arch/powerpc/kvm/book3s_pr.c

   1 /*
   2  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   3  *
   4  * Authors:
   5  *    Alexander Graf <agraf@suse.de>
   6  *    Kevin Wolf <mail@kevin-wolf.de>
   7  *    Paul Mackerras <paulus@samba.org>
   8  *
   9  * Description:
  10  * Functions relating to running KVM on Book 3S processors where
  11  * we don't have access to hypervisor mode, and we run the guest
  12  * in problem state (user mode).
  13  *
  14  * This file is derived from arch/powerpc/kvm/44x.c,
  15  * by Hollis Blanchard <hollisb@us.ibm.com>.
  16  *
  17  * This program is free software; you can redistribute it and/or modify
  18  * it under the terms of the GNU General Public License, version 2, as
  19  * published by the Free Software Foundation.
  20  */
  21
  22 #include <linux/kvm_host.h>
  23 #include <linux/export.h>
  24 #include <linux/err.h>
  25 #include <linux/slab.h>
  26
  27 #include <asm/reg.h>
  28 #include <asm/cputable.h>
  29 #include <asm/cacheflush.h>
  30 #include <asm/tlbflush.h>
  31 #include <asm/uaccess.h>
  32 #include <asm/io.h>
  33 #include <asm/kvm_ppc.h>
  34 #include <asm/kvm_book3s.h>
  35 #include <asm/mmu_context.h>
  36 #include <asm/switch_to.h>
  37 #include <asm/firmware.h>
  38 #include <asm/hvcall.h>
  39 #include <linux/gfp.h>
  40 #include <linux/sched.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/highmem.h>
  43
  44 #include "trace.h"
  45
  46 /* #define EXIT_DEBUG */
  47 /* #define DEBUG_EXT */
  48
  49 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  50                              ulong msr);
  51
  52 /* Some compatibility defines */
  53 #ifdef CONFIG_PPC_BOOK3S_32
  54 #define MSR_USER32 MSR_USER
  55 #define MSR_USER64 MSR_USER
  56 #define HW_PAGE_SIZE PAGE_SIZE
  57 #endif
  58
  59 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  60 {
  61 #ifdef CONFIG_PPC_BOOK3S_64
  62         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  63         memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
  64         memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
  65                sizeof(get_paca()->shadow_vcpu));
  66         svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
  67         svcpu_put(svcpu);
  68 #endif
  69         vcpu->cpu = smp_processor_id();
  70 #ifdef CONFIG_PPC_BOOK3S_32
  71         current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
  72 #endif
  73 }
  74
  75 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  76 {
  77 #ifdef CONFIG_PPC_BOOK3S_64
  78         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  79         memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
  80         memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  81                sizeof(get_paca()->shadow_vcpu));
  82         to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
  83         svcpu_put(svcpu);
  84 #endif
  85
  86         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
  87         vcpu->cpu = -1;
  88 }
  89
  90 int kvmppc_core_check_requests(struct kvm_vcpu *vcpu)
  91 {
  92         int r = 1; /* Indicate we want to get back into the guest */
  93
  94         /* We misuse TLB_FLUSH to indicate that we want to clear
  95            all shadow cache entries */
  96         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
  97                 kvmppc_mmu_pte_flush(vcpu, 0, 0);
  98
  99         return r;
 100 }
 101
 102 /************* MMU Notifiers *************/
 103
 104 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
 105 {
 106         trace_kvm_unmap_hva(hva);
 107
 108         /*
 109          * Flush all shadow tlb entries everywhere. This is slow, but
 110          * we are 100% sure that we catch the to be unmapped page
 111          */
 112         kvm_flush_remote_tlbs(kvm);
 113
 114         return 0;
 115 }
 116
 117 int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
 118 {
 119         /* kvm_unmap_hva flushes everything anyways */
 120         kvm_unmap_hva(kvm, start);
 121
 122         return 0;
 123 }
 124
 125 int kvm_age_hva(struct kvm *kvm, unsigned long hva)
 126 {
 127         /* XXX could be more clever ;) */
 128         return 0;
 129 }
 130
 131 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
 132 {
 133         /* XXX could be more clever ;) */
 134         return 0;
 135 }
 136
 137 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
 138 {
 139         /* The page will get remapped properly on its next fault */
 140         kvm_unmap_hva(kvm, hva);
 141 }
 142
 143 /*****************************************/
 144
 145 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 146 {
 147         ulong smsr = vcpu->arch.shared->msr;
 148
 149         /* Guest MSR values */
 150         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE;
 151         /* Process MSR values */
 152         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 153         /* External providers the guest reserved */
 154         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
 155         /* 64-bit Process MSR values */
 156 #ifdef CONFIG_PPC_BOOK3S_64
 157         smsr |= MSR_ISF | MSR_HV;
 158 #endif
 159         vcpu->arch.shadow_msr = smsr;
 160 }
 161
 162 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 163 {
 164         ulong old_msr = vcpu->arch.shared->msr;
 165
 166 #ifdef EXIT_DEBUG
 167         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 168 #endif
 169
 170         msr &= to_book3s(vcpu)->msr_mask;
 171         vcpu->arch.shared->msr = msr;
 172         kvmppc_recalc_shadow_msr(vcpu);
 173
 174         if (msr & MSR_POW) {
 175                 if (!vcpu->arch.pending_exceptions) {
 176                         kvm_vcpu_block(vcpu);
 177                         clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 178                         vcpu->stat.halt_wakeup++;
 179
 180                         /* Unset POW bit after we woke up */
 181                         msr &= ~MSR_POW;
 182                         vcpu->arch.shared->msr = msr;
 183                 }
 184         }
 185
 186         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 187                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 188                 kvmppc_mmu_flush_segments(vcpu);
 189                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 190
 191                 /* Preload magic page segment when in kernel mode */
 192                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 193                         struct kvm_vcpu_arch *a = &vcpu->arch;
 194
 195                         if (msr & MSR_DR)
 196                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 197                         else
 198                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 199                 }
 200         }
 201
 202         /*
 203          * When switching from 32 to 64-bit, we may have a stale 32-bit
 204          * magic page around, we need to flush it. Typically 32-bit magic
 205          * page will be instanciated when calling into RTAS. Note: We
 206          * assume that such transition only happens while in kernel mode,
 207          * ie, we never transition from user 32-bit to kernel 64-bit with
 208          * a 32-bit magic page around.
