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 <linux/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 #define HPTE_R_M   _PAGE_COHERENT
  64 #endif
  65
  66 static bool kvmppc_is_split_real(struct kvm_vcpu *vcpu)
  67 {
  68         ulong msr = kvmppc_get_msr(vcpu);
  69         return (msr & (MSR_IR|MSR_DR)) == MSR_DR;
  70 }
  71
  72 static void kvmppc_fixup_split_real(struct kvm_vcpu *vcpu)
  73 {
  74         ulong msr = kvmppc_get_msr(vcpu);
  75         ulong pc = kvmppc_get_pc(vcpu);
  76
  77         /* We are in DR only split real mode */
  78         if ((msr & (MSR_IR|MSR_DR)) != MSR_DR)
  79                 return;
  80
  81         /* We have not fixed up the guest already */
  82         if (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK)
  83                 return;
  84
  85         /* The code is in fixupable address space */
  86         if (pc & SPLIT_HACK_MASK)
  87                 return;
  88
  89         vcpu->arch.hflags |= BOOK3S_HFLAG_SPLIT_HACK;
  90         kvmppc_set_pc(vcpu, pc | SPLIT_HACK_OFFS);
  91 }
  92
  93 void kvmppc_unfixup_split_real(struct kvm_vcpu *vcpu);
  94
  95 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
  96 {
  97 #ifdef CONFIG_PPC_BOOK3S_64
  98         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  99         memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
 100         svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
 101         svcpu->in_use = 0;
 102         svcpu_put(svcpu);
 103 #endif
 104
 105         /* Disable AIL if supported */
 106         if (cpu_has_feature(CPU_FTR_HVMODE) &&
 107             cpu_has_feature(CPU_FTR_ARCH_207S))
 108                 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) & ~LPCR_AIL);
 109
 110         vcpu->cpu = smp_processor_id();
 111 #ifdef CONFIG_PPC_BOOK3S_32
 112         current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
 113 #endif
 114
 115         if (kvmppc_is_split_real(vcpu))
 116                 kvmppc_fixup_split_real(vcpu);
 117 }
 118
 119 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
 120 {
 121 #ifdef CONFIG_PPC_BOOK3S_64
 122         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 123         if (svcpu->in_use) {
 124                 kvmppc_copy_from_svcpu(vcpu);
 125         }
 126         memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
 127         to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
 128         svcpu_put(svcpu);
 129 #endif
 130
 131         if (kvmppc_is_split_real(vcpu))
 132                 kvmppc_unfixup_split_real(vcpu);
 133
 134         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
 135         kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
 136
 137         /* Enable AIL if supported */
 138         if (cpu_has_feature(CPU_FTR_HVMODE) &&
 139             cpu_has_feature(CPU_FTR_ARCH_207S))
 140                 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_AIL_3);
 141
 142         vcpu->cpu = -1;
 143 }
 144
 145 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
 146 void kvmppc_copy_to_svcpu(struct kvm_vcpu *vcpu)
 147 {
 148         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 149
 150         svcpu->gpr[0] = vcpu->arch.gpr[0];
 151         svcpu->gpr[1] = vcpu->arch.gpr[1];
 152         svcpu->gpr[2] = vcpu->arch.gpr[2];
 153         svcpu->gpr[3] = vcpu->arch.gpr[3];
 154         svcpu->gpr[4] = vcpu->arch.gpr[4];
 155         svcpu->gpr[5] = vcpu->arch.gpr[5];
 156         svcpu->gpr[6] = vcpu->arch.gpr[6];
 157         svcpu->gpr[7] = vcpu->arch.gpr[7];
 158         svcpu->gpr[8] = vcpu->arch.gpr[8];
 159         svcpu->gpr[9] = vcpu->arch.gpr[9];
 160         svcpu->gpr[10] = vcpu->arch.gpr[10];
 161         svcpu->gpr[11] = vcpu->arch.gpr[11];
 162         svcpu->gpr[12] = vcpu->arch.gpr[12];
 163         svcpu->gpr[13] = vcpu->arch.gpr[13];
 164         svcpu->cr  = vcpu->arch.cr;
 165         svcpu->xer = vcpu->arch.xer;
 166         svcpu->ctr = vcpu->arch.ctr;
 167         svcpu->lr  = vcpu->arch.lr;
 168         svcpu->pc  = vcpu->arch.pc;
 169 #ifdef CONFIG_PPC_BOOK3S_64
 170         svcpu->shadow_fscr = vcpu->arch.shadow_fscr;
 171 #endif
 172         /*
 173          * Now also save the current time base value. We use this
 174          * to find the guest purr and spurr value.
 175          */
 176         vcpu->arch.entry_tb = get_tb();
 177         vcpu->arch.entry_vtb = get_vtb();
 178         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 179                 vcpu->arch.entry_ic = mfspr(SPRN_IC);
 180         svcpu->in_use = true;
 181
 182         svcpu_put(svcpu);
 183 }
 184
 185 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
 186 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu)
 187 {
 188         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 189
 190         /*
 191          * Maybe we were already preempted and synced the svcpu from
 192          * our preempt notifiers. Don't bother touching this svcpu then.
 193          */
 194         if (!svcpu->in_use)
 195                 goto out;
 196
 197         vcpu->arch.gpr[0] = svcpu->gpr[0];
 198         vcpu->arch.gpr[1] = svcpu->gpr[1];
 199         vcpu->arch.gpr[2] = svcpu->gpr[2];
 200         vcpu->arch.gpr[3] = svcpu->gpr[3];
 201         vcpu->arch.gpr[4] = svcpu->gpr[4];
 202         vcpu->arch.gpr[5] = svcpu->gpr[5];
 203         vcpu->arch.gpr[6] = svcpu->gpr[6];
 204         vcpu->arch.gpr[7] = svcpu->gpr[7];
 205         vcpu->arch.gpr[8] = svcpu->gpr[8];
 206         vcpu->arch.gpr[9] = svcpu->gpr[9];
 207         vcpu->arch.gpr[10] = svcpu->gpr[10];
 208         vcpu->arch.gpr[11] = svcpu->gpr[11];
 209         vcpu->arch.gpr[12] = svcpu->gpr[12];
 210         vcpu->arch.gpr[13] = svcpu->gpr[13];
 211         vcpu->arch.cr  = svcpu->cr;
 212         vcpu->arch.xer = svcpu->xer;
 213         vcpu->arch.ctr = svcpu->ctr;
 214         vcpu->arch.lr  = svcpu->lr;
 215         vcpu->arch.pc  = svcpu->pc;
 216         vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
 217         vcpu->arch.fault_dar   = svcpu->fault_dar;
 218         vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
 219         vcpu->arch.last_inst   = svcpu->last_inst;
 220 #ifdef CONFIG_PPC_BOOK3S_64
 221         vcpu->arch.shadow_fscr = svcpu->shadow_fscr;
 222 #endif
 223         /*
 224          * Update purr and spurr using time base on exit.
