arch/powerpc/kvm/book3s_pr.c

   1 /*
   2  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   3  *
   4  * Authors:
   5  *    Alexander Graf <agraf@suse.de>
   6  *    Kevin Wolf <mail@kevin-wolf.de>
   7  *    Paul Mackerras <paulus@samba.org>
   8  *
   9  * Description:
  10  * Functions relating to running KVM on Book 3S processors where
  11  * we don't have access to hypervisor mode, and we run the guest
  12  * in problem state (user mode).
  13  *
  14  * This file is derived from arch/powerpc/kvm/44x.c,
  15  * by Hollis Blanchard <hollisb@us.ibm.com>.
  16  *
  17  * This program is free software; you can redistribute it and/or modify
  18  * it under the terms of the GNU General Public License, version 2, as
  19  * published by the Free Software Foundation.
  20  */
  21
  22 #include <linux/kvm_host.h>
  23 #include <linux/export.h>
  24 #include <linux/err.h>
  25 #include <linux/slab.h>
  26
  27 #include <asm/reg.h>
  28 #include <asm/cputable.h>
  29 #include <asm/cacheflush.h>
  30 #include <asm/tlbflush.h>
  31 #include <asm/uaccess.h>
  32 #include <asm/io.h>
  33 #include <asm/kvm_ppc.h>
  34 #include <asm/kvm_book3s.h>
  35 #include <asm/mmu_context.h>
  36 #include <asm/switch_to.h>
  37 #include <asm/firmware.h>
  38 #include <asm/hvcall.h>
  39 #include <linux/gfp.h>
  40 #include <linux/sched.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/highmem.h>
  43 #include <linux/module.h>
  44
  45 #include "book3s.h"
  46
  47 #define CREATE_TRACE_POINTS
  48 #include "trace_pr.h"
  49
  50 /* #define EXIT_DEBUG */
  51 /* #define DEBUG_EXT */
  52
  53 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  54                              ulong msr);
  55
  56 /* Some compatibility defines */
  57 #ifdef CONFIG_PPC_BOOK3S_32
  58 #define MSR_USER32 MSR_USER
  59 #define MSR_USER64 MSR_USER
  60 #define HW_PAGE_SIZE PAGE_SIZE
  61 #endif
  62
  63 static void kvmppc_core_vcpu_load_pr(struct kvm_vcpu *vcpu, int cpu)
  64 {
  65 #ifdef CONFIG_PPC_BOOK3S_64
  66         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  67         memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
  68         svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
  69         svcpu->in_use = 0;
  70         svcpu_put(svcpu);
  71 #endif
  72         vcpu->cpu = smp_processor_id();
  73 #ifdef CONFIG_PPC_BOOK3S_32
  74         current->thread.kvm_shadow_vcpu = vcpu->arch.shadow_vcpu;
  75 #endif
  76 }
  77
  78 static void kvmppc_core_vcpu_put_pr(struct kvm_vcpu *vcpu)
  79 {
  80 #ifdef CONFIG_PPC_BOOK3S_64
  81         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  82         if (svcpu->in_use) {
  83                 kvmppc_copy_from_svcpu(vcpu, svcpu);
  84         }
  85         memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
  86         to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
  87         svcpu_put(svcpu);
  88 #endif
  89
  90         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
  91         vcpu->cpu = -1;
  92 }
  93
  94 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
  95 void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu *svcpu,
  96                           struct kvm_vcpu *vcpu)
  97 {
  98         svcpu->gpr[0] = vcpu->arch.gpr[0];
  99         svcpu->gpr[1] = vcpu->arch.gpr[1];
 100         svcpu->gpr[2] = vcpu->arch.gpr[2];
 101         svcpu->gpr[3] = vcpu->arch.gpr[3];
 102         svcpu->gpr[4] = vcpu->arch.gpr[4];
 103         svcpu->gpr[5] = vcpu->arch.gpr[5];
 104         svcpu->gpr[6] = vcpu->arch.gpr[6];
 105         svcpu->gpr[7] = vcpu->arch.gpr[7];
 106         svcpu->gpr[8] = vcpu->arch.gpr[8];
 107         svcpu->gpr[9] = vcpu->arch.gpr[9];
 108         svcpu->gpr[10] = vcpu->arch.gpr[10];
 109         svcpu->gpr[11] = vcpu->arch.gpr[11];
 110         svcpu->gpr[12] = vcpu->arch.gpr[12];
 111         svcpu->gpr[13] = vcpu->arch.gpr[13];
 112         svcpu->cr  = vcpu->arch.cr;
 113         svcpu->xer = vcpu->arch.xer;
 114         svcpu->ctr = vcpu->arch.ctr;
 115         svcpu->lr  = vcpu->arch.lr;
 116         svcpu->pc  = vcpu->arch.pc;
 117         svcpu->in_use = true;
 118 }
 119
 120 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
 121 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu,
 122                             struct kvmppc_book3s_shadow_vcpu *svcpu)
 123 {
 124         /*
 125          * vcpu_put would just call us again because in_use hasn't
 126          * been updated yet.
 127          */
 128         preempt_disable();
 129
 130         /*
 131          * Maybe we were already preempted and synced the svcpu from
 132          * our preempt notifiers. Don't bother touching this svcpu then.
 133          */
 134         if (!svcpu->in_use)
 135                 goto out;
 136
 137         vcpu->arch.gpr[0] = svcpu->gpr[0];
 138         vcpu->arch.gpr[1] = svcpu->gpr[1];
 139         vcpu->arch.gpr[2] = svcpu->gpr[2];
 140         vcpu->arch.gpr[3] = svcpu->gpr[3];
 141         vcpu->arch.gpr[4] = svcpu->gpr[4];
 142         vcpu->arch.gpr[5] = svcpu->gpr[5];
 143         vcpu->arch.gpr[6] = svcpu->gpr[6];
 144         vcpu->arch.gpr[7] = svcpu->gpr[7];
 145         vcpu->arch.gpr[8] = svcpu->gpr[8];
 146         vcpu->arch.gpr[9] = svcpu->gpr[9];
 147         vcpu->arch.gpr[10] = svcpu->gpr[10];
 148         vcpu->arch.gpr[11] = svcpu->gpr[11];
 149         vcpu->arch.gpr[12] = svcpu->gpr[12];
 150         vcpu->arch.gpr[13] = svcpu->gpr[13];
 151         vcpu->arch.cr  = svcpu->cr;
 152         vcpu->arch.xer = svcpu->xer;
 153         vcpu->arch.ctr = svcpu->ctr;
 154         vcpu->arch.lr  = svcpu->lr;
 155         vcpu->arch.pc  = svcpu->pc;
 156         vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
 157         vcpu->arch.fault_dar   = svcpu->fault_dar;
 158         vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
 159         vcpu->arch.last_inst   = svcpu->last_inst;
 160         svcpu->in_use = false;
 161
 162 out:
 163         preempt_enable();
 164 }
 165
 166 static int kvmppc_core_check_requests_pr(struct kvm_vcpu *vcpu)
 167 {
 168         int r = 1; /* Indicate we want to get back into the guest */
 169
 170         /* We misuse TLB_FLUSH to indicate that we want to clear
 171            all shadow cache entries */
 172         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
 173                 kvmppc_mmu_pte_flush(vcpu, 0, 0);
 174
 175         return r;
 176 }
 177
 178 /************* MMU Notifiers *************/
 179 static void do_kvm_unmap_hva(struct kvm *kvm, unsigned long start,
 180                              unsigned long end)
 181 {
 182         long i;
 183         struct kvm_vcpu *vcpu;
 184         struct kvm_memslots *slots;
 185         struct kvm_memory_slot *memslot;
 186
 187         slots = kvm_memslots(kvm);
 188         kvm_for_each_memslot(memslot, slots) {
 189                 unsigned long hva_start, hva_end;
 190                 gfn_t gfn, gfn_end;
 191
 192                 hva_start = max(start, memslot->userspace_addr);
 193                 hva_end = min(end, memslot->userspace_addr +
 194                                         (memslot->npages << PAGE_SHIFT));
 195                 if (hva_start >= hva_end)
 196                         continue;
 197                 /*
 198                  * {gfn(page) | page intersects with [hva_start, hva_end)} =
 199                  * {gfn, gfn+1, ..., gfn_end-1}.
