arch/powerpc/kvm/book3s.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  *
   8  * Description:
   9  * This file is derived from arch/powerpc/kvm/44x.c,
  10  * by Hollis Blanchard <hollisb@us.ibm.com>.
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License, version 2, as
  14  * published by the Free Software Foundation.
  15  */
  16
  17 #include <linux/kvm_host.h>
  18 #include <linux/err.h>
  19
  20 #include <asm/reg.h>
  21 #include <asm/cputable.h>
  22 #include <asm/cacheflush.h>
  23 #include <asm/tlbflush.h>
  24 #include <asm/uaccess.h>
  25 #include <asm/io.h>
  26 #include <asm/kvm_ppc.h>
  27 #include <asm/kvm_book3s.h>
  28 #include <asm/mmu_context.h>
  29 #include <linux/sched.h>
  30 #include <linux/vmalloc.h>
  31
  32 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  33
  34 /* #define EXIT_DEBUG */
  35 /* #define EXIT_DEBUG_SIMPLE */
  36 /* #define DEBUG_EXT */
  37
  38 static void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr);
  39
  40 struct kvm_stats_debugfs_item debugfs_entries[] = {
  41         { "exits",       VCPU_STAT(sum_exits) },
  42         { "mmio",        VCPU_STAT(mmio_exits) },
  43         { "sig",         VCPU_STAT(signal_exits) },
  44         { "sysc",        VCPU_STAT(syscall_exits) },
  45         { "inst_emu",    VCPU_STAT(emulated_inst_exits) },
  46         { "dec",         VCPU_STAT(dec_exits) },
  47         { "ext_intr",    VCPU_STAT(ext_intr_exits) },
  48         { "queue_intr",  VCPU_STAT(queue_intr) },
  49         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  50         { "pf_storage",  VCPU_STAT(pf_storage) },
  51         { "sp_storage",  VCPU_STAT(sp_storage) },
  52         { "pf_instruc",  VCPU_STAT(pf_instruc) },
  53         { "sp_instruc",  VCPU_STAT(sp_instruc) },
  54         { "ld",          VCPU_STAT(ld) },
  55         { "ld_slow",     VCPU_STAT(ld_slow) },
  56         { "st",          VCPU_STAT(st) },
  57         { "st_slow",     VCPU_STAT(st_slow) },
  58         { NULL }
  59 };
  60
  61 void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
  62 {
  63 }
  64
  65 void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
  66 {
  67 }
  68
  69 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  70 {
  71         memcpy(get_paca()->kvm_slb, to_book3s(vcpu)->slb_shadow, sizeof(get_paca()->kvm_slb));
  72         memcpy(&get_paca()->shadow_vcpu, &to_book3s(vcpu)->shadow_vcpu,
  73                sizeof(get_paca()->shadow_vcpu));
  74         get_paca()->kvm_slb_max = to_book3s(vcpu)->slb_shadow_max;
  75 }
  76
  77 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  78 {
  79         memcpy(to_book3s(vcpu)->slb_shadow, get_paca()->kvm_slb, sizeof(get_paca()->kvm_slb));
  80         memcpy(&to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  81                sizeof(get_paca()->shadow_vcpu));
  82         to_book3s(vcpu)->slb_shadow_max = get_paca()->kvm_slb_max;
  83
  84         kvmppc_giveup_ext(vcpu, MSR_FP);
  85         kvmppc_giveup_ext(vcpu, MSR_VEC);
  86         kvmppc_giveup_ext(vcpu, MSR_VSX);
  87 }
  88
  89 #if defined(EXIT_DEBUG)
  90 static u32 kvmppc_get_dec(struct kvm_vcpu *vcpu)
  91 {
  92         u64 jd = mftb() - vcpu->arch.dec_jiffies;
  93         return vcpu->arch.dec - jd;
  94 }
  95 #endif
  96
  97 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
  98 {
  99         ulong old_msr = vcpu->arch.msr;
 100
 101 #ifdef EXIT_DEBUG
 102         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 103 #endif
 104         msr &= to_book3s(vcpu)->msr_mask;
 105         vcpu->arch.msr = msr;
 106         vcpu->arch.shadow_msr = msr | MSR_USER32;
 107         vcpu->arch.shadow_msr &= (MSR_FE0 | MSR_USER64 | MSR_SE | MSR_BE |
 108                                   MSR_DE | MSR_FE1);
 109         vcpu->arch.shadow_msr |= (msr & vcpu->arch.guest_owned_ext);
 110
 111         if (msr & (MSR_WE|MSR_POW)) {
 112                 if (!vcpu->arch.pending_exceptions) {
 113                         kvm_vcpu_block(vcpu);
 114                         vcpu->stat.halt_wakeup++;
 115                 }
 116         }
 117
 118         if (((vcpu->arch.msr & (MSR_IR|MSR_DR)) != (old_msr & (MSR_IR|MSR_DR))) ||
 119             (vcpu->arch.msr & MSR_PR) != (old_msr & MSR_PR)) {
 120                 kvmppc_mmu_flush_segments(vcpu);