 209          */
 210         if (vcpu->arch.magic_page_pa &&
 211             !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
 212                 /* going from RTAS to normal kernel code */
 213                 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
 214                                      ~0xFFFUL);
 215         }
 216
 217         /* Preload FPU if it's enabled */
 218         if (vcpu->arch.shared->msr & MSR_FP)
 219                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 220 }
 221
 222 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 223 {
 224         u32 host_pvr;
 225
 226         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 227         vcpu->arch.pvr = pvr;
 228 #ifdef CONFIG_PPC_BOOK3S_64
 229         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 230                 kvmppc_mmu_book3s_64_init(vcpu);
 231                 if (!to_book3s(vcpu)->hior_explicit)
 232                         to_book3s(vcpu)->hior = 0xfff00000;
 233                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 234                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 235         } else
 236 #endif
 237         {
 238                 kvmppc_mmu_book3s_32_init(vcpu);
 239                 if (!to_book3s(vcpu)->hior_explicit)
 240                         to_book3s(vcpu)->hior = 0;
 241                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 242                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 243         }
 244
 245         kvmppc_sanity_check(vcpu);
 246
 247         /* If we are in hypervisor level on 970, we can tell the CPU to
 248          * treat DCBZ as 32 bytes store */
 249         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 250         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 251             !strcmp(cur_cpu_spec->platform, "ppc970"))
 252                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 253
 254         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 255            really needs them in a VM on Cell and force disable them. */
 256         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 257                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 258
 259 #ifdef CONFIG_PPC_BOOK3S_32
 260         /* 32 bit Book3S always has 32 byte dcbz */
 261         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 262 #endif
 263
 264         /* On some CPUs we can execute paired single operations natively */
 265         asm ( "mfpvr %0" : "=r"(host_pvr));
 266         switch (host_pvr) {
 267         case 0x00080200:        /* lonestar 2.0 */
 268         case 0x00088202:        /* lonestar 2.2 */
 269         case 0x70000100:        /* gekko 1.0 */
 270         case 0x00080100:        /* gekko 2.0 */
 271         case 0x00083203:        /* gekko 2.3a */
 272         case 0x00083213:        /* gekko 2.3b */
 273         case 0x00083204:        /* gekko 2.4 */
 274         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 275         case 0x00087200:        /* broadway */
 276                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 277                 /* Enable HID2.PSE - in case we need it later */
 278                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 279         }
 280 }
 281
 282 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 283  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 284  * emulate 32 bytes dcbz length.
 285  *
 286  * The Book3s_64 inventors also realized this case and implemented a special bit
 287  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 288  *
 289  * My approach here is to patch the dcbz instruction on executing pages.
 290  */
 291 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 292 {
 293         struct page *hpage;
 294         u64 hpage_offset;
 295         u32 *page;
 296         int i;
 297
 298         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 299         if (is_error_page(hpage))
 300                 return;
 301
 302         hpage_offset = pte->raddr & ~PAGE_MASK;
 303         hpage_offset &= ~0xFFFULL;
 304         hpage_offset /= 4;
 305
 306         get_page(hpage);
 307         page = kmap_atomic(hpage);
 308
 309         /* patch dcbz into reserved instruction, so we trap */
 310         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 311                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 312                         page[i] &= 0xfffffff7;
 313
 314         kunmap_atomic(page);
 315         put_page(hpage);
 316 }
 317
 318 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 319 {
 320         ulong mp_pa = vcpu->arch.magic_page_pa;
 321
 322         if (!(vcpu->arch.shared->msr & MSR_SF))
 323                 mp_pa = (uint32_t)mp_pa;
 324
 325         if (unlikely(mp_pa) &&
 326             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 327                 return 1;
 328         }
 329
 330         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 331 }
 332
 333 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 334                             ulong eaddr, int vec)
 335 {
 336         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 337         int r = RESUME_GUEST;
 338         int relocated;
 339         int page_found = 0;
 340         struct kvmppc_pte pte;
 341         bool is_mmio = false;
 342         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 343         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 344         u64 vsid;
 345
 346         relocated = data ? dr : ir;
 347
 348         /* Resolve real address if translation turned on */
 349         if (relocated) {
 350                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 351         } else {
 352                 pte.may_execute = true;
 353                 pte.may_read = true;
 354                 pte.may_write = true;
 355                 pte.raddr = eaddr & KVM_PAM;
 356                 pte.eaddr = eaddr;
 357                 pte.vpage = eaddr >> 12;
 358         }
 359
 360         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 361         case 0:
 362                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 363                 break;
 364         case MSR_DR:
 365         case MSR_IR:
 366                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 367
 368                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 369                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 370                 else
 371                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 372                 pte.vpage |= vsid;
 373
 374                 if (vsid == -1)
 375                         page_found = -EINVAL;
 376                 break;
 377         }
 378
 379         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 380            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 381                 /*
 382                  * If we do the dcbz hack, we have to NX on every execution,
 383                  * so we can patch the executing code. This renders our guest
 384                  * NX-less.
