arch/powerpc/kvm/book3s_pr.c

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