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 <linux/gfp.h>
  37 #include <linux/sched.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/highmem.h>
  40
  41 #include "trace.h"
  42
  43 /* #define EXIT_DEBUG */
  44 /* #define DEBUG_EXT */
  45
  46 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  47                              ulong msr);
  48
  49 /* Some compatibility defines */
  50 #ifdef CONFIG_PPC_BOOK3S_32
  51 #define MSR_USER32 MSR_USER
  52 #define MSR_USER64 MSR_USER
  53 #define HW_PAGE_SIZE PAGE_SIZE
  54 #endif
  55
  56 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  57 {
  58 #ifdef CONFIG_PPC_BOOK3S_64
  59         memcpy(to_svcpu(vcpu)->slb, to_book3s(vcpu)->slb_shadow, sizeof(to_svcpu(vcpu)->slb));
  60         memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
  61                sizeof(get_paca()->shadow_vcpu));
  62         to_svcpu(vcpu)->slb_max = to_book3s(vcpu)->slb_shadow_max;
  63 #endif
  64
  65 #ifdef CONFIG_PPC_BOOK3S_32
  66         current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
  67 #endif
  68 }
  69
  70 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  71 {
  72 #ifdef CONFIG_PPC_BOOK3S_64
  73         memcpy(to_book3s(vcpu)->slb_shadow, to_svcpu(vcpu)->slb, sizeof(to_svcpu(vcpu)->slb));
  74         memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  75                sizeof(get_paca()->shadow_vcpu));
  76         to_book3s(vcpu)->slb_shadow_max = to_svcpu(vcpu)->slb_max;
  77 #endif
  78
  79         kvmppc_giveup_ext(vcpu, MSR_FP);
  80         kvmppc_giveup_ext(vcpu, MSR_VEC);
  81         kvmppc_giveup_ext(vcpu, MSR_VSX);
  82 }
  83
  84 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
  85 {
  86         ulong smsr = vcpu->arch.shared->msr;
  87
  88         /* Guest MSR values */
  89         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE;
  90         /* Process MSR values */
  91         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
  92         /* External providers the guest reserved */
  93         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
  94         /* 64-bit Process MSR values */
  95 #ifdef CONFIG_PPC_BOOK3S_64
  96         smsr |= MSR_ISF | MSR_HV;
  97 #endif
  98         vcpu->arch.shadow_msr = smsr;
  99 }
 100
 101 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 102 {
 103         ulong old_msr = vcpu->arch.shared->msr;
 104
 105 #ifdef EXIT_DEBUG
 106         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 107 #endif
 108
 109         msr &= to_book3s(vcpu)->msr_mask;
 110         vcpu->arch.shared->msr = msr;
 111         kvmppc_recalc_shadow_msr(vcpu);
 112
 113         if (msr & MSR_POW) {
 114                 if (!vcpu->arch.pending_exceptions) {
 115                         kvm_vcpu_block(vcpu);
 116                         vcpu->stat.halt_wakeup++;
 117
 118                         /* Unset POW bit after we woke up */
 119                         msr &= ~MSR_POW;
 120                         vcpu->arch.shared->msr = msr;
 121                 }
 122         }
 123
 124         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 125                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 126                 kvmppc_mmu_flush_segments(vcpu);
 127                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 128
 129                 /* Preload magic page segment when in kernel mode */
 130                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 131                         struct kvm_vcpu_arch *a = &vcpu->arch;
 132
 133                         if (msr & MSR_DR)
 134                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 135                         else
 136                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 137                 }
 138         }
 139
 140         /* Preload FPU if it's enabled */
 141         if (vcpu->arch.shared->msr & MSR_FP)
 142                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 143 }
 144
 145 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 146 {
 147         u32 host_pvr;
 148
 149         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 150         vcpu->arch.pvr = pvr;
 151 #ifdef CONFIG_PPC_BOOK3S_64
 152         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 153                 kvmppc_mmu_book3s_64_init(vcpu);
 154                 to_book3s(vcpu)->hior = 0xfff00000;
 155                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 156                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 157         } else
 158 #endif
 159         {
 160                 kvmppc_mmu_book3s_32_init(vcpu);
 161                 to_book3s(vcpu)->hior = 0;
 162                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 163                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 164         }
 165
 166         kvmppc_sanity_check(vcpu);
 167
 168         /* If we are in hypervisor level on 970, we can tell the CPU to
 169          * treat DCBZ as 32 bytes store */
 170         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 171         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 172             !strcmp(cur_cpu_spec->platform, "ppc970"))
 173                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 174
 175         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 176            really needs them in a VM on Cell and force disable them. */
 177         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 178                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 179
 180 #ifdef CONFIG_PPC_BOOK3S_32
 181         /* 32 bit Book3S always has 32 byte dcbz */
 182         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 183 #endif
 184
 185         /* On some CPUs we can execute paired single operations natively */
 186         asm ( "mfpvr %0" : "=r"(host_pvr));
 187         switch (host_pvr) {
 188         case 0x00080200:        /* lonestar 2.0 */
 189         case 0x00088202:        /* lonestar 2.2 */
 190         case 0x70000100:        /* gekko 1.0 */
 191         case 0x00080100:        /* gekko 2.0 */
 192         case 0x00083203:        /* gekko 2.3a */
 193         case 0x00083213:        /* gekko 2.3b */
 194         case 0x00083204:        /* gekko 2.4 */
 195         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 196         case 0x00087200:        /* broadway */
 197                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 198                 /* Enable HID2.PSE - in case we need it later */
 199                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 200         }
 201 }
 202
 203 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 204  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 205  * emulate 32 bytes dcbz length.