 385                  */
 386                 pte.may_execute = !data;
 387         }
 388
 389         if (page_found == -ENOENT) {
 390                 /* Page not found in guest PTE entries */
 391                 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 392                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 393                 vcpu->arch.shared->dsisr = svcpu->fault_dsisr;
 394                 vcpu->arch.shared->msr |=
 395                         (svcpu->shadow_srr1 & 0x00000000f8000000ULL);
 396                 svcpu_put(svcpu);
 397                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 398         } else if (page_found == -EPERM) {
 399                 /* Storage protection */
 400                 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 401                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 402                 vcpu->arch.shared->dsisr = svcpu->fault_dsisr & ~DSISR_NOHPTE;
 403                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 404                 vcpu->arch.shared->msr |=
 405                         svcpu->shadow_srr1 & 0x00000000f8000000ULL;
 406                 svcpu_put(svcpu);
 407                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 408         } else if (page_found == -EINVAL) {
 409                 /* Page not found in guest SLB */
 410                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 411                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 412         } else if (!is_mmio &&
 413                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 414                 /* The guest's PTE is not mapped yet. Map on the host */
 415                 kvmppc_mmu_map_page(vcpu, &pte);
 416                 if (data)
 417                         vcpu->stat.sp_storage++;
 418                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 419                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 420                         kvmppc_patch_dcbz(vcpu, &pte);
 421         } else {
 422                 /* MMIO */
 423                 vcpu->stat.mmio_exits++;
 424                 vcpu->arch.paddr_accessed = pte.raddr;
 425                 vcpu->arch.vaddr_accessed = pte.eaddr;
 426                 r = kvmppc_emulate_mmio(run, vcpu);
 427                 if ( r == RESUME_HOST_NV )
 428                         r = RESUME_HOST;
 429         }
 430
 431         return r;
 432 }
 433
 434 static inline int get_fpr_index(int i)
 435 {
 436         return i * TS_FPRWIDTH;
 437 }
 438
 439 /* Give up external provider (FPU, Altivec, VSX) */
 440 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 441 {
 442         struct thread_struct *t = &current->thread;
 443         u64 *vcpu_fpr = vcpu->arch.fpr;
 444 #ifdef CONFIG_VSX
 445         u64 *vcpu_vsx = vcpu->arch.vsr;
 446 #endif
 447         u64 *thread_fpr = (u64*)t->fpr;
 448         int i;
 449
 450         /*
 451          * VSX instructions can access FP and vector registers, so if
 452          * we are giving up VSX, make sure we give up FP and VMX as well.
 453          */
 454         if (msr & MSR_VSX)
 455                 msr |= MSR_FP | MSR_VEC;
 456
 457         msr &= vcpu->arch.guest_owned_ext;
 458         if (!msr)
 459                 return;
 460
 461 #ifdef DEBUG_EXT
 462         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 463 #endif
 464
 465         if (msr & MSR_FP) {
 466                 /*
 467                  * Note that on CPUs with VSX, giveup_fpu stores
 468                  * both the traditional FP registers and the added VSX
 469                  * registers into thread.fpr[].
