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/setup.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         to_book3s(vcpu)->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         case PVR_POWER8E:
 452         case PVR_POWER8NVL:
 453                 vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
 454                         BOOK3S_HFLAG_NEW_TLBIE;
 455                 break;
 456         }
 457
 458 #ifdef CONFIG_PPC_BOOK3S_32
 459         /* 32 bit Book3S always has 32 byte dcbz */
 460         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 461 #endif
 462
 463         /* On some CPUs we can execute paired single operations natively */
 464         asm ( "mfpvr %0" : "=r"(host_pvr));
 465         switch (host_pvr) {
 466         case 0x00080200:        /* lonestar 2.0 */
 467         case 0x00088202:        /* lonestar 2.2 */
 468         case 0x70000100:        /* gekko 1.0 */
 469         case 0x00080100:        /* gekko 2.0 */
 470         case 0x00083203:        /* gekko 2.3a */
 471         case 0x00083213:        /* gekko 2.3b */
 472         case 0x00083204:        /* gekko 2.4 */
 473         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 474         case 0x00087200:        /* broadway */
 475                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 476                 /* Enable HID2.PSE - in case we need it later */
 477                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 478         }
 479 }
 480
 481 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 482  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 483  * emulate 32 bytes dcbz length.
 484  *
 485  * The Book3s_64 inventors also realized this case and implemented a special bit
 486  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 487  *
 488  * My approach here is to patch the dcbz instruction on executing pages.
 489  */
 490 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 491 {
 492         struct page *hpage;
 493         u64 hpage_offset;
 494         u32 *page;
 495         int i;
 496
 497         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 498         if (is_error_page(hpage))
 499                 return;
 500
 501         hpage_offset = pte->raddr & ~PAGE_MASK;
 502         hpage_offset &= ~0xFFFULL;
 503         hpage_offset /= 4;
 504
 505         get_page(hpage);
 506         page = kmap_atomic(hpage);
 507
 508         /* patch dcbz into reserved instruction, so we trap */
 509         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 510                 if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
 511                         page[i] &= cpu_to_be32(0xfffffff7);
 512
 513         kunmap_atomic(page);
 514         put_page(hpage);
 515 }
 516
 517 static bool kvmppc_visible_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
 518 {
 519         ulong mp_pa = vcpu->arch.magic_page_pa;
 520
 521         if (!(kvmppc_get_msr(vcpu) & MSR_SF))
 522                 mp_pa = (uint32_t)mp_pa;
 523
 524         gpa &= ~0xFFFULL;
 525         if (unlikely(mp_pa) && unlikely((mp_pa & KVM_PAM) == (gpa & KVM_PAM))) {
 526                 return true;
 527         }
 528
 529         return kvm_is_visible_gfn(vcpu->kvm, gpa >> PAGE_SHIFT);
 530 }
 531
 532 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 533                             ulong eaddr, int vec)
 534 {
 535         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 536         bool iswrite = false;
 537         int r = RESUME_GUEST;
 538         int relocated;
 539         int page_found = 0;
 540         struct kvmppc_pte pte;
 541         bool is_mmio = false;
 542         bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
 543         bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
 544         u64 vsid;
 545
 546         relocated = data ? dr : ir;
 547         if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
 548                 iswrite = true;
 549
 550         /* Resolve real address if translation turned on */
 551         if (relocated) {
 552                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
 553         } else {
 554                 pte.may_execute = true;
 555                 pte.may_read = true;
 556                 pte.may_write = true;
 557                 pte.raddr = eaddr & KVM_PAM;
 558                 pte.eaddr = eaddr;
 559                 pte.vpage = eaddr >> 12;
 560                 pte.page_size = MMU_PAGE_64K;
 561         }
 562
 563         switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
 564         case 0:
 565                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 566                 break;
 567         case MSR_DR:
 568                 if (!data &&
 569                     (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
 570                     ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
 571                         pte.raddr &= ~SPLIT_HACK_MASK;
 572                 /* fall through */
 573         case MSR_IR:
 574                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 575
 576                 if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
 577                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 578                 else
 579                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 580                 pte.vpage |= vsid;
 581
 582                 if (vsid == -1)
 583                         page_found = -EINVAL;
 584                 break;
 585         }
 586
 587         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 588            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 589                 /*
 590                  * If we do the dcbz hack, we have to NX on every execution,
 591                  * so we can patch the executing code. This renders our guest
 592                  * NX-less.
 593                  */
 594                 pte.may_execute = !data;
 595         }
 596
 597         if (page_found == -ENOENT) {
 598                 /* Page not found in guest PTE entries */
 599                 u64 ssrr1 = vcpu->arch.shadow_srr1;
 600                 u64 msr = kvmppc_get_msr(vcpu);
 601                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 602                 kvmppc_set_dsisr(vcpu, vcpu->arch.fault_dsisr);
 603                 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
 604                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 605         } else if (page_found == -EPERM) {
 606                 /* Storage protection */
 607                 u32 dsisr = vcpu->arch.fault_dsisr;
 608                 u64 ssrr1 = vcpu->arch.shadow_srr1;
 609                 u64 msr = kvmppc_get_msr(vcpu);
 610                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 611                 dsisr = (dsisr & ~DSISR_NOHPTE) | DSISR_PROTFAULT;
 612                 kvmppc_set_dsisr(vcpu, dsisr);
 613                 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
 614                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 615         } else if (page_found == -EINVAL) {
 616                 /* Page not found in guest SLB */
 617                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 618                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 619         } else if (!is_mmio &&
 620                    kvmppc_visible_gpa(vcpu, pte.raddr)) {
 621                 if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
 622                         /*
 623                          * There is already a host HPTE there, presumably
 624                          * a read-only one for a page the guest thinks
 625                          * is writable, so get rid of it first.
