1655                 info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1656                 info->sps[i].enc[0].page_shift = 16;
1657                 info->sps[i].enc[0].pte_enc = 1;
1658                 ++i;
1659         }
1660
1661         /* Standard 16M large page size segment */
1662         info->sps[i].page_shift = 24;
1663         info->sps[i].slb_enc = SLB_VSID_L;
1664         info->sps[i].enc[0].page_shift = 24;
1665         info->sps[i].enc[0].pte_enc = 0;
1666
1667         return 0;
1668 }
1669 #else
1670 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1671                                          struct kvm_ppc_smmu_info *info)
1672 {
1673         /* We should not get called */
1674         BUG();
1675 }
1676 #endif /* CONFIG_PPC64 */
1677
1678 static unsigned int kvm_global_user_count = 0;
1679 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1680
1681 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1682 {
1683         mutex_init(&kvm->arch.hpt_mutex);
1684
1685 #ifdef CONFIG_PPC_BOOK3S_64
1686         /* Start out with the default set of hcalls enabled */
1687         kvmppc_pr_init_default_hcalls(kvm);
1688 #endif
1689
1690         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1691                 spin_lock(&kvm_global_user_count_lock);
1692                 if (++kvm_global_user_count == 1)
1693                         pSeries_disable_reloc_on_exc();
1694                 spin_unlock(&kvm_global_user_count_lock);
1695         }
1696         return 0;
1697 }
1698
1699 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1700 {
1701 #ifdef CONFIG_PPC64
1702         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1703 #endif
1704
1705         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1706                 spin_lock(&kvm_global_user_count_lock);
1707                 BUG_ON(kvm_global_user_count == 0);
1708                 if (--kvm_global_user_count == 0)
1709                         pSeries_enable_reloc_on_exc();
1710                 spin_unlock(&kvm_global_user_count_lock);
1711         }
1712 }
1713
1714 static int kvmppc_core_check_processor_compat_pr(void)
1715 {
1716         /*
1717          * Disable KVM for Power9 untill the required bits merged.
1718          */
1719         if (cpu_has_feature(CPU_FTR_ARCH_300))
1720                 return -EIO;
1721         return 0;
1722 }
1723
1724 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1725                                  unsigned int ioctl, unsigned long arg)
1726 {
1727         return -ENOTTY;
1728 }
1729
1730 static struct kvmppc_ops kvm_ops_pr = {
1731         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1732         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1733         .get_one_reg = kvmppc_get_one_reg_pr,
1734         .set_one_reg = kvmppc_set_one_reg_pr,
1735         .vcpu_load   = kvmppc_core_vcpu_load_pr,
1736         .vcpu_put    = kvmppc_core_vcpu_put_pr,
1737         .set_msr     = kvmppc_set_msr_pr,
1738         .vcpu_run    = kvmppc_vcpu_run_pr,
1739         .vcpu_create = kvmppc_core_vcpu_create_pr,
1740         .vcpu_free   = kvmppc_core_vcpu_free_pr,
1741         .check_requests = kvmppc_core_check_requests_pr,
1742         .get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1743         .flush_memslot = kvmppc_core_flush_memslot_pr,
1744         .prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1745         .commit_memory_region = kvmppc_core_commit_memory_region_pr,
1746         .unmap_hva = kvm_unmap_hva_pr,
1747         .unmap_hva_range = kvm_unmap_hva_range_pr,
1748         .age_hva  = kvm_age_hva_pr,
1749         .test_age_hva = kvm_test_age_hva_pr,
1750         .set_spte_hva = kvm_set_spte_hva_pr,
1751         .mmu_destroy  = kvmppc_mmu_destroy_pr,
1752         .free_memslot = kvmppc_core_free_memslot_pr,
1753         .create_memslot = kvmppc_core_create_memslot_pr,
1754         .init_vm = kvmppc_core_init_vm_pr,
1755         .destroy_vm = kvmppc_core_destroy_vm_pr,
1756         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1757         .emulate_op = kvmppc_core_emulate_op_pr,
1758         .emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1759         .emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1760         .fast_vcpu_kick = kvm_vcpu_kick,
1761         .arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1762 #ifdef CONFIG_PPC_BOOK3S_64
1763         .hcall_implemented = kvmppc_hcall_impl_pr,
1764 #endif
1765 };
1766
1767
1768 int kvmppc_book3s_init_pr(void)
1769 {
1770         int r;
1771
1772         r = kvmppc_core_check_processor_compat_pr();
1773         if (r < 0)
1774                 return r;
1775
1776         kvm_ops_pr.owner = THIS_MODULE;
1777         kvmppc_pr_ops = &kvm_ops_pr;
1778
1779         r = kvmppc_mmu_hpte_sysinit();
1780         return r;
1781 }
1782
1783 void kvmppc_book3s_exit_pr(void)
1784 {
1785         kvmppc_pr_ops = NULL;
1786         kvmppc_mmu_hpte_sysexit();
1787 }
1788
1789 /*
1790  * We only support separate modules for book3s 64
1791  */
1792 #ifdef CONFIG_PPC_BOOK3S_64
1793
1794 module_init(kvmppc_book3s_init_pr);
1795 module_exit(kvmppc_book3s_exit_pr);
1796
1797 MODULE_LICENSE("GPL");
1798 MODULE_ALIAS_MISCDEV(KVM_MINOR);
1799 MODULE_ALIAS("devname:kvm");
1800 #endif