 470                  */
 471                 if (current->thread.regs->msr & MSR_FP)
 472                         giveup_fpu(current);
 473                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 474                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 475
 476                 vcpu->arch.fpscr = t->fpscr.val;
 477
 478 #ifdef CONFIG_VSX
 479                 if (cpu_has_feature(CPU_FTR_VSX))
 480                         for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
 481                                 vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 482 #endif
 483         }
 484
 485 #ifdef CONFIG_ALTIVEC
 486         if (msr & MSR_VEC) {
 487                 if (current->thread.regs->msr & MSR_VEC)
 488                         giveup_altivec(current);
 489                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 490                 vcpu->arch.vscr = t->vscr;
 491         }
 492 #endif
 493
 494         vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
 495         kvmppc_recalc_shadow_msr(vcpu);
 496 }
 497
 498 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 499 {
 500         ulong srr0 = kvmppc_get_pc(vcpu);
 501         u32 last_inst = kvmppc_get_last_inst(vcpu);
 502         int ret;
 503
 504         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 505         if (ret == -ENOENT) {
 506                 ulong msr = vcpu->arch.shared->msr;
 507
 508                 msr = kvmppc_set_field(msr, 33, 33, 1);
 509                 msr = kvmppc_set_field(msr, 34, 36, 0);
 510                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 511                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 512                 return EMULATE_AGAIN;
 513         }
 514
 515         return EMULATE_DONE;
 516 }
 517
 518 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 519 {
 520
 521         /* Need to do paired single emulation? */
 522         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 523                 return EMULATE_DONE;
 524
 525         /* Read out the instruction */
 526         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 527                 /* Need to emulate */
 528                 return EMULATE_FAIL;
 529
 530         return EMULATE_AGAIN;
 531 }
 532
 533 /* Handle external providers (FPU, Altivec, VSX) */
 534 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 535                              ulong msr)
 536 {
 537         struct thread_struct *t = &current->thread;
 538         u64 *vcpu_fpr = vcpu->arch.fpr;
 539 #ifdef CONFIG_VSX
 540         u64 *vcpu_vsx = vcpu->arch.vsr;
 541 #endif
 542         u64 *thread_fpr = (u64*)t->fpr;
 543         int i;
 544
 545         /* When we have paired singles, we emulate in software */
 546         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 547                 return RESUME_GUEST;
 548
 549         if (!(vcpu->arch.shared->msr & msr)) {
 550                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 551                 return RESUME_GUEST;
 552         }
 553
 554         if (msr == MSR_VSX) {
 555                 /* No VSX?  Give an illegal instruction interrupt */
 556 #ifdef CONFIG_VSX
 557                 if (!cpu_has_feature(CPU_FTR_VSX))
 558 #endif
 559                 {
 560                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 561                         return RESUME_GUEST;
 562                 }
 563
 564                 /*
 565                  * We have to load up all the FP and VMX registers before
 566                  * we can let the guest use VSX instructions.
 567                  */
 568                 msr = MSR_FP | MSR_VEC | MSR_VSX;
 569         }
 570
 571         /* See if we already own all the ext(s) needed */
 572         msr &= ~vcpu->arch.guest_owned_ext;
 573         if (!msr)
 574                 return RESUME_GUEST;
 575
 576 #ifdef DEBUG_EXT
 577         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 578 #endif
 579
 580         if (msr & MSR_FP) {
 581                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 582                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 583 #ifdef CONFIG_VSX
 584                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
 585                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 586 #endif
 587                 t->fpscr.val = vcpu->arch.fpscr;
 588                 t->fpexc_mode = 0;
 589                 kvmppc_load_up_fpu();
 590         }
 591
 592         if (msr & MSR_VEC) {
 593 #ifdef CONFIG_ALTIVEC
 594                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 595                 t->vscr = vcpu->arch.vscr;
 596                 t->vrsave = -1;
 597                 kvmppc_load_up_altivec();
 598 #endif
 599         }
 600
 601         current->thread.regs->msr |= msr;
 602         vcpu->arch.guest_owned_ext |= msr;
 603         kvmppc_recalc_shadow_msr(vcpu);
 604
 605         return RESUME_GUEST;
 606 }
 607
 608 /*
 609  * Kernel code using FP or VMX could have flushed guest state to
 610  * the thread_struct; if so, get it back now.
 611  */
 612 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
 613 {
 614         unsigned long lost_ext;
 615
 616         lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
 617         if (!lost_ext)
 618                 return;
 619
 620         if (lost_ext & MSR_FP)
 621                 kvmppc_load_up_fpu();
 622         if (lost_ext & MSR_VEC)
 623                 kvmppc_load_up_altivec();
 624         current->thread.regs->msr |= lost_ext;
 625 }
 626
 627 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 628                        unsigned int exit_nr)
 629 {
 630         int r = RESUME_HOST;
 631         int s;
 632
 633         vcpu->stat.sum_exits++;
 634
 635         run->exit_reason = KVM_EXIT_UNKNOWN;
 636         run->ready_for_interrupt_injection = 1;
 637
 638         /* We get here with MSR.EE=1 */
 639
 640         trace_kvm_exit(exit_nr, vcpu);
 641         kvm_guest_exit();
 642
 643         switch (exit_nr) {
 644         case BOOK3S_INTERRUPT_INST_STORAGE:
 645         {
 646                 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 647                 ulong shadow_srr1 = svcpu->shadow_srr1;
 648                 vcpu->stat.pf_instruc++;
 649
 650 #ifdef CONFIG_PPC_BOOK3S_32
 651                 /* We set segments as unused segments when invalidating them. So
 652                  * treat the respective fault as segment fault. */
 653                 if (svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT] == SR_INVALID) {
 654                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 655                         r = RESUME_GUEST;
 656                         svcpu_put(svcpu);
 657                         break;
 658                 }
 659 #endif
 660                 svcpu_put(svcpu);
 661
 662                 /* only care about PTEG not found errors, but leave NX alone */
 663                 if (shadow_srr1 & 0x40000000) {
 664                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 665                         vcpu->stat.sp_instruc++;
 666                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 667                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 668                         /*
 669                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 670                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 671                          *     that no guest that needs the dcbz hack does NX.