 121                 kvmppc_mmu_map_segment(vcpu, vcpu->arch.pc);
 122         }
 123 }
 124
 125 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
 126 {
 127         vcpu->arch.srr0 = vcpu->arch.pc;
 128         vcpu->arch.srr1 = vcpu->arch.msr | flags;
 129         vcpu->arch.pc = to_book3s(vcpu)->hior + vec;
 130         vcpu->arch.mmu.reset_msr(vcpu);
 131 }
 132
 133 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
 134 {
 135         unsigned int prio;
 136
 137         switch (vec) {
 138         case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET;         break;
 139         case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK;        break;
 140         case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE;         break;
 141         case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT;         break;
 142         case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE;         break;
 143         case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT;         break;
 144         case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL;             break;
 145         case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT;            break;
 146         case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM;              break;
 147         case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL;           break;
 148         case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER;          break;
 149         case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL;              break;
 150         case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG;                break;
 151         case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC;              break;
 152         case 0xf40: prio = BOOK3S_IRQPRIO_VSX;                  break;
 153         default:    prio = BOOK3S_IRQPRIO_MAX;                  break;
 154         }
 155
 156         return prio;
 157 }
 158
 159 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
 160                                           unsigned int vec)
 161 {
 162         clear_bit(kvmppc_book3s_vec2irqprio(vec),
 163                   &vcpu->arch.pending_exceptions);
 164 }
 165
 166 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
 167 {
 168         vcpu->stat.queue_intr++;
 169
 170         set_bit(kvmppc_book3s_vec2irqprio(vec),
 171                 &vcpu->arch.pending_exceptions);
 172 #ifdef EXIT_DEBUG
 173         printk(KERN_INFO "Queueing interrupt %x\n", vec);
 174 #endif
 175 }
 176
 177
 178 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong flags)
 179 {
 180         to_book3s(vcpu)->prog_flags = flags;
 181         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
 182 }
 183
 184 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 185 {
 186         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 187 }
 188
 189 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 190 {
 191         return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
 192 }
 193
 194 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 195 {
 196         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 197 }
 198
 199 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 200                                 struct kvm_interrupt *irq)
 201 {
 202         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 203 }
 204
 205 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
 206 {
 207         int deliver = 1;
 208         int vec = 0;
 209         ulong flags = 0ULL;
 210
 211         switch (priority) {
 212         case BOOK3S_IRQPRIO_DECREMENTER:
 213                 deliver = vcpu->arch.msr & MSR_EE;
 214                 vec = BOOK3S_INTERRUPT_DECREMENTER;
 215                 break;
 216         case BOOK3S_IRQPRIO_EXTERNAL:
 217                 deliver = vcpu->arch.msr & MSR_EE;
 218                 vec = BOOK3S_INTERRUPT_EXTERNAL;
 219                 break;
 220         case BOOK3S_IRQPRIO_SYSTEM_RESET:
 221                 vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
 222                 break;
 223         case BOOK3S_IRQPRIO_MACHINE_CHECK:
 224                 vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
 225                 break;
 226         case BOOK3S_IRQPRIO_DATA_STORAGE:
 227                 vec = BOOK3S_INTERRUPT_DATA_STORAGE;
 228                 break;
