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/err.h>
  24 #include <linux/slab.h>
  25
  26 #include <asm/reg.h>
  27 #include <asm/cputable.h>
  28 #include <asm/cacheflush.h>
  29 #include <asm/tlbflush.h>
  30 #include <asm/uaccess.h>
  31 #include <asm/io.h>
  32 #include <asm/kvm_ppc.h>
  33 #include <asm/kvm_book3s.h>
  34 #include <asm/mmu_context.h>
  35 #include <linux/gfp.h>
  36 #include <linux/sched.h>
  37 #include <linux/vmalloc.h>
  38 #include <linux/highmem.h>
  39
  40 #include "trace.h"
  41
  42 /* #define EXIT_DEBUG */
  43 /* #define DEBUG_EXT */
  44
  45 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  46                              ulong msr);
  47
  48 /* Some compatibility defines */
  49 #ifdef CONFIG_PPC_BOOK3S_32
  50 #define MSR_USER32 MSR_USER
  51 #define MSR_USER64 MSR_USER
  52 #define HW_PAGE_SIZE PAGE_SIZE
  53 #endif
  54
  55 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  56 {
  57 #ifdef CONFIG_PPC_BOOK3S_64
  58         memcpy(to_svcpu(vcpu)->slb, to_book3s(vcpu)->slb_shadow, sizeof(to_svcpu(vcpu)->slb));
  59         memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
  60                sizeof(get_paca()->shadow_vcpu));
  61         to_svcpu(vcpu)->slb_max = to_book3s(vcpu)->slb_shadow_max;
  62 #endif
  63
  64 #ifdef CONFIG_PPC_BOOK3S_32
  65         current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
  66 #endif
  67 }
  68
  69 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  70 {
  71 #ifdef CONFIG_PPC_BOOK3S_64
  72         memcpy(to_book3s(vcpu)->slb_shadow, to_svcpu(vcpu)->slb, sizeof(to_svcpu(vcpu)->slb));
  73         memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  74                sizeof(get_paca()->shadow_vcpu));
  75         to_book3s(vcpu)->slb_shadow_max = to_svcpu(vcpu)->slb_max;
  76 #endif
  77
  78         kvmppc_giveup_ext(vcpu, MSR_FP);
  79         kvmppc_giveup_ext(vcpu, MSR_VEC);
  80         kvmppc_giveup_ext(vcpu, MSR_VSX);
  81 }
  82
  83 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
  84 {
  85         ulong smsr = vcpu->arch.shared->msr;
  86
  87         /* Guest MSR values */
  88         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE;
  89         /* Process MSR values */
  90         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
  91         /* External providers the guest reserved */
  92         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
  93         /* 64-bit Process MSR values */
  94 #ifdef CONFIG_PPC_BOOK3S_64
  95         smsr |= MSR_ISF | MSR_HV;
  96 #endif
  97         vcpu->arch.shadow_msr = smsr;
  98 }
  99
 100 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 101 {
 102         ulong old_msr = vcpu->arch.shared->msr;
 103
 104 #ifdef EXIT_DEBUG
 105         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 106 #endif
 107
 108         msr &= to_book3s(vcpu)->msr_mask;
 109         vcpu->arch.shared->msr = msr;
 110         kvmppc_recalc_shadow_msr(vcpu);
 111
 112         if (msr & MSR_POW) {
 113                 if (!vcpu->arch.pending_exceptions) {
 114                         kvm_vcpu_block(vcpu);
 115                         vcpu->stat.halt_wakeup++;
 116
 117                         /* Unset POW bit after we woke up */
 118                         msr &= ~MSR_POW;
 119                         vcpu->arch.shared->msr = msr;
 120                 }
 121         }
 122
 123         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 124                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 125                 kvmppc_mmu_flush_segments(vcpu);
 126                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 127
 128                 /* Preload magic page segment when in kernel mode */
 129                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 130                         struct kvm_vcpu_arch *a = &vcpu->arch;
 131
 132                         if (msr & MSR_DR)
 133                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 134                         else
 135                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 136                 }
 137         }
 138
 139         /* Preload FPU if it's enabled */
 140         if (vcpu->arch.shared->msr & MSR_FP)
 141                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 142 }
 143
 144 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 145 {
 146         u32 host_pvr;
 147
 148         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 149         vcpu->arch.pvr = pvr;
 150 #ifdef CONFIG_PPC_BOOK3S_64
 151         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 152                 kvmppc_mmu_book3s_64_init(vcpu);
 153                 if (!to_book3s(vcpu)->hior_sregs)
 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                 if (!to_book3s(vcpu)->hior_sregs)
 162                         to_book3s(vcpu)->hior = 0;
 163                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 164                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 165         }
 166
 167         kvmppc_sanity_check(vcpu);
 168
 169         /* If we are in hypervisor level on 970, we can tell the CPU to
 170          * treat DCBZ as 32 bytes store */
 171         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 172         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 173             !strcmp(cur_cpu_spec->platform, "ppc970"))