 672                          */
 673                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 674                         r = RESUME_GUEST;
 675                 } else {
 676                         vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
 677                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 678                         r = RESUME_GUEST;
 679                 }
 680                 break;
 681         }
 682         case BOOK3S_INTERRUPT_DATA_STORAGE:
 683         {
 684                 ulong dar = kvmppc_get_fault_dar(vcpu);
 685                 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 686                 u32 fault_dsisr = svcpu->fault_dsisr;
 687                 vcpu->stat.pf_storage++;
 688
 689 #ifdef CONFIG_PPC_BOOK3S_32
 690                 /* We set segments as unused segments when invalidating them. So
 691                  * treat the respective fault as segment fault. */
 692                 if ((svcpu->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 693                         kvmppc_mmu_map_segment(vcpu, dar);
 694                         r = RESUME_GUEST;
 695                         svcpu_put(svcpu);
 696                         break;
 697                 }
 698 #endif
 699                 svcpu_put(svcpu);
 700
 701                 /* The only case we need to handle is missing shadow PTEs */
 702                 if (fault_dsisr & DSISR_NOHPTE) {
 703                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 704                 } else {
 705                         vcpu->arch.shared->dar = dar;
 706                         vcpu->arch.shared->dsisr = fault_dsisr;
 707                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 708                         r = RESUME_GUEST;
 709                 }
 710                 break;
 711         }
 712         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 713                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 714                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 715                         kvmppc_book3s_queue_irqprio(vcpu,
 716                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 717                 }
 718                 r = RESUME_GUEST;
 719                 break;
 720         case BOOK3S_INTERRUPT_INST_SEGMENT:
 721                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 722                         kvmppc_book3s_queue_irqprio(vcpu,
 723                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 724                 }
 725                 r = RESUME_GUEST;
 726                 break;
 727         /* We're good on these - the host merely wanted to get our attention */
 728         case BOOK3S_INTERRUPT_DECREMENTER:
 729         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 730                 vcpu->stat.dec_exits++;
 731                 r = RESUME_GUEST;
 732                 break;
 733         case BOOK3S_INTERRUPT_EXTERNAL:
 734         case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
 735         case BOOK3S_INTERRUPT_EXTERNAL_HV:
 736                 vcpu->stat.ext_intr_exits++;
 737                 r = RESUME_GUEST;
 738                 break;
 739         case BOOK3S_INTERRUPT_PERFMON:
 740                 r = RESUME_GUEST;
 741                 break;
 742         case BOOK3S_INTERRUPT_PROGRAM:
 743         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 744         {
 745                 enum emulation_result er;
 746                 struct kvmppc_book3s_shadow_vcpu *svcpu;
 747                 ulong flags;
 748
 749 program_interrupt:
 750                 svcpu = svcpu_get(vcpu);
 751                 flags = svcpu->shadow_srr1 & 0x1f0000ull;
 752                 svcpu_put(svcpu);
 753
 754                 if (vcpu->arch.shared->msr & MSR_PR) {
 755 #ifdef EXIT_DEBUG
 756                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 757 #endif
 758                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 759                             (INS_DCBZ & 0xfffffff7)) {
 760                                 kvmppc_core_queue_program(vcpu, flags);
 761                                 r = RESUME_GUEST;
 762                                 break;
 763                         }
 764                 }
 765
 766                 vcpu->stat.emulated_inst_exits++;
 767                 er = kvmppc_emulate_instruction(run, vcpu);
 768                 switch (er) {
 769                 case EMULATE_DONE:
 770                         r = RESUME_GUEST_NV;
 771                         break;
 772                 case EMULATE_AGAIN:
 773                         r = RESUME_GUEST;
 774                         break;
 775                 case EMULATE_FAIL:
 776                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 777                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 778                         kvmppc_core_queue_program(vcpu, flags);
 779                         r = RESUME_GUEST;
 780                         break;
 781                 case EMULATE_DO_MMIO:
 782                         run->exit_reason = KVM_EXIT_MMIO;
 783                         r = RESUME_HOST_NV;
 784                         break;
 785                 case EMULATE_EXIT_USER:
 786                         r = RESUME_HOST_NV;
 787                         break;
 788                 default:
 789                         BUG();
 790                 }
 791                 break;
 792         }
 793         case BOOK3S_INTERRUPT_SYSCALL:
 794                 if (vcpu->arch.papr_enabled &&
 795                     (kvmppc_get_last_sc(vcpu) == 0x44000022) &&
 796                     !(vcpu->arch.shared->msr & MSR_PR)) {
 797                         /* SC 1 papr hypercalls */
 798                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
 799                         int i;
 800
 801 #ifdef CONFIG_KVM_BOOK3S_64_PR
 802                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
 803                                 r = RESUME_GUEST;
 804                                 break;
 805                         }
 806 #endif
 807
 808                         run->papr_hcall.nr = cmd;
 809                         for (i = 0; i < 9; ++i) {
 810                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
 811                                 run->papr_hcall.args[i] = gpr;
 812                         }
 813                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
 814                         vcpu->arch.hcall_needed = 1;
 815                         r = RESUME_HOST;
 816                 } else if (vcpu->arch.osi_enabled &&
 817                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 818                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 819                         /* MOL hypercalls */
 820                         u64 *gprs = run->osi.gprs;
 821                         int i;
 822