 200                  */
 201                 gfn = hva_to_gfn_memslot(hva_start, memslot);
 202                 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
 203                 kvm_for_each_vcpu(i, vcpu, kvm)
 204                         kvmppc_mmu_pte_pflush(vcpu, gfn << PAGE_SHIFT,
 205                                               gfn_end << PAGE_SHIFT);
 206         }
 207 }
 208
 209 static int kvm_unmap_hva_pr(struct kvm *kvm, unsigned long hva)
 210 {
 211         trace_kvm_unmap_hva(hva);
 212
 213         do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
 214
 215         return 0;
 216 }
 217
 218 static int kvm_unmap_hva_range_pr(struct kvm *kvm, unsigned long start,
 219                                   unsigned long end)
 220 {
 221         do_kvm_unmap_hva(kvm, start, end);
 222
 223         return 0;
 224 }
 225
 226 static int kvm_age_hva_pr(struct kvm *kvm, unsigned long hva)
 227 {
 228         /* XXX could be more clever ;) */
 229         return 0;
 230 }
 231
 232 static int kvm_test_age_hva_pr(struct kvm *kvm, unsigned long hva)
 233 {
 234         /* XXX could be more clever ;) */
 235         return 0;
 236 }
 237
 238 static void kvm_set_spte_hva_pr(struct kvm *kvm, unsigned long hva, pte_t pte)
 239 {
 240         /* The page will get remapped properly on its next fault */
 241         do_kvm_unmap_hva(kvm, hva, hva + PAGE_SIZE);
 242 }
 243
 244 /*****************************************/
 245
 246 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 247 {
 248         ulong smsr = vcpu->arch.shared->msr;
 249
 250         /* Guest MSR values */
 251         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE;
 252         /* Process MSR values */
 253         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 254         /* External providers the guest reserved */
 255         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
 256         /* 64-bit Process MSR values */
 257 #ifdef CONFIG_PPC_BOOK3S_64
 258         smsr |= MSR_ISF | MSR_HV;
 259 #endif
 260         vcpu->arch.shadow_msr = smsr;
 261 }
 262
 263 static void kvmppc_set_msr_pr(struct kvm_vcpu *vcpu, u64 msr)
 264 {
 265         ulong old_msr = vcpu->arch.shared->msr;
 266
 267 #ifdef EXIT_DEBUG
 268         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 269 #endif
 270
 271         msr &= to_book3s(vcpu)->msr_mask;
 272         vcpu->arch.shared->msr = msr;
 273         kvmppc_recalc_shadow_msr(vcpu);
 274
 275         if (msr & MSR_POW) {
 276                 if (!vcpu->arch.pending_exceptions) {
 277                         kvm_vcpu_block(vcpu);
 278                         clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 279                         vcpu->stat.halt_wakeup++;
 280
 281                         /* Unset POW bit after we woke up */
 282                         msr &= ~MSR_POW;
 283                         vcpu->arch.shared->msr = msr;
 284                 }
 285         }
 286
 287         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 288                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 289                 kvmppc_mmu_flush_segments(vcpu);
 290                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 291
 292                 /* Preload magic page segment when in kernel mode */
 293                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 294                         struct kvm_vcpu_arch *a = &vcpu->arch;
 295
 296                         if (msr & MSR_DR)
 297                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 298                         else
 299                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 300                 }
 301         }
 302
 303         /*
 304          * When switching from 32 to 64-bit, we may have a stale 32-bit
 305          * magic page around, we need to flush it. Typically 32-bit magic
 306          * page will be instanciated when calling into RTAS. Note: We
 307          * assume that such transition only happens while in kernel mode,
 308          * ie, we never transition from user 32-bit to kernel 64-bit with
 309          * a 32-bit magic page around.
 310          */
 311         if (vcpu->arch.magic_page_pa &&
 312             !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
 313                 /* going from RTAS to normal kernel code */
 314                 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
 315                                      ~0xFFFUL);
 316         }
 317
 318         /* Preload FPU if it's enabled */
 319         if (vcpu->arch.shared->msr & MSR_FP)
 320                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 321 }
 322
 323 void kvmppc_set_pvr_pr(struct kvm_vcpu *vcpu, u32 pvr)
 324 {
 325         u32 host_pvr;
 326
 327         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 328         vcpu->arch.pvr = pvr;
 329 #ifdef CONFIG_PPC_BOOK3S_64
 330         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 331                 kvmppc_mmu_book3s_64_init(vcpu);
 332                 if (!to_book3s(vcpu)->hior_explicit)
 333                         to_book3s(vcpu)->hior = 0xfff00000;
 334                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 335                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 336         } else
 337 #endif
 338         {
 339                 kvmppc_mmu_book3s_32_init(vcpu);
 340                 if (!to_book3s(vcpu)->hior_explicit)
 341                         to_book3s(vcpu)->hior = 0;
 342                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 343                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 344         }
 345
 346         kvmppc_sanity_check(vcpu);
 347
 348         /* If we are in hypervisor level on 970, we can tell the CPU to
 349          * treat DCBZ as 32 bytes store */
 350         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 351         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 352             !strcmp(cur_cpu_spec->platform, "ppc970"))
 353                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 354
 355         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 356            really needs them in a VM on Cell and force disable them. */
 357         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 358                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 359
 360         /*
 361          * If they're asking for POWER6 or later, set the flag
 362          * indicating that we can do multiple large page sizes
 363          * and 1TB segments.
