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                 if (!to_book3s(vcpu)->hior_sregs)
 155                         to_book3s(vcpu)->hior = 0xfff00000;
 156                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 157                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 158         } else
 159 #endif
 160         {
 161                 kvmppc_mmu_book3s_32_init(vcpu);
 162                 if (!to_book3s(vcpu)->hior_sregs)
 163                         to_book3s(vcpu)->hior = 0;
 164                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 165                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 166         }
 167
 168         kvmppc_sanity_check(vcpu);
 169
 170         /* If we are in hypervisor level on 970, we can tell the CPU to
 171          * treat DCBZ as 32 bytes store */
 172         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 173         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 174             !strcmp(cur_cpu_spec->platform, "ppc970"))
 175                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 176
 177         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 178            really needs them in a VM on Cell and force disable them. */
 179         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 180                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 181
 182 #ifdef CONFIG_PPC_BOOK3S_32
 183         /* 32 bit Book3S always has 32 byte dcbz */
 184         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 185 #endif
 186
 187         /* On some CPUs we can execute paired single operations natively */
 188         asm ( "mfpvr %0" : "=r"(host_pvr));
 189         switch (host_pvr) {
 190         case 0x00080200:        /* lonestar 2.0 */
 191         case 0x00088202:        /* lonestar 2.2 */
 192         case 0x70000100:        /* gekko 1.0 */
 193         case 0x00080100:        /* gekko 2.0 */
 194         case 0x00083203:        /* gekko 2.3a */
 195         case 0x00083213:        /* gekko 2.3b */
 196         case 0x00083204:        /* gekko 2.4 */
 197         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 198         case 0x00087200:        /* broadway */
 199                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 200                 /* Enable HID2.PSE - in case we need it later */
 201                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 202         }
 203 }
 204
 205 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 206  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 207  * emulate 32 bytes dcbz length.
 208  *
 209  * The Book3s_64 inventors also realized this case and implemented a special bit
 210  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 211  *
 212  * My approach here is to patch the dcbz instruction on executing pages.
 213  */
 214 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 215 {
 216         struct page *hpage;
 217         u64 hpage_offset;
 218         u32 *page;
 219         int i;
 220
 221         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 222         if (is_error_page(hpage)) {
 223                 kvm_release_page_clean(hpage);
 224                 return;
 225         }
 226
 227         hpage_offset = pte->raddr & ~PAGE_MASK;
 228         hpage_offset &= ~0xFFFULL;
 229         hpage_offset /= 4;
 230
 231         get_page(hpage);
 232         page = kmap_atomic(hpage, KM_USER0);
 233
 234         /* patch dcbz into reserved instruction, so we trap */
 235         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 236                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 237                         page[i] &= 0xfffffff7;
 238
 239         kunmap_atomic(page, KM_USER0);
 240         put_page(hpage);
 241 }
 242
 243 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 244 {
 245         ulong mp_pa = vcpu->arch.magic_page_pa;
 246
 247         if (unlikely(mp_pa) &&
 248             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 249                 return 1;
 250         }
 251
 252         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 253 }
 254
 255 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 256                             ulong eaddr, int vec)
 257 {
 258         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 259         int r = RESUME_GUEST;
 260         int relocated;
 261         int page_found = 0;
 262         struct kvmppc_pte pte;
 263         bool is_mmio = false;
 264         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 265         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 266         u64 vsid;
 267
 268         relocated = data ? dr : ir;
 269
 270         /* Resolve real address if translation turned on */
 271         if (relocated) {
 272                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 273         } else {
 274                 pte.may_execute = true;
 275                 pte.may_read = true;
 276                 pte.may_write = true;
 277                 pte.raddr = eaddr & KVM_PAM;
 278                 pte.eaddr = eaddr;
 279                 pte.vpage = eaddr >> 12;
 280         }
 281
 282         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 283         case 0:
 284                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 285                 break;
 286         case MSR_DR:
 287         case MSR_IR:
 288                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 289
 290                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 291                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 292                 else
 293                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 294                 pte.vpage |= vsid;
 295
 296                 if (vsid == -1)
 297                         page_found = -EINVAL;
 298                 break;
 299         }
 300
 301         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 302            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 303                 /*
 304                  * If we do the dcbz hack, we have to NX on every execution,
 305                  * so we can patch the executing code. This renders our guest
 306                  * NX-less.