 225          */
 226         vcpu->arch.purr += get_tb() - vcpu->arch.entry_tb;
 227         vcpu->arch.spurr += get_tb() - vcpu->arch.entry_tb;
 228         to_book3s(vcpu)->vtb += get_vtb() - vcpu->arch.entry_vtb;
 229         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 230                 vcpu->arch.ic += mfspr(SPRN_IC) - vcpu->arch.entry_ic;
 231         svcpu->in_use = false;
 232
 233 out:
 234         svcpu_put(svcpu);
 235 }
 236
 237 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
 238 {
 239         int r = 1; /* Indicate we want to get back into the guest */
 240
 241         /* We misuse TLB_FLUSH to indicate that we want to clear
 242            all shadow cache entries */
 243         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
 244                 kvmppc_mmu_pte_flush(vcpu, 0, 0);
 245
 246         return r;
 247 }
 248
 249 /************* MMU Notifiers *************/
 250 static void do_kvm_unmap_hva(struct kvm *kvm, unsigned long start,
 251                              unsigned long end)
 252 {
 253         long i;
 254         struct kvm_vcpu *vcpu;
 255         struct kvm_memslots *slots;
 256         struct kvm_memory_slot *memslot;
 257
 258         slots = kvm_memslots(kvm);
 259         kvm_for_each_memslot(memslot, slots) {
 260                 unsigned long hva_start, hva_end;
 261                 gfn_t gfn, gfn_end;
 262
 263                 hva_start = max(start, memslot->userspace_addr);
 264                 hva_end = min(end, memslot->userspace_addr +
 265                                         (memslot->npages << PAGE_SHIFT));
 266                 if (hva_start >= hva_end)
 267                         continue;
 268                 /*
 269                  * {gfn(page) | page intersects with [hva_start, hva_end)} =
 270                  * {gfn, gfn+1, ..., gfn_end-1}.
 271                  */
 272                 gfn = hva_to_gfn_memslot(hva_start, memslot);
 273                 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
 274                 kvm_for_each_vcpu(i, vcpu, kvm)
 275                         kvmppc_mmu_pte_pflush(vcpu, gfn << PAGE_SHIFT,
 276                                               gfn_end << PAGE_SHIFT);
 277         }
 278 }
 279
 280 static int kvm_unmap_hva_range_pr(struct kvm *kvm, unsigned long start,
 281                                   unsigned long end)
 282 {
 283         do_kvm_unmap_hva(kvm, start, end);
 284
 285         return 0;
 286 }
 287
 288 static int kvm_age_hva_pr(struct kvm *kvm, unsigned long start,
 289                           unsigned long end)
 290 {
 291         /* XXX could be more clever ;) */
 292         return 0;
 293 }
 294
 295 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
 296 {
 297         /* XXX could be more clever ;) */
 298         return 0;
 299 }
 300
 301 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
 302 {
 303         /* The page will get remapped properly on its next fault */
 304         do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
 305 }
 306
 307 /*****************************************/
 308
 309 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 310 {
 311         ulong guest_msr = kvmppc_get_msr(vcpu);
 312         ulong smsr = guest_msr;
 313
 314         /* Guest MSR values */
 315         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_LE;
 316         /* Process MSR values */
 317         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 318         /* External providers the guest reserved */
 319         smsr |= (guest_msr & vcpu->arch.guest_owned_ext);
 320         /* 64-bit Process MSR values */
 321 #ifdef CONFIG_PPC_BOOK3S_64
 322         smsr |= MSR_ISF | MSR_HV;
 323 #endif
 324         vcpu->arch.shadow_msr = smsr;
 325 }
 326
 327 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
 328 {
 329         ulong old_msr = kvmppc_get_msr(vcpu);
 330
 331 #ifdef EXIT_DEBUG
 332         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 333 #endif
 334
 335         msr &= to_book3s(vcpu)->msr_mask;
 336         kvmppc_set_msr_fast(vcpu, msr);
 337         kvmppc_recalc_shadow_msr(vcpu);
 338
 339         if (msr & MSR_POW) {
 340                 if (!vcpu->arch.pending_exceptions) {
 341                         kvm_vcpu_block(vcpu);
 342                         kvm_clear_request(KVM_REQ_UNHALT, vcpu);
 343                         vcpu->stat.halt_wakeup++;
 344
 345                         /* Unset POW bit after we woke up */
 346                         msr &= ~MSR_POW;
 347                         kvmppc_set_msr_fast(vcpu, msr);
 348                 }
 349         }
 350
 351         if (kvmppc_is_split_real(vcpu))
 352                 kvmppc_fixup_split_real(vcpu);
 353         else
 354                 kvmppc_unfixup_split_real(vcpu);
 355
 356         if ((kvmppc_get_msr(vcpu) & (MSR_PR|MSR_IR|MSR_DR)) !=
 357                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 358                 kvmppc_mmu_flush_segments(vcpu);
 359                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 360
 361                 /* Preload magic page segment when in kernel mode */
 362                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 363                         struct kvm_vcpu_arch *a = &vcpu->arch;
 364
 365                         if (msr & MSR_DR)
 366                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 367                         else
 368                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 369                 }
 370         }
 371
 372         /*
 373          * When switching from 32 to 64-bit, we may have a stale 32-bit
 374          * magic page around, we need to flush it. Typically 32-bit magic
 375          * page will be instanciated when calling into RTAS. Note: We
 376          * assume that such transition only happens while in kernel mode,
 377          * ie, we never transition from user 32-bit to kernel 64-bit with
 378          * a 32-bit magic page around.
 379          */
 380         if (vcpu->arch.magic_page_pa &&
 381             !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
 382                 /* going from RTAS to normal kernel code */
 383                 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
 384                                      ~0xFFFUL);
 385         }
 386
 387         /* Preload FPU if it's enabled */
 388         if (kvmppc_get_msr(vcpu) & MSR_FP)
 389                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 390 }
 391
 392 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
 393 {
 394         u32 host_pvr;
 395
 396         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 397         vcpu->arch.pvr = pvr;
 398 #ifdef CONFIG_PPC_BOOK3S_64
 399         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 400                 kvmppc_mmu_book3s_64_init(vcpu);
 401                 if (!to_book3s(vcpu)->hior_explicit)
 402                         to_book3s(vcpu)->hior = 0xfff00000;
 403                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 404                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 405         } else
 406 #endif
 407         {
 408                 kvmppc_mmu_book3s_32_init(vcpu);
 409                 if (!to_book3s(vcpu)->hior_explicit)
 410                         to_book3s(vcpu)->hior = 0;
 411                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 412                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 413         }
 414
 415         kvmppc_sanity_check(vcpu);
 416
 417         /* If we are in hypervisor level on 970, we can tell the CPU to
 418          * treat DCBZ as 32 bytes store */
 419         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 420         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 421             !strcmp(cur_cpu_spec->platform, "ppc970"))
 422                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 423
 424         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 425            really needs them in a VM on Cell and force disable them. */
 426         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 427                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 428
 429         /*
 430          * If they're asking for POWER6 or later, set the flag
 431          * indicating that we can do multiple large page sizes
 432          * and 1TB segments.
 433          * Also set the flag that indicates that tlbie has the large
 434          * page bit in the RB operand instead of the instruction.
 435          */
 436         switch (PVR_VER(pvr)) {
 437         case PVR_POWER6:
 438         case PVR_POWER7:
 439         case PVR_POWER7p:
 440         case PVR_POWER8:
 441         case PVR_POWER8E:
 442         case PVR_POWER8NVL:
 443                 vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
 444                         BOOK3S_HFLAG_NEW_TLBIE;
 445                 break;
 446         }
 447
 448 #ifdef CONFIG_PPC_BOOK3S_32
 449         /* 32 bit Book3S always has 32 byte dcbz */
 450         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 451 #endif
 452
 453         /* On some CPUs we can execute paired single operations natively */
 454         asm ( "mfpvr %0" : "=r"(host_pvr));
 455         switch (host_pvr) {
 456         case 0x00080200:        /* lonestar 2.0 */
 457         case 0x00088202:        /* lonestar 2.2 */
 458         case 0x70000100:        /* gekko 1.0 */
 459         case 0x00080100:        /* gekko 2.0 */
 460         case 0x00083203:        /* gekko 2.3a */
 461         case 0x00083213:        /* gekko 2.3b */
 462         case 0x00083204:        /* gekko 2.4 */
 463         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 464         case 0x00087200:        /* broadway */
 465                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 466                 /* Enable HID2.PSE - in case we need it later */
 467                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 468         }
 469 }
 470
 471 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 472  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 473  * emulate 32 bytes dcbz length.
