arch/powerpc/kvm/book3s_pr.c

   1 /*
   2  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   3  *
   4  * Authors:
   5  *    Alexander Graf <agraf@suse.de>
   6  *    Kevin Wolf <mail@kevin-wolf.de>
   7  *    Paul Mackerras <paulus@samba.org>
   8  *
   9  * Description:
  10  * Functions relating to running KVM on Book 3S processors where
  11  * we don't have access to hypervisor mode, and we run the guest
  12  * in problem state (user mode).
  13  *
  14  * This file is derived from arch/powerpc/kvm/44x.c,
  15  * by Hollis Blanchard <hollisb@us.ibm.com>.
  16  *
  17  * This program is free software; you can redistribute it and/or modify
  18  * it under the terms of the GNU General Public License, version 2, as
  19  * published by the Free Software Foundation.
  20  */
  21
  22 #include <linux/kvm_host.h>
  23 #include <linux/export.h>
  24 #include <linux/err.h>
  25 #include <linux/slab.h>
  26
  27 #include <asm/reg.h>
  28 #include <asm/cputable.h>
  29 #include <asm/cacheflush.h>
  30 #include <asm/tlbflush.h>
  31 #include <asm/uaccess.h>
  32 #include <asm/io.h>
  33 #include <asm/kvm_ppc.h>
  34 #include <asm/kvm_book3s.h>
  35 #include <asm/mmu_context.h>
  36 #include <asm/switch_to.h>
  37 #include <asm/firmware.h>
  38 #include <asm/hvcall.h>
  39 #include <linux/gfp.h>
  40 #include <linux/sched.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/highmem.h>
  43
  44 #include "book3s.h"
  45
  46 #define CREATE_TRACE_POINTS
  47 #include "trace_pr.h"
  48
  49 /* #define EXIT_DEBUG */
  50 /* #define DEBUG_EXT */
  51
  52 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  53                              ulong msr);
  54
  55 /* Some compatibility defines */
  56 #ifdef CONFIG_PPC_BOOK3S_32
  57 #define MSR_USER32 MSR_USER
  58 #define MSR_USER64 MSR_USER
  59 #define HW_PAGE_SIZE PAGE_SIZE
  60 #endif
  61
  62 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
  63 {
  64 #ifdef CONFIG_PPC_BOOK3S_64
  65         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  66         memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
  67         svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
  68         svcpu_put(svcpu);
  69 #endif
  70         vcpu->cpu = smp_processor_id();
  71 #ifdef CONFIG_PPC_BOOK3S_32
  72         current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
  73 #endif
  74 }
  75
  76 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
  77 {
  78 #ifdef CONFIG_PPC_BOOK3S_64
  79         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  80         memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
  81         to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
  82         svcpu_put(svcpu);
  83 #endif
  84
  85         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
  86         vcpu->cpu = -1;
  87 }
  88
  89 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
  90 void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu *svcpu,
  91                           struct kvm_vcpu *vcpu)
  92 {
  93         svcpu->gpr[0] = vcpu->arch.gpr[0];
  94         svcpu->gpr[1] = vcpu->arch.gpr[1];
  95         svcpu->gpr[2] = vcpu->arch.gpr[2];
  96         svcpu->gpr[3] = vcpu->arch.gpr[3];
  97         svcpu->gpr[4] = vcpu->arch.gpr[4];
  98         svcpu->gpr[5] = vcpu->arch.gpr[5];
  99         svcpu->gpr[6] = vcpu->arch.gpr[6];
 100         svcpu->gpr[7] = vcpu->arch.gpr[7];
 101         svcpu->gpr[8] = vcpu->arch.gpr[8];
 102         svcpu->gpr[9] = vcpu->arch.gpr[9];
 103         svcpu->gpr[10] = vcpu->arch.gpr[10];
 104         svcpu->gpr[11] = vcpu->arch.gpr[11];
 105         svcpu->gpr[12] = vcpu->arch.gpr[12];
 106         svcpu->gpr[13] = vcpu->arch.gpr[13];
 107         svcpu->cr  = vcpu->arch.cr;
 108         svcpu->xer = vcpu->arch.xer;
 109         svcpu->ctr = vcpu->arch.ctr;
 110         svcpu->lr  = vcpu->arch.lr;
 111         svcpu->pc  = vcpu->arch.pc;
 112 }
 113
 114 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
 115 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu,
 116                             struct kvmppc_book3s_shadow_vcpu *svcpu)
 117 {
 118         vcpu->arch.gpr[0] = svcpu->gpr[0];
 119         vcpu->arch.gpr[1] = svcpu->gpr[1];
 120         vcpu->arch.gpr[2] = svcpu->gpr[2];
 121         vcpu->arch.gpr[3] = svcpu->gpr[3];
 122         vcpu->arch.gpr[4] = svcpu->gpr[4];
 123         vcpu->arch.gpr[5] = svcpu->gpr[5];
 124         vcpu->arch.gpr[6] = svcpu->gpr[6];
 125         vcpu->arch.gpr[7] = svcpu->gpr[7];
 126         vcpu->arch.gpr[8] = svcpu->gpr[8];
 127         vcpu->arch.gpr[9] = svcpu->gpr[9];
 128         vcpu->arch.gpr[10] = svcpu->gpr[10];
 129         vcpu->arch.gpr[11] = svcpu->gpr[11];
 130         vcpu->arch.gpr[12] = svcpu->gpr[12];
 131         vcpu->arch.gpr[13] = svcpu->gpr[13];
 132         vcpu->arch.cr  = svcpu->cr;
 133         vcpu->arch.xer = svcpu->xer;
 134         vcpu->arch.ctr = svcpu->ctr;
 135         vcpu->arch.lr  = svcpu->lr;
 136         vcpu->arch.pc  = svcpu->pc;
 137         vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
 138         vcpu->arch.fault_dar   = svcpu->fault_dar;
 139         vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
 140         vcpu->arch.last_inst   = svcpu->last_inst;
 141 }
 142
 143 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
 144 {
 145         int r = 1; /* Indicate we want to get back into the guest */
 146
 147         /* We misuse TLB_FLUSH to indicate that we want to clear
 148            all shadow cache entries */
 149         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
 150                 kvmppc_mmu_pte_flush(vcpu, 0, 0);
 151
 152         return r;
 153 }
 154
 155 /************* MMU Notifiers *************/
 156 static void do_kvm_unmap_hva(struct kvm *kvm, unsigned long start,
 157                              unsigned long end)
 158 {
 159         long i;
 160         struct kvm_vcpu *vcpu;
 161         struct kvm_memslots *slots;
 162         struct kvm_memory_slot *memslot;
 163
 164         slots = kvm_memslots(kvm);
 165         kvm_for_each_memslot(memslot, slots) {
 166                 unsigned long hva_start, hva_end;
 167                 gfn_t gfn, gfn_end;
 168
 169                 hva_start = max(start, memslot->userspace_addr);
 170                 hva_end = min(end, memslot->userspace_addr +
 171                                         (memslot->npages << PAGE_SHIFT));
 172                 if (hva_start >= hva_end)
 173                         continue;
 174                 /*
 175                  * {gfn(page) | page intersects with [hva_start, hva_end)} =
 176                  * {gfn, gfn+1, ..., gfn_end-1}.