 206  *
 207  * The Book3s_64 inventors also realized this case and implemented a special bit
 208  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 209  *
 210  * My approach here is to patch the dcbz instruction on executing pages.
 211  */
 212 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 213 {
 214         struct page *hpage;
 215         u64 hpage_offset;
 216         u32 *page;
 217         int i;
 218
 219         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 220         if (is_error_page(hpage)) {
 221                 kvm_release_page_clean(hpage);
 222                 return;
 223         }
 224
 225         hpage_offset = pte->raddr & ~PAGE_MASK;
 226         hpage_offset &= ~0xFFFULL;
 227         hpage_offset /= 4;
 228
 229         get_page(hpage);
 230         page = kmap_atomic(hpage, KM_USER0);
 231
 232         /* patch dcbz into reserved instruction, so we trap */
 233         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 234                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 235                         page[i] &= 0xfffffff7;
 236
 237         kunmap_atomic(page, KM_USER0);
 238         put_page(hpage);
 239 }
 240
 241 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 242 {
 243         ulong mp_pa = vcpu->arch.magic_page_pa;
 244
 245         if (unlikely(mp_pa) &&
 246             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 247                 return 1;
 248         }
 249
 250         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 251 }
 252
 253 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 254                             ulong eaddr, int vec)
 255 {
 256         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 257         int r = RESUME_GUEST;
 258         int relocated;
 259         int page_found = 0;
 260         struct kvmppc_pte pte;
 261         bool is_mmio = false;
 262         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 263         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 264         u64 vsid;
 265
 266         relocated = data ? dr : ir;
 267
 268         /* Resolve real address if translation turned on */
 269         if (relocated) {
 270                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 271         } else {
 272                 pte.may_execute = true;
 273                 pte.may_read = true;
 274                 pte.may_write = true;
 275                 pte.raddr = eaddr & KVM_PAM;
 276                 pte.eaddr = eaddr;
 277                 pte.vpage = eaddr >> 12;
 278         }
 279
 280         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 281         case 0:
 282                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 283                 break;
 284         case MSR_DR:
 285         case MSR_IR:
 286                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 287
 288                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 289                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 290                 else
 291                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 292                 pte.vpage |= vsid;
 293
 294                 if (vsid == -1)
 295                         page_found = -EINVAL;
 296                 break;
 297         }
 298
 299         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 300            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 301                 /*
 302                  * If we do the dcbz hack, we have to NX on every execution,
 303                  * so we can patch the executing code. This renders our guest
 304                  * NX-less.