 474  *
 475  * The Book3s_64 inventors also realized this case and implemented a special bit
 476  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 477  *
 478  * My approach here is to patch the dcbz instruction on executing pages.
 479  */
 480 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 481 {
 482         struct page *hpage;
 483         u64 hpage_offset;
 484         u32 *page;
 485         int i;
 486
 487         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 488         if (is_error_page(hpage))
 489                 return;
 490
 491         hpage_offset = pte->raddr & ~PAGE_MASK;
 492         hpage_offset &= ~0xFFFULL;
 493         hpage_offset /= 4;
 494
 495         get_page(hpage);
 496         page = kmap_atomic(hpage);
 497
 498         /* patch dcbz into reserved instruction, so we trap */
 499         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 500                 if ((be32_to_cpu(page[i]) & 0xff0007ff) == INS_DCBZ)
 501                         page[i] &= cpu_to_be32(0xfffffff7);
 502
 503         kunmap_atomic(page);
 504         put_page(hpage);
 505 }
 506
 507 static bool kvmppc_visible_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
 508 {
 509         ulong mp_pa = vcpu->arch.magic_page_pa;
 510
 511         if (!(kvmppc_get_msr(vcpu) & MSR_SF))
 512                 mp_pa = (uint32_t)mp_pa;
 513
 514         gpa &= ~0xFFFULL;
 515         if (unlikely(mp_pa) && unlikely((mp_pa & KVM_PAM) == (gpa & KVM_PAM))) {
 516                 return true;
 517         }
 518
 519         return kvm_is_visible_gfn(vcpu->kvm, gpa >> PAGE_SHIFT);
 520 }
 521
 522 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 523                             ulong eaddr, int vec)
 524 {
 525         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 526         bool iswrite = false;
 527         int r = RESUME_GUEST;
 528         int relocated;
 529         int page_found = 0;
 530         struct kvmppc_pte pte = { 0 };
 531         bool dr = (kvmppc_get_msr(vcpu) & MSR_DR) ? true : false;
 532         bool ir = (kvmppc_get_msr(vcpu) & MSR_IR) ? true : false;
 533         u64 vsid;
 534
 535         relocated = data ? dr : ir;
 536         if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
 537                 iswrite = true;
 538
 539         /* Resolve real address if translation turned on */
 540         if (relocated) {
 541                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
 542         } else {
 543                 pte.may_execute = true;
 544                 pte.may_read = true;
 545                 pte.may_write = true;
 546                 pte.raddr = eaddr & KVM_PAM;
 547                 pte.eaddr = eaddr;
 548                 pte.vpage = eaddr >> 12;
 549                 pte.page_size = MMU_PAGE_64K;
 550                 pte.wimg = HPTE_R_M;
 551         }
 552
 553         switch (kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) {
 554         case 0:
 555                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 556                 break;
 557         case MSR_DR:
 558                 if (!data &&
 559                     (vcpu->arch.hflags & BOOK3S_HFLAG_SPLIT_HACK) &&
 560                     ((pte.raddr & SPLIT_HACK_MASK) == SPLIT_HACK_OFFS))
 561                         pte.raddr &= ~SPLIT_HACK_MASK;
 562                 /* fall through */
 563         case MSR_IR:
 564                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 565
 566                 if ((kvmppc_get_msr(vcpu) & (MSR_DR|MSR_IR)) == MSR_DR)
 567                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 568                 else
 569                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 570                 pte.vpage |= vsid;
 571
 572                 if (vsid == -1)
 573                         page_found = -EINVAL;
 574                 break;
 575         }
 576
 577         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 578            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 579                 /*
 580                  * If we do the dcbz hack, we have to NX on every execution,
 581                  * so we can patch the executing code. This renders our guest
 582                  * NX-less.
 583                  */
 584                 pte.may_execute = !data;
 585         }
 586
 587         if (page_found == -ENOENT) {
 588                 /* Page not found in guest PTE entries */
 589                 u64 ssrr1 = vcpu->arch.shadow_srr1;
 590                 u64 msr = kvmppc_get_msr(vcpu);
 591                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 592                 kvmppc_set_dsisr(vcpu, vcpu->arch.fault_dsisr);
 593                 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
 594                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 595         } else if (page_found == -EPERM) {
 596                 /* Storage protection */
 597                 u32 dsisr = vcpu->arch.fault_dsisr;
 598                 u64 ssrr1 = vcpu->arch.shadow_srr1;
 599                 u64 msr = kvmppc_get_msr(vcpu);
 600                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 601                 dsisr = (dsisr & ~DSISR_NOHPTE) | DSISR_PROTFAULT;
 602                 kvmppc_set_dsisr(vcpu, dsisr);
 603                 kvmppc_set_msr_fast(vcpu, msr | (ssrr1 & 0xf8000000ULL));
 604                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 605         } else if (page_found == -EINVAL) {
 606                 /* Page not found in guest SLB */
 607                 kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
 608                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 609         } else if (kvmppc_visible_gpa(vcpu, pte.raddr)) {
 610                 if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
 611                         /*
 612                          * There is already a host HPTE there, presumably
 613                          * a read-only one for a page the guest thinks
 614                          * is writable, so get rid of it first.
 615                          */
 616                         kvmppc_mmu_unmap_page(vcpu, &pte);
 617                 }
 618                 /* The guest's PTE is not mapped yet. Map on the host */
 619                 if (kvmppc_mmu_map_page(vcpu, &pte, iswrite) == -EIO) {
 620                         /* Exit KVM if mapping failed */
 621                         run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 622                         return RESUME_HOST;
 623                 }
 624                 if (data)
 625                         vcpu->stat.sp_storage++;
 626                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 627                          (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 628                         kvmppc_patch_dcbz(vcpu, &pte);
 629         } else {
 630                 /* MMIO */
 631                 vcpu->stat.mmio_exits++;
 632                 vcpu->arch.paddr_accessed = pte.raddr;
 633                 vcpu->arch.vaddr_accessed = pte.eaddr;
 634                 r = kvmppc_emulate_mmio(run, vcpu);
 635                 if ( r == RESUME_HOST_NV )
 636                         r = RESUME_HOST;
 637         }
 638
 639         return r;
 640 }
 641
 642 /* Give up external provider (FPU, Altivec, VSX) */
 643 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 644 {
 645         struct thread_struct *t = &current->thread;
 646
 647         /*
 648          * VSX instructions can access FP and vector registers, so if
 649          * we are giving up VSX, make sure we give up FP and VMX as well.
 650          */
 651         if (msr & MSR_VSX)
 652                 msr |= MSR_FP | MSR_VEC;
 653
 654         msr &= vcpu->arch.guest_owned_ext;
 655         if (!msr)
 656                 return;
 657
 658 #ifdef DEBUG_EXT
 659         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 660 #endif
 661
 662         if (msr & MSR_FP) {
 663                 /*
 664                  * Note that on CPUs with VSX, giveup_fpu stores
 665                  * both the traditional FP registers and the added VSX
 666                  * registers into thread.fp_state.fpr[].