 307                  */
 308                 pte.may_execute = !data;
 309         }
 310
 311         if (page_found == -ENOENT) {
 312                 /* Page not found in guest PTE entries */
 313                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 314                 vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 315                 vcpu->arch.shared->msr |=
 316                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 317                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 318         } else if (page_found == -EPERM) {
 319                 /* Storage protection */
 320                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 321                 vcpu->arch.shared->dsisr =
 322                         to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
 323                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 324                 vcpu->arch.shared->msr |=
 325                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 326                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 327         } else if (page_found == -EINVAL) {
 328                 /* Page not found in guest SLB */
 329                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 330                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 331         } else if (!is_mmio &&
 332                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 333                 /* The guest's PTE is not mapped yet. Map on the host */
 334                 kvmppc_mmu_map_page(vcpu, &pte);
 335                 if (data)
 336                         vcpu->stat.sp_storage++;
 337                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 338                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 339                         kvmppc_patch_dcbz(vcpu, &pte);
 340         } else {
 341                 /* MMIO */
 342                 vcpu->stat.mmio_exits++;
 343                 vcpu->arch.paddr_accessed = pte.raddr;
 344                 r = kvmppc_emulate_mmio(run, vcpu);
 345                 if ( r == RESUME_HOST_NV )
 346                         r = RESUME_HOST;
 347         }
 348
 349         return r;
 350 }
 351
 352 static inline int get_fpr_index(int i)
 353 {
 354 #ifdef CONFIG_VSX
 355         i *= 2;
 356 #endif
 357         return i;
 358 }
 359
 360 /* Give up external provider (FPU, Altivec, VSX) */
 361 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 362 {
 363         struct thread_struct *t = &current->thread;
 364         u64 *vcpu_fpr = vcpu->arch.fpr;
 365 #ifdef CONFIG_VSX
 366         u64 *vcpu_vsx = vcpu->arch.vsr;
 367 #endif
 368         u64 *thread_fpr = (u64*)t->fpr;
 369         int i;
 370
 371         if (!(vcpu->arch.guest_owned_ext & msr))
 372                 return;
 373
 374 #ifdef DEBUG_EXT
 375         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 376 #endif
 377
 378         switch (msr) {
 379         case MSR_FP:
 380                 giveup_fpu(current);
 381                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 382                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 383
 384                 vcpu->arch.fpscr = t->fpscr.val;
 385                 break;
 386         case MSR_VEC:
 387 #ifdef CONFIG_ALTIVEC
 388                 giveup_altivec(current);
 389                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 390                 vcpu->arch.vscr = t->vscr;
 391 #endif
 392                 break;
 393         case MSR_VSX:
 394 #ifdef CONFIG_VSX
 395                 __giveup_vsx(current);
 396                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 397                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 398 #endif
 399                 break;
 400         default:
 401                 BUG();
 402         }
 403
 404         vcpu->arch.guest_owned_ext &= ~msr;
 405         current->thread.regs->msr &= ~msr;
 406         kvmppc_recalc_shadow_msr(vcpu);
 407 }
 408
 409 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 410 {
 411         ulong srr0 = kvmppc_get_pc(vcpu);
 412         u32 last_inst = kvmppc_get_last_inst(vcpu);
 413         int ret;
 414
 415         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 416         if (ret == -ENOENT) {
 417                 ulong msr = vcpu->arch.shared->msr;
 418
 419                 msr = kvmppc_set_field(msr, 33, 33, 1);
 420                 msr = kvmppc_set_field(msr, 34, 36, 0);
 421                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 422                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 423                 return EMULATE_AGAIN;