 626                          */
 627                         kvmppc_mmu_unmap_page(vcpu, &pte);
 628                 }
 629                 /* The guest's PTE is not mapped yet. Map on the host */
 630                 if (kvmppc_mmu_map_page(vcpu, &pte, iswrite) == -EIO) {
 631                         /* Exit KVM if mapping failed */
 632                         run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 633                         return RESUME_HOST;
 634                 }
 635                 if (data)
 636                         vcpu->stat.sp_storage++;
 637                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 638                          (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 639                         kvmppc_patch_dcbz(vcpu, &pte);
 640         } else {
 641                 /* MMIO */
 642                 vcpu->stat.mmio_exits++;
 643                 vcpu->arch.paddr_accessed = pte.raddr;
 644                 vcpu->arch.vaddr_accessed = pte.eaddr;
 645                 r = kvmppc_emulate_mmio(run, vcpu);
 646                 if ( r == RESUME_HOST_NV )
 647                         r = RESUME_HOST;
 648         }
 649
 650         return r;
 651 }
 652
 653 /* Give up external provider (FPU, Altivec, VSX) */
 654 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 655 {
 656         struct thread_struct *t = &current->thread;
 657
 658         /*
 659          * VSX instructions can access FP and vector registers, so if
 660          * we are giving up VSX, make sure we give up FP and VMX as well.
 661          */
 662         if (msr & MSR_VSX)
 663                 msr |= MSR_FP | MSR_VEC;
 664
 665         msr &= vcpu->arch.guest_owned_ext;
 666         if (!msr)
 667                 return;
 668
 669 #ifdef DEBUG_EXT
 670         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 671 #endif
 672
 673         if (msr & MSR_FP) {
 674                 /*
 675                  * Note that on CPUs with VSX, giveup_fpu stores
 676                  * both the traditional FP registers and the added VSX
 677                  * registers into thread.fp_state.fpr[].
 678                  */
 679                 if (t->regs->msr & MSR_FP)
 680                         giveup_fpu(current);
 681                 t->fp_save_area = NULL;
 682         }
 683
 684 #ifdef CONFIG_ALTIVEC
 685         if (msr & MSR_VEC) {
 686                 if (current->thread.regs->msr & MSR_VEC)
 687                         giveup_altivec(current);
 688                 t->vr_save_area = NULL;
 689         }
 690 #endif
 691
 692         vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
 693         kvmppc_recalc_shadow_msr(vcpu);
 694 }
 695
 696 /* Give up facility (TAR / EBB / DSCR) */
 697 static void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
 698 {
 699 #ifdef CONFIG_PPC_BOOK3S_64
 700         if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
 701                 /* Facility not available to the guest, ignore giveup request*/
 702                 return;
 703         }
 704
 705         switch (fac) {
 706         case FSCR_TAR_LG:
 707                 vcpu->arch.tar = mfspr(SPRN_TAR);
 708                 mtspr(SPRN_TAR, current->thread.tar);
 709                 vcpu->arch.shadow_fscr &= ~FSCR_TAR;
 710                 break;
 711         }
 712 #endif
 713 }
 714
 715 /* Handle external providers (FPU, Altivec, VSX) */
 716 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 717                              ulong msr)
 718 {
 719         struct thread_struct *t = &current->thread;
 720
 721         /* When we have paired singles, we emulate in software */
 722         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 723                 return RESUME_GUEST;
 724
 725         if (!(kvmppc_get_msr(vcpu) & msr)) {
 726                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 727                 return RESUME_GUEST;
 728         }
 729
 730         if (msr == MSR_VSX) {
 731                 /* No VSX?  Give an illegal instruction interrupt */
 732 #ifdef CONFIG_VSX
 733                 if (!cpu_has_feature(CPU_FTR_VSX))
 734 #endif
 735                 {
 736                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 737                         return RESUME_GUEST;
 738                 }
 739
 740                 /*
 741                  * We have to load up all the FP and VMX registers before
 742                  * we can let the guest use VSX instructions.
 743                  */
 744                 msr = MSR_FP | MSR_VEC | MSR_VSX;
 745         }
 746
 747         /* See if we already own all the ext(s) needed */
 748         msr &= ~vcpu->arch.guest_owned_ext;
 749         if (!msr)
 750                 return RESUME_GUEST;
 751
 752 #ifdef DEBUG_EXT
 753         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 754 #endif
 755
 756         if (msr & MSR_FP) {
 757                 preempt_disable();
 758                 enable_kernel_fp();
 759                 load_fp_state(&vcpu->arch.fp);
 760                 disable_kernel_fp();
 761                 t->fp_save_area = &vcpu->arch.fp;
 762                 preempt_enable();
 763         }
 764
 765         if (msr & MSR_VEC) {
 766 #ifdef CONFIG_ALTIVEC
 767                 preempt_disable();
 768                 enable_kernel_altivec();
 769                 load_vr_state(&vcpu->arch.vr);
 770                 disable_kernel_altivec();
 771                 t->vr_save_area = &vcpu->arch.vr;
 772                 preempt_enable();
 773 #endif
 774         }
 775
 776         t->regs->msr |= msr;
 777         vcpu->arch.guest_owned_ext |= msr;
 778         kvmppc_recalc_shadow_msr(vcpu);
 779
 780         return RESUME_GUEST;
 781 }
 782
 783 /*
 784  * Kernel code using FP or VMX could have flushed guest state to
 785  * the thread_struct; if so, get it back now.