 667                  */
 668                 if (t->regs->msr & MSR_FP)
 669                         giveup_fpu(current);
 670                 t->fp_save_area = NULL;
 671         }
 672
 673 #ifdef CONFIG_ALTIVEC
 674         if (msr & MSR_VEC) {
 675                 if (current->thread.regs->msr & MSR_VEC)
 676                         giveup_altivec(current);
 677                 t->vr_save_area = NULL;
 678         }
 679 #endif
 680
 681         vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
 682         kvmppc_recalc_shadow_msr(vcpu);
 683 }
 684
 685 /* Give up facility (TAR / EBB / DSCR) */
 686 static void kvmppc_giveup_fac(struct kvm_vcpu *vcpu, ulong fac)
 687 {
 688 #ifdef CONFIG_PPC_BOOK3S_64
 689         if (!(vcpu->arch.shadow_fscr & (1ULL << fac))) {
 690                 /* Facility not available to the guest, ignore giveup request*/
 691                 return;
 692         }
 693
 694         switch (fac) {
 695         case FSCR_TAR_LG:
 696                 vcpu->arch.tar = mfspr(SPRN_TAR);
 697                 mtspr(SPRN_TAR, current->thread.tar);
 698                 vcpu->arch.shadow_fscr &= ~FSCR_TAR;
 699                 break;
 700         }
 701 #endif
 702 }
 703
 704 /* Handle external providers (FPU, Altivec, VSX) */
 705 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 706                              ulong msr)
 707 {
 708         struct thread_struct *t = &current->thread;
 709
 710         /* When we have paired singles, we emulate in software */
 711         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 712                 return RESUME_GUEST;
 713
 714         if (!(kvmppc_get_msr(vcpu) & msr)) {
 715                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 716                 return RESUME_GUEST;
 717         }
 718
 719         if (msr == MSR_VSX) {
 720                 /* No VSX?  Give an illegal instruction interrupt */
 721 #ifdef CONFIG_VSX
 722                 if (!cpu_has_feature(CPU_FTR_VSX))
 723 #endif
 724                 {
 725                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 726                         return RESUME_GUEST;
 727                 }
 728
 729                 /*
 730                  * We have to load up all the FP and VMX registers before
 731                  * we can let the guest use VSX instructions.
 732                  */
 733                 msr = MSR_FP | MSR_VEC | MSR_VSX;
 734         }
 735
 736         /* See if we already own all the ext(s) needed */
 737         msr &= ~vcpu->arch.guest_owned_ext;
 738         if (!msr)
 739                 return RESUME_GUEST;
 740
 741 #ifdef DEBUG_EXT
 742         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 743 #endif
 744
 745         if (msr & MSR_FP) {
 746                 preempt_disable();
 747                 enable_kernel_fp();
 748                 load_fp_state(&vcpu->arch.fp);
 749                 disable_kernel_fp();
 750                 t->fp_save_area = &vcpu->arch.fp;
 751                 preempt_enable();
 752         }
 753
 754         if (msr & MSR_VEC) {
 755 #ifdef CONFIG_ALTIVEC
 756                 preempt_disable();
 757                 enable_kernel_altivec();
 758                 load_vr_state(&vcpu->arch.vr);
 759                 disable_kernel_altivec();
 760                 t->vr_save_area = &vcpu->arch.vr;
 761                 preempt_enable();
 762 #endif
 763         }
 764
 765         t->regs->msr |= msr;
 766         vcpu->arch.guest_owned_ext |= msr;
 767         kvmppc_recalc_shadow_msr(vcpu);
 768
 769         return RESUME_GUEST;
 770 }
 771
 772 /*
 773  * Kernel code using FP or VMX could have flushed guest state to
 774  * the thread_struct; if so, get it back now.
 775  */
 776 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
 777 {
 778         unsigned long lost_ext;
 779
 780         lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
 781         if (!lost_ext)
 782                 return;
 783
 784         if (lost_ext & MSR_FP) {
 785                 preempt_disable();
 786                 enable_kernel_fp();
 787                 load_fp_state(&vcpu->arch.fp);
 788                 disable_kernel_fp();
 789                 preempt_enable();
 790         }
 791 #ifdef CONFIG_ALTIVEC
 792         if (lost_ext & MSR_VEC) {
 793                 preempt_disable();
 794                 enable_kernel_altivec();
 795                 load_vr_state(&vcpu->arch.vr);
 796                 disable_kernel_altivec();
 797                 preempt_enable();
 798         }
 799 #endif
 800         current->thread.regs->msr |= lost_ext;
 801 }
 802
 803 #ifdef CONFIG_PPC_BOOK3S_64
 804
 805 static void kvmppc_trigger_fac_interrupt(struct kvm_vcpu *vcpu, ulong fac)
 806 {
 807         /* Inject the Interrupt Cause field and trigger a guest interrupt */
 808         vcpu->arch.fscr &= ~(0xffULL << 56);
 809         vcpu->arch.fscr |= (fac << 56);
 810         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_FAC_UNAVAIL);
 811 }
 812
 813 static void kvmppc_emulate_fac(struct kvm_vcpu *vcpu, ulong fac)
 814 {
 815         enum emulation_result er = EMULATE_FAIL;
 816
 817         if (!(kvmppc_get_msr(vcpu) & MSR_PR))
 818                 er = kvmppc_emulate_instruction(vcpu->run, vcpu);
 819
 820         if ((er != EMULATE_DONE) && (er != EMULATE_AGAIN)) {
 821                 /* Couldn't emulate, trigger interrupt in guest */
 822                 kvmppc_trigger_fac_interrupt(vcpu, fac);
 823         }
 824 }
 825
 826 /* Enable facilities (TAR, EBB, DSCR) for the guest */
 827 static int kvmppc_handle_fac(struct kvm_vcpu *vcpu, ulong fac)
 828 {
 829         bool guest_fac_enabled;
 830         BUG_ON(!cpu_has_feature(CPU_FTR_ARCH_207S));
 831
 832         /*
 833          * Not every facility is enabled by FSCR bits, check whether the
 834          * guest has this facility enabled at all.
 835          */
 836         switch (fac) {
 837         case FSCR_TAR_LG:
 838         case FSCR_EBB_LG:
 839                 guest_fac_enabled = (vcpu->arch.fscr & (1ULL << fac));
 840                 break;
 841         case FSCR_TM_LG:
 842                 guest_fac_enabled = kvmppc_get_msr(vcpu) & MSR_TM;
 843                 break;
 844         default:
 845                 guest_fac_enabled = false;
 846                 break;
 847         }
 848
 849         if (!guest_fac_enabled) {
 850                 /* Facility not enabled by the guest */
 851                 kvmppc_trigger_fac_interrupt(vcpu, fac);
 852                 return RESUME_GUEST;
 853         }
 854
 855         switch (fac) {
 856         case FSCR_TAR_LG:
 857                 /* TAR switching isn't lazy in Linux yet */
 858                 current->thread.tar = mfspr(SPRN_TAR);
 859                 mtspr(SPRN_TAR, vcpu->arch.tar);
 860                 vcpu->arch.shadow_fscr |= FSCR_TAR;
 861                 break;
 862         default:
 863                 kvmppc_emulate_fac(vcpu, fac);
 864                 break;
 865         }
 866
 867         return RESUME_GUEST;
 868 }
 869
 870 void kvmppc_set_fscr(struct kvm_vcpu *vcpu, u64 fscr)
 871 {
 872         if ((vcpu->arch.fscr & FSCR_TAR) && !(fscr & FSCR_TAR)) {
 873                 /* TAR got dropped, drop it in shadow too */
 874                 kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
 875         }
 876         vcpu->arch.fscr = fscr;
 877 }
 878 #endif
 879
 880 static void kvmppc_setup_debug(struct kvm_vcpu *vcpu)
 881 {
 882         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
 883                 u64 msr = kvmppc_get_msr(vcpu);
 884
 885                 kvmppc_set_msr(vcpu, msr | MSR_SE);
 886         }
 887 }
 888
 889 static void kvmppc_clear_debug(struct kvm_vcpu *vcpu)
 890 {
 891         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
 892                 u64 msr = kvmppc_get_msr(vcpu);
 893
 894                 kvmppc_set_msr(vcpu, msr & ~MSR_SE);
 895         }
 896 }
 897
 898 static int kvmppc_exit_pr_progint(struct kvm_run *run, struct kvm_vcpu *vcpu,
 899                                   unsigned int exit_nr)
 900 {
 901         enum emulation_result er;
 902         ulong flags;
 903         u32 last_inst;
 904         int emul, r;
 905
 906         /*
 907          * shadow_srr1 only contains valid flags if we came here via a program
 908          * exception. The other exceptions (emulation assist, FP unavailable,
 909          * etc.) do not provide flags in SRR1, so use an illegal-instruction
 910          * exception when injecting a program interrupt into the guest.