 424         }
 425
 426         return EMULATE_DONE;
 427 }
 428
 429 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 430 {
 431
 432         /* Need to do paired single emulation? */
 433         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 434                 return EMULATE_DONE;
 435
 436         /* Read out the instruction */
 437         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 438                 /* Need to emulate */
 439                 return EMULATE_FAIL;
 440
 441         return EMULATE_AGAIN;
 442 }
 443
 444 /* Handle external providers (FPU, Altivec, VSX) */
 445 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 446                              ulong msr)
 447 {
 448         struct thread_struct *t = &current->thread;
 449         u64 *vcpu_fpr = vcpu->arch.fpr;
 450 #ifdef CONFIG_VSX
 451         u64 *vcpu_vsx = vcpu->arch.vsr;
 452 #endif
 453         u64 *thread_fpr = (u64*)t->fpr;
 454         int i;
 455
 456         /* When we have paired singles, we emulate in software */
 457         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 458                 return RESUME_GUEST;
 459
 460         if (!(vcpu->arch.shared->msr & msr)) {
 461                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 462                 return RESUME_GUEST;
 463         }
 464
 465         /* We already own the ext */
 466         if (vcpu->arch.guest_owned_ext & msr) {
 467                 return RESUME_GUEST;
 468         }
 469
 470 #ifdef DEBUG_EXT
 471         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 472 #endif
 473
 474         current->thread.regs->msr |= msr;
 475
 476         switch (msr) {
 477         case MSR_FP:
 478                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 479                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 480
 481                 t->fpscr.val = vcpu->arch.fpscr;
 482                 t->fpexc_mode = 0;
 483                 kvmppc_load_up_fpu();
 484                 break;
 485         case MSR_VEC:
 486 #ifdef CONFIG_ALTIVEC
 487                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 488                 t->vscr = vcpu->arch.vscr;
 489                 t->vrsave = -1;
 490                 kvmppc_load_up_altivec();
 491 #endif
 492                 break;
 493         case MSR_VSX:
 494 #ifdef CONFIG_VSX
 495                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 496                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 497                 kvmppc_load_up_vsx();
 498 #endif
 499                 break;
 500         default:
 501                 BUG();
 502         }
 503
 504         vcpu->arch.guest_owned_ext |= msr;
 505
 506         kvmppc_recalc_shadow_msr(vcpu);
 507
 508         return RESUME_GUEST;
 509 }
 510
 511 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 512                        unsigned int exit_nr)
 513 {
 514         int r = RESUME_HOST;
 515
 516         vcpu->stat.sum_exits++;
 517
 518         run->exit_reason = KVM_EXIT_UNKNOWN;
 519         run->ready_for_interrupt_injection = 1;
 520
 521         trace_kvm_book3s_exit(exit_nr, vcpu);
 522         kvm_resched(vcpu);
 523         switch (exit_nr) {
 524         case BOOK3S_INTERRUPT_INST_STORAGE:
 525                 vcpu->stat.pf_instruc++;
 526
 527 #ifdef CONFIG_PPC_BOOK3S_32
 528                 /* We set segments as unused segments when invalidating them. So
 529                  * treat the respective fault as segment fault. */
 530                 if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
 531                     == SR_INVALID) {
 532                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 533                         r = RESUME_GUEST;
 534                         break;
 535                 }
 536 #endif
 537
 538                 /* only care about PTEG not found errors, but leave NX alone */
 539                 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
 540                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 541                         vcpu->stat.sp_instruc++;
 542                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 543                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 544                         /*
 545                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 546                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 547                          *     that no guest that needs the dcbz hack does NX.