 174                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 175
 176         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 177            really needs them in a VM on Cell and force disable them. */
 178         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 179                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 180
 181 #ifdef CONFIG_PPC_BOOK3S_32
 182         /* 32 bit Book3S always has 32 byte dcbz */
 183         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 184 #endif
 185
 186         /* On some CPUs we can execute paired single operations natively */
 187         asm ( "mfpvr %0" : "=r"(host_pvr));
 188         switch (host_pvr) {
 189         case 0x00080200:        /* lonestar 2.0 */
 190         case 0x00088202:        /* lonestar 2.2 */
 191         case 0x70000100:        /* gekko 1.0 */
 192         case 0x00080100:        /* gekko 2.0 */
 193         case 0x00083203:        /* gekko 2.3a */
 194         case 0x00083213:        /* gekko 2.3b */
 195         case 0x00083204:        /* gekko 2.4 */
 196         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 197         case 0x00087200:        /* broadway */
 198                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 199                 /* Enable HID2.PSE - in case we need it later */
 200                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 201         }
 202 }
 203
 204 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 205  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 206  * emulate 32 bytes dcbz length.
 207  *
 208  * The Book3s_64 inventors also realized this case and implemented a special bit
 209  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 210  *
 211  * My approach here is to patch the dcbz instruction on executing pages.
 212  */
 213 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 214 {
 215         struct page *hpage;
 216         u64 hpage_offset;
 217         u32 *page;
 218         int i;
 219
 220         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 221         if (is_error_page(hpage)) {
 222                 kvm_release_page_clean(hpage);
 223                 return;
 224         }
 225
 226         hpage_offset = pte->raddr & ~PAGE_MASK;
 227         hpage_offset &= ~0xFFFULL;
 228         hpage_offset /= 4;
 229
 230         get_page(hpage);
 231         page = kmap_atomic(hpage, KM_USER0);
 232
 233         /* patch dcbz into reserved instruction, so we trap */
 234         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 235                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 236                         page[i] &= 0xfffffff7;
 237
 238         kunmap_atomic(page, KM_USER0);
 239         put_page(hpage);
 240 }
 241
 242 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 243 {
 244         ulong mp_pa = vcpu->arch.magic_page_pa;
 245
 246         if (unlikely(mp_pa) &&
 247             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 248                 return 1;
 249         }
 250
 251         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 252 }
 253
 254 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 255                             ulong eaddr, int vec)
 256 {
 257         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 258         int r = RESUME_GUEST;
 259         int relocated;
 260         int page_found = 0;
 261         struct kvmppc_pte pte;
 262         bool is_mmio = false;
 263         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 264         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 265         u64 vsid;
 266
 267         relocated = data ? dr : ir;
 268
 269         /* Resolve real address if translation turned on */
 270         if (relocated) {
 271                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 272         } else {
 273                 pte.may_execute = true;
 274                 pte.may_read = true;
 275                 pte.may_write = true;
 276                 pte.raddr = eaddr & KVM_PAM;
 277                 pte.eaddr = eaddr;
 278                 pte.vpage = eaddr >> 12;
 279         }
 280
 281         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 282         case 0:
 283                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 284                 break;
 285         case MSR_DR:
 286         case MSR_IR:
 287                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 288
 289                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 290                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 291                 else
 292                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 293                 pte.vpage |= vsid;
 294
 295                 if (vsid == -1)
 296                         page_found = -EINVAL;
 297                 break;
 298         }
 299
 300         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 301            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 302                 /*
 303                  * If we do the dcbz hack, we have to NX on every execution,
 304                  * so we can patch the executing code. This renders our guest
 305                  * NX-less.