 911          */
 912         if (exit_nr == BOOK3S_INTERRUPT_PROGRAM)
 913                 flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
 914         else
 915                 flags = SRR1_PROGILL;
 916
 917         emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
 918         if (emul != EMULATE_DONE)
 919                 return RESUME_GUEST;
 920
 921         if (kvmppc_get_msr(vcpu) & MSR_PR) {
 922 #ifdef EXIT_DEBUG
 923                 pr_info("Userspace triggered 0x700 exception at\n 0x%lx (0x%x)\n",
 924                         kvmppc_get_pc(vcpu), last_inst);
 925 #endif
 926                 if ((last_inst & 0xff0007ff) != (INS_DCBZ & 0xfffffff7)) {
 927                         kvmppc_core_queue_program(vcpu, flags);
 928                         return RESUME_GUEST;
 929                 }
 930         }
 931
 932         vcpu->stat.emulated_inst_exits++;
 933         er = kvmppc_emulate_instruction(run, vcpu);
 934         switch (er) {
 935         case EMULATE_DONE:
 936                 r = RESUME_GUEST_NV;
 937                 break;
 938         case EMULATE_AGAIN:
 939                 r = RESUME_GUEST;
 940                 break;
 941         case EMULATE_FAIL:
 942                 pr_crit("%s: emulation at %lx failed (%08x)\n",
 943                         __func__, kvmppc_get_pc(vcpu), last_inst);
 944                 kvmppc_core_queue_program(vcpu, flags);
 945                 r = RESUME_GUEST;
 946                 break;
 947         case EMULATE_DO_MMIO:
 948                 run->exit_reason = KVM_EXIT_MMIO;
 949                 r = RESUME_HOST_NV;
 950                 break;
 951         case EMULATE_EXIT_USER:
 952                 r = RESUME_HOST_NV;
 953                 break;
 954         default:
 955                 BUG();
 956         }
 957
 958         return r;
 959 }
 960
 961 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
 962                           unsigned int exit_nr)
 963 {
 964         int r = RESUME_HOST;
 965         int s;
 966
 967         vcpu->stat.sum_exits++;
 968
 969         run->exit_reason = KVM_EXIT_UNKNOWN;
 970         run->ready_for_interrupt_injection = 1;
 971
 972         /* We get here with MSR.EE=1 */
 973
 974         trace_kvm_exit(exit_nr, vcpu);
 975         guest_exit();
 976
 977         switch (exit_nr) {
 978         case BOOK3S_INTERRUPT_INST_STORAGE:
 979         {
 980                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 981                 vcpu->stat.pf_instruc++;
 982
 983                 if (kvmppc_is_split_real(vcpu))
 984                         kvmppc_fixup_split_real(vcpu);
 985
 986 #ifdef CONFIG_PPC_BOOK3S_32
 987                 /* We set segments as unused segments when invalidating them. So
 988                  * treat the respective fault as segment fault. */
 989                 {
 990                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 991                         u32 sr;
 992
 993                         svcpu = svcpu_get(vcpu);
 994                         sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
 995                         svcpu_put(svcpu);
 996                         if (sr == SR_INVALID) {
 997                                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 998                                 r = RESUME_GUEST;
 999                                 break;
1000                         }
1001                 }
1002 #endif
1003
1004                 /* only care about PTEG not found errors, but leave NX alone */
1005                 if (shadow_srr1 & 0x40000000) {
1006                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
1007                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
1008                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1009                         vcpu->stat.sp_instruc++;
1010                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
1011                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
1012                         /*
1013                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
1014                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
1015                          *     that no guest that needs the dcbz hack does NX.
1016                          */
1017                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
1018                         r = RESUME_GUEST;
1019                 } else {
1020                         u64 msr = kvmppc_get_msr(vcpu);
1021                         msr |= shadow_srr1 & 0x58000000;
1022                         kvmppc_set_msr_fast(vcpu, msr);
1023                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1024                         r = RESUME_GUEST;
1025                 }
1026                 break;
1027         }
1028         case BOOK3S_INTERRUPT_DATA_STORAGE:
1029         {
1030                 ulong dar = kvmppc_get_fault_dar(vcpu);
1031                 u32 fault_dsisr = vcpu->arch.fault_dsisr;
1032                 vcpu->stat.pf_storage++;
1033
1034 #ifdef CONFIG_PPC_BOOK3S_32
1035                 /* We set segments as unused segments when invalidating them. So
1036                  * treat the respective fault as segment fault. */
1037                 {
1038                         struct kvmppc_book3s_shadow_vcpu *svcpu;
1039                         u32 sr;
1040
1041                         svcpu = svcpu_get(vcpu);
1042                         sr = svcpu->sr[dar >> SID_SHIFT];
1043                         svcpu_put(svcpu);
1044                         if (sr == SR_INVALID) {
1045                                 kvmppc_mmu_map_segment(vcpu, dar);
1046                                 r = RESUME_GUEST;
1047                                 break;
1048                         }
1049                 }
1050 #endif
1051
1052                 /*
1053                  * We need to handle missing shadow PTEs, and
1054                  * protection faults due to us mapping a page read-only
1055                  * when the guest thinks it is writable.