 786  */
 787 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
 788 {
 789         unsigned long lost_ext;
 790
 791         lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
 792         if (!lost_ext)
 793                 return;
 794
 795         if (lost_ext & MSR_FP) {
 796                 preempt_disable();
 797                 enable_kernel_fp();
 798                 load_fp_state(&vcpu->arch.fp);
 799                 disable_kernel_fp();
 800                 preempt_enable();
 801         }
 802 #ifdef CONFIG_ALTIVEC
 803         if (lost_ext & MSR_VEC) {
 804                 preempt_disable();
 805                 enable_kernel_altivec();
 806                 load_vr_state(&vcpu->arch.vr);
 807                 disable_kernel_altivec();
 808                 preempt_enable();
 809         }
 810 #endif
 811         current->thread.regs->msr |= lost_ext;
 812 }
 813
 814 #ifdef CONFIG_PPC_BOOK3S_64
 815
 816 static void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
 817 {
 818         /* Inject the Interrupt Cause field and trigger a guest interrupt */
 819         vcpu->arch.fscr &= ~(0xffULL << 56);
 820         vcpu->arch.fscr |= (fac << 56);
 821         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
 822 }
 823
 824 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
 825 {
 826         enum emulation_result er = EMULATE_FAIL;
 827
 828         if (!(kvmppc_get_msr(vcpu) & MSR_PR))
 829                 er = kvmppc_emulate_instruction(vcpu->run, vcpu);
 830
 831         if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
 832                 /* Couldn't emulate, trigger interrupt in guest */
 833                 kvmppc_trigger_fac_interrupt(vcpu, fac);
 834         }
 835 }
 836
 837 /* Enable facilities (TAR, EBB, DSCR) for the guest */
 838 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
 839 {
 840         bool guest_fac_enabled;
 841         BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
 842
 843         /*
 844          * Not every facility is enabled by FSCR bits, check whether the
 845          * guest has this facility enabled at all.
 846          */
 847         switch (fac) {
 848         case FSCR_TAR_LG:
 849         case FSCR_EBB_LG:
 850                 guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
 851                 break;
 852         case FSCR_TM_LG:
 853                 guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
 854                 break;
 855         default:
 856                 guest_fac_enabled = false;
 857                 break;
 858         }
 859
 860         if (!guest_fac_enabled) {
 861                 /* Facility not enabled by the guest */
 862                 kvmppc_trigger_fac_interrupt(vcpu, fac);
 863                 return RESUME_GUEST;
 864         }
 865
 866         switch (fac) {
 867         case FSCR_TAR_LG:
 868                 /* TAR switching isn't lazy in Linux yet */
 869                 current->thread.tar = mfspr(SPRN_TAR);
 870                 mtspr(SPRN_TAR, vcpu->arch.tar);
 871                 vcpu->arch.shadow_fscr |= FSCR_TAR;
 872                 break;
 873         default:
 874                 kvmppc_emulate_fac(vcpu, fac);
 875                 break;
 876         }
 877
 878         return RESUME_GUEST;
 879 }
 880
 881 void kvmppc_set_fscr(struct kvm_vcpu *vcpu, u64 fscr)
 882 {
 883         if ((vcpu->arch.fscr & FSCR_TAR) && !(fscr & FSCR_TAR)) {
 884                 /* TAR got dropped, drop it in shadow too */
 885                 kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
 886         }
 887         vcpu->arch.fscr = fscr;
 888 }
 889 #endif
 890
 891 static void kvmppc_setup_debug(struct kvm_vcpu *vcpu)
 892 {
 893         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
 894                 u64 msr = kvmppc_get_msr(vcpu);
 895
 896                 kvmppc_set_msr(vcpu, msr | MSR_SE);
 897         }
 898 }
 899
 900 static void kvmppc_clear_debug(struct kvm_vcpu *vcpu)
 901 {
 902         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
 903                 u64 msr = kvmppc_get_msr(vcpu);
 904
 905                 kvmppc_set_msr(vcpu, msr & ~MSR_SE);
 906         }
 907 }
 908
 909 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
 910                           unsigned int exit_nr)
 911 {
 912         int r = RESUME_HOST;
 913         int s;
 914
 915         vcpu->stat.sum_exits++;
 916
 917         run->exit_reason = KVM_EXIT_UNKNOWN;
 918         run->ready_for_interrupt_injection = 1;
 919
 920         /* We get here with MSR.EE=1 */
 921
 922         trace_kvm_exit(exit_nr, vcpu);
 923         guest_exit();
 924
 925         switch (exit_nr) {
 926         case BOOK3S_INTERRUPT_INST_STORAGE:
 927         {
 928                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 929                 vcpu->stat.pf_instruc++;
 930
 931                 if (kvmppc_is_split_real(vcpu))
 932                         kvmppc_fixup_split_real(vcpu);
 933
 934 #ifdef CONFIG_PPC_BOOK3S_32
 935                 /* We set segments as unused segments when invalidating them. So
 936                  * treat the respective fault as segment fault. */
 937                 {
 938                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 939                         u32 sr;
 940
 941                         svcpu = svcpu_get(vcpu);
 942                         sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
 943                         svcpu_put(svcpu);
 944                         if (sr == SR_INVALID) {
 945                                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 946                                 r = RESUME_GUEST;
 947                                 break;
 948                         }
 949                 }
 950 #endif
 951
 952                 /* only care about PTEG not found errors, but leave NX alone */
 953                 if (shadow_srr1 & 0x40000000) {
 954                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
 955                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 956                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 957                         vcpu->stat.sp_instruc++;
 958                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 959                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 960                         /*
 961                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 962                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 963                          *     that no guest that needs the dcbz hack does NX.