 305                  */
 306                 pte.may_execute = !data;
 307         }
 308
 309         if (page_found == -ENOENT) {
 310                 /* Page not found in guest PTE entries */
 311                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 312                 vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 313                 vcpu->arch.shared->msr |=
 314                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 315                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 316         } else if (page_found == -EPERM) {
 317                 /* Storage protection */
 318                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 319                 vcpu->arch.shared->dsisr =
 320                         to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
 321                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 322                 vcpu->arch.shared->msr |=
 323                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 324                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 325         } else if (page_found == -EINVAL) {
 326                 /* Page not found in guest SLB */
 327                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 328                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 329         } else if (!is_mmio &&
 330                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 331                 /* The guest's PTE is not mapped yet. Map on the host */
 332                 kvmppc_mmu_map_page(vcpu, &pte);
 333                 if (data)
 334                         vcpu->stat.sp_storage++;
 335                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 336                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 337                         kvmppc_patch_dcbz(vcpu, &pte);
 338         } else {
 339                 /* MMIO */
 340                 vcpu->stat.mmio_exits++;
 341                 vcpu->arch.paddr_accessed = pte.raddr;
 342                 r = kvmppc_emulate_mmio(run, vcpu);
 343                 if ( r == RESUME_HOST_NV )
 344                         r = RESUME_HOST;
 345         }
 346
 347         return r;
 348 }
 349
 350 static inline int get_fpr_index(int i)
 351 {
 352 #ifdef CONFIG_VSX
 353         i *= 2;
 354 #endif
 355         return i;
 356 }
 357
 358 /* Give up external provider (FPU, Altivec, VSX) */
 359 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 360 {
 361         struct thread_struct *t = &current->thread;
 362         u64 *vcpu_fpr = vcpu->arch.fpr;
 363 #ifdef CONFIG_VSX
 364         u64 *vcpu_vsx = vcpu->arch.vsr;
 365 #endif
 366         u64 *thread_fpr = (u64*)t->fpr;
 367         int i;
 368
 369         if (!(vcpu->arch.guest_owned_ext & msr))
 370                 return;
 371
 372 #ifdef DEBUG_EXT
 373         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 374 #endif
 375
 376         switch (msr) {
 377         case MSR_FP:
 378                 giveup_fpu(current);
 379                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 380                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 381
 382                 vcpu->arch.fpscr = t->fpscr.val;
 383                 break;
 384         case MSR_VEC:
 385 #ifdef CONFIG_ALTIVEC
 386                 giveup_altivec(current);
 387                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 388                 vcpu->arch.vscr = t->vscr;
 389 #endif
 390                 break;
 391         case MSR_VSX:
 392 #ifdef CONFIG_VSX
 393                 __giveup_vsx(current);
 394                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 395                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 396 #endif
 397                 break;
 398         default:
 399                 BUG();
 400         }
 401
 402         vcpu->arch.guest_owned_ext &= ~msr;
 403         current->thread.regs->msr &= ~msr;
 404         kvmppc_recalc_shadow_msr(vcpu);
 405 }
 406
 407 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 408 {
 409         ulong srr0 = kvmppc_get_pc(vcpu);
 410         u32 last_inst = kvmppc_get_last_inst(vcpu);
 411         int ret;
 412
 413         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 414         if (ret == -ENOENT) {
 415                 ulong msr = vcpu->arch.shared->msr;
 416
 417                 msr = kvmppc_set_field(msr, 33, 33, 1);
 418                 msr = kvmppc_set_field(msr, 34, 36, 0);
 419                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 420                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 421                 return EMULATE_AGAIN;
 422         }
 423
 424         return EMULATE_DONE;
 425 }
 426
 427 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 428 {
 429
 430         /* Need to do paired single emulation? */
 431         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 432                 return EMULATE_DONE;
 433
 434         /* Read out the instruction */
 435         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 436                 /* Need to emulate */
 437                 return EMULATE_FAIL;
 438
 439         return EMULATE_AGAIN;
 440 }
 441
 442 /* Handle external providers (FPU, Altivec, VSX) */
 443 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 444                              ulong msr)
 445 {
 446         struct thread_struct *t = &current->thread;
 447         u64 *vcpu_fpr = vcpu->arch.fpr;
 448 #ifdef CONFIG_VSX
 449         u64 *vcpu_vsx = vcpu->arch.vsr;
 450 #endif
 451         u64 *thread_fpr = (u64*)t->fpr;
 452         int i;
 453
 454         /* When we have paired singles, we emulate in software */
 455         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 456                 return RESUME_GUEST;