 306                  */
 307                 pte.may_execute = !data;
 308         }
 309
 310         if (page_found == -ENOENT) {
 311                 /* Page not found in guest PTE entries */
 312                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 313                 vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 314                 vcpu->arch.shared->msr |=
 315                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 316                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 317         } else if (page_found == -EPERM) {
 318                 /* Storage protection */
 319                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 320                 vcpu->arch.shared->dsisr =
 321                         to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
 322                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 323                 vcpu->arch.shared->msr |=
 324                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 325                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 326         } else if (page_found == -EINVAL) {
 327                 /* Page not found in guest SLB */
 328                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 329                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 330         } else if (!is_mmio &&
 331                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 332                 /* The guest's PTE is not mapped yet. Map on the host */
 333                 kvmppc_mmu_map_page(vcpu, &pte);
 334                 if (data)
 335                         vcpu->stat.sp_storage++;
 336                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 337                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 338                         kvmppc_patch_dcbz(vcpu, &pte);
 339         } else {
 340                 /* MMIO */
 341                 vcpu->stat.mmio_exits++;
 342                 vcpu->arch.paddr_accessed = pte.raddr;
 343                 r = kvmppc_emulate_mmio(run, vcpu);
 344                 if ( r == RESUME_HOST_NV )
 345                         r = RESUME_HOST;
 346         }
 347
 348         return r;
 349 }
 350
 351 static inline int get_fpr_index(int i)
 352 {
 353 #ifdef CONFIG_VSX
 354         i *= 2;
 355 #endif
 356         return i;
 357 }
 358
 359 /* Give up external provider (FPU, Altivec, VSX) */
 360 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 361 {
 362         struct thread_struct *t = &current->thread;
 363         u64 *vcpu_fpr = vcpu->arch.fpr;
 364 #ifdef CONFIG_VSX
 365         u64 *vcpu_vsx = vcpu->arch.vsr;
 366 #endif
 367         u64 *thread_fpr = (u64*)t->fpr;
 368         int i;
 369
 370         if (!(vcpu->arch.guest_owned_ext & msr))
 371                 return;
 372
 373 #ifdef DEBUG_EXT
 374         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 375 #endif
 376
 377         switch (msr) {
 378         case MSR_FP:
 379                 giveup_fpu(current);
 380                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 381                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 382
 383                 vcpu->arch.fpscr = t->fpscr.val;
 384                 break;
 385         case MSR_VEC:
 386 #ifdef CONFIG_ALTIVEC
 387                 giveup_altivec(current);
 388                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 389                 vcpu->arch.vscr = t->vscr;
 390 #endif
 391                 break;
 392         case MSR_VSX:
 393 #ifdef CONFIG_VSX
 394                 __giveup_vsx(current);
 395                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 396                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 397 #endif
 398                 break;
 399         default:
 400                 BUG();
 401         }
 402
 403         vcpu->arch.guest_owned_ext &= ~msr;
 404         current->thread.regs->msr &= ~msr;
 405         kvmppc_recalc_shadow_msr(vcpu);
 406 }
 407
 408 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 409 {
 410         ulong srr0 = kvmppc_get_pc(vcpu);
 411         u32 last_inst = kvmppc_get_last_inst(vcpu);
 412         int ret;
 413
 414         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 415         if (ret == -ENOENT) {
 416                 ulong msr = vcpu->arch.shared->msr;
 417
 418                 msr = kvmppc_set_field(msr, 33, 33, 1);
 419                 msr = kvmppc_set_field(msr, 34, 36, 0);
 420                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 421                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 422                 return EMULATE_AGAIN;