 548                          */
 549                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 550                         r = RESUME_GUEST;
 551                 } else {
 552                         vcpu->arch.shared->msr |=
 553                                 to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
 554                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 555                         r = RESUME_GUEST;
 556                 }
 557                 break;
 558         case BOOK3S_INTERRUPT_DATA_STORAGE:
 559         {
 560                 ulong dar = kvmppc_get_fault_dar(vcpu);
 561                 vcpu->stat.pf_storage++;
 562
 563 #ifdef CONFIG_PPC_BOOK3S_32
 564                 /* We set segments as unused segments when invalidating them. So
 565                  * treat the respective fault as segment fault. */
 566                 if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 567                         kvmppc_mmu_map_segment(vcpu, dar);
 568                         r = RESUME_GUEST;
 569                         break;
 570                 }
 571 #endif
 572
 573                 /* The only case we need to handle is missing shadow PTEs */
 574                 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
 575                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 576                 } else {
 577                         vcpu->arch.shared->dar = dar;
 578                         vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 579                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 580                         r = RESUME_GUEST;
 581                 }
 582                 break;
 583         }
 584         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 585                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 586                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 587                         kvmppc_book3s_queue_irqprio(vcpu,
 588                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 589                 }
 590                 r = RESUME_GUEST;
 591                 break;
 592         case BOOK3S_INTERRUPT_INST_SEGMENT:
 593                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 594                         kvmppc_book3s_queue_irqprio(vcpu,
 595                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 596                 }
 597                 r = RESUME_GUEST;
 598                 break;
 599         /* We're good on these - the host merely wanted to get our attention */
 600         case BOOK3S_INTERRUPT_DECREMENTER:
 601                 vcpu->stat.dec_exits++;
 602                 r = RESUME_GUEST;
 603                 break;
 604         case BOOK3S_INTERRUPT_EXTERNAL:
 605                 vcpu->stat.ext_intr_exits++;
 606                 r = RESUME_GUEST;
 607                 break;
 608         case BOOK3S_INTERRUPT_PERFMON:
 609                 r = RESUME_GUEST;
 610                 break;
 611         case BOOK3S_INTERRUPT_PROGRAM:
 612         {
 613                 enum emulation_result er;
 614                 ulong flags;
 615
 616 program_interrupt:
 617                 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
 618
 619                 if (vcpu->arch.shared->msr & MSR_PR) {
 620 #ifdef EXIT_DEBUG
 621                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 622 #endif
 623                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 624                             (INS_DCBZ & 0xfffffff7)) {
 625                                 kvmppc_core_queue_program(vcpu, flags);
 626                                 r = RESUME_GUEST;
 627                                 break;
 628                         }
 629                 }
 630
 631                 vcpu->stat.emulated_inst_exits++;
 632                 er = kvmppc_emulate_instruction(run, vcpu);
 633                 switch (er) {
 634                 case EMULATE_DONE:
 635                         r = RESUME_GUEST_NV;
 636                         break;
 637                 case EMULATE_AGAIN:
 638                         r = RESUME_GUEST;
 639                         break;
 640                 case EMULATE_FAIL:
 641                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 642                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 643                         kvmppc_core_queue_program(vcpu, flags);
 644                         r = RESUME_GUEST;
 645                         break;
 646                 case EMULATE_DO_MMIO:
 647                         run->exit_reason = KVM_EXIT_MMIO;
 648                         r = RESUME_HOST_NV;
 649                         break;
 650                 default:
 651                         BUG();
 652                 }
 653                 break;
 654         }
 655         case BOOK3S_INTERRUPT_SYSCALL:
 656                 if (vcpu->arch.papr_enabled &&
 657                     (kvmppc_get_last_inst(vcpu) == 0x44000022) &&