 364          * Also set the flag that indicates that tlbie has the large
 365          * page bit in the RB operand instead of the instruction.
 366          */
 367         switch (PVR_VER(pvr)) {
 368         case PVR_POWER6:
 369         case PVR_POWER7:
 370         case PVR_POWER7p:
 371         case PVR_POWER8:
 372                 vcpu->arch.hflags |= BOOK3S_HFLAG_MULTI_PGSIZE |
 373                         BOOK3S_HFLAG_NEW_TLBIE;
 374                 break;
 375         }
 376
 377 #ifdef CONFIG_PPC_BOOK3S_32
 378         /* 32 bit Book3S always has 32 byte dcbz */
 379         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 380 #endif
 381
 382         /* On some CPUs we can execute paired single operations natively */
 383         asm ( "mfpvr %0" : "=r"(host_pvr));
 384         switch (host_pvr) {
 385         case 0x00080200:        /* lonestar 2.0 */
 386         case 0x00088202:        /* lonestar 2.2 */
 387         case 0x70000100:        /* gekko 1.0 */
 388         case 0x00080100:        /* gekko 2.0 */
 389         case 0x00083203:        /* gekko 2.3a */
 390         case 0x00083213:        /* gekko 2.3b */
 391         case 0x00083204:        /* gekko 2.4 */
 392         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 393         case 0x00087200:        /* broadway */
 394                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 395                 /* Enable HID2.PSE - in case we need it later */
 396                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 397         }
 398 }
 399
 400 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 401  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 402  * emulate 32 bytes dcbz length.
 403  *
 404  * The Book3s_64 inventors also realized this case and implemented a special bit
 405  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 406  *
 407  * My approach here is to patch the dcbz instruction on executing pages.
 408  */
 409 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 410 {
 411         struct page *hpage;
 412         u64 hpage_offset;
 413         u32 *page;
 414         int i;
 415
 416         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 417         if (is_error_page(hpage))
 418                 return;
 419
 420         hpage_offset = pte->raddr & ~PAGE_MASK;
 421         hpage_offset &= ~0xFFFULL;
 422         hpage_offset /= 4;
 423
 424         get_page(hpage);
 425         page = kmap_atomic(hpage);
 426
 427         /* patch dcbz into reserved instruction, so we trap */
 428         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 429                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 430                         page[i] &= 0xfffffff7;
 431
 432         kunmap_atomic(page);
 433         put_page(hpage);
 434 }
 435
 436 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 437 {
 438         ulong mp_pa = vcpu->arch.magic_page_pa;
 439
 440         if (!(vcpu->arch.shared->msr & MSR_SF))
 441                 mp_pa = (uint32_t)mp_pa;
 442
 443         if (unlikely(mp_pa) &&
 444             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 445                 return 1;
 446         }
 447
 448         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 449 }
 450
 451 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 452                             ulong eaddr, int vec)
 453 {
 454         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 455         bool iswrite = false;
 456         int r = RESUME_GUEST;
 457         int relocated;
 458         int page_found = 0;
 459         struct kvmppc_pte pte;
 460         bool is_mmio = false;
 461         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 462         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 463         u64 vsid;
 464
 465         relocated = data ? dr : ir;
 466         if (data && (vcpu->arch.fault_dsisr & DSISR_ISSTORE))
 467                 iswrite = true;
 468
 469         /* Resolve real address if translation turned on */
 470         if (relocated) {
 471                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data, iswrite);
 472         } else {
 473                 pte.may_execute = true;
 474                 pte.may_read = true;
 475                 pte.may_write = true;
 476                 pte.raddr = eaddr & KVM_PAM;
 477                 pte.eaddr = eaddr;
 478                 pte.vpage = eaddr >> 12;
 479                 pte.page_size = MMU_PAGE_64K;
 480         }
 481
 482         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 483         case 0:
 484                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 485                 break;
 486         case MSR_DR:
 487         case MSR_IR:
 488                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 489
 490                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 491                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 492                 else
 493                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 494                 pte.vpage |= vsid;
 495
 496                 if (vsid == -1)
 497                         page_found = -EINVAL;
 498                 break;
 499         }
 500
 501         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 502            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 503                 /*
 504                  * If we do the dcbz hack, we have to NX on every execution,
 505                  * so we can patch the executing code. This renders our guest
 506                  * NX-less.
 507                  */
 508                 pte.may_execute = !data;
 509         }
 510
 511         if (page_found == -ENOENT) {
 512                 /* Page not found in guest PTE entries */
 513                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 514                 vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr;
 515                 vcpu->arch.shared->msr |=
 516                         vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
 517                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 518         } else if (page_found == -EPERM) {
 519                 /* Storage protection */
 520                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 521                 vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
 522                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 523                 vcpu->arch.shared->msr |=
 524                         vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
 525                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 526         } else if (page_found == -EINVAL) {
 527                 /* Page not found in guest SLB */
 528                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 529                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 530         } else if (!is_mmio &&
 531                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 532                 if (data && !(vcpu->arch.fault_dsisr & DSISR_NOHPTE)) {
 533                         /*
 534                          * There is already a host HPTE there, presumably
 535                          * a read-only one for a page the guest thinks
 536                          * is writable, so get rid of it first.
 537                          */
 538                         kvmppc_mmu_unmap_page(vcpu, &pte);
 539                 }
 540                 /* The guest's PTE is not mapped yet. Map on the host */
 541                 kvmppc_mmu_map_page(vcpu, &pte, iswrite);
 542                 if (data)
 543                         vcpu->stat.sp_storage++;
 544                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 545                          (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 546                         kvmppc_patch_dcbz(vcpu, &pte);
 547         } else {
 548                 /* MMIO */
 549                 vcpu->stat.mmio_exits++;
 550                 vcpu->arch.paddr_accessed = pte.raddr;
 551                 vcpu->arch.vaddr_accessed = pte.eaddr;
 552                 r = kvmppc_emulate_mmio(run, vcpu);
 553                 if ( r == RESUME_HOST_NV )
 554                         r = RESUME_HOST;
 555         }
 556
 557         return r;
 558 }
 559
 560 static inline int get_fpr_index(int i)
 561 {
 562         return i * TS_FPRWIDTH;
 563 }
 564
 565 /* Give up external provider (FPU, Altivec, VSX) */
 566 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 567 {
 568         struct thread_struct *t = &current->thread;
 569         u64 *vcpu_fpr = vcpu->arch.fpr;
 570 #ifdef CONFIG_VSX
 571         u64 *vcpu_vsx = vcpu->arch.vsr;
 572 #endif
 573         u64 *thread_fpr = &t->fp_state.fpr[0][0];
 574         int i;
 575
 576         /*
 577          * VSX instructions can access FP and vector registers, so if
 578          * we are giving up VSX, make sure we give up FP and VMX as well.