 423         }
 424
 425         return EMULATE_DONE;
 426 }
 427
 428 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 429 {
 430
 431         /* Need to do paired single emulation? */
 432         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 433                 return EMULATE_DONE;
 434
 435         /* Read out the instruction */
 436         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 437                 /* Need to emulate */
 438                 return EMULATE_FAIL;
 439
 440         return EMULATE_AGAIN;
 441 }
 442
 443 /* Handle external providers (FPU, Altivec, VSX) */
 444 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 445                              ulong msr)
 446 {
 447         struct thread_struct *t = &current->thread;
 448         u64 *vcpu_fpr = vcpu->arch.fpr;
 449 #ifdef CONFIG_VSX
 450         u64 *vcpu_vsx = vcpu->arch.vsr;
 451 #endif
 452         u64 *thread_fpr = (u64*)t->fpr;
 453         int i;
 454
 455         /* When we have paired singles, we emulate in software */
 456         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 457                 return RESUME_GUEST;
 458
 459         if (!(vcpu->arch.shared->msr & msr)) {
 460                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 461                 return RESUME_GUEST;
 462         }
 463
 464         /* We already own the ext */
 465         if (vcpu->arch.guest_owned_ext & msr) {
 466                 return RESUME_GUEST;
 467         }
 468
 469 #ifdef DEBUG_EXT
 470         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 471 #endif
 472
 473         current->thread.regs->msr |= msr;
 474
 475         switch (msr) {
 476         case MSR_FP:
 477                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 478                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 479
 480                 t->fpscr.val = vcpu->arch.fpscr;
 481                 t->fpexc_mode = 0;
 482                 kvmppc_load_up_fpu();
 483                 break;
 484         case MSR_VEC:
 485 #ifdef CONFIG_ALTIVEC
 486                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 487                 t->vscr = vcpu->arch.vscr;
 488                 t->vrsave = -1;
 489                 kvmppc_load_up_altivec();
 490 #endif
 491                 break;
 492         case MSR_VSX:
 493 #ifdef CONFIG_VSX
 494                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 495                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 496                 kvmppc_load_up_vsx();
 497 #endif
 498                 break;
 499         default:
 500                 BUG();
 501         }
 502
 503         vcpu->arch.guest_owned_ext |= msr;
 504
 505         kvmppc_recalc_shadow_msr(vcpu);
 506
 507         return RESUME_GUEST;
 508 }
 509
 510 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 511                        unsigned int exit_nr)
 512 {
 513         int r = RESUME_HOST;
 514
 515         vcpu->stat.sum_exits++;
 516
 517         run->exit_reason = KVM_EXIT_UNKNOWN;
 518         run->ready_for_interrupt_injection = 1;
 519
 520         trace_kvm_book3s_exit(exit_nr, vcpu);
 521         kvm_resched(vcpu);
 522         switch (exit_nr) {
 523         case BOOK3S_INTERRUPT_INST_STORAGE:
 524                 vcpu->stat.pf_instruc++;
 525
 526 #ifdef CONFIG_PPC_BOOK3S_32
 527                 /* We set segments as unused segments when invalidating them. So
 528                  * treat the respective fault as segment fault. */
 529                 if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
 530                     == SR_INVALID) {
 531                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 532                         r = RESUME_GUEST;
 533                         break;
 534                 }
 535 #endif
 536
 537                 /* only care about PTEG not found errors, but leave NX alone */
 538                 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
 539                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 540                         vcpu->stat.sp_instruc++;
 541                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 542                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 543                         /*
 544                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 545                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 546                          *     that no guest that needs the dcbz hack does NX.