 229         case BOOK3S_IRQPRIO_INST_STORAGE:
 230                 vec = BOOK3S_INTERRUPT_INST_STORAGE;
 231                 break;
 232         case BOOK3S_IRQPRIO_DATA_SEGMENT:
 233                 vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
 234                 break;
 235         case BOOK3S_IRQPRIO_INST_SEGMENT:
 236                 vec = BOOK3S_INTERRUPT_INST_SEGMENT;
 237                 break;
 238         case BOOK3S_IRQPRIO_ALIGNMENT:
 239                 vec = BOOK3S_INTERRUPT_ALIGNMENT;
 240                 break;
 241         case BOOK3S_IRQPRIO_PROGRAM:
 242                 vec = BOOK3S_INTERRUPT_PROGRAM;
 243                 flags = to_book3s(vcpu)->prog_flags;
 244                 break;
 245         case BOOK3S_IRQPRIO_VSX:
 246                 vec = BOOK3S_INTERRUPT_VSX;
 247                 break;
 248         case BOOK3S_IRQPRIO_ALTIVEC:
 249                 vec = BOOK3S_INTERRUPT_ALTIVEC;
 250                 break;
 251         case BOOK3S_IRQPRIO_FP_UNAVAIL:
 252                 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
 253                 break;
 254         case BOOK3S_IRQPRIO_SYSCALL:
 255                 vec = BOOK3S_INTERRUPT_SYSCALL;
 256                 break;
 257         case BOOK3S_IRQPRIO_DEBUG:
 258                 vec = BOOK3S_INTERRUPT_TRACE;
 259                 break;
 260         case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
 261                 vec = BOOK3S_INTERRUPT_PERFMON;
 262                 break;
 263         default:
 264                 deliver = 0;
 265                 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
 266                 break;
 267         }
 268
 269 #if 0
 270         printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
 271 #endif
 272
 273         if (deliver)
 274                 kvmppc_inject_interrupt(vcpu, vec, flags);
 275
 276         return deliver;
 277 }
 278
 279 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 280 {
 281         unsigned long *pending = &vcpu->arch.pending_exceptions;
 282         unsigned int priority;
 283
 284 #ifdef EXIT_DEBUG
 285         if (vcpu->arch.pending_exceptions)
 286                 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
 287 #endif
 288         priority = __ffs(*pending);
 289         while (priority <= (sizeof(unsigned int) * 8)) {
 290                 if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
 291                     (priority != BOOK3S_IRQPRIO_DECREMENTER)) {
 292                         /* DEC interrupts get cleared by mtdec */
 293                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 294                         break;
 295                 }
 296
 297                 priority = find_next_bit(pending,
 298                                          BITS_PER_BYTE * sizeof(*pending),
 299                                          priority + 1);
 300         }
 301 }
 302
 303 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 304 {
 305         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 306         vcpu->arch.pvr = pvr;
 307         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 308                 kvmppc_mmu_book3s_64_init(vcpu);
 309                 to_book3s(vcpu)->hior = 0xfff00000;
 310                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 311         } else {
 312                 kvmppc_mmu_book3s_32_init(vcpu);
 313                 to_book3s(vcpu)->hior = 0;
 314                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 315         }
 316
 317         /* If we are in hypervisor level on 970, we can tell the CPU to
 318          * treat DCBZ as 32 bytes store */
 319         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 320         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 321             !strcmp(cur_cpu_spec->platform, "ppc970"))
 322                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 323
 324 }
 325
 326 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 327  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 328  * emulate 32 bytes dcbz length.
 329  *
 330  * The Book3s_64 inventors also realized this case and implemented a special bit
 331  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 332  *
 333  * My approach here is to patch the dcbz instruction on executing pages.