 579          */
 580         if (msr & MSR_VSX)
 581                 msr |= MSR_FP | MSR_VEC;
 582
 583         msr &= vcpu->arch.guest_owned_ext;
 584         if (!msr)
 585                 return;
 586
 587 #ifdef DEBUG_EXT
 588         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 589 #endif
 590
 591         if (msr & MSR_FP) {
 592                 /*
 593                  * Note that on CPUs with VSX, giveup_fpu stores
 594                  * both the traditional FP registers and the added VSX
 595                  * registers into thread.fp_state.fpr[].
 596                  */
 597                 if (current->thread.regs->msr & MSR_FP)
 598                         giveup_fpu(current);
 599                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 600                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 601
 602                 vcpu->arch.fpscr = t->fp_state.fpscr;
 603
 604 #ifdef CONFIG_VSX
 605                 if (cpu_has_feature(CPU_FTR_VSX))
 606                         for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
 607                                 vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 608 #endif
 609         }
 610
 611 #ifdef CONFIG_ALTIVEC
 612         if (msr & MSR_VEC) {
 613                 if (current->thread.regs->msr & MSR_VEC)
 614                         giveup_altivec(current);
 615                 memcpy(vcpu->arch.vr, t->vr_state.vr, sizeof(vcpu->arch.vr));
 616                 vcpu->arch.vscr = t->vr_state.vscr;
 617         }
 618 #endif
 619
 620         vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
 621         kvmppc_recalc_shadow_msr(vcpu);
 622 }
 623
 624 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 625 {
 626         ulong srr0 = kvmppc_get_pc(vcpu);
 627         u32 last_inst = kvmppc_get_last_inst(vcpu);
 628         int ret;
 629
 630         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 631         if (ret == -ENOENT) {
 632                 ulong msr = vcpu->arch.shared->msr;
 633
 634                 msr = kvmppc_set_field(msr, 33, 33, 1);
 635                 msr = kvmppc_set_field(msr, 34, 36, 0);
 636                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 637                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 638                 return EMULATE_AGAIN;
 639         }
 640
 641         return EMULATE_DONE;
 642 }
 643
 644 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 645 {
 646
 647         /* Need to do paired single emulation? */
 648         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 649                 return EMULATE_DONE;
 650
 651         /* Read out the instruction */
 652         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 653                 /* Need to emulate */
 654                 return EMULATE_FAIL;
 655
 656         return EMULATE_AGAIN;
 657 }
 658
 659 /* Handle external providers (FPU, Altivec, VSX) */
 660 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 661                              ulong msr)
 662 {
 663         struct thread_struct *t = &current->thread;
 664         u64 *vcpu_fpr = vcpu->arch.fpr;
 665 #ifdef CONFIG_VSX
 666         u64 *vcpu_vsx = vcpu->arch.vsr;
 667 #endif
 668         u64 *thread_fpr = &t->fp_state.fpr[0][0];
 669         int i;
 670
 671         /* When we have paired singles, we emulate in software */
 672         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 673                 return RESUME_GUEST;
 674
 675         if (!(vcpu->arch.shared->msr & msr)) {
 676                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 677                 return RESUME_GUEST;
 678         }
 679
 680         if (msr == MSR_VSX) {
 681                 /* No VSX?  Give an illegal instruction interrupt */
 682 #ifdef CONFIG_VSX
 683                 if (!cpu_has_feature(CPU_FTR_VSX))
 684 #endif
 685                 {
 686                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 687                         return RESUME_GUEST;
 688                 }
 689
 690                 /*
 691                  * We have to load up all the FP and VMX registers before
 692                  * we can let the guest use VSX instructions.
 693                  */
 694                 msr = MSR_FP | MSR_VEC | MSR_VSX;
 695         }
 696
 697         /* See if we already own all the ext(s) needed */
 698         msr &= ~vcpu->arch.guest_owned_ext;
 699         if (!msr)
 700                 return RESUME_GUEST;
 701
 702 #ifdef DEBUG_EXT
 703         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 704 #endif
 705
 706         if (msr & MSR_FP) {
 707                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 708                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 709 #ifdef CONFIG_VSX
 710                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
 711                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 712 #endif
 713                 t->fp_state.fpscr = vcpu->arch.fpscr;
 714                 t->fpexc_mode = 0;
 715                 kvmppc_load_up_fpu();
 716         }
 717
 718         if (msr & MSR_VEC) {
 719 #ifdef CONFIG_ALTIVEC
 720                 memcpy(t->vr_state.vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 721                 t->vr_state.vscr = vcpu->arch.vscr;
 722                 t->vrsave = -1;
 723                 kvmppc_load_up_altivec();
 724 #endif
 725         }
 726
 727         current->thread.regs->msr |= msr;
 728         vcpu->arch.guest_owned_ext |= msr;
 729         kvmppc_recalc_shadow_msr(vcpu);
 730
 731         return RESUME_GUEST;
 732 }
 733
 734 /*
 735  * Kernel code using FP or VMX could have flushed guest state to
 736  * the thread_struct; if so, get it back now.
 737  */
 738 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
 739 {
 740         unsigned long lost_ext;
 741
 742         lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
 743         if (!lost_ext)
 744                 return;
 745
 746         if (lost_ext & MSR_FP)
 747                 kvmppc_load_up_fpu();
 748 #ifdef CONFIG_ALTIVEC
 749         if (lost_ext & MSR_VEC)
 750                 kvmppc_load_up_altivec();
 751 #endif
 752         current->thread.regs->msr |= lost_ext;
 753 }
 754
 755 int kvmppc_handle_exit_pr(struct kvm_run *run, struct kvm_vcpu *vcpu,
 756                           unsigned int exit_nr)
 757 {
 758         int r = RESUME_HOST;
 759         int s;
 760
 761         vcpu->stat.sum_exits++;
 762
 763         run->exit_reason = KVM_EXIT_UNKNOWN;
 764         run->ready_for_interrupt_injection = 1;
 765
 766         /* We get here with MSR.EE=1 */
 767
 768         trace_kvm_exit(exit_nr, vcpu);
 769         kvm_guest_exit();
 770
 771         switch (exit_nr) {
 772         case BOOK3S_INTERRUPT_INST_STORAGE:
 773         {
 774                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 775                 vcpu->stat.pf_instruc++;
 776
 777 #ifdef CONFIG_PPC_BOOK3S_32
 778                 /* We set segments as unused segments when invalidating them. So
 779                  * treat the respective fault as segment fault. */
 780                 {
 781                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 782                         u32 sr;
 783
 784                         svcpu = svcpu_get(vcpu);
 785                         sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
 786                         svcpu_put(svcpu);
 787                         if (sr == SR_INVALID) {
 788                                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 789                                 r = RESUME_GUEST;
 790                                 break;
 791                         }
 792                 }
 793 #endif
 794
 795                 /* only care about PTEG not found errors, but leave NX alone */
 796                 if (shadow_srr1 & 0x40000000) {
 797                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
 798                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 799                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 800                         vcpu->stat.sp_instruc++;
 801                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 802                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 803                         /*
 804                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 805                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 806                          *     that no guest that needs the dcbz hack does NX.