 964                          */
 965                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 966                         r = RESUME_GUEST;
 967                 } else {
 968                         u64 msr = kvmppc_get_msr(vcpu);
 969                         msr |= shadow_srr1 & 0x58000000;
 970                         kvmppc_set_msr_fast(vcpu, msr);
 971                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 972                         r = RESUME_GUEST;
 973                 }
 974                 break;
 975         }
 976         case BOOK3S_INTERRUPT_DATA_STORAGE:
 977         {
 978                 ulong dar = kvmppc_get_fault_dar(vcpu);
 979                 u32 fault_dsisr = vcpu->arch.fault_dsisr;
 980                 vcpu->stat.pf_storage++;
 981
 982 #ifdef CONFIG_PPC_BOOK3S_32
 983                 /* We set segments as unused segments when invalidating them. So
 984                  * treat the respective fault as segment fault. */
 985                 {
 986                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 987                         u32 sr;
 988
 989                         svcpu = svcpu_get(vcpu);
 990                         sr = svcpu->sr[dar >> SID_SHIFT];
 991                         svcpu_put(svcpu);
 992                         if (sr == SR_INVALID) {
 993                                 kvmppc_mmu_map_segment(vcpu, dar);
 994                                 r = RESUME_GUEST;
 995                                 break;
 996                         }
 997                 }
 998 #endif
 999
1000                 /*
1001                  * We need to handle missing shadow PTEs, and
1002                  * protection faults due to us mapping a page read-only
1003                  * when the guest thinks it is writable.
1004                  */
1005                 if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
1006                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
1007                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
1008                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1009                 } else {
1010                         kvmppc_set_dar(vcpu, dar);
1011                         kvmppc_set_dsisr(vcpu, fault_dsisr);
1012                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1013                         r = RESUME_GUEST;
1014                 }
1015                 break;
1016         }
1017         case BOOK3S_INTERRUPT_DATA_SEGMENT:
1018                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
1019                         kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
1020                         kvmppc_book3s_queue_irqprio(vcpu,
1021                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
1022                 }
1023                 r = RESUME_GUEST;
1024                 break;
1025         case BOOK3S_INTERRUPT_INST_SEGMENT:
1026                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1027                         kvmppc_book3s_queue_irqprio(vcpu,
1028                                 BOOK3S_INTERRUPT_INST_SEGMENT);
1029                 }
1030                 r = RESUME_GUEST;
1031                 break;
1032         /* We're good on these - the host merely wanted to get our attention */
1033         case BOOK3S_INTERRUPT_DECREMENTER:
1034         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1035         case BOOK3S_INTERRUPT_DOORBELL:
1036         case BOOK3S_INTERRUPT_H_DOORBELL:
1037                 vcpu->stat.dec_exits++;
1038                 r = RESUME_GUEST;
1039                 break;
1040         case BOOK3S_INTERRUPT_EXTERNAL:
1041         case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
1042         case BOOK3S_INTERRUPT_EXTERNAL_HV:
1043                 vcpu->stat.ext_intr_exits++;
1044                 r = RESUME_GUEST;
1045                 break;
1046         case BOOK3S_INTERRUPT_PERFMON:
1047                 r = RESUME_GUEST;
1048                 break;
1049         case BOOK3S_INTERRUPT_PROGRAM:
1050         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1051         {
1052                 enum emulation_result er;
1053                 ulong flags;
1054                 u32 last_inst;
1055                 int emul;
1056
1057 program_interrupt:
1058                 /*
1059                  * shadow_srr1 only contains valid flags if we came here via
1060                  * a program exception. The other exceptions (emulation assist,
1061                  * FP unavailable, etc.) do not provide flags in SRR1, so use
1062                  * an illegal-instruction exception when injecting a program
1063                  * interrupt into the guest.
1064                  */
1065                 if (exit_nr == BOOK3S_INTERRUPT_PROGRAM)
1066                         flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
1067                 else
1068                         flags = SRR1_PROGILL;
1069
1070                 emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1071                 if (emul != EMULATE_DONE) {
1072                         r = RESUME_GUEST;
1073                         break;
1074                 }
1075
1076                 if (kvmppc_get_msr(vcpu) & MSR_PR) {
1077 #ifdef EXIT_DEBUG
1078                         pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
1079                                 kvmppc_get_pc(vcpu), last_inst);
1080 #endif
1081                         if ((last_inst & 0xff0007ff) !=
1082                             (INS_DCBZ & 0xfffffff7)) {
1083                                 kvmppc_core_queue_program(vcpu, flags);
1084                                 r = RESUME_GUEST;
1085                                 break;
1086                         }
1087                 }
1088
1089                 vcpu->stat.emulated_inst_exits++;
1090                 er = kvmppc_emulate_instruction(run, vcpu);
1091                 switch (er) {
1092                 case EMULATE_DONE:
1093                         r = RESUME_GUEST_NV;
1094                         break;
1095                 case EMULATE_AGAIN:
1096                         r = RESUME_GUEST;
1097                         break;
1098                 case EMULATE_FAIL:
1099                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
1100                                __func__, kvmppc_get_pc(vcpu), last_inst);
1101                         kvmppc_core_queue_program(vcpu, flags);
1102                         r = RESUME_GUEST;
1103                         break;
1104                 case EMULATE_DO_MMIO:
1105                         run->exit_reason = KVM_EXIT_MMIO;
1106                         r = RESUME_HOST_NV;
1107                         break;
1108                 case EMULATE_EXIT_USER:
1109                         r = RESUME_HOST_NV;
1110                         break;
1111                 default:
1112                         BUG();
1113                 }
1114                 break;
1115         }
1116         case BOOK3S_INTERRUPT_SYSCALL:
1117         {
1118                 u32 last_sc;
1119                 int emul;