 457
 458         if (!(vcpu->arch.shared->msr & msr)) {
 459                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 460                 return RESUME_GUEST;
 461         }
 462
 463         /* We already own the ext */
 464         if (vcpu->arch.guest_owned_ext & msr) {
 465                 return RESUME_GUEST;
 466         }
 467
 468 #ifdef DEBUG_EXT
 469         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 470 #endif
 471
 472         current->thread.regs->msr |= msr;
 473
 474         switch (msr) {
 475         case MSR_FP:
 476                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 477                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 478
 479                 t->fpscr.val = vcpu->arch.fpscr;
 480                 t->fpexc_mode = 0;
 481                 kvmppc_load_up_fpu();
 482                 break;
 483         case MSR_VEC:
 484 #ifdef CONFIG_ALTIVEC
 485                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 486                 t->vscr = vcpu->arch.vscr;
 487                 t->vrsave = -1;
 488                 kvmppc_load_up_altivec();
 489 #endif
 490                 break;
 491         case MSR_VSX:
 492 #ifdef CONFIG_VSX
 493                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 494                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 495                 kvmppc_load_up_vsx();
 496 #endif
 497                 break;
 498         default:
 499                 BUG();
 500         }
 501
 502         vcpu->arch.guest_owned_ext |= msr;
 503
 504         kvmppc_recalc_shadow_msr(vcpu);
 505
 506         return RESUME_GUEST;
 507 }
 508
 509 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 510                        unsigned int exit_nr)
 511 {
 512         int r = RESUME_HOST;
 513
 514         vcpu->stat.sum_exits++;
 515
 516         run->exit_reason = KVM_EXIT_UNKNOWN;
 517         run->ready_for_interrupt_injection = 1;
 518
 519         trace_kvm_book3s_exit(exit_nr, vcpu);
 520         kvm_resched(vcpu);
 521         switch (exit_nr) {
 522         case BOOK3S_INTERRUPT_INST_STORAGE:
 523                 vcpu->stat.pf_instruc++;
 524
 525 #ifdef CONFIG_PPC_BOOK3S_32
 526                 /* We set segments as unused segments when invalidating them. So
 527                  * treat the respective fault as segment fault. */
 528                 if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
 529                     == SR_INVALID) {
 530                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 531                         r = RESUME_GUEST;
 532                         break;
 533                 }
 534 #endif
 535
 536                 /* only care about PTEG not found errors, but leave NX alone */
 537                 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
 538                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 539                         vcpu->stat.sp_instruc++;
 540                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 541                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 542                         /*
 543                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 544                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 545                          *     that no guest that needs the dcbz hack does NX.
 546                          */
 547                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 548                         r = RESUME_GUEST;
 549                 } else {
 550                         vcpu->arch.shared->msr |=
 551                                 to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
 552                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 553                         r = RESUME_GUEST;
 554                 }
 555                 break;
 556         case BOOK3S_INTERRUPT_DATA_STORAGE:
 557         {
 558                 ulong dar = kvmppc_get_fault_dar(vcpu);
 559                 vcpu->stat.pf_storage++;
 560
 561 #ifdef CONFIG_PPC_BOOK3S_32
 562                 /* We set segments as unused segments when invalidating them. So
 563                  * treat the respective fault as segment fault. */
 564                 if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 565                         kvmppc_mmu_map_segment(vcpu, dar);
 566                         r = RESUME_GUEST;
 567                         break;
 568                 }
 569 #endif
 570
 571                 /* The only case we need to handle is missing shadow PTEs */
 572                 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
 573                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 574                 } else {
 575                         vcpu->arch.shared->dar = dar;
 576                         vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 577                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 578                         r = RESUME_GUEST;
 579                 }
 580                 break;
 581         }
 582         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 583                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 584                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 585                         kvmppc_book3s_queue_irqprio(vcpu,
 586                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 587                 }
 588                 r = RESUME_GUEST;
 589                 break;
 590         case BOOK3S_INTERRUPT_INST_SEGMENT:
 591                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 592                         kvmppc_book3s_queue_irqprio(vcpu,
 593                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 594                 }