 823                         run->exit_reason = KVM_EXIT_OSI;
 824                         for (i = 0; i < 32; i++)
 825                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 826                         vcpu->arch.osi_needed = 1;
 827                         r = RESUME_HOST_NV;
 828                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 829                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 830                         /* KVM PV hypercalls */
 831                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 832                         r = RESUME_GUEST;
 833                 } else {
 834                         /* Guest syscalls */
 835                         vcpu->stat.syscall_exits++;
 836                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 837                         r = RESUME_GUEST;
 838                 }
 839                 break;
 840         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 841         case BOOK3S_INTERRUPT_ALTIVEC:
 842         case BOOK3S_INTERRUPT_VSX:
 843         {
 844                 int ext_msr = 0;
 845
 846                 switch (exit_nr) {
 847                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 848                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 849                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 850                 }
 851
 852                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
 853                 case EMULATE_DONE:
 854                         /* everything ok - let's enable the ext */
 855                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
 856                         break;
 857                 case EMULATE_FAIL:
 858                         /* we need to emulate this instruction */
 859                         goto program_interrupt;
 860                         break;
 861                 default:
 862                         /* nothing to worry about - go again */
 863                         break;
 864                 }
 865                 break;
 866         }
 867         case BOOK3S_INTERRUPT_ALIGNMENT:
 868                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
 869                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
 870                                 kvmppc_get_last_inst(vcpu));
 871                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
 872                                 kvmppc_get_last_inst(vcpu));
 873                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 874                 }
 875                 r = RESUME_GUEST;
 876                 break;
 877         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 878         case BOOK3S_INTERRUPT_TRACE:
 879                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 880                 r = RESUME_GUEST;
 881                 break;
 882         default:
 883         {
 884                 struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 885                 ulong shadow_srr1 = svcpu->shadow_srr1;
 886                 svcpu_put(svcpu);
 887                 /* Ugh - bork here! What did we get? */
 888                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
 889                         exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
 890                 r = RESUME_HOST;
 891                 BUG();
 892                 break;
 893         }
 894         }
 895
 896         if (!(r & RESUME_HOST)) {
 897                 /* To avoid clobbering exit_reason, only check for signals if
 898                  * we aren't already exiting to userspace for some other
 899                  * reason. */
 900
 901                 /*
 902                  * Interrupts could be timers for the guest which we have to
 903                  * inject again, so let's postpone them until we're in the guest
 904                  * and if we really did time things so badly, then we just exit
 905                  * again due to a host external interrupt.
 906                  */
 907                 local_irq_disable();
 908                 s = kvmppc_prepare_to_enter(vcpu);
 909                 if (s <= 0) {
 910                         local_irq_enable();
 911                         r = s;
 912                 } else {
 913                         kvmppc_fix_ee_before_entry();
 914                 }
 915                 kvmppc_handle_lost_ext(vcpu);
 916         }
 917
 918         trace_kvm_book3s_reenter(r, vcpu);
 919
 920         return r;
 921 }
 922
 923 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 924                                   struct kvm_sregs *sregs)
 925 {
 926         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 927         int i;
 928
 929         sregs->pvr = vcpu->arch.pvr;
 930
 931         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 932         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 933                 for (i = 0; i < 64; i++) {
 934                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
 935                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 936                 }
 937         } else {
 938                 for (i = 0; i < 16; i++)
 939                         sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
 940
 941                 for (i = 0; i < 8; i++) {
 942                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 943                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 944                 }
 945         }
 946
 947         return 0;
 948 }
 949
 950 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 951                                   struct kvm_sregs *sregs)
 952 {
 953         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 954         int i;
 955
 956         kvmppc_set_pvr(vcpu, sregs->pvr);
 957
 958         vcpu3s->sdr1 = sregs->u.s.sdr1;
 959         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 960                 for (i = 0; i < 64; i++) {
 961                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 962                                                     sregs->u.s.ppc64.slb[i].slbe);
 963                 }
 964         } else {
 965                 for (i = 0; i < 16; i++) {
 966                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 967                 }
 968                 for (i = 0; i < 8; i++) {
 969                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 970                                        (u32)sregs->u.s.ppc32.ibat[i]);
 971                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 972                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 973                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 974                                        (u32)sregs->u.s.ppc32.dbat[i]);
 975                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 976                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 977                 }