 177                  */
 178                 gfn = hva_to_gfn_memslot(hva_start, memslot);
 179                 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
 180                 kvm_for_each_vcpu(i, vcpu, kvm)
 181                         kvmppc_mmu_pte_pflush(vcpu, gfn << PAGE_SHIFT,
 182                                               gfn_end << PAGE_SHIFT);
 183         }
 184 }
 185
 186 static int kvm_unmap_hva_pr(struct kvm *kvm, unsigned long hva)
 187 {
 188         trace_kvm_unmap_hva(hva);
 189
 190         do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
 191
 192         return 0;
 193 }
 194
 195 static int kvm_unmap_hva_range_pr(struct kvm *kvm, unsigned long start,
 196                                   unsigned long end)
 197 {
 198         do_kvm_unmap_hva(kvm, start, end);
 199
 200         return 0;
 201 }
 202
 203 static int kvm_age_hva_pr(struct kvm *kvm, unsigned long hva)
 204 {
 205         /* XXX could be more clever ;) */
 206         return 0;
 207 }
 208
 209 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
 210 {
 211         /* XXX could be more clever ;) */
 212         return 0;
 213 }
 214
 215 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
 216 {
 217         /* The page will get remapped properly on its next fault */
 218         do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
 219 }
 220
 221 /*****************************************/
 222
 223 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 224 {
 225         ulong smsr = vcpu->arch.shared->msr;
 226
 227         /* Guest MSR values */
 228         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE;
 229         /* Process MSR values */
 230         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 231         /* External providers the guest reserved */
 232         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
 233         /* 64-bit Process MSR values */
 234 #ifdef CONFIG_PPC_BOOK3S_64
 235         smsr |= MSR_ISF | MSR_HV;
 236 #endif
 237         vcpu->arch.shadow_msr = smsr;
 238 }
 239
 240 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
 241 {
 242         ulong old_msr = vcpu->arch.shared->msr;
 243
 244 #ifdef EXIT_DEBUG
 245         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 246 #endif
 247
 248         msr &= to_book3s(vcpu)->msr_mask;
 249         vcpu->arch.shared->msr = msr;
 250         kvmppc_recalc_shadow_msr(vcpu);
 251
 252         if (msr & MSR_POW) {
 253                 if (!vcpu->arch.pending_exceptions) {
 254                         kvm_vcpu_block(vcpu);
 255                         clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 256                         vcpu->stat.halt_wakeup++;
 257
 258                         /* Unset POW bit after we woke up */
 259                         msr &= ~MSR_POW;
 260                         vcpu->arch.shared->msr = msr;
 261                 }
 262         }
 263
 264         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 265                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 266                 kvmppc_mmu_flush_segments(vcpu);
 267                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 268
 269                 /* Preload magic page segment when in kernel mode */
 270                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 271                         struct kvm_vcpu_arch *a = &vcpu->arch;
 272
 273                         if (msr & MSR_DR)
 274                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 275                         else
 276                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 277                 }
 278         }
 279
 280         /*
 281          * When switching from 32 to 64-bit, we may have a stale 32-bit
 282          * magic page around, we need to flush it. Typically 32-bit magic
 283          * page will be instanciated when calling into RTAS. Note: We
 284          * assume that such transition only happens while in kernel mode,
 285          * ie, we never transition from user 32-bit to kernel 64-bit with
 286          * a 32-bit magic page around.
 287          */
 288         if (vcpu->arch.magic_page_pa &&
 289             !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
 290                 /* going from RTAS to normal kernel code */
 291                 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
 292                                      ~0xFFFUL);
 293         }
 294
 295         /* Preload FPU if it's enabled */
 296         if (vcpu->arch.shared->msr & MSR_FP)
 297                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 298 }
 299
 300 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
 301 {
 302         u32 host_pvr;
 303
 304         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 305         vcpu->arch.pvr = pvr;
 306 #ifdef CONFIG_PPC_BOOK3S_64
 307         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 308                 kvmppc_mmu_book3s_64_init(vcpu);
 309                 if (!to_book3s(vcpu)->hior_explicit)
 310                         to_book3s(vcpu)->hior = 0xfff00000;
 311                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 312                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 313         } else
 314 #endif
 315         {
 316                 kvmppc_mmu_book3s_32_init(vcpu);
 317                 if (!to_book3s(vcpu)->hior_explicit)
 318                         to_book3s(vcpu)->hior = 0;
 319                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 320                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 321         }
 322
 323         kvmppc_sanity_check(vcpu);
 324
 325         /* If we are in hypervisor level on 970, we can tell the CPU to
 326          * treat DCBZ as 32 bytes store */
 327         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 328         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 329             !strcmp(cur_cpu_spec->platform, "ppc970"))
 330                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 331
 332         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 333            really needs them in a VM on Cell and force disable them. */
 334         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 335                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 336
 337         /*
 338          * If they're asking for POWER6 or later, set the flag
 339          * indicating that we can do multiple large page sizes
 340          * and 1TB segments.
 341          * Also set the flag that indicates that tlbie has the large
 342          * page bit in the RB operand instead of the instruction.