1056                  */
1057                 if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
1058                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
1059                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
1060                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1061                 } else {
1062                         kvmppc_set_dar(vcpu, dar);
1063                         kvmppc_set_dsisr(vcpu, fault_dsisr);
1064                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1065                         r = RESUME_GUEST;
1066                 }
1067                 break;
1068         }
1069         case BOOK3S_INTERRUPT_DATA_SEGMENT:
1070                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
1071                         kvmppc_set_dar(vcpu, kvmppc_get_fault_dar(vcpu));
1072                         kvmppc_book3s_queue_irqprio(vcpu,
1073                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
1074                 }
1075                 r = RESUME_GUEST;
1076                 break;
1077         case BOOK3S_INTERRUPT_INST_SEGMENT:
1078                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
1079                         kvmppc_book3s_queue_irqprio(vcpu,
1080                                 BOOK3S_INTERRUPT_INST_SEGMENT);
1081                 }
1082                 r = RESUME_GUEST;
1083                 break;
1084         /* We're good on these - the host merely wanted to get our attention */
1085         case BOOK3S_INTERRUPT_DECREMENTER:
1086         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1087         case BOOK3S_INTERRUPT_DOORBELL:
1088         case BOOK3S_INTERRUPT_H_DOORBELL:
1089                 vcpu->stat.dec_exits++;
1090                 r = RESUME_GUEST;
1091                 break;
1092         case BOOK3S_INTERRUPT_EXTERNAL:
1093         case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
1094         case BOOK3S_INTERRUPT_EXTERNAL_HV:
1095                 vcpu->stat.ext_intr_exits++;
1096                 r = RESUME_GUEST;
1097                 break;
1098         case BOOK3S_INTERRUPT_PERFMON:
1099                 r = RESUME_GUEST;
1100                 break;
1101         case BOOK3S_INTERRUPT_PROGRAM:
1102         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1103                 r = kvmppc_exit_pr_progint(run, vcpu, exit_nr);
1104                 break;
1105         case BOOK3S_INTERRUPT_SYSCALL:
1106         {
1107                 u32 last_sc;
1108                 int emul;
1109
1110                 /* Get last sc for papr */
1111                 if (vcpu->arch.papr_enabled) {
1112                         /* The sc instuction points SRR0 to the next inst */
1113                         emul = kvmppc_get_last_inst(vcpu, INST_SC, &last_sc);
1114                         if (emul != EMULATE_DONE) {
1115                                 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) - 4);
1116                                 r = RESUME_GUEST;
1117                                 break;
1118                         }
1119                 }
1120
1121                 if (vcpu->arch.papr_enabled &&
1122                     (last_sc == 0x44000022) &&
1123                     !(kvmppc_get_msr(vcpu) & MSR_PR)) {
1124                         /* SC 1 papr hypercalls */
1125                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
1126                         int i;
1127
1128 #ifdef CONFIG_PPC_BOOK3S_64
1129                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
1130                                 r = RESUME_GUEST;
1131                                 break;
1132                         }
1133 #endif
1134
1135                         run->papr_hcall.nr = cmd;
1136                         for (i = 0; i < 9; ++i) {
1137                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
1138                                 run->papr_hcall.args[i] = gpr;
1139                         }
1140                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
1141                         vcpu->arch.hcall_needed = 1;
1142                         r = RESUME_HOST;
1143                 } else if (vcpu->arch.osi_enabled &&
1144                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1145                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1146                         /* MOL hypercalls */
1147                         u64 *gprs = run->osi.gprs;
1148                         int i;
1149
1150                         run->exit_reason = KVM_EXIT_OSI;
1151                         for (i = 0; i < 32; i++)
1152                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
1153                         vcpu->arch.osi_needed = 1;
1154                         r = RESUME_HOST_NV;
1155                 } else if (!(kvmppc_get_msr(vcpu) & MSR_PR) &&
1156                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1157                         /* KVM PV hypercalls */
1158                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1159                         r = RESUME_GUEST;
1160                 } else {
1161                         /* Guest syscalls */
1162                         vcpu->stat.syscall_exits++;
1163                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1164                         r = RESUME_GUEST;
1165                 }
1166                 break;
1167         }
1168         case BOOK3S_INTERRUPT_FP_UNAVAIL:
1169         case BOOK3S_INTERRUPT_ALTIVEC:
1170         case BOOK3S_INTERRUPT_VSX:
1171         {
1172                 int ext_msr = 0;
1173                 int emul;
1174                 u32 last_inst;
1175
1176                 if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE) {
1177                         /* Do paired single instruction emulation */
1178                         emul = kvmppc_get_last_inst(vcpu, INST_GENERIC,
1179                                                     &last_inst);
1180                         if (emul == EMULATE_DONE)
1181                                 r = kvmppc_exit_pr_progint(run, vcpu, exit_nr);
1182                         else
1183                                 r = RESUME_GUEST;
1184
1185                         break;
1186                 }
1187
1188                 /* Enable external provider */
1189                 switch (exit_nr) {
1190                 case BOOK3S_INTERRUPT_FP_UNAVAIL:
1191                         ext_msr = MSR_FP;
1192                         break;
1193
1194                 case BOOK3S_INTERRUPT_ALTIVEC:
1195                         ext_msr = MSR_VEC;
1196                         break;
1197
1198                 case BOOK3S_INTERRUPT_VSX:
1199                         ext_msr = MSR_VSX;
1200                         break;
1201                 }
1202
1203                 r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1204                 break;
1205         }
1206         case BOOK3S_INTERRUPT_ALIGNMENT:
1207         {
1208                 u32 last_inst;
1209                 int emul = kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst);
1210
1211                 if (emul == EMULATE_DONE) {
1212                         u32 dsisr;
1213                         u64 dar;
1214
1215                         dsisr = kvmppc_alignment_dsisr(vcpu, last_inst);
1216                         dar = kvmppc_alignment_dar(vcpu, last_inst);
1217
1218                         kvmppc_set_dsisr(vcpu, dsisr);
1219                         kvmppc_set_dar(vcpu, dar);
1220
1221                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1222                 }
1223                 r = RESUME_GUEST;
1224                 break;
1225         }
1226 #ifdef CONFIG_PPC_BOOK3S_64
1227         case BOOK3S_INTERRUPT_FAC_UNAVAIL:
1228                 kvmppc_handle_fac(vcpu, vcpu->arch.shadow_fscr >> 56);
1229                 r = RESUME_GUEST;
1230                 break;
1231 #endif
1232         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1233                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1234                 r = RESUME_GUEST;
1235                 break;
1236         case BOOK3S_INTERRUPT_TRACE:
1237                 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
1238                         run->exit_reason = KVM_EXIT_DEBUG;
1239                         r = RESUME_HOST;
1240                 } else {
1241                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1242                         r = RESUME_GUEST;
1243                 }
1244                 break;
1245         default:
1246         {
1247                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1248                 /* Ugh - bork here! What did we get? */
1249                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1250                         exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1251                 r = RESUME_HOST;
1252                 BUG();
1253                 break;
1254         }
1255         }
1256
1257         if (!(r & RESUME_HOST)) {
1258                 /* To avoid clobbering exit_reason, only check for signals if
1259                  * we aren't already exiting to userspace for some other
1260                  * reason. */
1261
1262                 /*
1263                  * Interrupts could be timers for the guest which we have to
1264                  * inject again, so let's postpone them until we're in the guest
1265                  * and if we really did time things so badly, then we just exit
1266                  * again due to a host external interrupt.