 807                          */
 808                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 809                         r = RESUME_GUEST;
 810                 } else {
 811                         vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
 812                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 813                         r = RESUME_GUEST;
 814                 }
 815                 break;
 816         }
 817         case BOOK3S_INTERRUPT_DATA_STORAGE:
 818         {
 819                 ulong dar = kvmppc_get_fault_dar(vcpu);
 820                 u32 fault_dsisr = vcpu->arch.fault_dsisr;
 821                 vcpu->stat.pf_storage++;
 822
 823 #ifdef CONFIG_PPC_BOOK3S_32
 824                 /* We set segments as unused segments when invalidating them. So
 825                  * treat the respective fault as segment fault. */
 826                 {
 827                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 828                         u32 sr;
 829
 830                         svcpu = svcpu_get(vcpu);
 831                         sr = svcpu->sr[dar >> SID_SHIFT];
 832                         svcpu_put(svcpu);
 833                         if (sr == SR_INVALID) {
 834                                 kvmppc_mmu_map_segment(vcpu, dar);
 835                                 r = RESUME_GUEST;
 836                                 break;
 837                         }
 838                 }
 839 #endif
 840
 841                 /*
 842                  * We need to handle missing shadow PTEs, and
 843                  * protection faults due to us mapping a page read-only
 844                  * when the guest thinks it is writable.
 845                  */
 846                 if (fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT)) {
 847                         int idx = srcu_read_lock(&vcpu->kvm->srcu);
 848                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 849                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 850                 } else {
 851                         vcpu->arch.shared->dar = dar;
 852                         vcpu->arch.shared->dsisr = fault_dsisr;
 853                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 854                         r = RESUME_GUEST;
 855                 }
 856                 break;
 857         }
 858         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 859                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 860                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 861                         kvmppc_book3s_queue_irqprio(vcpu,
 862                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 863                 }
 864                 r = RESUME_GUEST;
 865                 break;
 866         case BOOK3S_INTERRUPT_INST_SEGMENT:
 867                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 868                         kvmppc_book3s_queue_irqprio(vcpu,
 869                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 870                 }
 871                 r = RESUME_GUEST;
 872                 break;
 873         /* We're good on these - the host merely wanted to get our attention */
 874         case BOOK3S_INTERRUPT_DECREMENTER:
 875         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 876                 vcpu->stat.dec_exits++;
 877                 r = RESUME_GUEST;
 878                 break;
 879         case BOOK3S_INTERRUPT_EXTERNAL:
 880         case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
 881         case BOOK3S_INTERRUPT_EXTERNAL_HV:
 882                 vcpu->stat.ext_intr_exits++;
 883                 r = RESUME_GUEST;
 884                 break;
 885         case BOOK3S_INTERRUPT_PERFMON:
 886                 r = RESUME_GUEST;
 887                 break;
 888         case BOOK3S_INTERRUPT_PROGRAM:
 889         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 890         {
 891                 enum emulation_result er;
 892                 ulong flags;
 893
 894 program_interrupt:
 895                 flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
 896
 897                 if (vcpu->arch.shared->msr & MSR_PR) {
 898 #ifdef EXIT_DEBUG
 899                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 900 #endif
 901                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 902                             (INS_DCBZ & 0xfffffff7)) {
 903                                 kvmppc_core_queue_program(vcpu, flags);
 904                                 r = RESUME_GUEST;
 905                                 break;
 906                         }
 907                 }
 908
 909                 vcpu->stat.emulated_inst_exits++;
 910                 er = kvmppc_emulate_instruction(run, vcpu);
 911                 switch (er) {
 912                 case EMULATE_DONE:
 913                         r = RESUME_GUEST_NV;
 914                         break;
 915                 case EMULATE_AGAIN:
 916                         r = RESUME_GUEST;
 917                         break;
 918                 case EMULATE_FAIL:
 919                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 920                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 921                         kvmppc_core_queue_program(vcpu, flags);
 922                         r = RESUME_GUEST;
 923                         break;
 924                 case EMULATE_DO_MMIO:
 925                         run->exit_reason = KVM_EXIT_MMIO;
 926                         r = RESUME_HOST_NV;
 927                         break;
 928                 case EMULATE_EXIT_USER:
 929                         r = RESUME_HOST_NV;
 930                         break;
 931                 default:
 932                         BUG();
 933                 }
 934                 break;
 935         }
 936         case BOOK3S_INTERRUPT_SYSCALL:
 937                 if (vcpu->arch.papr_enabled &&
 938                     (kvmppc_get_last_sc(vcpu) == 0x44000022) &&
 939                     !(vcpu->arch.shared->msr & MSR_PR)) {
 940                         /* SC 1 papr hypercalls */
 941                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
 942                         int i;
 943
 944 #ifdef CONFIG_PPC_BOOK3S_64
 945                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
 946                                 r = RESUME_GUEST;
 947                                 break;
 948                         }
 949 #endif
 950
 951                         run->papr_hcall.nr = cmd;
 952                         for (i = 0; i < 9; ++i) {
 953                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
 954                                 run->papr_hcall.args[i] = gpr;
 955                         }
 956                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
 957                         vcpu->arch.hcall_needed = 1;
 958                         r = RESUME_HOST;
 959                 } else if (vcpu->arch.osi_enabled &&
 960                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 961                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 962                         /* MOL hypercalls */
 963                         u64 *gprs = run->osi.gprs;
 964                         int i;
 965
 966                         run->exit_reason = KVM_EXIT_OSI;
 967                         for (i = 0; i < 32; i++)
 968                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 969                         vcpu->arch.osi_needed = 1;
 970                         r = RESUME_HOST_NV;
 971                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 972                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 973                         /* KVM PV hypercalls */
 974                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 975                         r = RESUME_GUEST;
 976                 } else {
 977                         /* Guest syscalls */
 978                         vcpu->stat.syscall_exits++;
 979                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 980                         r = RESUME_GUEST;
 981                 }
 982                 break;
 983         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 984         case BOOK3S_INTERRUPT_ALTIVEC:
 985         case BOOK3S_INTERRUPT_VSX:
 986         {
 987                 int ext_msr = 0;
 988
 989                 switch (exit_nr) {
 990                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 991                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 992                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 993                 }
 994
 995                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
 996                 case EMULATE_DONE:
 997                         /* everything ok - let's enable the ext */
 998                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
 999                         break;
1000                 case EMULATE_FAIL:
1001                         /* we need to emulate this instruction */
1002                         goto program_interrupt;
1003                         break;
1004                 default:
1005                         /* nothing to worry about - go again */
1006                         break;
1007                 }
1008                 break;
1009         }
1010         case BOOK3S_INTERRUPT_ALIGNMENT:
1011                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
1012                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
1013                                 kvmppc_get_last_inst(vcpu));
1014                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
1015                                 kvmppc_get_last_inst(vcpu));
1016                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1017                 }
1018                 r = RESUME_GUEST;
1019                 break;
1020         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1021         case BOOK3S_INTERRUPT_TRACE:
1022                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1023                 r = RESUME_GUEST;
1024                 break;
1025         default:
1026         {
1027                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
1028                 /* Ugh - bork here! What did we get? */
1029                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1030                         exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
1031                 r = RESUME_HOST;
1032                 BUG();
1033                 break;
1034         }
1035         }
1036
1037         if (!(r & RESUME_HOST)) {
1038                 /* To avoid clobbering exit_reason, only check for signals if
1039                  * we aren't already exiting to userspace for some other
1040                  * reason. */
1041
1042                 /*
1043                  * Interrupts could be timers for the guest which we have to
1044                  * inject again, so let's postpone them until we're in the guest
1045                  * and if we really did time things so badly, then we just exit
1046                  * again due to a host external interrupt.