1120
1121                 /* Get last sc for papr */
1122                 if (vcpu->arch.papr_enabled) {
1123                         /* The sc instuction points SRR0 to the next inst */
1124                         emul = kvmppc_get_last_inst(vcpu, INST_SC, &last_sc);
1125                         if (emul != EMULATE_DONE) {
1126                                 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) - 4);
1127                                 r = RESUME_GUEST;
1128                                 break;
1129                         }
1130                 }
1131
1132                 if (vcpu->arch.papr_enabled &&
1133                     (last_sc == 0x44000022) &&
1134                     !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1135                         /* SC 1 papr hypercalls */
1136                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
1137                         int i;
1138
1139 #ifdef CONFIG_PPC_BOOK3S_64
1140                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1141                                 r = RESUME_GUEST;
1142                                 break;
1143                         }
1144 #endif
1145
1146                         run->papr_hcall.nr = cmd;
1147                         for (i = 0; i < 9; ++i) {
1148                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1149                                 run->papr_hcall.args[i] = gpr;
1150                         }
1151                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
1152                         vcpu->arch.hcall_needed = 1;
1153                         r = RESUME_HOST;
1154                 } else if (vcpu->arch.osi_enabled &&
1155                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1156                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1157                         /* MOL hypercalls */
1158                         u64 *gprs = run->osi.gprs;
1159                         int i;
1160
1161                         run->exit_reason = KVM_EXIT_OSI;
1162                         for (i = 0; i < 32; i++)
1163                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
1164                         vcpu->arch.osi_needed = 1;
1165                         r = RESUME_HOST_NV;
1166                 } else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1167                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1168                         /* KVM PV hypercalls */
1169                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1170                         r = RESUME_GUEST;
1171                 } else {
1172                         /* Guest syscalls */
1173                         vcpu->stat.syscall_exits++;
1174                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1175                         r = RESUME_GUEST;
1176                 }
1177                 break;
1178         }
1179         case BOOK3S_INTERRUPT_FP_UNAVAIL:
1180         case BOOK3S_INTERRUPT_ALTIVEC:
1181         case BOOK3S_INTERRUPT_VSX:
1182         {
1183                 int ext_msr = 0;
1184                 int emul;
1185                 u32 last_inst;
1186
1187                 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) {
1188                         /* Do paired single instruction emulation */
1189                         emul = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1190                                                     &last_inst);
1191                         if (emul == EMULATE_DONE)
1192                                 goto program_interrupt;
1193                         else
1194                                 r = RESUME_GUEST;
1195
1196                         break;
1197                 }
1198
1199                 /* Enable external provider */
1200                 switch (exit_nr) {
1201                 case BOOK3S_INTERRUPT_FP_UNAVAIL:
1202                         ext_msr = MSR_FP;
1203                         break;
1204
1205                 case BOOK3S_INTERRUPT_ALTIVEC:
1206                         ext_msr = MSR_VEC;
1207                         break;
1208
1209                 case BOOK3S_INTERRUPT_VSX:
1210                         ext_msr = MSR_VSX;
1211                         break;
1212                 }
1213
1214                 r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1215                 break;
1216         }
1217         case BOOK3S_INTERRUPT_ALIGNMENT:
1218         {
1219                 u32 last_inst;
1220                 int emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1221
1222                 if (emul == EMULATE_DONE) {
1223                         u32 dsisr;
1224                         u64 dar;
1225
1226                         dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1227                         dar = kvmppc_alignment_dar(vcpu, last_inst);
1228
1229                         kvmppc_set_dsisr(vcpu, dsisr);
1230                         kvmppc_set_dar(vcpu, dar);
1231
1232                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1233                 }
1234                 r = RESUME_GUEST;
1235                 break;
1236         }
1237 #ifdef CONFIG_PPC_BOOK3S_64
1238         case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1239                 kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1240                 r = RESUME_GUEST;
1241                 break;
1242 #endif
1243         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1244                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1245                 r = RESUME_GUEST;
1246                 break;
1247         case BOOK3S_INTERRUPT_TRACE:
1248                 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1249                         run->exit_reason = KVM_EXIT_DEBUG;
1250                         r = RESUME_HOST;
1251                 } else {
1252                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1253                         r = RESUME_GUEST;
1254                 }
1255                 break;
1256         default:
1257         {
1258                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1259                 /* Ugh - bork here! What did we get? */
1260                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1261                         exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1262                 r = RESUME_HOST;
1263                 BUG();
1264                 break;
1265         }
1266         }
1267
1268         if (!(r & RESUME_HOST)) {
1269                 /* To avoid clobbering exit_reason, only check for signals if
1270                  * we aren't already exiting to userspace for some other
1271                  * reason. */
1272
1273                 /*
1274                  * Interrupts could be timers for the guest which we have to
1275                  * inject again, so let's postpone them until we're in the guest
1276                  * and if we really did time things so badly, then we just exit
1277                  * again due to a host external interrupt.