 978         }
 979
 980         /* Flush the MMU after messing with the segments */
 981         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 982
 983         return 0;
 984 }
 985
 986 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
 987 {
 988         int r = 0;
 989
 990         switch (id) {
 991         case KVM_REG_PPC_HIOR:
 992                 *val = get_reg_val(id, to_book3s(vcpu)->hior);
 993                 break;
 994 #ifdef CONFIG_VSX
 995         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
 996                 long int i = id - KVM_REG_PPC_VSR0;
 997
 998                 if (!cpu_has_feature(CPU_FTR_VSX)) {
 999                         r = -ENXIO;
1000                         break;
1001                 }
1002                 val->vsxval[0] = vcpu->arch.fpr[i];
1003                 val->vsxval[1] = vcpu->arch.vsr[i];
1004                 break;
1005         }
1006 #endif /* CONFIG_VSX */
1007         default:
1008                 r = -EINVAL;
1009                 break;
1010         }
1011
1012         return r;
1013 }
1014
1015 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
1016 {
1017         int r = 0;
1018
1019         switch (id) {
1020         case KVM_REG_PPC_HIOR:
1021                 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1022                 to_book3s(vcpu)->hior_explicit = true;
1023                 break;
1024 #ifdef CONFIG_VSX
1025         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1026                 long int i = id - KVM_REG_PPC_VSR0;
1027
1028                 if (!cpu_has_feature(CPU_FTR_VSX)) {
1029                         r = -ENXIO;
1030                         break;
1031                 }
1032                 vcpu->arch.fpr[i] = val->vsxval[0];
1033                 vcpu->arch.vsr[i] = val->vsxval[1];
1034                 break;
1035         }
1036 #endif /* CONFIG_VSX */
1037         default:
1038                 r = -EINVAL;
1039                 break;
1040         }
1041
1042         return r;
1043 }
1044
1045 int kvmppc_core_check_processor_compat(void)
1046 {
1047         return 0;
1048 }
1049
1050 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1051 {
1052         struct kvmppc_vcpu_book3s *vcpu_book3s;
1053         struct kvm_vcpu *vcpu;
1054         int err = -ENOMEM;
1055         unsigned long p;
1056
1057         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1058         if (!vcpu_book3s)
1059                 goto out;
1060
1061         vcpu_book3s->shadow_vcpu =
1062                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
1063         if (!vcpu_book3s->shadow_vcpu)
1064                 goto free_vcpu;
1065
1066         vcpu = &vcpu_book3s->vcpu;
1067         err = kvm_vcpu_init(vcpu, kvm, id);
1068         if (err)
1069                 goto free_shadow_vcpu;
1070
1071         err = -ENOMEM;
1072         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1073         if (!p)
1074                 goto uninit_vcpu;
1075         /* the real shared page fills the last 4k of our page */
1076         vcpu->arch.shared = (void *)(p + PAGE_SIZE - 4096);
1077
1078 #ifdef CONFIG_PPC_BOOK3S_64
1079         /* default to book3s_64 (970fx) */
1080         vcpu->arch.pvr = 0x3C0301;
1081 #else
1082         /* default to book3s_32 (750) */
1083         vcpu->arch.pvr = 0x84202;
1084 #endif
1085         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1086         vcpu->arch.slb_nr = 64;
1087
1088         vcpu->arch.shadow_msr = MSR_USER64;
1089
1090         err = kvmppc_mmu_init(vcpu);
1091         if (err < 0)
1092                 goto uninit_vcpu;
1093
1094         return vcpu;
1095
1096 uninit_vcpu:
1097         kvm_vcpu_uninit(vcpu);
1098 free_shadow_vcpu:
1099         kfree(vcpu_book3s->shadow_vcpu);
1100 free_vcpu:
1101         vfree(vcpu_book3s);
1102 out:
1103         return ERR_PTR(err);
1104 }
1105
1106 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1107 {
1108         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1109
1110         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1111         kvm_vcpu_uninit(vcpu);
1112         kfree(vcpu_book3s->shadow_vcpu);
1113         vfree(vcpu_book3s);
1114 }
1115
1116 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1117 {
1118         int ret;
1119         double fpr[32][TS_FPRWIDTH];
1120         unsigned int fpscr;
1121         int fpexc_mode;
1122 #ifdef CONFIG_ALTIVEC
1123         vector128 vr[32];
1124         vector128 vscr;
1125         unsigned long uninitialized_var(vrsave);
1126         int used_vr;
1127 #endif
1128 #ifdef CONFIG_VSX
1129         int used_vsr;
1130 #endif
1131         ulong ext_msr;
1132
1133         /* Check if we can run the vcpu at all */
1134         if (!vcpu->arch.sane) {
1135                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1136                 ret = -EINVAL;
1137                 goto out;
1138         }
1139
1140         /*
1141          * Interrupts could be timers for the guest which we have to inject
1142          * again, so let's postpone them until we're in the guest and if we
1143          * really did time things so badly, then we just exit again due to
1144          * a host external interrupt.
1145          */
1146         local_irq_disable();
1147         ret = kvmppc_prepare_to_enter(vcpu);
1148         if (ret <= 0) {
1149                 local_irq_enable();
1150                 goto out;
1151         }
1152
1153         /* Save FPU state in stack */
1154         if (current->thread.regs->msr & MSR_FP)
1155                 giveup_fpu(current);
1156         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1157         fpscr = current->thread.fpscr.val;
1158         fpexc_mode = current->thread.fpexc_mode;
1159
1160 #ifdef CONFIG_ALTIVEC
1161         /* Save Altivec state in stack */
1162         used_vr = current->thread.used_vr;
1163         if (used_vr) {
1164                 if (current->thread.regs->msr & MSR_VEC)
1165                         giveup_altivec(current);
1166                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1167                 vscr = current->thread.vscr;
1168                 vrsave = current->thread.vrsave;
1169         }
1170 #endif
1171
1172 #ifdef CONFIG_VSX
1173         /* Save VSX state in stack */
1174         used_vsr = current->thread.used_vsr;
1175         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1176                 __giveup_vsx(current);
1177 #endif
1178
1179         /* Remember the MSR with disabled extensions */
1180         ext_msr = current->thread.regs->msr;
1181
1182         /* Preload FPU if it's enabled */
1183         if (vcpu->arch.shared->msr & MSR_FP)
1184                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1185
1186         kvmppc_fix_ee_before_entry();
1187
1188         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1189
1190         /* No need for kvm_guest_exit. It's done in handle_exit.