 658                     !(vcpu->arch.shared->msr & MSR_PR)) {
 659                         /* SC 1 papr hypercalls */
 660                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
 661                         int i;
 662
 663                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
 664                                 r = RESUME_GUEST;
 665                                 break;
 666                         }
 667
 668                         run->papr_hcall.nr = cmd;
 669                         for (i = 0; i < 9; ++i) {
 670                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
 671                                 run->papr_hcall.args[i] = gpr;
 672                         }
 673                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
 674                         vcpu->arch.hcall_needed = 1;
 675                         r = RESUME_HOST;
 676                 } else if (vcpu->arch.osi_enabled &&
 677                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 678                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 679                         /* MOL hypercalls */
 680                         u64 *gprs = run->osi.gprs;
 681                         int i;
 682
 683                         run->exit_reason = KVM_EXIT_OSI;
 684                         for (i = 0; i < 32; i++)
 685                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 686                         vcpu->arch.osi_needed = 1;
 687                         r = RESUME_HOST_NV;
 688                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 689                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 690                         /* KVM PV hypercalls */
 691                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 692                         r = RESUME_GUEST;
 693                 } else {
 694                         /* Guest syscalls */
 695                         vcpu->stat.syscall_exits++;
 696                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 697                         r = RESUME_GUEST;
 698                 }
 699                 break;
 700         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 701         case BOOK3S_INTERRUPT_ALTIVEC:
 702         case BOOK3S_INTERRUPT_VSX:
 703         {
 704                 int ext_msr = 0;
 705
 706                 switch (exit_nr) {
 707                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 708                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 709                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 710                 }
 711
 712                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
 713                 case EMULATE_DONE:
 714                         /* everything ok - let's enable the ext */
 715                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
 716                         break;
 717                 case EMULATE_FAIL:
 718                         /* we need to emulate this instruction */
 719                         goto program_interrupt;
 720                         break;
 721                 default:
 722                         /* nothing to worry about - go again */
 723                         break;
 724                 }
 725                 break;
 726         }
 727         case BOOK3S_INTERRUPT_ALIGNMENT:
 728                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
 729                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
 730                                 kvmppc_get_last_inst(vcpu));
 731                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
 732                                 kvmppc_get_last_inst(vcpu));
 733                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 734                 }
 735                 r = RESUME_GUEST;
 736                 break;
 737         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 738         case BOOK3S_INTERRUPT_TRACE:
 739                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 740                 r = RESUME_GUEST;
 741                 break;
 742         default:
 743                 /* Ugh - bork here! What did we get? */
 744                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
 745                         exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
 746                 r = RESUME_HOST;
 747                 BUG();
 748                 break;
 749         }
 750
 751
 752         if (!(r & RESUME_HOST)) {
 753                 /* To avoid clobbering exit_reason, only check for signals if
 754                  * we aren't already exiting to userspace for some other
 755                  * reason. */
 756                 if (signal_pending(current)) {
 757 #ifdef EXIT_DEBUG
 758                         printk(KERN_EMERG "KVM: Going back to host\n");
 759 #endif
 760                         vcpu->stat.signal_exits++;
 761                         run->exit_reason = KVM_EXIT_INTR;