 547                          */
 548                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 549                         r = RESUME_GUEST;
 550                 } else {
 551                         vcpu->arch.shared->msr |=
 552                                 to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
 553                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 554                         r = RESUME_GUEST;
 555                 }
 556                 break;
 557         case BOOK3S_INTERRUPT_DATA_STORAGE:
 558         {
 559                 ulong dar = kvmppc_get_fault_dar(vcpu);
 560                 vcpu->stat.pf_storage++;
 561
 562 #ifdef CONFIG_PPC_BOOK3S_32
 563                 /* We set segments as unused segments when invalidating them. So
 564                  * treat the respective fault as segment fault. */
 565                 if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 566                         kvmppc_mmu_map_segment(vcpu, dar);
 567                         r = RESUME_GUEST;
 568                         break;
 569                 }
 570 #endif
 571
 572                 /* The only case we need to handle is missing shadow PTEs */
 573                 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
 574                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 575                 } else {
 576                         vcpu->arch.shared->dar = dar;
 577                         vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 578                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 579                         r = RESUME_GUEST;
 580                 }
 581                 break;
 582         }
 583         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 584                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 585                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 586                         kvmppc_book3s_queue_irqprio(vcpu,
 587                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 588                 }
 589                 r = RESUME_GUEST;
 590                 break;
 591         case BOOK3S_INTERRUPT_INST_SEGMENT:
 592                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 593                         kvmppc_book3s_queue_irqprio(vcpu,
 594                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 595                 }
 596                 r = RESUME_GUEST;
 597                 break;
 598         /* We're good on these - the host merely wanted to get our attention */
 599         case BOOK3S_INTERRUPT_DECREMENTER:
 600                 vcpu->stat.dec_exits++;
 601                 r = RESUME_GUEST;
 602                 break;
 603         case BOOK3S_INTERRUPT_EXTERNAL:
 604                 vcpu->stat.ext_intr_exits++;
 605                 r = RESUME_GUEST;
 606                 break;
 607         case BOOK3S_INTERRUPT_PERFMON:
 608                 r = RESUME_GUEST;
 609                 break;
 610         case BOOK3S_INTERRUPT_PROGRAM:
 611         {
 612                 enum emulation_result er;
 613                 ulong flags;
 614
 615 program_interrupt:
 616                 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
 617
 618                 if (vcpu->arch.shared->msr & MSR_PR) {
 619 #ifdef EXIT_DEBUG
 620                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 621 #endif
 622                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 623                             (INS_DCBZ & 0xfffffff7)) {
 624                                 kvmppc_core_queue_program(vcpu, flags);
 625                                 r = RESUME_GUEST;
 626                                 break;
 627                         }
 628                 }
 629
 630                 vcpu->stat.emulated_inst_exits++;
 631                 er = kvmppc_emulate_instruction(run, vcpu);
 632                 switch (er) {
 633                 case EMULATE_DONE:
 634                         r = RESUME_GUEST_NV;
 635                         break;
 636                 case EMULATE_AGAIN:
 637                         r = RESUME_GUEST;
 638                         break;
 639                 case EMULATE_FAIL:
 640                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 641                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 642                         kvmppc_core_queue_program(vcpu, flags);
 643                         r = RESUME_GUEST;
 644                         break;
 645                 case EMULATE_DO_MMIO:
 646                         run->exit_reason = KVM_EXIT_MMIO;
 647                         r = RESUME_HOST_NV;
 648                         break;
 649                 default:
 650                         BUG();
 651                 }
 652                 break;
 653         }
 654         case BOOK3S_INTERRUPT_SYSCALL:
 655                 if (vcpu->arch.papr_enabled &&
 656                     (kvmppc_get_last_inst(vcpu) == 0x44000022) &&