 595                 r = RESUME_GUEST;
 596                 break;
 597         /* We're good on these - the host merely wanted to get our attention */
 598         case BOOK3S_INTERRUPT_DECREMENTER:
 599                 vcpu->stat.dec_exits++;
 600                 r = RESUME_GUEST;
 601                 break;
 602         case BOOK3S_INTERRUPT_EXTERNAL:
 603                 vcpu->stat.ext_intr_exits++;
 604                 r = RESUME_GUEST;
 605                 break;
 606         case BOOK3S_INTERRUPT_PERFMON:
 607                 r = RESUME_GUEST;
 608                 break;
 609         case BOOK3S_INTERRUPT_PROGRAM:
 610         {
 611                 enum emulation_result er;
 612                 ulong flags;
 613
 614 program_interrupt:
 615                 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
 616
 617                 if (vcpu->arch.shared->msr & MSR_PR) {
 618 #ifdef EXIT_DEBUG
 619                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 620 #endif
 621                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 622                             (INS_DCBZ & 0xfffffff7)) {
 623                                 kvmppc_core_queue_program(vcpu, flags);
 624                                 r = RESUME_GUEST;
 625                                 break;
 626                         }
 627                 }
 628
 629                 vcpu->stat.emulated_inst_exits++;
 630                 er = kvmppc_emulate_instruction(run, vcpu);
 631                 switch (er) {
 632                 case EMULATE_DONE:
 633                         r = RESUME_GUEST_NV;
 634                         break;
 635                 case EMULATE_AGAIN:
 636                         r = RESUME_GUEST;
 637                         break;
 638                 case EMULATE_FAIL:
 639                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 640                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 641                         kvmppc_core_queue_program(vcpu, flags);
 642                         r = RESUME_GUEST;
 643                         break;
 644                 case EMULATE_DO_MMIO:
 645                         run->exit_reason = KVM_EXIT_MMIO;
 646                         r = RESUME_HOST_NV;
 647                         break;
 648                 default:
 649                         BUG();
 650                 }
 651                 break;
 652         }
 653         case BOOK3S_INTERRUPT_SYSCALL:
 654                 if (vcpu->arch.papr_enabled &&
 655                     (kvmppc_get_last_inst(vcpu) == 0x44000022) &&
 656                     !(vcpu->arch.shared->msr & MSR_PR)) {
 657                         /* SC 1 papr hypercalls */
 658                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
 659                         int i;
 660
 661                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
 662                                 r = RESUME_GUEST;
 663                                 break;
 664                         }
 665
 666                         run->papr_hcall.nr = cmd;
 667                         for (i = 0; i < 9; ++i) {
 668                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
 669                                 run->papr_hcall.args[i] = gpr;
 670                         }
 671                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
 672                         vcpu->arch.hcall_needed = 1;
 673                         r = RESUME_HOST;
 674                 } else if (vcpu->arch.osi_enabled &&
 675                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 676                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 677                         /* MOL hypercalls */
 678                         u64 *gprs = run->osi.gprs;
 679                         int i;
 680
 681                         run->exit_reason = KVM_EXIT_OSI;
 682                         for (i = 0; i < 32; i++)
 683                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 684                         vcpu->arch.osi_needed = 1;
 685                         r = RESUME_HOST_NV;
 686                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 687                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 688                         /* KVM PV hypercalls */
 689                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 690                         r = RESUME_GUEST;
 691                 } else {
 692                         /* Guest syscalls */
 693                         vcpu->stat.syscall_exits++;
 694                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 695                         r = RESUME_GUEST;
 696                 }
 697                 break;
 698         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 699         case BOOK3S_INTERRUPT_ALTIVEC:
 700         case BOOK3S_INTERRUPT_VSX:
 701         {
 702                 int ext_msr = 0;
 703
 704                 switch (exit_nr) {
 705                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 706                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 707                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 708                 }
 709
 710                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
 711                 case EMULATE_DONE:
 712                         /* everything ok - let's enable the ext */
 713                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
 714                         break;
 715                 case EMULATE_FAIL:
 716                         /* we need to emulate this instruction */
 717                         goto program_interrupt;
 718                         break;
 719                 default:
 720                         /* nothing to worry about - go again */
 721                         break;
 722                 }
 723                 break;
 724         }
 725         case BOOK3S_INTERRUPT_ALIGNMENT:
 726                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
 727                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
 728                                 kvmppc_get_last_inst(vcpu));
 729                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
 730                                 kvmppc_get_last_inst(vcpu));
 731                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 732                 }
 733                 r = RESUME_GUEST;
 734                 break;
 735         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 736         case BOOK3S_INTERRUPT_TRACE:
 737                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 738                 r = RESUME_GUEST;
 739                 break;
 740         default:
 741                 /* Ugh - bork here! What did we get? */
 742                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
 743                         exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
 744                 r = RESUME_HOST;
 745                 BUG();
 746                 break;
 747         }
 748
 749
 750         if (!(r & RESUME_HOST)) {
 751                 /* To avoid clobbering exit_reason, only check for signals if
 752                  * we aren't already exiting to userspace for some other
 753                  * reason. */
 754                 if (signal_pending(current)) {
 755 #ifdef EXIT_DEBUG
 756                         printk(KERN_EMERG "KVM: Going back to host\n");
 757 #endif
 758                         vcpu->stat.signal_exits++;
 759                         run->exit_reason = KVM_EXIT_INTR;
 760                         r = -EINTR;
 761                 } else {
 762                         /* In case an interrupt came in that was triggered
 763                          * from userspace (like DEC), we need to check what
 764                          * to inject now! */
 765                         kvmppc_core_deliver_interrupts(vcpu);
 766                 }
 767         }
 768
 769         trace_kvm_book3s_reenter(r, vcpu);
 770
 771         return r;
 772 }
 773
 774 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 775                                   struct kvm_sregs *sregs)
 776 {
 777         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 778         int i;
 779
 780         sregs->pvr = vcpu->arch.pvr;
 781
 782         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 783         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 784                 for (i = 0; i < 64; i++) {
 785                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
 786                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 787                 }
 788         } else {
 789                 for (i = 0; i < 16; i++)
 790                         sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
 791
 792                 for (i = 0; i < 8; i++) {
 793                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 794                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 795                 }
 796         }
 797
 798         return 0;
 799 }
 800
 801 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 802                                   struct kvm_sregs *sregs)
 803 {
 804         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 805         int i;
 806
 807         kvmppc_set_pvr(vcpu, sregs->pvr);
 808
 809         vcpu3s->sdr1 = sregs->u.s.sdr1;
 810         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 811                 for (i = 0; i < 64; i++) {
 812                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 813                                                     sregs->u.s.ppc64.slb[i].slbe);
 814                 }
 815         } else {
 816                 for (i = 0; i < 16; i++) {
 817                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 818                 }
 819                 for (i = 0; i < 8; i++) {
 820                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 821                                        (u32)sregs->u.s.ppc32.ibat[i]);
 822                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 823                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 824                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 825                                        (u32)sregs->u.s.ppc32.dbat[i]);
 826                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 827                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 828                 }
 829         }
 830
 831         /* Flush the MMU after messing with the segments */
 832         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 833
 834         return 0;
 835 }
 836
 837 int kvmppc_core_check_processor_compat(void)
 838 {
 839         return 0;
 840 }
 841
 842 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 843 {
 844         struct kvmppc_vcpu_book3s *vcpu_book3s;
 845         struct kvm_vcpu *vcpu;
 846         int err = -ENOMEM;
 847         unsigned long p;
 848
 849         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
 850         if (!vcpu_book3s)
 851                 goto out;
 852
 853         vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
 854                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
 855         if (!vcpu_book3s->shadow_vcpu)
 856                 goto free_vcpu;
 857
 858         vcpu = &vcpu_book3s->vcpu;
 859         err = kvm_vcpu_init(vcpu, kvm, id);
 860         if (err)
 861                 goto free_shadow_vcpu;