1047                  */
1048                 local_irq_disable();
1049                 s = kvmppc_prepare_to_enter(vcpu);
1050                 if (s <= 0) {
1051                         local_irq_enable();
1052                         r = s;
1053                 } else {
1054                         kvmppc_fix_ee_before_entry();
1055                 }
1056                 kvmppc_handle_lost_ext(vcpu);
1057         }
1058
1059         trace_kvm_book3s_reenter(r, vcpu);
1060
1061         return r;
1062 }
1063
1064 static int kvm_arch_vcpu_ioctl_get_sregs_pr(struct kvm_vcpu *vcpu,
1065                                             struct kvm_sregs *sregs)
1066 {
1067         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1068         int i;
1069
1070         sregs->pvr = vcpu->arch.pvr;
1071
1072         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1073         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1074                 for (i = 0; i < 64; i++) {
1075                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
1076                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1077                 }
1078         } else {
1079                 for (i = 0; i < 16; i++)
1080                         sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
1081
1082                 for (i = 0; i < 8; i++) {
1083                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1084                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1085                 }
1086         }
1087
1088         return 0;
1089 }
1090
1091 static int kvm_arch_vcpu_ioctl_set_sregs_pr(struct kvm_vcpu *vcpu,
1092                                             struct kvm_sregs *sregs)
1093 {
1094         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1095         int i;
1096
1097         kvmppc_set_pvr_pr(vcpu, sregs->pvr);
1098
1099         vcpu3s->sdr1 = sregs->u.s.sdr1;
1100         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1101                 for (i = 0; i < 64; i++) {
1102                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1103                                                     sregs->u.s.ppc64.slb[i].slbe);
1104                 }
1105         } else {
1106                 for (i = 0; i < 16; i++) {
1107                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1108                 }
1109                 for (i = 0; i < 8; i++) {
1110                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1111                                        (u32)sregs->u.s.ppc32.ibat[i]);
1112                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1113                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1114                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1115                                        (u32)sregs->u.s.ppc32.dbat[i]);
1116                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1117                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1118                 }
1119         }
1120
1121         /* Flush the MMU after messing with the segments */
1122         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1123
1124         return 0;
1125 }
1126
1127 static int kvmppc_get_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1128                                  union kvmppc_one_reg *val)
1129 {
1130         int r = 0;
1131
1132         switch (id) {
1133         case KVM_REG_PPC_HIOR:
1134                 *val = get_reg_val(id, to_book3s(vcpu)->hior);
1135                 break;
1136 #ifdef CONFIG_VSX
1137         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1138                 long int i = id - KVM_REG_PPC_VSR0;
1139
1140                 if (!cpu_has_feature(CPU_FTR_VSX)) {
1141                         r = -ENXIO;
1142                         break;
1143                 }
1144                 val->vsxval[0] = vcpu->arch.fpr[i];
1145                 val->vsxval[1] = vcpu->arch.vsr[i];
1146                 break;
1147         }
1148 #endif /* CONFIG_VSX */
1149         default:
1150                 r = -EINVAL;
1151                 break;
1152         }
1153
1154         return r;
1155 }
1156
1157 static int kvmppc_set_one_reg_pr(struct kvm_vcpu *vcpu, u64 id,
1158                                  union kvmppc_one_reg *val)
1159 {
1160         int r = 0;
1161
1162         switch (id) {
1163         case KVM_REG_PPC_HIOR:
1164                 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1165                 to_book3s(vcpu)->hior_explicit = true;
1166                 break;
1167 #ifdef CONFIG_VSX
1168         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1169                 long int i = id - KVM_REG_PPC_VSR0;
1170
1171                 if (!cpu_has_feature(CPU_FTR_VSX)) {
1172                         r = -ENXIO;
1173                         break;
1174                 }
1175                 vcpu->arch.fpr[i] = val->vsxval[0];
1176                 vcpu->arch.vsr[i] = val->vsxval[1];
1177                 break;
1178         }
1179 #endif /* CONFIG_VSX */
1180         default:
1181                 r = -EINVAL;
1182                 break;
1183         }
1184
1185         return r;
1186 }
1187
1188 static struct kvm_vcpu *kvmppc_core_vcpu_create_pr(struct kvm *kvm,
1189                                                    unsigned int id)
1190 {
1191         struct kvmppc_vcpu_book3s *vcpu_book3s;
1192         struct kvm_vcpu *vcpu;
1193         int err = -ENOMEM;
1194         unsigned long p;
1195
1196         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1197         if (!vcpu)
1198                 goto out;
1199
1200         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1201         if (!vcpu_book3s)
1202                 goto free_vcpu;
1203         vcpu->arch.book3s = vcpu_book3s;
1204
1205 #ifdef CONFIG_KVM_BOOK3S_32
1206         vcpu->arch.shadow_vcpu =
1207                 kzalloc(sizeof(*vcpu->arch.shadow_vcpu), GFP_KERNEL);
1208         if (!vcpu->arch.shadow_vcpu)
1209                 goto free_vcpu3s;
1210 #endif
1211
1212         err = kvm_vcpu_init(vcpu, kvm, id);
1213         if (err)
1214                 goto free_shadow_vcpu;
1215
1216         err = -ENOMEM;
1217         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1218         if (!p)
1219                 goto uninit_vcpu;
1220         /* the real shared page fills the last 4k of our page */
1221         vcpu->arch.shared = (void *)(p + PAGE_SIZE - 4096);
1222
1223 #ifdef CONFIG_PPC_BOOK3S_64
1224         /*
1225          * Default to the same as the host if we're on sufficiently
1226          * recent machine that we have 1TB segments;
1227          * otherwise default to PPC970FX.