 343          */
 344         switch (PVR_VER(pvr)) {
 345         case PVR_POWER6:
 346         case PVR_POWER7:
 347         case PVR_POWER7p:
 348         case PVR_POWER8:
 349                 vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
 350                         BOOK3S_HFLAG_NEW_TLBIE;
 351                 break;
 352         }
 353
 354 #ifdef CONFIG_PPC_BOOK3S_32
 355         /* 32 bit Book3S always has 32 byte dcbz */
 356         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 357 #endif
 358
 359         /* On some CPUs we can execute paired single operations natively */
 360         asm ( "mfpvr %0" : "=r"(host_pvr));
 361         switch (host_pvr) {
 362         case 0x00080200:        /* lonestar 2.0 */
 363         case 0x00088202:        /* lonestar 2.2 */
 364         case 0x70000100:        /* gekko 1.0 */
 365         case 0x00080100:        /* gekko 2.0 */
 366         case 0x00083203:        /* gekko 2.3a */
 367         case 0x00083213:        /* gekko 2.3b */
 368         case 0x00083204:        /* gekko 2.4 */
 369         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 370         case 0x00087200:        /* broadway */
 371                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 372                 /* Enable HID2.PSE - in case we need it later */
 373                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 374         }
 375 }
 376
 377 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 378  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 379  * emulate 32 bytes dcbz length.
 380  *
 381  * The Book3s_64 inventors also realized this case and implemented a special bit
 382  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 383  *
 384  * My approach here is to patch the dcbz instruction on executing pages.
 385  */
 386 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 387 {
 388         struct page *hpage;
 389         u64 hpage_offset;
 390         u32 *page;
 391         int i;
 392
 393         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 394         if (is_error_page(hpage))
 395                 return;
 396
 397         hpage_offset = pte->raddr & ~PAGE_MASK;
 398         hpage_offset &= ~0xFFFULL;
 399         hpage_offset /= 4;
 400
 401         get_page(hpage);
 402         page = kmap_atomic(hpage);
 403
 404         /* patch dcbz into reserved instruction, so we trap */
 405         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 406                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 407                         page[i] &= 0xfffffff7;
 408
 409         kunmap_atomic(page);
 410         put_page(hpage);
 411 }
 412
 413 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 414 {
 415         ulong mp_pa = vcpu->arch.magic_page_pa;
 416
 417         if (!(vcpu->arch.shared->msr & MSR_SF))
 418                 mp_pa = (uint32_t)mp_pa;
 419
 420         if (unlikely(mp_pa) &&
 421             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 422                 return 1;
 423         }
 424
 425         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 426 }
 427
 428 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 429                             ulong eaddr, int vec)
 430 {
 431         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 432         bool iswrite = false;
 433         int r = RESUME_GUEST;
 434         int relocated;
 435         int page_found = 0;
 436         struct kvmppc_pte pte;
 437         bool is_mmio = false;
 438         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 439         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 440         u64 vsid;
 441
 442         relocated = data ? dr : ir;
 443         if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
 444                 iswrite = true;
 445
 446         /* Resolve real address if translation turned on */
 447         if (relocated) {
 448                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
 449         } else {
 450                 pte.may_execute = true;
 451                 pte.may_read = true;
 452                 pte.may_write = true;
 453                 pte.raddr = eaddr & KVM_PAM;
 454                 pte.eaddr = eaddr;
 455                 pte.vpage = eaddr >> 12;
 456                 pte.page_size = MMU_PAGE_64K;
 457         }
 458
 459         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 460         case 0:
 461                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 462                 break;
 463         case MSR_DR:
 464         case MSR_IR:
 465                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 466
 467                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 468                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 469                 else
 470                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 471                 pte.vpage |= vsid;
 472
 473                 if (vsid == -1)
 474                         page_found = -EINVAL;
 475                 break;
 476         }
 477
 478         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 479            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 480                 /*
 481                  * If we do the dcbz hack, we have to NX on every execution,
 482                  * so we can patch the executing code. This renders our guest
 483                  * NX-less.
 484                  */
 485                 pte.may_execute = !data;
 486         }
 487
 488         if (page_found == -ENOENT) {
 489                 /* Page not found in guest PTE entries */
 490                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 491                 vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr;
 492                 vcpu->arch.shared->msr |=
 493                         vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
 494                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 495         } else if (page_found == -EPERM) {
 496                 /* Storage protection */
 497                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 498                 vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
 499                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 500                 vcpu->arch.shared->msr |=
 501                         vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
 502                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 503         } else if (page_found == -EINVAL) {
 504                 /* Page not found in guest SLB */
 505                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 506                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 507         } else if (!is_mmio &&
 508                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 509                 if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
 510                         /*
 511                          * There is already a host HPTE there, presumably
 512                          * a read-only one for a page the guest thinks
 513                          * is writable, so get rid of it first.
 514                          */
 515                         kvmppc_mmu_unmap_page(vcpu, &pte);
 516                 }
 517                 /* The guest's PTE is not mapped yet. Map on the host */
 518                 kvmppc_mmu_map_page(vcpu, &pte, iswrite);
 519                 if (data)
 520                         vcpu->stat.sp_storage++;
 521                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 522                          (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 523                         kvmppc_patch_dcbz(vcpu, &pte);
 524         } else {
 525                 /* MMIO */
 526                 vcpu->stat.mmio_exits++;
 527                 vcpu->arch.paddr_accessed = pte.raddr;
 528                 vcpu->arch.vaddr_accessed = pte.eaddr;
 529                 r = kvmppc_emulate_mmio(run, vcpu);
 530                 if ( r == RESUME_HOST_NV )
 531                         r = RESUME_HOST;
 532         }
 533
 534         return r;
 535 }
 536
 537 static inline int get_fpr_index(int i)
 538 {
 539         return i * TS_FPRWIDTH;
 540 }
 541
 542 /* Give up external provider (FPU, Altivec, VSX) */
 543 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 544 {
 545         struct thread_struct *t = &current->thread;
 546         u64 *vcpu_fpr = vcpu->arch.fpr;
 547 #ifdef CONFIG_VSX
 548         u64 *vcpu_vsx = vcpu->arch.vsr;
 549 #endif
 550         u64 *thread_fpr = (u64*)t->fpr;
 551         int i;
 552
 553         /*
 554          * VSX instructions can access FP and vector registers, so if
 555          * we are giving up VSX, make sure we give up FP and VMX as well.
 556          */
 557         if (msr & MSR_VSX)
 558                 msr |= MSR_FP | MSR_VEC;
 559
 560         msr &= vcpu->arch.guest_owned_ext;
 561         if (!msr)
 562                 return;
 563
 564 #ifdef DEBUG_EXT
 565         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 566 #endif
 567
 568         if (msr & MSR_FP) {
 569                 /*
 570                  * Note that on CPUs with VSX, giveup_fpu stores
 571                  * both the traditional FP registers and the added VSX
 572                  * registers into thread.fpr[].