 657                     !(vcpu->arch.shared->msr & MSR_PR)) {
 658                         /* SC 1 papr hypercalls */
 659                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
 660                         int i;
 661
 662                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
 663                                 r = RESUME_GUEST;
 664                                 break;
 665                         }
 666
 667                         run->papr_hcall.nr = cmd;
 668                         for (i = 0; i < 9; ++i) {
 669                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
 670                                 run->papr_hcall.args[i] = gpr;
 671                         }
 672                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
 673                         vcpu->arch.hcall_needed = 1;
 674                         r = RESUME_HOST;
 675                 } else if (vcpu->arch.osi_enabled &&
 676                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 677                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 678                         /* MOL hypercalls */
 679                         u64 *gprs = run->osi.gprs;
 680                         int i;
 681
 682                         run->exit_reason = KVM_EXIT_OSI;
 683                         for (i = 0; i < 32; i++)
 684                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 685                         vcpu->arch.osi_needed = 1;
 686                         r = RESUME_HOST_NV;
 687                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 688                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 689                         /* KVM PV hypercalls */
 690                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 691                         r = RESUME_GUEST;
 692                 } else {
 693                         /* Guest syscalls */
 694                         vcpu->stat.syscall_exits++;
 695                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 696                         r = RESUME_GUEST;
 697                 }
 698                 break;
 699         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 700         case BOOK3S_INTERRUPT_ALTIVEC:
 701         case BOOK3S_INTERRUPT_VSX:
 702         {
 703                 int ext_msr = 0;
 704
 705                 switch (exit_nr) {
 706                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 707                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 708                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 709                 }
 710
 711                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
 712                 case EMULATE_DONE:
 713                         /* everything ok - let's enable the ext */
 714                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
 715                         break;
 716                 case EMULATE_FAIL:
 717                         /* we need to emulate this instruction */
 718                         goto program_interrupt;
 719                         break;
 720                 default:
 721                         /* nothing to worry about - go again */
 722                         break;
 723                 }
 724                 break;
 725         }
 726         case BOOK3S_INTERRUPT_ALIGNMENT:
 727                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
 728                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
 729                                 kvmppc_get_last_inst(vcpu));
 730                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
 731                                 kvmppc_get_last_inst(vcpu));
 732                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 733                 }
 734                 r = RESUME_GUEST;
 735                 break;
 736         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 737         case BOOK3S_INTERRUPT_TRACE:
 738                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 739                 r = RESUME_GUEST;
 740                 break;
 741         default:
 742                 /* Ugh - bork here! What did we get? */
 743                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
 744                         exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
 745                 r = RESUME_HOST;
 746                 BUG();
 747                 break;
 748         }
 749
 750
 751         if (!(r & RESUME_HOST)) {
 752                 /* To avoid clobbering exit_reason, only check for signals if
 753                  * we aren't already exiting to userspace for some other
 754                  * reason. */
 755                 if (signal_pending(current)) {
 756 #ifdef EXIT_DEBUG
 757                         printk(KERN_EMERG "KVM: Going back to host\n");
 758 #endif
 759                         vcpu->stat.signal_exits++;
 760                         run->exit_reason = KVM_EXIT_INTR;