1228          */
1229         vcpu->arch.pvr = 0x3C0301;
1230         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1231                 vcpu->arch.pvr = mfspr(SPRN_PVR);
1232 #else
1233         /* default to book3s_32 (750) */
1234         vcpu->arch.pvr = 0x84202;
1235 #endif
1236         kvmppc_set_pvr_pr(vcpu, vcpu->arch.pvr);
1237         vcpu->arch.slb_nr = 64;
1238
1239         vcpu->arch.shadow_msr = MSR_USER64;
1240
1241         err = kvmppc_mmu_init(vcpu);
1242         if (err < 0)
1243                 goto uninit_vcpu;
1244
1245         return vcpu;
1246
1247 uninit_vcpu:
1248         kvm_vcpu_uninit(vcpu);
1249 free_shadow_vcpu:
1250 #ifdef CONFIG_KVM_BOOK3S_32
1251         kfree(vcpu->arch.shadow_vcpu);
1252 free_vcpu3s:
1253 #endif
1254         vfree(vcpu_book3s);
1255 free_vcpu:
1256         kmem_cache_free(kvm_vcpu_cache, vcpu);
1257 out:
1258         return ERR_PTR(err);
1259 }
1260
1261 static void kvmppc_core_vcpu_free_pr(struct kvm_vcpu *vcpu)
1262 {
1263         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1264
1265         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1266         kvm_vcpu_uninit(vcpu);
1267 #ifdef CONFIG_KVM_BOOK3S_32
1268         kfree(vcpu->arch.shadow_vcpu);
1269 #endif
1270         vfree(vcpu_book3s);
1271         kmem_cache_free(kvm_vcpu_cache, vcpu);
1272 }
1273
1274 static int kvmppc_vcpu_run_pr(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1275 {
1276         int ret;
1277         struct thread_fp_state fp;
1278         int fpexc_mode;
1279 #ifdef CONFIG_ALTIVEC
1280         struct thread_vr_state vr;
1281         unsigned long uninitialized_var(vrsave);
1282         int used_vr;
1283 #endif
1284 #ifdef CONFIG_VSX
1285         int used_vsr;
1286 #endif
1287         ulong ext_msr;
1288
1289         /* Check if we can run the vcpu at all */
1290         if (!vcpu->arch.sane) {
1291                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1292                 ret = -EINVAL;
1293                 goto out;
1294         }
1295
1296         /*
1297          * Interrupts could be timers for the guest which we have to inject
1298          * again, so let's postpone them until we're in the guest and if we
1299          * really did time things so badly, then we just exit again due to
1300          * a host external interrupt.
1301          */
1302         local_irq_disable();
1303         ret = kvmppc_prepare_to_enter(vcpu);
1304         if (ret <= 0) {
1305                 local_irq_enable();
1306                 goto out;
1307         }
1308
1309         /* Save FPU state in stack */
1310         if (current->thread.regs->msr & MSR_FP)
1311                 giveup_fpu(current);
1312         fp = current->thread.fp_state;
1313         fpexc_mode = current->thread.fpexc_mode;
1314
1315 #ifdef CONFIG_ALTIVEC
1316         /* Save Altivec state in stack */
1317         used_vr = current->thread.used_vr;
1318         if (used_vr) {
1319                 if (current->thread.regs->msr & MSR_VEC)
1320                         giveup_altivec(current);
1321                 vr = current->thread.vr_state;
1322                 vrsave = current->thread.vrsave;
1323         }
1324 #endif
1325
1326 #ifdef CONFIG_VSX
1327         /* Save VSX state in stack */
1328         used_vsr = current->thread.used_vsr;
1329         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1330                 __giveup_vsx(current);
1331 #endif
1332
1333         /* Remember the MSR with disabled extensions */
1334         ext_msr = current->thread.regs->msr;
1335
1336         /* Preload FPU if it's enabled */
1337         if (vcpu->arch.shared->msr & MSR_FP)
1338                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1339
1340         kvmppc_fix_ee_before_entry();
1341
1342         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1343
1344         /* No need for kvm_guest_exit. It's done in handle_exit.
1345            We also get here with interrupts enabled. */
1346
1347         /* Make sure we save the guest FPU/Altivec/VSX state */
1348         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1349
1350         current->thread.regs->msr = ext_msr;
1351
1352         /* Restore FPU/VSX state from stack */
1353         current->thread.fp_state = fp;
1354         current->thread.fpexc_mode = fpexc_mode;
1355
1356 #ifdef CONFIG_ALTIVEC
1357         /* Restore Altivec state from stack */
1358         if (used_vr && current->thread.used_vr) {
1359                 current->thread.vr_state = vr;
1360                 current->thread.vrsave = vrsave;
1361         }
1362         current->thread.used_vr = used_vr;
1363 #endif
1364
1365 #ifdef CONFIG_VSX
1366         current->thread.used_vsr = used_vsr;
1367 #endif
1368
1369 out:
1370         vcpu->mode = OUTSIDE_GUEST_MODE;
1371         return ret;
1372 }
1373
1374 /*
1375  * Get (and clear) the dirty memory log for a memory slot.
1376  */
1377 static int kvm_vm_ioctl_get_dirty_log_pr(struct kvm *kvm,
1378                                          struct kvm_dirty_log *log)
1379 {
1380         struct kvm_memory_slot *memslot;
1381         struct kvm_vcpu *vcpu;
1382         ulong ga, ga_end;
1383         int is_dirty = 0;
1384         int r;
1385         unsigned long n;
1386
1387         mutex_lock(&kvm->slots_lock);
1388
1389         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1390         if (r)
1391                 goto out;
1392
1393         /* If nothing is dirty, don't bother messing with page tables. */
1394         if (is_dirty) {
1395                 memslot = id_to_memslot(kvm->memslots, log->slot);
1396
1397                 ga = memslot->base_gfn << PAGE_SHIFT;
1398                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1399
1400                 kvm_for_each_vcpu(n, vcpu, kvm)
1401                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1402
1403                 n = kvm_dirty_bitmap_bytes(memslot);
1404                 memset(memslot->dirty_bitmap, 0, n);
1405         }
1406
1407         r = 0;
1408 out:
1409         mutex_unlock(&kvm->slots_lock);
1410         return r;
1411 }
1412
1413 static void kvmppc_core_flush_memslot_pr(struct kvm *kvm,
1414                                          struct kvm_memory_slot *memslot)
1415 {
1416         return;
1417 }
1418
1419 static int kvmppc_core_prepare_memory_region_pr(struct kvm *kvm,
1420                                         struct kvm_memory_slot *memslot,
1421                                         struct kvm_userspace_memory_region *mem)
1422 {
1423         return 0;
1424 }
1425
1426 static void kvmppc_core_commit_memory_region_pr(struct kvm *kvm,
1427                                 struct kvm_userspace_memory_region *mem,
1428                                 const struct kvm_memory_slot *old)
1429 {
1430         return;
1431 }
1432
1433 static void kvmppc_core_free_memslot_pr(struct kvm_memory_slot *free,
1434                                         struct kvm_memory_slot *dont)
1435 {
1436         return;
1437 }
1438
1439 static int kvmppc_core_create_memslot_pr(struct kvm_memory_slot *slot,
1440                                          unsigned long npages)
1441 {
1442         return 0;
1443 }
1444
1445
1446 #ifdef CONFIG_PPC64
1447 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1448                                          struct kvm_ppc_smmu_info *info)
1449 {
1450         long int i;
1451         struct kvm_vcpu *vcpu;
1452
1453         info->flags = 0;
1454
1455         /* SLB is always 64 entries */
1456         info->slb_size = 64;
1457
1458         /* Standard 4k base page size segment */
1459         info->sps[0].page_shift = 12;
1460         info->sps[0].slb_enc = 0;
1461         info->sps[0].enc[0].page_shift = 12;
1462         info->sps[0].enc[0].pte_enc = 0;
1463
1464         /*
1465          * 64k large page size.