 573                  */
 574                 if (current->thread.regs->msr & MSR_FP)
 575                         giveup_fpu(current);
 576                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 577                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 578
 579                 vcpu->arch.fpscr = t->fpscr.val;
 580
 581 #ifdef CONFIG_VSX
 582                 if (cpu_has_feature(CPU_FTR_VSX))
 583                         for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
 584                                 vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 585 #endif
 586         }
 587
 588 #ifdef CONFIG_ALTIVEC
 589         if (msr & MSR_VEC) {
 590                 if (current->thread.regs->msr & MSR_VEC)
 591                         giveup_altivec(current);
 592                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 593                 vcpu->arch.vscr = t->vscr;
 594         }
 595 #endif
 596
 597         vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
 598         kvmppc_recalc_shadow_msr(vcpu);
 599 }
 600
 601 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 602 {
 603         ulong srr0 = kvmppc_get_pc(vcpu);
 604         u32 last_inst = kvmppc_get_last_inst(vcpu);
 605         int ret;
 606
 607         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 608         if (ret == -ENOENT) {
 609                 ulong msr = vcpu->arch.shared->msr;
 610
 611                 msr = kvmppc_set_field(msr, 33, 33, 1);
 612                 msr = kvmppc_set_field(msr, 34, 36, 0);
 613                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 614                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 615                 return EMULATE_AGAIN;
 616         }
 617
 618         return EMULATE_DONE;
 619 }
 620
 621 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 622 {
 623
 624         /* Need to do paired single emulation? */
 625         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 626                 return EMULATE_DONE;
 627
 628         /* Read out the instruction */
 629         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 630                 /* Need to emulate */
 631                 return EMULATE_FAIL;
 632
 633         return EMULATE_AGAIN;
 634 }
 635
 636 /* Handle external providers (FPU, Altivec, VSX) */
 637 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 638                              ulong msr)
 639 {
 640         struct thread_struct *t = &current->thread;
 641         u64 *vcpu_fpr = vcpu->arch.fpr;
 642 #ifdef CONFIG_VSX
 643         u64 *vcpu_vsx = vcpu->arch.vsr;
 644 #endif
 645         u64 *thread_fpr = (u64*)t->fpr;
 646         int i;
 647
 648         /* When we have paired singles, we emulate in software */
 649         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 650                 return RESUME_GUEST;
 651
 652         if (!(vcpu->arch.shared->msr & msr)) {
 653                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 654                 return RESUME_GUEST;
 655         }
 656
 657         if (msr == MSR_VSX) {
 658                 /* No VSX?  Give an illegal instruction interrupt */
 659 #ifdef CONFIG_VSX
 660                 if (!cpu_has_feature(CPU_FTR_VSX))
 661 #endif
 662                 {
 663                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 664                         return RESUME_GUEST;
 665                 }
 666
 667                 /*
 668                  * We have to load up all the FP and VMX registers before
 669                  * we can let the guest use VSX instructions.
 670                  */
 671                 msr = MSR_FP | MSR_VEC | MSR_VSX;
 672         }
 673
 674         /* See if we already own all the ext(s) needed */
 675         msr &= ~vcpu->arch.guest_owned_ext;
 676         if (!msr)
 677                 return RESUME_GUEST;
 678
 679 #ifdef DEBUG_EXT
 680         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 681 #endif
 682
 683         if (msr & MSR_FP) {
 684                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 685                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 686 #ifdef CONFIG_VSX
 687                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
 688                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 689 #endif
 690                 t->fpscr.val = vcpu->arch.fpscr;
 691                 t->fpexc_mode = 0;
 692                 kvmppc_load_up_fpu();
 693         }
 694
 695         if (msr & MSR_VEC) {
 696 #ifdef CONFIG_ALTIVEC
 697                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 698                 t->vscr = vcpu->arch.vscr;
 699                 t->vrsave = -1;
 700                 kvmppc_load_up_altivec();
 701 #endif
 702         }
 703
 704         current->thread.regs->msr |= msr;
 705         vcpu->arch.guest_owned_ext |= msr;
 706         kvmppc_recalc_shadow_msr(vcpu);
 707
 708         return RESUME_GUEST;
 709 }
 710
 711 /*
 712  * Kernel code using FP or VMX could have flushed guest state to
 713  * the thread_struct; if so, get it back now.
 714  */
 715 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
 716 {
 717         unsigned long lost_ext;
 718
 719         lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
 720         if (!lost_ext)
 721                 return;
 722
 723         if (lost_ext & MSR_FP)
 724                 kvmppc_load_up_fpu();
 725 #ifdef CONFIG_ALTIVEC
 726         if (lost_ext & MSR_VEC)
 727                 kvmppc_load_up_altivec();
 728 #endif
 729         current->thread.regs->msr |= lost_ext;
 730 }
 731
 732 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
 733                           unsigned int exit_nr)
 734 {
 735         int r = RESUME_HOST;
 736         int s;
 737
 738         vcpu->stat.sum_exits++;
 739
 740         run->exit_reason = KVM_EXIT_UNKNOWN;
 741         run->ready_for_interrupt_injection = 1;
 742
 743         /* We get here with MSR.EE=1 */
 744
 745         trace_kvm_exit(exit_nr, vcpu);
 746         kvm_guest_exit();
 747
 748         switch (exit_nr) {
 749         case BOOK3S_INTERRUPT_INST_STORAGE:
 750         {
 751                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 752                 vcpu->stat.pf_instruc++;
 753
 754 #ifdef CONFIG_PPC_BOOK3S_32
 755                 /* We set segments as unused segments when invalidating them. So
 756                  * treat the respective fault as segment fault. */
 757                 {
 758                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 759                         u32 sr;
 760
 761                         svcpu = svcpu_get(vcpu);
 762                         sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
 763                         svcpu_put(svcpu);
 764                         if (sr == SR_INVALID) {
 765                                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 766                                 r = RESUME_GUEST;
 767                                 break;
 768                         }
 769                 }
 770 #endif
 771
 772                 /* only care about PTEG not found errors, but leave NX alone */
 773                 if (shadow_srr1 & 0x40000000) {
 774                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
 775                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 776                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 777                         vcpu->stat.sp_instruc++;
 778                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 779                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 780                         /*
 781                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 782                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 783                          *     that no guest that needs the dcbz hack does NX.