1267                  */
1268                 s = kvmppc_prepare_to_enter(vcpu);
1269                 if (s <= 0)
1270                         r = s;
1271                 else {
1272                         /* interrupts now hard-disabled */
1273                         kvmppc_fix_ee_before_entry();
1274                 }
1275
1276                 kvmppc_handle_lost_ext(vcpu);
1277         }
1278
1279         trace_kvm_book3s_reenter(r, vcpu);
1280
1281         return r;
1282 }
1283
1284 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1285                                             struct kvm_sregs *sregs)
1286 {
1287         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1288         int i;
1289
1290         sregs->pvr = vcpu->arch.pvr;
1291
1292         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1293         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1294                 for (i = 0; i < 64; i++) {
1295                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1296                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1297                 }
1298         } else {
1299                 for (i = 0; i < 16; i++)
1300                         sregs->u.s.ppc32.sr[i] = kvmppc_get_sr(vcpu, i);
1301
1302                 for (i = 0; i < 8; i++) {
1303                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1304                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1305                 }
1306         }
1307
1308         return 0;
1309 }
1310
1311 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1312                                             struct kvm_sregs *sregs)
1313 {
1314         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1315         int i;
1316
1317         kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1318
1319         vcpu3s->sdr1 = sregs->u.s.sdr1;
1320 #ifdef CONFIG_PPC_BOOK3S_64
1321         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1322                 /* Flush all SLB entries */
1323                 vcpu->arch.mmu.slbmte(vcpu, 0, 0);
1324                 vcpu->arch.mmu.slbia(vcpu);
1325
1326                 for (i = 0; i < 64; i++) {
1327                         u64 rb = sregs->u.s.ppc64.slb[i].slbe;
1328                         u64 rs = sregs->u.s.ppc64.slb[i].slbv;
1329
1330                         if (rb & SLB_ESID_V)
1331                                 vcpu->arch.mmu.slbmte(vcpu, rs, rb);
1332                 }
1333         } else
1334 #endif
1335         {
1336                 for (i = 0; i < 16; i++) {
1337                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1338                 }
1339                 for (i = 0; i < 8; i++) {
1340                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1341                                        (u32)sregs->u.s.ppc32.ibat[i]);
1342                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1343                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1344                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1345                                        (u32)sregs->u.s.ppc32.dbat[i]);
1346                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1347                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1348                 }
1349         }
1350
1351         /* Flush the MMU after messing with the segments */
1352         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1353
1354         return 0;
1355 }
1356
1357 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1358                                  union kvmppc_one_reg *val)
1359 {
1360         int r = 0;
1361
1362         switch (id) {
1363         case KVM_REG_PPC_DEBUG_INST:
1364                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1365                 break;
1366         case KVM_REG_PPC_HIOR:
1367                 *val = get_reg_val(id, to_book3s(vcpu)->hior);
1368                 break;
1369         case KVM_REG_PPC_VTB:
1370                 *val = get_reg_val(id, to_book3s(vcpu)->vtb);
1371                 break;
1372         case KVM_REG_PPC_LPCR:
1373         case KVM_REG_PPC_LPCR_64:
1374                 /*
1375                  * We are only interested in the LPCR_ILE bit
1376                  */
1377                 if (vcpu->arch.intr_msr & MSR_LE)
1378                         *val = get_reg_val(id, LPCR_ILE);
1379                 else
1380                         *val = get_reg_val(id, 0);
1381                 break;
1382         default:
1383                 r = -EINVAL;
1384                 break;
1385         }
1386
1387         return r;
1388 }
1389
1390 static void kvmppc_set_lpcr_pr(struct kvm_vcpu *vcpu, u64 new_lpcr)
1391 {
1392         if (new_lpcr & LPCR_ILE)
1393                 vcpu->arch.intr_msr |= MSR_LE;
1394         else
1395                 vcpu->arch.intr_msr &= ~MSR_LE;
1396 }
1397
1398 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1399                                  union kvmppc_one_reg *val)
1400 {
1401         int r = 0;
1402
1403         switch (id) {
1404         case KVM_REG_PPC_HIOR:
1405                 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1406                 to_book3s(vcpu)->hior_explicit = true;
1407                 break;
1408         case KVM_REG_PPC_VTB:
1409                 to_book3s(vcpu)->vtb = set_reg_val(id, *val);
1410                 break;
1411         case KVM_REG_PPC_LPCR:
1412         case KVM_REG_PPC_LPCR_64:
1413                 kvmppc_set_lpcr_pr(vcpu, set_reg_val(id, *val));
1414                 break;
1415         default:
1416                 r = -EINVAL;
1417                 break;
1418         }
1419
1420         return r;
1421 }
1422
1423 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1424                                                    unsigned int id)
1425 {
1426         struct kvmppc_vcpu_book3s *vcpu_book3s;
1427         struct kvm_vcpu *vcpu;
1428         int err = -ENOMEM;
1429         unsigned long p;
1430
1431         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1432         if (!vcpu)
1433                 goto out;
1434
1435         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1436         if (!vcpu_book3s)
1437                 goto free_vcpu;
1438         vcpu->arch.book3s = vcpu_book3s;
1439
1440 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1441         vcpu->arch.shadow_vcpu =
1442                 kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1443         if (!vcpu->arch.shadow_vcpu)
1444                 goto free_vcpu3s;
1445 #endif
1446
1447         err = kvm_vcpu_init(vcpu, kvm, id);
1448         if (err)
1449                 goto free_shadow_vcpu;
1450
1451         err = -ENOMEM;
1452         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1453         if (!p)
1454                 goto uninit_vcpu;
1455         vcpu->arch.shared = (void *)p;
1456 #ifdef CONFIG_PPC_BOOK3S_64
1457         /* Always start the shared struct in native endian mode */
1458 #ifdef __BIG_ENDIAN__
1459         vcpu->arch.shared_big_endian = true;
1460 #else
1461         vcpu->arch.shared_big_endian = false;
1462 #endif
1463
1464         /*
1465          * Default to the same as the host if we're on sufficiently
1466          * recent machine that we have 1TB segments;
1467          * otherwise default to PPC970FX.
1468          */
1469         vcpu->arch.pvr = 0x3C0301;
1470         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1471                 vcpu->arch.pvr = mfspr(SPRN_PVR);
1472         vcpu->arch.intr_msr = MSR_SF;
1473 #else
1474         /* default to book3s_32 (750) */
1475         vcpu->arch.pvr = 0x84202;
1476 #endif
1477         kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1478         vcpu->arch.slb_nr = 64;
1479
1480         vcpu->arch.shadow_msr = MSR_USER64 & ~MSR_LE;
1481
1482         err = kvmppc_mmu_init(vcpu);
1483         if (err < 0)
1484                 goto uninit_vcpu;
1485
1486         return vcpu;
1487
1488 uninit_vcpu:
1489         kvm_vcpu_uninit(vcpu);
1490 free_shadow_vcpu:
1491 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1492         kfree(vcpu->arch.shadow_vcpu);
1493 free_vcpu3s:
1494 #endif
1495         vfree(vcpu_book3s);
1496 free_vcpu:
1497         kmem_cache_free(kvm_vcpu_cache, vcpu);
1498 out:
1499         return ERR_PTR(err);
1500 }
1501
1502 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1503 {
1504         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1505
1506         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1507         kvm_vcpu_uninit(vcpu);
1508 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
1509         kfree(vcpu->arch.shadow_vcpu);
1510 #endif
1511         vfree(vcpu_book3s);
1512         kmem_cache_free(kvm_vcpu_cache, vcpu);
1513 }
1514
1515 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1516 {
1517         int ret;
1518 #ifdef CONFIG_ALTIVEC
1519         unsigned long uninitialized_var(vrsave);
1520 #endif
1521
1522         /* Check if we can run the vcpu at all */
1523         if (!vcpu->arch.sane) {
1524                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1525                 ret = -EINVAL;
1526                 goto out;
1527         }
1528
1529         kvmppc_setup_debug(vcpu);
1530
1531         /*
1532          * Interrupts could be timers for the guest which we have to inject
1533          * again, so let's postpone them until we're in the guest and if we
1534          * really did time things so badly, then we just exit again due to
1535          * a host external interrupt.
1536          */
1537         ret = kvmppc_prepare_to_enter(vcpu);
1538         if (ret <= 0)
1539                 goto out;
1540         /* interrupts now hard-disabled */
1541
1542         /* Save FPU, Altivec and VSX state */
1543         giveup_all(current);
1544
1545         /* Preload FPU if it's enabled */
1546         if (kvmppc_get_msr(vcpu) & MSR_FP)
1547                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1548
1549         kvmppc_fix_ee_before_entry();
1550
1551         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1552
1553         kvmppc_clear_debug(vcpu);
1554
1555         /* No need for guest_exit. It's done in handle_exit.
1556            We also get here with interrupts enabled. */
1557
1558         /* Make sure we save the guest FPU/Altivec/VSX state */
1559         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1560
1561         /* Make sure we save the guest TAR/EBB/DSCR state */
1562         kvmppc_giveup_fac(vcpu, FSCR_TAR_LG);
1563
1564 out:
1565         vcpu->mode = OUTSIDE_GUEST_MODE;
1566         return ret;
1567 }
1568
1569 /*
1570  * Get (and clear) the dirty memory log for a memory slot.