 761                         r = -EINTR;
 762                 } else {
 763                         /* In case an interrupt came in that was triggered
 764                          * from userspace (like DEC), we need to check what
 765                          * to inject now! */
 766                         kvmppc_core_deliver_interrupts(vcpu);
 767                 }
 768         }
 769
 770         trace_kvm_book3s_reenter(r, vcpu);
 771
 772         return r;
 773 }
 774
 775 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 776                                   struct kvm_sregs *sregs)
 777 {
 778         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 779         int i;
 780
 781         sregs->pvr = vcpu->arch.pvr;
 782
 783         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 784         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 785                 for (i = 0; i < 64; i++) {
 786                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
 787                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 788                 }
 789         } else {
 790                 for (i = 0; i < 16; i++)
 791                         sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
 792
 793                 for (i = 0; i < 8; i++) {
 794                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 795                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 796                 }
 797         }
 798
 799         if (sregs->u.s.flags & KVM_SREGS_S_HIOR)
 800                 sregs->u.s.hior = to_book3s(vcpu)->hior;
 801
 802         return 0;
 803 }
 804
 805 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 806                                   struct kvm_sregs *sregs)
 807 {
 808         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 809         int i;
 810
 811         kvmppc_set_pvr(vcpu, sregs->pvr);
 812
 813         vcpu3s->sdr1 = sregs->u.s.sdr1;
 814         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 815                 for (i = 0; i < 64; i++) {
 816                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 817                                                     sregs->u.s.ppc64.slb[i].slbe);
 818                 }
 819         } else {
 820                 for (i = 0; i < 16; i++) {
 821                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 822                 }
 823                 for (i = 0; i < 8; i++) {
 824                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 825                                        (u32)sregs->u.s.ppc32.ibat[i]);
 826                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 827                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 828                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 829                                        (u32)sregs->u.s.ppc32.dbat[i]);
 830                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 831                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 832                 }
 833         }
 834
 835         /* Flush the MMU after messing with the segments */
 836         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 837
 838         if (sregs->u.s.flags & KVM_SREGS_S_HIOR) {
 839                 to_book3s(vcpu)->hior_sregs = true;
 840                 to_book3s(vcpu)->hior = sregs->u.s.hior;
 841         }
 842
 843         return 0;
 844 }
 845
 846 int kvmppc_core_check_processor_compat(void)
 847 {
 848         return 0;
 849 }
 850
 851 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 852 {
 853         struct kvmppc_vcpu_book3s *vcpu_book3s;
 854         struct kvm_vcpu *vcpu;
 855         int err = -ENOMEM;
 856         unsigned long p;
 857
 858         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
 859         if (!vcpu_book3s)
 860                 goto out;
 861
 862         vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
 863                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
 864         if (!vcpu_book3s->shadow_vcpu)
 865                 goto free_vcpu;
 866
 867         vcpu = &vcpu_book3s->vcpu;
 868         err = kvm_vcpu_init(vcpu, kvm, id);
 869         if (err)
 870                 goto free_shadow_vcpu;
 871
 872         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
 873         /* the real shared page fills the last 4k of our page */
 874         vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
 875         if (!p)
 876                 goto uninit_vcpu;
 877
 878 #ifdef CONFIG_PPC_BOOK3S_64
 879         /* default to book3s_64 (970fx) */
 880         vcpu->arch.pvr = 0x3C0301;
 881 #else
 882         /* default to book3s_32 (750) */
 883         vcpu->arch.pvr = 0x84202;
 884 #endif
 885         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 886         vcpu->arch.slb_nr = 64;
 887
 888         vcpu->arch.shadow_msr = MSR_USER64;
 889
 890         err = kvmppc_mmu_init(vcpu);
 891         if (err < 0)
 892                 goto uninit_vcpu;