 784                          */
 785                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 786                         r = RESUME_GUEST;
 787                 } else {
 788                         vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
 789                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 790                         r = RESUME_GUEST;
 791                 }
 792                 break;
 793         }
 794         case BOOK3S_INTERRUPT_DATA_STORAGE:
 795         {
 796                 ulong dar = kvmppc_get_fault_dar(vcpu);
 797                 u32 fault_dsisr = vcpu->arch.fault_dsisr;
 798                 vcpu->stat.pf_storage++;
 799
 800 #ifdef CONFIG_PPC_BOOK3S_32
 801                 /* We set segments as unused segments when invalidating them. So
 802                  * treat the respective fault as segment fault. */
 803                 {
 804                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 805                         u32 sr;
 806
 807                         svcpu = svcpu_get(vcpu);
 808                         sr = svcpu->sr[dar >> SID_SHIFT];
 809                         svcpu_put(svcpu);
 810                         if (sr == SR_INVALID) {
 811                                 kvmppc_mmu_map_segment(vcpu, dar);
 812                                 r = RESUME_GUEST;
 813                                 break;
 814                         }
 815                 }
 816 #endif
 817
 818                 /*
 819                  * We need to handle missing shadow PTEs, and
 820                  * protection faults due to us mapping a page read-only
 821                  * when the guest thinks it is writable.
 822                  */
 823                 if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
 824                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
 825                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 826                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 827                 } else {
 828                         vcpu->arch.shared->dar = dar;
 829                         vcpu->arch.shared->dsisr = fault_dsisr;
 830                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 831                         r = RESUME_GUEST;
 832                 }
 833                 break;
 834         }
 835         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 836                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 837                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 838                         kvmppc_book3s_queue_irqprio(vcpu,
 839                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 840                 }
 841                 r = RESUME_GUEST;
 842                 break;
 843         case BOOK3S_INTERRUPT_INST_SEGMENT:
 844                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 845                         kvmppc_book3s_queue_irqprio(vcpu,
 846                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 847                 }
 848                 r = RESUME_GUEST;
 849                 break;
 850         /* We're good on these - the host merely wanted to get our attention */
 851         case BOOK3S_INTERRUPT_DECREMENTER:
 852         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 853                 vcpu->stat.dec_exits++;
 854                 r = RESUME_GUEST;
 855                 break;
 856         case BOOK3S_INTERRUPT_EXTERNAL:
 857         case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
 858         case BOOK3S_INTERRUPT_EXTERNAL_HV:
 859                 vcpu->stat.ext_intr_exits++;
 860                 r = RESUME_GUEST;
 861                 break;
 862         case BOOK3S_INTERRUPT_PERFMON:
 863                 r = RESUME_GUEST;
 864                 break;
 865         case BOOK3S_INTERRUPT_PROGRAM:
 866         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 867         {
 868                 enum emulation_result er;
 869                 ulong flags;
 870
 871 program_interrupt:
 872                 flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
 873
 874                 if (vcpu->arch.shared->msr & MSR_PR) {
 875 #ifdef EXIT_DEBUG
 876                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 877 #endif
 878                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 879                             (INS_DCBZ & 0xfffffff7)) {
 880                                 kvmppc_core_queue_program(vcpu, flags);
 881                                 r = RESUME_GUEST;
 882                                 break;
 883                         }
 884                 }
 885
 886                 vcpu->stat.emulated_inst_exits++;
 887                 er = kvmppc_emulate_instruction(run, vcpu);
 888                 switch (er) {
 889                 case EMULATE_DONE:
 890                         r = RESUME_GUEST_NV;
 891                         break;
 892                 case EMULATE_AGAIN:
 893                         r = RESUME_GUEST;
 894                         break;
 895                 case EMULATE_FAIL:
 896                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 897                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 898                         kvmppc_core_queue_program(vcpu, flags);
 899                         r = RESUME_GUEST;
 900                         break;
 901                 case EMULATE_DO_MMIO:
 902                         run->exit_reason = KVM_EXIT_MMIO;
 903                         r = RESUME_HOST_NV;
 904                         break;
 905                 case EMULATE_EXIT_USER:
 906                         r = RESUME_HOST_NV;
 907                         break;
 908                 default:
 909                         BUG();
 910                 }
 911                 break;
 912         }
 913         case BOOK3S_INTERRUPT_SYSCALL:
 914                 if (vcpu->arch.papr_enabled &&
 915                     (kvmppc_get_last_sc(vcpu) == 0x44000022) &&
 916                     !(vcpu->arch.shared->msr & MSR_PR)) {
 917                         /* SC 1 papr hypercalls */
 918                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
 919                         int i;
 920
 921 #ifdef CONFIG_KVM_BOOK3S_64_PR
 922                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
 923                                 r = RESUME_GUEST;
 924                                 break;
 925                         }
 926 #endif
 927
 928                         run->papr_hcall.nr = cmd;
 929                         for (i = 0; i < 9; ++i) {
 930                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
 931                                 run->papr_hcall.args[i] = gpr;
 932                         }
 933                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
 934                         vcpu->arch.hcall_needed = 1;
 935                         r = RESUME_HOST;
 936                 } else if (vcpu->arch.osi_enabled &&
 937                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 938                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 939                         /* MOL hypercalls */
 940                         u64 *gprs = run->osi.gprs;
 941                         int i;
 942
 943                         run->exit_reason = KVM_EXIT_OSI;
 944                         for (i = 0; i < 32; i++)
 945                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 946                         vcpu->arch.osi_needed = 1;
 947                         r = RESUME_HOST_NV;
 948                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 949                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 950                         /* KVM PV hypercalls */
 951                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 952                         r = RESUME_GUEST;
 953                 } else {
 954                         /* Guest syscalls */
 955                         vcpu->stat.syscall_exits++;
 956                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 957                         r = RESUME_GUEST;
 958                 }
 959                 break;
 960         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 961         case BOOK3S_INTERRUPT_ALTIVEC:
 962         case BOOK3S_INTERRUPT_VSX:
 963         {
 964                 int ext_msr = 0;
 965
 966                 switch (exit_nr) {
 967                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 968                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 969                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 970                 }
 971
 972                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
 973                 case EMULATE_DONE:
 974                         /* everything ok - let's enable the ext */
 975                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
 976                         break;
 977                 case EMULATE_FAIL:
 978                         /* we need to emulate this instruction */
 979                         goto program_interrupt;
 980                         break;
 981                 default:
 982                         /* nothing to worry about - go again */
 983                         break;
 984                 }
 985                 break;
 986         }
 987         case BOOK3S_INTERRUPT_ALIGNMENT:
 988                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
 989                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
 990                                 kvmppc_get_last_inst(vcpu));
 991                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
 992                                 kvmppc_get_last_inst(vcpu));
 993                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 994                 }
 995                 r = RESUME_GUEST;
 996                 break;
 997         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 998         case BOOK3S_INTERRUPT_TRACE:
 999                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1000                 r = RESUME_GUEST;
1001                 break;
1002         default:
1003         {
1004                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1005                 /* Ugh - bork here! What did we get? */
1006                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1007                         exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1008                 r = RESUME_HOST;
1009                 BUG();
1010                 break;
1011         }
1012         }
1013
1014         if (!(r & RESUME_HOST)) {
1015                 /* To avoid clobbering exit_reason, only check for signals if
1016                  * we aren't already exiting to userspace for some other
1017                  * reason. */
1018
1019                 /*
1020                  * Interrupts could be timers for the guest which we have to
1021                  * inject again, so let's postpone them until we're in the guest
1022                  * and if we really did time things so badly, then we just exit
1023                  * again due to a host external interrupt.