1571  */
1572 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1573                                          struct kvm_dirty_log *log)
1574 {
1575         struct kvm_memslots *slots;
1576         struct kvm_memory_slot *memslot;
1577         struct kvm_vcpu *vcpu;
1578         ulong ga, ga_end;
1579         int is_dirty = 0;
1580         int r;
1581         unsigned long n;
1582
1583         mutex_lock(&kvm->slots_lock);
1584
1585         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1586         if (r)
1587                 goto out;
1588
1589         /* If nothing is dirty, don't bother messing with page tables. */
1590         if (is_dirty) {
1591                 slots = kvm_memslots(kvm);
1592                 memslot = id_to_memslot(slots, log->slot);
1593
1594                 ga = memslot->base_gfn << PAGE_SHIFT;
1595                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1596
1597                 kvm_for_each_vcpu(n, vcpu, kvm)
1598                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1599
1600                 n = kvm_dirty_bitmap_bytes(memslot);
1601                 memset(memslot->dirty_bitmap, 0, n);
1602         }
1603
1604         r = 0;
1605 out:
1606         mutex_unlock(&kvm->slots_lock);
1607         return r;
1608 }
1609
1610 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1611                                          struct kvm_memory_slot *memslot)
1612 {
1613         return;
1614 }
1615
1616 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1617                                         struct kvm_memory_slot *memslot,
1618                                         const struct kvm_userspace_memory_region *mem)
1619 {
1620         return 0;
1621 }
1622
1623 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1624                                 const struct kvm_userspace_memory_region *mem,
1625                                 const struct kvm_memory_slot *old,
1626                                 const struct kvm_memory_slot *new)
1627 {
1628         return;
1629 }
1630
1631 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1632                                         struct kvm_memory_slot *dont)
1633 {
1634         return;
1635 }
1636
1637 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1638                                          unsigned long npages)
1639 {
1640         return 0;
1641 }
1642
1643
1644 #ifdef CONFIG_PPC64
1645 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1646                                          struct kvm_ppc_smmu_info *info)
1647 {
1648         long int i;
1649         struct kvm_vcpu *vcpu;
1650
1651         info->flags = 0;
1652
1653         /* SLB is always 64 entries */
1654         info->slb_size = 64;
1655
1656         /* Standard 4k base page size segment */
1657         info->sps[0].page_shift = 12;
1658         info->sps[0].slb_enc = 0;
1659         info->sps[0].enc[0].page_shift = 12;
1660         info->sps[0].enc[0].pte_enc = 0;
1661
1662         /*
1663          * 64k large page size.
1664          * We only want to put this in if the CPUs we're emulating
1665          * support it, but unfortunately we don't have a vcpu easily
1666          * to hand here to test.  Just pick the first vcpu, and if
1667          * that doesn't exist yet, report the minimum capability,
1668          * i.e., no 64k pages.
1669          * 1T segment support goes along with 64k pages.
1670          */
1671         i = 1;
1672         vcpu = kvm_get_vcpu(kvm, 0);
1673         if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1674                 info->flags = KVM_PPC_1T_SEGMENTS;
1675                 info->sps[i].page_shift = 16;
1676                 info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1677                 info->sps[i].enc[0].page_shift = 16;
1678                 info->sps[i].enc[0].pte_enc = 1;
1679                 ++i;
1680         }
1681
1682         /* Standard 16M large page size segment */
1683         info->sps[i].page_shift = 24;
1684         info->sps[i].slb_enc = SLB_VSID_L;
1685         info->sps[i].enc[0].page_shift = 24;
1686         info->sps[i].enc[0].pte_enc = 0;
1687
1688         return 0;
1689 }
1690 #else
1691 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1692                                          struct kvm_ppc_smmu_info *info)
1693 {
1694         /* We should not get called */
1695         BUG();
1696 }
1697 #endif /* CONFIG_PPC64 */
1698
1699 static unsigned int kvm_global_user_count = 0;
1700 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1701
1702 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1703 {
1704         mutex_init(&kvm->arch.hpt_mutex);
1705
1706 #ifdef CONFIG_PPC_BOOK3S_64
1707         /* Start out with the default set of hcalls enabled */
1708         kvmppc_pr_init_default_hcalls(kvm);
1709 #endif
1710
1711         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1712                 spin_lock(&kvm_global_user_count_lock);
1713                 if (++kvm_global_user_count == 1)
1714                         pseries_disable_reloc_on_exc();
1715                 spin_unlock(&kvm_global_user_count_lock);
1716         }
1717         return 0;
1718 }
1719
1720 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1721 {
1722 #ifdef CONFIG_PPC64
1723         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1724 #endif
1725
1726         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1727                 spin_lock(&kvm_global_user_count_lock);
1728                 BUG_ON(kvm_global_user_count == 0);
1729                 if (--kvm_global_user_count == 0)
1730                         pseries_enable_reloc_on_exc();
1731                 spin_unlock(&kvm_global_user_count_lock);
1732         }
1733 }
1734
1735 static int kvmppc_core_check_processor_compat_pr(void)
1736 {
1737         /*
1738          * Disable KVM for Power9 untill the required bits merged.
1739          */
1740         if (cpu_has_feature(CPU_FTR_ARCH_300))
1741                 return -EIO;
1742         return 0;
1743 }
1744
1745 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1746                                  unsigned int ioctl, unsigned long arg)
1747 {
1748         return -ENOTTY;
1749 }
1750
1751 static struct kvmppc_ops kvm_ops_pr = {
1752         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1753         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1754         .get_one_reg = kvmppc_get_one_reg_pr,
1755         .set_one_reg = kvmppc_set_one_reg_pr,
1756         .vcpu_load   = kvmppc_core_vcpu_load_pr,
1757         .vcpu_put    = kvmppc_core_vcpu_put_pr,
1758         .set_msr     = kvmppc_set_msr_pr,
1759         .vcpu_run    = kvmppc_vcpu_run_pr,
1760         .vcpu_create = kvmppc_core_vcpu_create_pr,
1761         .vcpu_free   = kvmppc_core_vcpu_free_pr,
1762         .check_requests = kvmppc_core_check_requests_pr,
1763         .get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1764         .flush_memslot = kvmppc_core_flush_memslot_pr,
1765         .prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1766         .commit_memory_region = kvmppc_core_commit_memory_region_pr,
1767         .unmap_hva_range = kvm_unmap_hva_range_pr,
1768         .age_hva  = kvm_age_hva_pr,
1769         .test_age_hva = kvm_test_age_hva_pr,
1770         .set_spte_hva = kvm_set_spte_hva_pr,
1771         .mmu_destroy  = kvmppc_mmu_destroy_pr,
1772         .free_memslot = kvmppc_core_free_memslot_pr,
1773         .create_memslot = kvmppc_core_create_memslot_pr,
1774         .init_vm = kvmppc_core_init_vm_pr,
1775         .destroy_vm = kvmppc_core_destroy_vm_pr,
1776         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1777         .emulate_op = kvmppc_core_emulate_op_pr,
1778         .emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1779         .emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1780         .fast_vcpu_kick = kvm_vcpu_kick,
1781         .arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1782 #ifdef CONFIG_PPC_BOOK3S_64
1783         .hcall_implemented = kvmppc_hcall_impl_pr,
1784 #endif
1785 };
1786
1787
1788 int kvmppc_book3s_init_pr(void)
1789 {
1790         int r;
1791
1792         r = kvmppc_core_check_processor_compat_pr();
1793         if (r < 0)
1794                 return r;
1795
1796         kvm_ops_pr.owner = THIS_MODULE;
1797         kvmppc_pr_ops = &kvm_ops_pr;
1798
1799         r = kvmppc_mmu_hpte_sysinit();
1800         return r;
1801 }
1802
1803 void kvmppc_book3s_exit_pr(void)
1804 {
1805         kvmppc_pr_ops = NULL;
1806         kvmppc_mmu_hpte_sysexit();
1807 }
1808
1809 /*
1810  * We only support separate modules for book3s 64
1811  */
1812 #ifdef CONFIG_PPC_BOOK3S_64
1813
1814 module_init(kvmppc_book3s_init_pr);
1815 module_exit(kvmppc_book3s_exit_pr);
1816
1817 MODULE_LICENSE("GPL");
1818 MODULE_ALIAS_MISCDEV(KVM_MINOR);
1819 MODULE_ALIAS("devname:kvm");
1820 #endif