1278                  */
1279                 s = kvmppc_prepare_to_enter(vcpu);
1280                 if (s <= 0)
1281                         r = s;
1282                 else {
1283                         /* interrupts now hard-disabled */
1284                         kvmppc_fix_ee_before_entry();
1285                 }
1286
1287                 kvmppc_handle_lost_ext(vcpu);
1288         }
1289
1290         trace_kvm_book3s_reenter(r, vcpu);
1291
1292         return r;
1293 }
1294
1295 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1296                                             struct kvm_sregs *sregs)
1297 {
1298         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1299         int i;
1300
1301         sregs->pvr = vcpu->arch.pvr;
1302
1303         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1304         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1305                 for (i = 0; i < 64; i++) {
1306                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1307                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1308                 }
1309         } else {
1310                 for (i = 0; i < 16; i++)
1311                         sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1312
1313                 for (i = 0; i < 8; i++) {
1314                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1315                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1316                 }
1317         }
1318
1319         return 0;
1320 }
1321
1322 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1323                                             struct kvm_sregs *sregs)
1324 {
1325         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1326         int i;
1327
1328         kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1329
1330         vcpu3s->sdr1 = sregs->u.s.sdr1;
1331         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1332                 for (i = 0; i < 64; i++) {
1333                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1334                                                     sregs->u.s.ppc64.slb[i].slbe);
1335                 }
1336         } else {
1337                 for (i = 0; i < 16; i++) {
1338                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1339                 }
1340                 for (i = 0; i < 8; i++) {
1341                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1342                                        (u32)sregs->u.s.ppc32.ibat[i]);
1343                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1344                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1345                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1346                                        (u32)sregs->u.s.ppc32.dbat[i]);
1347                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1348                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1349                 }
1350         }
1351
1352         /* Flush the MMU after messing with the segments */
1353         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1354
1355         return 0;
1356 }
1357
1358 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1359                                  union kvmppc_one_reg *val)
1360 {
1361         int r = 0;
1362
1363         switch (id) {
1364         case KVM_REG_PPC_DEBUG_INST:
1365                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1366                 break;
1367         case KVM_REG_PPC_HIOR:
1368                 *val = get_reg_val(id, to_book3s(vcpu)->hior);
1369                 break;
1370         case KVM_REG_PPC_VTB:
1371                 *val = get_reg_val(id, to_book3s(vcpu)->vtb);
1372                 break;
1373         case KVM_REG_PPC_LPCR:
1374         case KVM_REG_PPC_LPCR_64:
1375                 /*
1376                  * We are only interested in the LPCR_ILE bit
1377                  */
1378                 if (vcpu->arch.intr_msr & MSR_LE)
1379                         *val = get_reg_val(id, LPCR_ILE);
1380                 else
1381                         *val = get_reg_val(id, 0);
1382                 break;
1383         default:
1384                 r = -EINVAL;
1385                 break;
1386         }
1387
1388         return r;
1389 }
1390
1391 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1392 {
1393         if (new_lpcr & LPCR_ILE)
1394                 vcpu->arch.intr_msr |= MSR_LE;
1395         else
1396                 vcpu->arch.intr_msr &= ~MSR_LE;
1397 }
1398
1399 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1400                                  union kvmppc_one_reg *val)
1401 {
1402         int r = 0;
1403
1404         switch (id) {
1405         case KVM_REG_PPC_HIOR:
1406                 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1407                 to_book3s(vcpu)->hior_explicit = true;
1408                 break;
1409         case KVM_REG_PPC_VTB:
1410                 to_book3s(vcpu)->vtb = set_reg_val(id, *val);
1411                 break;
1412         case KVM_REG_PPC_LPCR:
1413         case KVM_REG_PPC_LPCR_64:
1414                 kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1415                 break;
1416         default:
1417                 r = -EINVAL;
1418                 break;
1419         }
1420
1421         return r;
1422 }
1423
1424 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1425                                                    unsigned int id)
1426 {
1427         struct kvmppc_vcpu_book3s *vcpu_book3s;
1428         struct kvm_vcpu *vcpu;
1429         int err = -ENOMEM;
1430         unsigned long p;
1431
1432         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1433         if (!vcpu)
1434                 goto out;
1435
1436         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1437         if (!vcpu_book3s)
1438                 goto free_vcpu;
1439         vcpu->arch.book3s = vcpu_book3s;
1440
1441 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1442         vcpu->arch.shadow_vcpu =
1443                 kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1444         if (!vcpu->arch.shadow_vcpu)
1445                 goto free_vcpu3s;
1446 #endif
1447
1448         err = kvm_vcpu_init(vcpu, kvm, id);
1449         if (err)
1450                 goto free_shadow_vcpu;
1451
1452         err = -ENOMEM;
1453         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1454         if (!p)
1455                 goto uninit_vcpu;
1456         vcpu->arch.shared = (void *)p;
1457 #ifdef CONFIG_PPC_BOOK3S_64
1458         /* Always start the shared struct in native endian mode */
1459 #ifdef __BIG_ENDIAN__
1460         vcpu->arch.shared_big_endian = true;
1461 #else
1462         vcpu->arch.shared_big_endian = false;
1463 #endif
1464
1465         /*
1466          * Default to the same as the host if we're on sufficiently
1467          * recent machine that we have 1TB segments;
1468          * otherwise default to PPC970FX.