 762                         r = -EINTR;
 763                 } else {
 764                         /* In case an interrupt came in that was triggered
 765                          * from userspace (like DEC), we need to check what
 766                          * to inject now! */
 767                         kvmppc_core_deliver_interrupts(vcpu);
 768                 }
 769         }
 770
 771         trace_kvm_book3s_reenter(r, vcpu);
 772
 773         return r;
 774 }
 775
 776 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 777                                   struct kvm_sregs *sregs)
 778 {
 779         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 780         int i;
 781
 782         sregs->pvr = vcpu->arch.pvr;
 783
 784         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 785         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 786                 for (i = 0; i < 64; i++) {
 787                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
 788                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 789                 }
 790         } else {
 791                 for (i = 0; i < 16; i++)
 792                         sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
 793
 794                 for (i = 0; i < 8; i++) {
 795                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 796                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 797                 }
 798         }
 799
 800         if (sregs->u.s.flags & KVM_SREGS_S_HIOR)
 801                 sregs->u.s.hior = to_book3s(vcpu)->hior;
 802
 803         return 0;
 804 }
 805
 806 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 807                                   struct kvm_sregs *sregs)
 808 {
 809         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 810         int i;
 811
 812         kvmppc_set_pvr(vcpu, sregs->pvr);
 813
 814         vcpu3s->sdr1 = sregs->u.s.sdr1;
 815         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 816                 for (i = 0; i < 64; i++) {
 817                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 818                                                     sregs->u.s.ppc64.slb[i].slbe);
 819                 }
 820         } else {
 821                 for (i = 0; i < 16; i++) {
 822                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 823                 }
 824                 for (i = 0; i < 8; i++) {
 825                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 826                                        (u32)sregs->u.s.ppc32.ibat[i]);
 827                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 828                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 829                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 830                                        (u32)sregs->u.s.ppc32.dbat[i]);
 831                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 832                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 833                 }
 834         }
 835
 836         /* Flush the MMU after messing with the segments */
 837         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 838
 839         if (sregs->u.s.flags & KVM_SREGS_S_HIOR) {
 840                 to_book3s(vcpu)->hior_sregs = true;
 841                 to_book3s(vcpu)->hior = sregs->u.s.hior;
 842         }
 843
 844         return 0;
 845 }
 846
 847 int kvmppc_core_check_processor_compat(void)
 848 {
 849         return 0;
 850 }
 851
 852 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 853 {
 854         struct kvmppc_vcpu_book3s *vcpu_book3s;
 855         struct kvm_vcpu *vcpu;
 856         int err = -ENOMEM;
 857         unsigned long p;
 858
 859         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
 860         if (!vcpu_book3s)
 861                 goto out;
 862
 863         vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
 864                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
 865         if (!vcpu_book3s->shadow_vcpu)
 866                 goto free_vcpu;
 867
 868         vcpu = &vcpu_book3s->vcpu;
 869         err = kvm_vcpu_init(vcpu, kvm, id);
 870         if (err)
 871                 goto free_shadow_vcpu;
 872
 873         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
 874         /* the real shared page fills the last 4k of our page */
 875         vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
 876         if (!p)
 877                 goto uninit_vcpu;
 878
 879 #ifdef CONFIG_PPC_BOOK3S_64
 880         /* default to book3s_64 (970fx) */
 881         vcpu->arch.pvr = 0x3C0301;
 882 #else
 883         /* default to book3s_32 (750) */
 884         vcpu->arch.pvr = 0x84202;
 885 #endif
 886         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 887         vcpu->arch.slb_nr = 64;
 888
 889         vcpu->arch.shadow_msr = MSR_USER64;
 890
 891         err = kvmppc_mmu_init(vcpu);
 892         if (err < 0)
 893                 goto uninit_vcpu;