1466          * We only want to put this in if the CPUs we're emulating
1467          * support it, but unfortunately we don't have a vcpu easily
1468          * to hand here to test.  Just pick the first vcpu, and if
1469          * that doesn't exist yet, report the minimum capability,
1470          * i.e., no 64k pages.
1471          * 1T segment support goes along with 64k pages.
1472          */
1473         i = 1;
1474         vcpu = kvm_get_vcpu(kvm, 0);
1475         if (vcpu && (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
1476                 info->flags = KVM_PPC_1T_SEGMENTS;
1477                 info->sps[i].page_shift = 16;
1478                 info->sps[i].slb_enc = SLB_VSID_L | SLB_VSID_LP_01;
1479                 info->sps[i].enc[0].page_shift = 16;
1480                 info->sps[i].enc[0].pte_enc = 1;
1481                 ++i;
1482         }
1483
1484         /* Standard 16M large page size segment */
1485         info->sps[i].page_shift = 24;
1486         info->sps[i].slb_enc = SLB_VSID_L;
1487         info->sps[i].enc[0].page_shift = 24;
1488         info->sps[i].enc[0].pte_enc = 0;
1489
1490         return 0;
1491 }
1492 #else
1493 static int kvm_vm_ioctl_get_smmu_info_pr(struct kvm *kvm,
1494                                          struct kvm_ppc_smmu_info *info)
1495 {
1496         /* We should not get called */
1497         BUG();
1498 }
1499 #endif /* CONFIG_PPC64 */
1500
1501 static unsigned int kvm_global_user_count = 0;
1502 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1503
1504 static int kvmppc_core_init_vm_pr(struct kvm *kvm)
1505 {
1506         mutex_init(&kvm->arch.hpt_mutex);
1507
1508         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1509                 spin_lock(&kvm_global_user_count_lock);
1510                 if (++kvm_global_user_count == 1)
1511                         pSeries_disable_reloc_on_exc();
1512                 spin_unlock(&kvm_global_user_count_lock);
1513         }
1514         return 0;
1515 }
1516
1517 static void kvmppc_core_destroy_vm_pr(struct kvm *kvm)
1518 {
1519 #ifdef CONFIG_PPC64
1520         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1521 #endif
1522
1523         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1524                 spin_lock(&kvm_global_user_count_lock);
1525                 BUG_ON(kvm_global_user_count == 0);
1526                 if (--kvm_global_user_count == 0)
1527                         pSeries_enable_reloc_on_exc();
1528                 spin_unlock(&kvm_global_user_count_lock);
1529         }
1530 }
1531
1532 static int kvmppc_core_check_processor_compat_pr(void)
1533 {
1534         /* we are always compatible */
1535         return 0;
1536 }
1537
1538 static long kvm_arch_vm_ioctl_pr(struct file *filp,
1539                                  unsigned int ioctl, unsigned long arg)
1540 {
1541         return -ENOTTY;
1542 }
1543
1544 static struct kvmppc_ops kvm_ops_pr = {
1545         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_pr,
1546         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_pr,
1547         .get_one_reg = kvmppc_get_one_reg_pr,
1548         .set_one_reg = kvmppc_set_one_reg_pr,
1549         .vcpu_load   = kvmppc_core_vcpu_load_pr,
1550         .vcpu_put    = kvmppc_core_vcpu_put_pr,
1551         .set_msr     = kvmppc_set_msr_pr,
1552         .vcpu_run    = kvmppc_vcpu_run_pr,
1553         .vcpu_create = kvmppc_core_vcpu_create_pr,
1554         .vcpu_free   = kvmppc_core_vcpu_free_pr,
1555         .check_requests = kvmppc_core_check_requests_pr,
1556         .get_dirty_log = kvm_vm_ioctl_get_dirty_log_pr,
1557         .flush_memslot = kvmppc_core_flush_memslot_pr,
1558         .prepare_memory_region = kvmppc_core_prepare_memory_region_pr,
1559         .commit_memory_region = kvmppc_core_commit_memory_region_pr,
1560         .unmap_hva = kvm_unmap_hva_pr,
1561         .unmap_hva_range = kvm_unmap_hva_range_pr,
1562         .age_hva  = kvm_age_hva_pr,
1563         .test_age_hva = kvm_test_age_hva_pr,
1564         .set_spte_hva = kvm_set_spte_hva_pr,
1565         .mmu_destroy  = kvmppc_mmu_destroy_pr,
1566         .free_memslot = kvmppc_core_free_memslot_pr,
1567         .create_memslot = kvmppc_core_create_memslot_pr,
1568         .init_vm = kvmppc_core_init_vm_pr,
1569         .destroy_vm = kvmppc_core_destroy_vm_pr,
1570         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_pr,
1571         .emulate_op = kvmppc_core_emulate_op_pr,
1572         .emulate_mtspr = kvmppc_core_emulate_mtspr_pr,
1573         .emulate_mfspr = kvmppc_core_emulate_mfspr_pr,
1574         .fast_vcpu_kick = kvm_vcpu_kick,
1575         .arch_vm_ioctl  = kvm_arch_vm_ioctl_pr,
1576 };
1577
1578
1579 int kvmppc_book3s_init_pr(void)
1580 {
1581         int r;
1582
1583         r = kvmppc_core_check_processor_compat_pr();
1584         if (r < 0)
1585                 return r;
1586
1587         kvm_ops_pr.owner = THIS_MODULE;
1588         kvmppc_pr_ops = &kvm_ops_pr;
1589
1590         r = kvmppc_mmu_hpte_sysinit();
1591         return r;
1592 }
1593
1594 void kvmppc_book3s_exit_pr(void)
1595 {
1596         kvmppc_pr_ops = NULL;
1597         kvmppc_mmu_hpte_sysexit();
1598 }
1599
1600 /*
1601  * We only support separate modules for book3s 64
1602  */
1603 #ifdef CONFIG_PPC_BOOK3S_64
1604
1605 module_init(kvmppc_book3s_init_pr);
1606 module_exit(kvmppc_book3s_exit_pr);
1607
1608 MODULE_LICENSE("GPL");
1609 #endif