1469          */
1470         vcpu->arch.pvr = 0x3C0301;
1471         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1472                 vcpu->arch.pvr = mfspr(SPRN_PVR);
1473         vcpu->arch.intr_msr = MSR_SF;
1474 #else
1475         /* default to book3s_32 (750) */
1476         vcpu->arch.pvr = 0x84202;
1477 #endif
1478         kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1479         vcpu->arch.slb_nr = 64;
1480
1481         vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1482
1483         err = kvmppc_mmu_init(vcpu);
1484         if (err < 0)
1485                 goto free_shared_page;
1486
1487         return vcpu;
1488
1489 free_shared_page:
1490         free_page((unsigned long)vcpu->arch.shared);
1491 uninit_vcpu:
1492         kvm_vcpu_uninit(vcpu);
1493 free_shadow_vcpu:
1494 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1495         kfree(vcpu->arch.shadow_vcpu);
1496 free_vcpu3s:
1497 #endif
1498         vfree(vcpu_book3s);
1499 free_vcpu:
1500         kmem_cache_free(kvm_vcpu_cache, vcpu);
1501 out:
1502         return ERR_PTR(err);
1503 }
1504
1505 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1506 {
1507         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1508
1509         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1510         kvm_vcpu_uninit(vcpu);
1511 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1512         kfree(vcpu->arch.shadow_vcpu);
1513 #endif
1514         vfree(vcpu_book3s);
1515         kmem_cache_free(kvm_vcpu_cache, vcpu);
1516 }
1517
1518 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1519 {
1520         int ret;
1521 #ifdef CONFIG_ALTIVEC
1522         unsigned long uninitialized_var(vrsave);
1523 #endif
1524
1525         /* Check if we can run the vcpu at all */
1526         if (!vcpu->arch.sane) {
1527                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1528                 ret = -EINVAL;
1529                 goto out;
1530         }
1531
1532         kvmppc_setup_debug(vcpu);
1533
1534         /*
1535          * Interrupts could be timers for the guest which we have to inject
1536          * again, so let's postpone them until we're in the guest and if we
1537          * really did time things so badly, then we just exit again due to
1538          * a host external interrupt.
1539          */
1540         ret = kvmppc_prepare_to_enter(vcpu);
1541         if (ret <= 0)
1542                 goto out;
1543         /* interrupts now hard-disabled */
1544
1545         /* Save FPU, Altivec and VSX state */
1546         giveup_all(current);
1547
1548         /* Preload FPU if it's enabled */
1549         if (kvmppc_get_msr(vcpu) & MSR_FP)
1550                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1551
1552         kvmppc_fix_ee_before_entry();
1553
1554         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1555
1556         kvmppc_clear_debug(vcpu);
1557
1558         /* No need for guest_exit. It's done in handle_exit.
1559            We also get here with interrupts enabled. */
1560
1561         /* Make sure we save the guest FPU/Altivec/VSX state */
1562         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1563
1564         /* Make sure we save the guest TAR/EBB/DSCR state */
1565         kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1566
1567 out:
1568         vcpu->mode = OUTSIDE_GUEST_MODE;
1569         return ret;
1570 }
1571
1572 /*
1573  * Get (and clear) the dirty memory log for a memory slot.
1574  */
1575 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1576                                          struct kvm_dirty_log *log)
1577 {
1578         struct kvm_memslots *slots;
1579         struct kvm_memory_slot *memslot;
1580         struct kvm_vcpu *vcpu;
1581         ulong ga, ga_end;
1582         int is_dirty = 0;
1583         int r;
1584         unsigned long n;
1585
1586         mutex_lock(&kvm->slots_lock);
1587
1588         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1589         if (r)
1590                 goto out;
1591
1592         /* If nothing is dirty, don't bother messing with page tables. */
1593         if (is_dirty) {
1594                 slots = kvm_memslots(kvm);
1595                 memslot = id_to_memslot(slots, log->slot);
1596
1597                 ga = memslot->base_gfn << PAGE_SHIFT;
1598                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1599
1600                 kvm_for_each_vcpu(n, vcpu, kvm)
1601                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1602
1603                 n = kvm_dirty_bitmap_bytes(memslot);
1604                 memset(memslot->dirty_bitmap, 0, n);
1605         }
1606
1607         r = 0;
1608 out:
1609         mutex_unlock(&kvm->slots_lock);
1610         return r;
1611 }
1612
1613 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1614                                          struct kvm_memory_slot *memslot)
1615 {
1616         return;
1617 }
1618
1619 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1620                                         struct kvm_memory_slot *memslot,
1621                                         const struct kvm_userspace_memory_region *mem)
1622 {
1623         return 0;
1624 }
1625
1626 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1627                                 const struct kvm_userspace_memory_region *mem,
1628                                 const struct kvm_memory_slot *old,
1629                                 const struct kvm_memory_slot *new)
1630 {
1631         return;
1632 }
1633
1634 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1635                                         struct kvm_memory_slot *dont)
1636 {
1637         return;
1638 }
1639
1640 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1641                                          unsigned long npages)
1642 {
1643         return 0;
1644 }
1645
1646
1647 #ifdef CONFIG_PPC64
1648 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1649                                          struct kvm_ppc_smmu_info *info)
1650 {
1651         long int i;
1652         struct kvm_vcpu *vcpu;
1653
1654         info->flags = 0;
1655
1656         /* SLB is always 64 entries */
1657         info->slb_size = 64;
1658
1659         /* Standard 4k base page size segment */
1660         info->sps[0].page_shift = 12;
1661         info->sps[0].slb_enc = 0;
1662         info->sps[0].enc[0].page_shift = 12;
1663         info->sps[0].enc[0].pte_enc = 0;
1664
1665         /*
1666          * 64k large page size.
1667          * We only want to put this in if the CPUs we're emulating
1668          * support it, but unfortunately we don't have a vcpu easily
1669          * to hand here to test.  Just pick the first vcpu, and if
1670          * that doesn't exist yet, report the minimum capability,
1671          * i.e., no 64k pages.
1672          * 1T segment support goes along with 64k pages.