 894
 895         return vcpu;
 896
 897 uninit_vcpu:
 898         kvm_vcpu_uninit(vcpu);
 899 free_shadow_vcpu:
 900         kfree(vcpu_book3s->shadow_vcpu);
 901 free_vcpu:
 902         vfree(vcpu_book3s);
 903 out:
 904         return ERR_PTR(err);
 905 }
 906
 907 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 908 {
 909         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
 910
 911         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
 912         kvm_vcpu_uninit(vcpu);
 913         kfree(vcpu_book3s->shadow_vcpu);
 914         vfree(vcpu_book3s);
 915 }
 916
 917 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 918 {
 919         int ret;
 920         double fpr[32][TS_FPRWIDTH];
 921         unsigned int fpscr;
 922         int fpexc_mode;
 923 #ifdef CONFIG_ALTIVEC
 924         vector128 vr[32];
 925         vector128 vscr;
 926         unsigned long uninitialized_var(vrsave);
 927         int used_vr;
 928 #endif
 929 #ifdef CONFIG_VSX
 930         int used_vsr;
 931 #endif
 932         ulong ext_msr;
 933
 934         /* Check if we can run the vcpu at all */
 935         if (!vcpu->arch.sane) {
 936                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 937                 return -EINVAL;
 938         }
 939
 940         /* No need to go into the guest when all we do is going out */
 941         if (signal_pending(current)) {
 942                 kvm_run->exit_reason = KVM_EXIT_INTR;
 943                 return -EINTR;
 944         }
 945
 946         /* Save FPU state in stack */
 947         if (current->thread.regs->msr & MSR_FP)
 948                 giveup_fpu(current);
 949         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
 950         fpscr = current->thread.fpscr.val;
 951         fpexc_mode = current->thread.fpexc_mode;
 952
 953 #ifdef CONFIG_ALTIVEC
 954         /* Save Altivec state in stack */
 955         used_vr = current->thread.used_vr;
 956         if (used_vr) {
 957                 if (current->thread.regs->msr & MSR_VEC)
 958                         giveup_altivec(current);
 959                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
 960                 vscr = current->thread.vscr;
 961                 vrsave = current->thread.vrsave;
 962         }
 963 #endif
 964
 965 #ifdef CONFIG_VSX
 966         /* Save VSX state in stack */
 967         used_vsr = current->thread.used_vsr;
 968         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
 969                         __giveup_vsx(current);
 970 #endif
 971
 972         /* Remember the MSR with disabled extensions */
 973         ext_msr = current->thread.regs->msr;
 974
 975         /* Preload FPU if it's enabled */
 976         if (vcpu->arch.shared->msr & MSR_FP)
 977                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 978
 979         kvm_guest_enter();
 980
 981         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
 982
 983         kvm_guest_exit();
 984
 985         local_irq_disable();
 986
 987         current->thread.regs->msr = ext_msr;
 988
 989         /* Make sure we save the guest FPU/Altivec/VSX state */
 990         kvmppc_giveup_ext(vcpu, MSR_FP);
 991         kvmppc_giveup_ext(vcpu, MSR_VEC);
 992         kvmppc_giveup_ext(vcpu, MSR_VSX);
 993
 994         /* Restore FPU state from stack */
 995         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
 996         current->thread.fpscr.val = fpscr;
 997         current->thread.fpexc_mode = fpexc_mode;
 998
 999 #ifdef CONFIG_ALTIVEC
1000         /* Restore Altivec state from stack */
1001         if (used_vr && current->thread.used_vr) {
1002                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1003                 current->thread.vscr = vscr;
1004                 current->thread.vrsave = vrsave;
1005         }
1006         current->thread.used_vr = used_vr;
1007 #endif
1008
1009 #ifdef CONFIG_VSX
1010         current->thread.used_vsr = used_vsr;
1011 #endif
1012
1013         return ret;
1014 }
1015
1016 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1017                                       struct kvm_userspace_memory_region *mem)
1018 {
1019         return 0;
1020 }
1021
1022 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1023                                 struct kvm_userspace_memory_region *mem)
1024 {
1025 }
1026
1027 int kvmppc_core_init_vm(struct kvm *kvm)
1028 {
1029         return 0;
1030 }
1031
1032 void kvmppc_core_destroy_vm(struct kvm *kvm)
1033 {
1034 }
1035
1036 static int kvmppc_book3s_init(void)
1037 {
1038         int r;
1039
1040         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1041                      THIS_MODULE);
1042
1043         if (r)
1044                 return r;
1045
1046         r = kvmppc_mmu_hpte_sysinit();
1047
1048         return r;
1049 }
1050
1051 static void kvmppc_book3s_exit(void)
1052 {
1053         kvmppc_mmu_hpte_sysexit();
1054         kvm_exit();
1055 }
1056
1057 module_init(kvmppc_book3s_init);
1058 module_exit(kvmppc_book3s_exit);