1191            We also get here with interrupts enabled. */
1192
1193         /* Make sure we save the guest FPU/Altivec/VSX state */
1194         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1195
1196         current->thread.regs->msr = ext_msr;
1197
1198         /* Restore FPU/VSX state from stack */
1199         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1200         current->thread.fpscr.val = fpscr;
1201         current->thread.fpexc_mode = fpexc_mode;
1202
1203 #ifdef CONFIG_ALTIVEC
1204         /* Restore Altivec state from stack */
1205         if (used_vr && current->thread.used_vr) {
1206                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1207                 current->thread.vscr = vscr;
1208                 current->thread.vrsave = vrsave;
1209         }
1210         current->thread.used_vr = used_vr;
1211 #endif
1212
1213 #ifdef CONFIG_VSX
1214         current->thread.used_vsr = used_vsr;
1215 #endif
1216
1217 out:
1218         vcpu->mode = OUTSIDE_GUEST_MODE;
1219         return ret;
1220 }
1221
1222 /*
1223  * Get (and clear) the dirty memory log for a memory slot.
1224  */
1225 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1226                                       struct kvm_dirty_log *log)
1227 {
1228         struct kvm_memory_slot *memslot;
1229         struct kvm_vcpu *vcpu;
1230         ulong ga, ga_end;
1231         int is_dirty = 0;
1232         int r;
1233         unsigned long n;
1234
1235         mutex_lock(&kvm->slots_lock);
1236
1237         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1238         if (r)
1239                 goto out;
1240
1241         /* If nothing is dirty, don't bother messing with page tables. */
1242         if (is_dirty) {
1243                 memslot = id_to_memslot(kvm->memslots, log->slot);
1244
1245                 ga = memslot->base_gfn << PAGE_SHIFT;
1246                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1247
1248                 kvm_for_each_vcpu(n, vcpu, kvm)
1249                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1250
1251                 n = kvm_dirty_bitmap_bytes(memslot);
1252                 memset(memslot->dirty_bitmap, 0, n);
1253         }
1254
1255         r = 0;
1256 out:
1257         mutex_unlock(&kvm->slots_lock);
1258         return r;
1259 }
1260
1261 #ifdef CONFIG_PPC64
1262 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1263 {
1264         info->flags = KVM_PPC_1T_SEGMENTS;
1265
1266         /* SLB is always 64 entries */
1267         info->slb_size = 64;
1268
1269         /* Standard 4k base page size segment */
1270         info->sps[0].page_shift = 12;
1271         info->sps[0].slb_enc = 0;
1272         info->sps[0].enc[0].page_shift = 12;
1273         info->sps[0].enc[0].pte_enc = 0;
1274
1275         /* Standard 16M large page size segment */
1276         info->sps[1].page_shift = 24;
1277         info->sps[1].slb_enc = SLB_VSID_L;
1278         info->sps[1].enc[0].page_shift = 24;
1279         info->sps[1].enc[0].pte_enc = 0;
1280
1281         return 0;
1282 }
1283 #endif /* CONFIG_PPC64 */
1284
1285 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1286                               struct kvm_memory_slot *dont)
1287 {
1288 }
1289
1290 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1291                                unsigned long npages)
1292 {
1293         return 0;
1294 }
1295
1296 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1297                                       struct kvm_memory_slot *memslot,
1298                                       struct kvm_userspace_memory_region *mem)
1299 {
1300         return 0;
1301 }
1302
1303 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1304                                 struct kvm_userspace_memory_region *mem,
1305                                 const struct kvm_memory_slot *old)
1306 {
1307 }
1308
1309 void kvmppc_core_flush_memslot(struct kvm *kvm, struct kvm_memory_slot *memslot)
1310 {
1311 }
1312
1313 static unsigned int kvm_global_user_count = 0;
1314 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1315
1316 int kvmppc_core_init_vm(struct kvm *kvm)
1317 {
1318 #ifdef CONFIG_PPC64
1319         INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1320         INIT_LIST_HEAD(&kvm->arch.rtas_tokens);
1321 #endif
1322
1323         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1324                 spin_lock(&kvm_global_user_count_lock);
1325                 if (++kvm_global_user_count == 1)
1326                         pSeries_disable_reloc_on_exc();
1327                 spin_unlock(&kvm_global_user_count_lock);
1328         }
1329         return 0;
1330 }
1331
1332 void kvmppc_core_destroy_vm(struct kvm *kvm)
1333 {
1334 #ifdef CONFIG_PPC64
1335         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1336 #endif
1337
1338         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1339                 spin_lock(&kvm_global_user_count_lock);
1340                 BUG_ON(kvm_global_user_count == 0);
1341                 if (--kvm_global_user_count == 0)
1342                         pSeries_enable_reloc_on_exc();
1343                 spin_unlock(&kvm_global_user_count_lock);
1344         }
1345 }
1346
1347 static int kvmppc_book3s_init(void)
1348 {
1349         int r;
1350
1351         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1352                      THIS_MODULE);
1353
1354         if (r)
1355                 return r;
1356
1357         r = kvmppc_mmu_hpte_sysinit();
1358
1359         return r;
1360 }
1361
1362 static void kvmppc_book3s_exit(void)
1363 {
1364         kvmppc_mmu_hpte_sysexit();
1365         kvm_exit();
1366 }
1367
1368 module_init(kvmppc_book3s_init);
1369 module_exit(kvmppc_book3s_exit);