1024                  */
1025                 local_irq_disable();
1026                 s = kvmppc_prepare_to_enter(vcpu);
1027                 if (s <= 0) {
1028                         local_irq_enable();
1029                         r = s;
1030                 } else {
1031                         kvmppc_fix_ee_before_entry();
1032                 }
1033                 kvmppc_handle_lost_ext(vcpu);
1034         }
1035
1036         trace_kvm_book3s_reenter(r, vcpu);
1037
1038         return r;
1039 }
1040
1041 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1042                                             struct kvm_sregs *sregs)
1043 {
1044         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1045         int i;
1046
1047         sregs->pvr = vcpu->arch.pvr;
1048
1049         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1050         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1051                 for (i = 0; i < 64; i++) {
1052                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1053                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1054                 }
1055         } else {
1056                 for (i = 0; i < 16; i++)
1057                         sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
1058
1059                 for (i = 0; i < 8; i++) {
1060                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1061                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1062                 }
1063         }
1064
1065         return 0;
1066 }
1067
1068 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1069                                             struct kvm_sregs *sregs)
1070 {
1071         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1072         int i;
1073
1074         kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1075
1076         vcpu3s->sdr1 = sregs->u.s.sdr1;
1077         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1078                 for (i = 0; i < 64; i++) {
1079                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1080                                                     sregs->u.s.ppc64.slb[i].slbe);
1081                 }
1082         } else {
1083                 for (i = 0; i < 16; i++) {
1084                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1085                 }
1086                 for (i = 0; i < 8; i++) {
1087                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1088                                        (u32)sregs->u.s.ppc32.ibat[i]);
1089                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1090                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1091                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1092                                        (u32)sregs->u.s.ppc32.dbat[i]);
1093                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1094                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1095                 }
1096         }
1097
1098         /* Flush the MMU after messing with the segments */
1099         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1100
1101         return 0;
1102 }
1103
1104 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1105                                  union kvmppc_one_reg *val)
1106 {
1107         int r = 0;
1108
1109         switch (id) {
1110         case KVM_REG_PPC_HIOR:
1111                 *val = get_reg_val(id, to_book3s(vcpu)->hior);
1112                 break;
1113 #ifdef CONFIG_VSX
1114         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1115                 long int i = id - KVM_REG_PPC_VSR0;
1116
1117                 if (!cpu_has_feature(CPU_FTR_VSX)) {
1118                         r = -ENXIO;
1119                         break;
1120                 }
1121                 val->vsxval[0] = vcpu->arch.fpr[i];
1122                 val->vsxval[1] = vcpu->arch.vsr[i];
1123                 break;
1124         }
1125 #endif /* CONFIG_VSX */
1126         default:
1127                 r = -EINVAL;
1128                 break;
1129         }
1130
1131         return r;
1132 }
1133
1134 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1135                                  union kvmppc_one_reg *val)
1136 {
1137         int r = 0;
1138
1139         switch (id) {
1140         case KVM_REG_PPC_HIOR:
1141                 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1142                 to_book3s(vcpu)->hior_explicit = true;
1143                 break;
1144 #ifdef CONFIG_VSX
1145         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1146                 long int i = id - KVM_REG_PPC_VSR0;
1147
1148                 if (!cpu_has_feature(CPU_FTR_VSX)) {
1149                         r = -ENXIO;
1150                         break;
1151                 }
1152                 vcpu->arch.fpr[i] = val->vsxval[0];
1153                 vcpu->arch.vsr[i] = val->vsxval[1];
1154                 break;
1155         }
1156 #endif /* CONFIG_VSX */
1157         default:
1158                 r = -EINVAL;
1159                 break;
1160         }
1161
1162         return r;
1163 }
1164
1165 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1166                                                    unsigned int id)
1167 {
1168         struct kvmppc_vcpu_book3s *vcpu_book3s;
1169         struct kvm_vcpu *vcpu;
1170         int err = -ENOMEM;
1171         unsigned long p;
1172
1173         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1174         if (!vcpu)
1175                 goto out;
1176
1177         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1178         if (!vcpu_book3s)
1179                 goto free_vcpu;
1180         vcpu->arch.book3s = vcpu_book3s;
1181
1182 #ifdef CONFIG_KVM_BOOK3S_32
1183         vcpu->arch.shadow_vcpu =
1184                 kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1185         if (!vcpu->arch.shadow_vcpu)
1186                 goto free_vcpu3s;
1187 #endif
1188
1189         err = kvm_vcpu_init(vcpu, kvm, id);
1190         if (err)
1191                 goto free_shadow_vcpu;
1192
1193         err = -ENOMEM;
1194         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1195         if (!p)
1196                 goto uninit_vcpu;
1197         /* the real shared page fills the last 4k of our page */
1198         vcpu->arch.shared = (void *)(p + PAGE_SIZE - 4096);
1199
1200 #ifdef CONFIG_PPC_BOOK3S_64
1201         /*
1202          * Default to the same as the host if we're on sufficiently
1203          * recent machine that we have 1TB segments;
1204          * otherwise default to PPC970FX.