1673          */
1674         i = 1;
1675         vcpu = kvm_get_vcpu(kvm, 0);
1676         if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1677                 info->flags = KVM_PPC_1T_SEGMENTS;
1678                 info->sps[i].page_shift = 16;
1679                 info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1680                 info->sps[i].enc[0].page_shift = 16;
1681                 info->sps[i].enc[0].pte_enc = 1;
1682                 ++i;
1683         }
1684
1685         /* Standard 16M large page size segment */
1686         info->sps[i].page_shift = 24;
1687         info->sps[i].slb_enc = SLB_VSID_L;
1688         info->sps[i].enc[0].page_shift = 24;
1689         info->sps[i].enc[0].pte_enc = 0;
1690
1691         return 0;
1692 }
1693 #else
1694 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1695                                          struct kvm_ppc_smmu_info *info)
1696 {
1697         /* We should not get called */
1698         BUG();
1699 }
1700 #endif /* CONFIG_PPC64 */
1701
1702 static unsigned int kvm_global_user_count = 0;
1703 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1704
1705 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1706 {
1707         mutex_init(&kvm->arch.hpt_mutex);
1708
1709 #ifdef CONFIG_PPC_BOOK3S_64
1710         /* Start out with the default set of hcalls enabled */
1711         kvmppc_pr_init_default_hcalls(kvm);
1712 #endif
1713
1714         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1715                 spin_lock(&kvm_global_user_count_lock);
1716                 if (++kvm_global_user_count == 1)
1717                         pseries_disable_reloc_on_exc();
1718                 spin_unlock(&kvm_global_user_count_lock);
1719         }
1720         return 0;
1721 }
1722
1723 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1724 {
1725 #ifdef CONFIG_PPC64
1726         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1727 #endif
1728
1729         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1730                 spin_lock(&kvm_global_user_count_lock);
1731                 BUG_ON(kvm_global_user_count == 0);
1732                 if (--kvm_global_user_count == 0)
1733                         pseries_enable_reloc_on_exc();
1734                 spin_unlock(&kvm_global_user_count_lock);
1735         }
1736 }
1737
1738 static int kvmppc_core_check_processor_compat_pr(void)
1739 {
1740         /*
1741          * Disable KVM for Power9 untill the required bits merged.
1742          */
1743         if (cpu_has_feature(CPU_FTR_ARCH_300))
1744                 return -EIO;
1745         return 0;
1746 }
1747
1748 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1749                                  unsigned int ioctl, unsigned long arg)
1750 {
1751         return -ENOTTY;
1752 }
1753
1754 static struct kvmppc_ops kvm_ops_pr = {
1755         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1756         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1757         .get_one_reg = kvmppc_get_one_reg_pr,
1758         .set_one_reg = kvmppc_set_one_reg_pr,
1759         .vcpu_load   = kvmppc_core_vcpu_load_pr,
1760         .vcpu_put    = kvmppc_core_vcpu_put_pr,
1761         .set_msr     = kvmppc_set_msr_pr,
1762         .vcpu_run    = kvmppc_vcpu_run_pr,
1763         .vcpu_create = kvmppc_core_vcpu_create_pr,
1764         .vcpu_free   = kvmppc_core_vcpu_free_pr,
1765         .check_requests = kvmppc_core_check_requests_pr,
1766         .get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1767         .flush_memslot = kvmppc_core_flush_memslot_pr,
1768         .prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1769         .commit_memory_region = kvmppc_core_commit_memory_region_pr,
1770         .unmap_hva = kvm_unmap_hva_pr,
1771         .unmap_hva_range = kvm_unmap_hva_range_pr,
1772         .age_hva  = kvm_age_hva_pr,
1773         .test_age_hva = kvm_test_age_hva_pr,
1774         .set_spte_hva = kvm_set_spte_hva_pr,
1775         .mmu_destroy  = kvmppc_mmu_destroy_pr,
1776         .free_memslot = kvmppc_core_free_memslot_pr,
1777         .create_memslot = kvmppc_core_create_memslot_pr,
1778         .init_vm = kvmppc_core_init_vm_pr,
1779         .destroy_vm = kvmppc_core_destroy_vm_pr,
1780         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1781         .emulate_op = kvmppc_core_emulate_op_pr,
1782         .emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1783         .emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1784         .fast_vcpu_kick = kvm_vcpu_kick,
1785         .arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1786 #ifdef CONFIG_PPC_BOOK3S_64
1787         .hcall_implemented = kvmppc_hcall_impl_pr,
1788 #endif
1789 };
1790
1791
1792 int kvmppc_book3s_init_pr(void)
1793 {
1794         int r;
1795
1796         r = kvmppc_core_check_processor_compat_pr();
1797         if (r < 0)
1798                 return r;
1799
1800         kvm_ops_pr.owner = THIS_MODULE;
1801         kvmppc_pr_ops = &kvm_ops_pr;
1802
1803         r = kvmppc_mmu_hpte_sysinit();
1804         return r;
1805 }
1806
1807 void kvmppc_book3s_exit_pr(void)
1808 {
1809         kvmppc_pr_ops = NULL;
1810         kvmppc_mmu_hpte_sysexit();
1811 }
1812
1813 /*
1814  * We only support separate modules for book3s 64
1815  */
1816 #ifdef CONFIG_PPC_BOOK3S_64
1817
1818 module_init(kvmppc_book3s_init_pr);
1819 module_exit(kvmppc_book3s_exit_pr);
1820
1821 MODULE_LICENSE("GPL");
1822 MODULE_ALIAS_MISCDEV(KVM_MINOR);
1823 MODULE_ALIAS("devname:kvm");
1824 #endif