 862
 863         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
 864         /* the real shared page fills the last 4k of our page */
 865         vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
 866         if (!p)
 867                 goto uninit_vcpu;
 868
 869 #ifdef CONFIG_PPC_BOOK3S_64
 870         /* default to book3s_64 (970fx) */
 871         vcpu->arch.pvr = 0x3C0301;
 872 #else
 873         /* default to book3s_32 (750) */
 874         vcpu->arch.pvr = 0x84202;
 875 #endif
 876         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 877         vcpu->arch.slb_nr = 64;
 878
 879         vcpu->arch.shadow_msr = MSR_USER64;
 880
 881         err = kvmppc_mmu_init(vcpu);
 882         if (err < 0)
 883                 goto uninit_vcpu;
 884
 885         return vcpu;
 886
 887 uninit_vcpu:
 888         kvm_vcpu_uninit(vcpu);
 889 free_shadow_vcpu:
 890         kfree(vcpu_book3s->shadow_vcpu);
 891 free_vcpu:
 892         vfree(vcpu_book3s);
 893 out:
 894         return ERR_PTR(err);
 895 }
 896
 897 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 898 {
 899         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
 900
 901         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
 902         kvm_vcpu_uninit(vcpu);
 903         kfree(vcpu_book3s->shadow_vcpu);
 904         vfree(vcpu_book3s);
 905 }
 906
 907 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 908 {
 909         int ret;
 910         double fpr[32][TS_FPRWIDTH];
 911         unsigned int fpscr;
 912         int fpexc_mode;
 913 #ifdef CONFIG_ALTIVEC
 914         vector128 vr[32];
 915         vector128 vscr;
 916         unsigned long uninitialized_var(vrsave);
 917         int used_vr;
 918 #endif
 919 #ifdef CONFIG_VSX
 920         int used_vsr;
 921 #endif
 922         ulong ext_msr;
 923
 924         /* Check if we can run the vcpu at all */
 925         if (!vcpu->arch.sane) {
 926                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 927                 return -EINVAL;
 928         }
 929
 930         /* No need to go into the guest when all we do is going out */
 931         if (signal_pending(current)) {
 932                 kvm_run->exit_reason = KVM_EXIT_INTR;
 933                 return -EINTR;
 934         }
 935
 936         /* Save FPU state in stack */
 937         if (current->thread.regs->msr & MSR_FP)
 938                 giveup_fpu(current);
 939         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
 940         fpscr = current->thread.fpscr.val;
 941         fpexc_mode = current->thread.fpexc_mode;
 942
 943 #ifdef CONFIG_ALTIVEC
 944         /* Save Altivec state in stack */
 945         used_vr = current->thread.used_vr;
 946         if (used_vr) {
 947                 if (current->thread.regs->msr & MSR_VEC)
 948                         giveup_altivec(current);
 949                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
 950                 vscr = current->thread.vscr;
 951                 vrsave = current->thread.vrsave;
 952         }
 953 #endif
 954
 955 #ifdef CONFIG_VSX
 956         /* Save VSX state in stack */
 957         used_vsr = current->thread.used_vsr;
 958         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
 959                         __giveup_vsx(current);
 960 #endif
 961
 962         /* Remember the MSR with disabled extensions */
 963         ext_msr = current->thread.regs->msr;
 964
 965         /* Preload FPU if it's enabled */
 966         if (vcpu->arch.shared->msr & MSR_FP)
 967                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 968
 969         kvm_guest_enter();
 970
 971         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
 972
 973         kvm_guest_exit();
 974
 975         local_irq_disable();
 976
 977         current->thread.regs->msr = ext_msr;
 978
 979         /* Make sure we save the guest FPU/Altivec/VSX state */
 980         kvmppc_giveup_ext(vcpu, MSR_FP);
 981         kvmppc_giveup_ext(vcpu, MSR_VEC);
 982         kvmppc_giveup_ext(vcpu, MSR_VSX);
 983
 984         /* Restore FPU state from stack */
 985         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
 986         current->thread.fpscr.val = fpscr;
 987         current->thread.fpexc_mode = fpexc_mode;
 988
 989 #ifdef CONFIG_ALTIVEC
 990         /* Restore Altivec state from stack */
 991         if (used_vr && current->thread.used_vr) {
 992                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
 993                 current->thread.vscr = vscr;
 994                 current->thread.vrsave = vrsave;
 995         }
 996         current->thread.used_vr = used_vr;
 997 #endif
 998
 999 #ifdef CONFIG_VSX
1000         current->thread.used_vsr = used_vsr;
1001 #endif
1002
1003         return ret;
1004 }
1005
1006 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1007                                       struct kvm_userspace_memory_region *mem)
1008 {
1009         return 0;
1010 }
1011
1012 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1013                                 struct kvm_userspace_memory_region *mem)
1014 {
1015 }
1016
1017 int kvmppc_core_init_vm(struct kvm *kvm)
1018 {
1019         return 0;
1020 }
1021
1022 void kvmppc_core_destroy_vm(struct kvm *kvm)
1023 {
1024 }
1025
1026 static int kvmppc_book3s_init(void)
1027 {
1028         int r;
1029
1030         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1031                      THIS_MODULE);
1032
1033         if (r)
1034                 return r;
1035
1036         r = kvmppc_mmu_hpte_sysinit();
1037
1038         return r;
1039 }
1040
1041 static void kvmppc_book3s_exit(void)
1042 {
1043         kvmppc_mmu_hpte_sysexit();
1044         kvm_exit();
1045 }
1046
1047 module_init(kvmppc_book3s_init);
1048 module_exit(kvmppc_book3s_exit);