1205          */
1206         vcpu->arch.pvr = 0x3C0301;
1207         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1208                 vcpu->arch.pvr = mfspr(SPRN_PVR);
1209 #else
1210         /* default to book3s_32 (750) */
1211         vcpu->arch.pvr = 0x84202;
1212 #endif
1213         kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1214         vcpu->arch.slb_nr = 64;
1215
1216         vcpu->arch.shadow_msr = MSR_USER64;
1217
1218         err = kvmppc_mmu_init(vcpu);
1219         if (err < 0)
1220                 goto uninit_vcpu;
1221
1222         return vcpu;
1223
1224 uninit_vcpu:
1225         kvm_vcpu_uninit(vcpu);
1226 free_shadow_vcpu:
1227 #ifdef CONFIG_KVM_BOOK3S_32
1228         kfree(vcpu->arch.shadow_vcpu);
1229 free_vcpu3s:
1230 #endif
1231         vfree(vcpu_book3s);
1232 free_vcpu:
1233         kmem_cache_free(kvm_vcpu_cache, vcpu);
1234 out:
1235         return ERR_PTR(err);
1236 }
1237
1238 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1239 {
1240         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1241
1242         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1243         kvm_vcpu_uninit(vcpu);
1244 #ifdef CONFIG_KVM_BOOK3S_32
1245         kfree(vcpu->arch.shadow_vcpu);
1246 #endif
1247         vfree(vcpu_book3s);
1248         kmem_cache_free(kvm_vcpu_cache, vcpu);
1249 }
1250
1251 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1252 {
1253         int ret;
1254         double fpr[32][TS_FPRWIDTH];
1255         unsigned int fpscr;
1256         int fpexc_mode;
1257 #ifdef CONFIG_ALTIVEC
1258         vector128 vr[32];
1259         vector128 vscr;
1260         unsigned long uninitialized_var(vrsave);
1261         int used_vr;
1262 #endif
1263 #ifdef CONFIG_VSX
1264         int used_vsr;
1265 #endif
1266         ulong ext_msr;
1267
1268         /* Check if we can run the vcpu at all */
1269         if (!vcpu->arch.sane) {
1270                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1271                 ret = -EINVAL;
1272                 goto out;
1273         }
1274
1275         /*
1276          * Interrupts could be timers for the guest which we have to inject
1277          * again, so let's postpone them until we're in the guest and if we
1278          * really did time things so badly, then we just exit again due to
1279          * a host external interrupt.
1280          */
1281         local_irq_disable();
1282         ret = kvmppc_prepare_to_enter(vcpu);
1283         if (ret <= 0) {
1284                 local_irq_enable();
1285                 goto out;
1286         }
1287
1288         /* Save FPU state in stack */
1289         if (current->thread.regs->msr & MSR_FP)
1290                 giveup_fpu(current);
1291         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1292         fpscr = current->thread.fpscr.val;
1293         fpexc_mode = current->thread.fpexc_mode;
1294
1295 #ifdef CONFIG_ALTIVEC
1296         /* Save Altivec state in stack */
1297         used_vr = current->thread.used_vr;
1298         if (used_vr) {
1299                 if (current->thread.regs->msr & MSR_VEC)
1300                         giveup_altivec(current);
1301                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1302                 vscr = current->thread.vscr;
1303                 vrsave = current->thread.vrsave;
1304         }
1305 #endif
1306
1307 #ifdef CONFIG_VSX
1308         /* Save VSX state in stack */
1309         used_vsr = current->thread.used_vsr;
1310         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1311                 __giveup_vsx(current);
1312 #endif
1313
1314         /* Remember the MSR with disabled extensions */
1315         ext_msr = current->thread.regs->msr;
1316
1317         /* Preload FPU if it's enabled */
1318         if (vcpu->arch.shared->msr & MSR_FP)
1319                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1320
1321         kvmppc_fix_ee_before_entry();
1322
1323         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1324
1325         /* No need for kvm_guest_exit. It's done in handle_exit.
1326            We also get here with interrupts enabled. */
1327
1328         /* Make sure we save the guest FPU/Altivec/VSX state */
1329         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1330
1331         current->thread.regs->msr = ext_msr;
1332
1333         /* Restore FPU/VSX state from stack */
1334         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1335         current->thread.fpscr.val = fpscr;
1336         current->thread.fpexc_mode = fpexc_mode;
1337
1338 #ifdef CONFIG_ALTIVEC
1339         /* Restore Altivec state from stack */
1340         if (used_vr && current->thread.used_vr) {
1341                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1342                 current->thread.vscr = vscr;
1343                 current->thread.vrsave = vrsave;
1344         }
1345         current->thread.used_vr = used_vr;
1346 #endif
1347
1348 #ifdef CONFIG_VSX
1349         current->thread.used_vsr = used_vsr;
1350 #endif
1351
1352 out:
1353         vcpu->mode = OUTSIDE_GUEST_MODE;
1354         return ret;
1355 }
1356
1357 /*
1358  * Get (and clear) the dirty memory log for a memory slot.