 893
 894         return vcpu;
 895
 896 uninit_vcpu:
 897         kvm_vcpu_uninit(vcpu);
 898 free_shadow_vcpu:
 899         kfree(vcpu_book3s->shadow_vcpu);
 900 free_vcpu:
 901         vfree(vcpu_book3s);
 902 out:
 903         return ERR_PTR(err);
 904 }
 905
 906 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 907 {
 908         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
 909
 910         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
 911         kvm_vcpu_uninit(vcpu);
 912         kfree(vcpu_book3s->shadow_vcpu);
 913         vfree(vcpu_book3s);
 914 }
 915
 916 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 917 {
 918         int ret;
 919         double fpr[32][TS_FPRWIDTH];
 920         unsigned int fpscr;
 921         int fpexc_mode;
 922 #ifdef CONFIG_ALTIVEC
 923         vector128 vr[32];
 924         vector128 vscr;
 925         unsigned long uninitialized_var(vrsave);
 926         int used_vr;
 927 #endif
 928 #ifdef CONFIG_VSX
 929         int used_vsr;
 930 #endif
 931         ulong ext_msr;
 932
 933         /* Check if we can run the vcpu at all */
 934         if (!vcpu->arch.sane) {
 935                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 936                 return -EINVAL;
 937         }
 938
 939         /* No need to go into the guest when all we do is going out */
 940         if (signal_pending(current)) {
 941                 kvm_run->exit_reason = KVM_EXIT_INTR;
 942                 return -EINTR;
 943         }
 944
 945         /* Save FPU state in stack */
 946         if (current->thread.regs->msr & MSR_FP)
 947                 giveup_fpu(current);
 948         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
 949         fpscr = current->thread.fpscr.val;
 950         fpexc_mode = current->thread.fpexc_mode;
 951
 952 #ifdef CONFIG_ALTIVEC
 953         /* Save Altivec state in stack */
 954         used_vr = current->thread.used_vr;
 955         if (used_vr) {
 956                 if (current->thread.regs->msr & MSR_VEC)
 957                         giveup_altivec(current);
 958                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
 959                 vscr = current->thread.vscr;
 960                 vrsave = current->thread.vrsave;
 961         }
 962 #endif
 963
 964 #ifdef CONFIG_VSX
 965         /* Save VSX state in stack */
 966         used_vsr = current->thread.used_vsr;
 967         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
 968                         __giveup_vsx(current);
 969 #endif
 970
 971         /* Remember the MSR with disabled extensions */
 972         ext_msr = current->thread.regs->msr;
 973
 974         /* Preload FPU if it's enabled */
 975         if (vcpu->arch.shared->msr & MSR_FP)
 976                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 977
 978         kvm_guest_enter();
 979
 980         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
 981
 982         kvm_guest_exit();
 983
 984         local_irq_disable();
 985
 986         current->thread.regs->msr = ext_msr;
 987
 988         /* Make sure we save the guest FPU/Altivec/VSX state */
 989         kvmppc_giveup_ext(vcpu, MSR_FP);
 990         kvmppc_giveup_ext(vcpu, MSR_VEC);
 991         kvmppc_giveup_ext(vcpu, MSR_VSX);
 992
 993         /* Restore FPU state from stack */
 994         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
 995         current->thread.fpscr.val = fpscr;
 996         current->thread.fpexc_mode = fpexc_mode;
 997
 998 #ifdef CONFIG_ALTIVEC
 999         /* Restore Altivec state from stack */
1000         if (used_vr && current->thread.used_vr) {
1001                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1002                 current->thread.vscr = vscr;
1003                 current->thread.vrsave = vrsave;
1004         }
1005         current->thread.used_vr = used_vr;
1006 #endif
1007
1008 #ifdef CONFIG_VSX
1009         current->thread.used_vsr = used_vsr;
1010 #endif
1011
1012         return ret;
1013 }
1014
1015 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1016                                       struct kvm_userspace_memory_region *mem)
1017 {
1018         return 0;
1019 }
1020
1021 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1022                                 struct kvm_userspace_memory_region *mem)
1023 {
1024 }
1025
1026 int kvmppc_core_init_vm(struct kvm *kvm)
1027 {
1028         return 0;
1029 }
1030
1031 void kvmppc_core_destroy_vm(struct kvm *kvm)
1032 {
1033 }
1034
1035 static int kvmppc_book3s_init(void)
1036 {
1037         int r;
1038
1039         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1040                      THIS_MODULE);
1041
1042         if (r)
1043                 return r;
1044
1045         r = kvmppc_mmu_hpte_sysinit();
1046
1047         return r;
1048 }
1049
1050 static void kvmppc_book3s_exit(void)
1051 {
1052         kvmppc_mmu_hpte_sysexit();
1053         kvm_exit();
1054 }
1055
1056 module_init(kvmppc_book3s_init);
1057 module_exit(kvmppc_book3s_exit);