1359  */
1360 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1361                                          struct kvm_dirty_log *log)
1362 {
1363         struct kvm_memory_slot *memslot;
1364         struct kvm_vcpu *vcpu;
1365         ulong ga, ga_end;
1366         int is_dirty = 0;
1367         int r;
1368         unsigned long n;
1369
1370         mutex_lock(&kvm->slots_lock);
1371
1372         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1373         if (r)
1374                 goto out;
1375
1376         /* If nothing is dirty, don't bother messing with page tables. */
1377         if (is_dirty) {
1378                 memslot = id_to_memslot(kvm->memslots, log->slot);
1379
1380                 ga = memslot->base_gfn << PAGE_SHIFT;
1381                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1382
1383                 kvm_for_each_vcpu(n, vcpu, kvm)
1384                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1385
1386                 n = kvm_dirty_bitmap_bytes(memslot);
1387                 memset(memslot->dirty_bitmap, 0, n);
1388         }
1389
1390         r = 0;
1391 out:
1392         mutex_unlock(&kvm->slots_lock);
1393         return r;
1394 }
1395
1396 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1397                                          struct kvm_memory_slot *memslot)
1398 {
1399         return;
1400 }
1401
1402 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1403                                         struct kvm_memory_slot *memslot,
1404                                         struct kvm_userspace_memory_region *mem)
1405 {
1406         return 0;
1407 }
1408
1409 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1410                                 struct kvm_userspace_memory_region *mem,
1411                                 const struct kvm_memory_slot *old)
1412 {
1413         return;
1414 }
1415
1416 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1417                                         struct kvm_memory_slot *dont)
1418 {
1419         return;
1420 }
1421
1422 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1423                                          unsigned long npages)
1424 {
1425         return 0;
1426 }
1427
1428
1429 #ifdef CONFIG_PPC64
1430 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1431                                          struct kvm_ppc_smmu_info *info)
1432 {
1433         long int i;
1434         struct kvm_vcpu *vcpu;
1435
1436         info->flags = 0;
1437
1438         /* SLB is always 64 entries */
1439         info->slb_size = 64;
1440
1441         /* Standard 4k base page size segment */
1442         info->sps[0].page_shift = 12;
1443         info->sps[0].slb_enc = 0;
1444         info->sps[0].enc[0].page_shift = 12;
1445         info->sps[0].enc[0].pte_enc = 0;
1446
1447         /*
1448          * 64k large page size.
1449          * We only want to put this in if the CPUs we're emulating
1450          * support it, but unfortunately we don't have a vcpu easily
1451          * to hand here to test.  Just pick the first vcpu, and if
1452          * that doesn't exist yet, report the minimum capability,
1453          * i.e., no 64k pages.
1454          * 1T segment support goes along with 64k pages.
1455          */
1456         i = 1;
1457         vcpu = kvm_get_vcpu(kvm, 0);
1458         if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1459                 info->flags = KVM_PPC_1T_SEGMENTS;
1460                 info->sps[i].page_shift = 16;
1461                 info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1462                 info->sps[i].enc[0].page_shift = 16;
1463                 info->sps[i].enc[0].pte_enc = 1;
1464                 ++i;
1465         }
1466
1467         /* Standard 16M large page size segment */
1468         info->sps[i].page_shift = 24;
1469         info->sps[i].slb_enc = SLB_VSID_L;
1470         info->sps[i].enc[0].page_shift = 24;
1471         info->sps[i].enc[0].pte_enc = 0;
1472
1473         return 0;
1474 }
1475 #else
1476 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1477                                          struct kvm_ppc_smmu_info *info)
1478 {
1479         /* We should not get called */
1480         BUG();
1481 }
1482 #endif /* CONFIG_PPC64 */
1483
1484 static unsigned int kvm_global_user_count = 0;
1485 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1486
1487 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1488 {
1489         mutex_init(&kvm->arch.hpt_mutex);
1490
1491         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1492                 spin_lock(&kvm_global_user_count_lock);
1493                 if (++kvm_global_user_count == 1)
1494                         pSeries_disable_reloc_on_exc();
1495                 spin_unlock(&kvm_global_user_count_lock);
1496         }
1497         return 0;
1498 }
1499
1500 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1501 {
1502 #ifdef CONFIG_PPC64
1503         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1504 #endif
1505
1506         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1507                 spin_lock(&kvm_global_user_count_lock);
1508                 BUG_ON(kvm_global_user_count == 0);
1509                 if (--kvm_global_user_count == 0)
1510                         pSeries_enable_reloc_on_exc();
1511                 spin_unlock(&kvm_global_user_count_lock);
1512         }
1513 }
1514
1515 static int kvmppc_core_check_processor_compat_pr(void)
1516 {
1517         /* we are always compatible */
1518         return 0;
1519 }
1520
1521 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1522                                  unsigned int ioctl, unsigned long arg)
1523 {
1524         return -ENOTTY;
1525 }
1526
1527 static struct kvmppc_ops kvmppc_pr_ops = {
1528         .is_hv_enabled = false,
1529         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1530         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1531         .get_one_reg = kvmppc_get_one_reg_pr,
1532         .set_one_reg = kvmppc_set_one_reg_pr,
1533         .vcpu_load   = kvmppc_core_vcpu_load_pr,
1534         .vcpu_put    = kvmppc_core_vcpu_put_pr,
1535         .set_msr     = kvmppc_set_msr_pr,
1536         .vcpu_run    = kvmppc_vcpu_run_pr,
1537         .vcpu_create = kvmppc_core_vcpu_create_pr,
1538         .vcpu_free   = kvmppc_core_vcpu_free_pr,
1539         .check_requests = kvmppc_core_check_requests_pr,
1540         .get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1541         .flush_memslot = kvmppc_core_flush_memslot_pr,
1542         .prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1543         .commit_memory_region = kvmppc_core_commit_memory_region_pr,
1544         .unmap_hva = kvm_unmap_hva_pr,
1545         .unmap_hva_range = kvm_unmap_hva_range_pr,
1546         .age_hva  = kvm_age_hva_pr,
1547         .test_age_hva = kvm_test_age_hva_pr,
1548         .set_spte_hva = kvm_set_spte_hva_pr,
1549         .mmu_destroy  = kvmppc_mmu_destroy_pr,
1550         .free_memslot = kvmppc_core_free_memslot_pr,
1551         .create_memslot = kvmppc_core_create_memslot_pr,
1552         .init_vm = kvmppc_core_init_vm_pr,
1553         .destroy_vm = kvmppc_core_destroy_vm_pr,
1554         .check_processor_compat = kvmppc_core_check_processor_compat_pr,
1555         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1556         .emulate_op = kvmppc_core_emulate_op_pr,
1557         .emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1558         .emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1559         .fast_vcpu_kick = kvm_vcpu_kick,
1560         .arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1561 };
1562
1563 static int kvmppc_book3s_init_pr(void)
1564 {
1565         int r;
1566
1567         r = kvm_init(&kvmppc_pr_ops, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1568
1569         if (r)
1570                 return r;
1571
1572         r = kvmppc_mmu_hpte_sysinit();
1573
1574         return r;
1575 }
1576
1577 static void kvmppc_book3s_exit_pr(void)
1578 {
1579         kvmppc_mmu_hpte_sysexit();
1580         kvm_exit();
1581 }
1582
1583 module_init(kvmppc_book3s_init_pr);
1584 module_exit(kvmppc_book3s_exit_pr);