 334  */
 335 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 336 {
 337         bool touched = false;
 338         hva_t hpage;
 339         u32 *page;
 340         int i;
 341
 342         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 343         if (kvm_is_error_hva(hpage))
 344                 return;
 345
 346         hpage |= pte->raddr & ~PAGE_MASK;
 347         hpage &= ~0xFFFULL;
 348
 349         page = vmalloc(HW_PAGE_SIZE);
 350
 351         if (copy_from_user(page, (void __user *)hpage, HW_PAGE_SIZE))
 352                 goto out;
 353
 354         for (i=0; i < HW_PAGE_SIZE / 4; i++)
 355                 if ((page[i] & 0xff0007ff) == INS_DCBZ) {
 356                         page[i] &= 0xfffffff7; // reserved instruction, so we trap
 357                         touched = true;
 358                 }
 359
 360         if (touched)
 361                 copy_to_user((void __user *)hpage, page, HW_PAGE_SIZE);
 362
 363 out:
 364         vfree(page);
 365 }
 366
 367 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
 368                          struct kvmppc_pte *pte)
 369 {
 370         int relocated = (vcpu->arch.msr & (data ? MSR_DR : MSR_IR));
 371         int r;
 372
 373         if (relocated) {
 374                 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
 375         } else {
 376                 pte->eaddr = eaddr;
 377                 pte->raddr = eaddr & 0xffffffff;
 378                 pte->vpage = eaddr >> 12;
 379                 switch (vcpu->arch.msr & (MSR_DR|MSR_IR)) {
 380                 case 0:
 381                         pte->vpage |= VSID_REAL;
 382                 case MSR_DR:
 383                         pte->vpage |= VSID_REAL_DR;
 384                 case MSR_IR:
 385                         pte->vpage |= VSID_REAL_IR;
 386                 }
 387                 pte->may_read = true;
 388                 pte->may_write = true;
 389                 pte->may_execute = true;
 390                 r = 0;
 391         }
 392
 393         return r;
 394 }
 395
 396 static hva_t kvmppc_bad_hva(void)
 397 {
 398         return PAGE_OFFSET;
 399 }
 400
 401 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
 402                                bool read)
 403 {
 404         hva_t hpage;
 405
 406         if (read && !pte->may_read)
 407                 goto err;
 408
 409         if (!read && !pte->may_write)
 410                 goto err;
 411
 412         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 413         if (kvm_is_error_hva(hpage))
 414                 goto err;
 415
 416         return hpage | (pte->raddr & ~PAGE_MASK);
 417 err:
 418         return kvmppc_bad_hva();
 419 }
 420
 421 int kvmppc_st(struct kvm_vcpu *vcpu, ulong eaddr, int size, void *ptr)
 422 {
 423         struct kvmppc_pte pte;
 424         hva_t hva = eaddr;
 425
 426         vcpu->stat.st++;
 427
 428         if (kvmppc_xlate(vcpu, eaddr, false, &pte))
 429                 goto err;
 430
 431         hva = kvmppc_pte_to_hva(vcpu, &pte, false);
 432         if (kvm_is_error_hva(hva))
 433                 goto err;
 434
 435         if (copy_to_user((void __user *)hva, ptr, size)) {
 436                 printk(KERN_INFO "kvmppc_st at 0x%lx failed\n", hva);
 437                 goto err;
 438         }
 439
 440         return 0;
 441
 442 err:
 443         return -ENOENT;
 444 }
 445
 446 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong eaddr, int size, void *ptr,
 447                       bool data)
 448 {
 449         struct kvmppc_pte pte;
 450         hva_t hva = eaddr;
 451
 452         vcpu->stat.ld++;
 453
 454         if (kvmppc_xlate(vcpu, eaddr, data, &pte))
 455                 goto err;
 456
 457         hva = kvmppc_pte_to_hva(vcpu, &pte, true);
 458         if (kvm_is_error_hva(hva))
 459                 goto err;
 460
 461         if (copy_from_user(ptr, (void __user *)hva, size)) {
 462                 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
 463                 goto err;
 464         }
 465
 466         return 0;
 467
 468 err:
 469         return -ENOENT;
 470 }
 471
 472 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 473 {
 474         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 475 }
 476
 477 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 478                             ulong eaddr, int vec)
 479 {
 480         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 481         int r = RESUME_GUEST;
 482         int relocated;
 483         int page_found = 0;
 484         struct kvmppc_pte pte;
 485         bool is_mmio = false;
 486
 487         if ( vec == BOOK3S_INTERRUPT_DATA_STORAGE ) {
 488                 relocated = (vcpu->arch.msr & MSR_DR);
 489         } else {
 490                 relocated = (vcpu->arch.msr & MSR_IR);
 491         }
 492
 493         /* Resolve real address if translation turned on */
 494         if (relocated) {
 495                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 496         } else {
 497                 pte.may_execute = true;
 498                 pte.may_read = true;
 499                 pte.may_write = true;
 500                 pte.raddr = eaddr & 0xffffffff;
 501                 pte.eaddr = eaddr;
 502                 pte.vpage = eaddr >> 12;
 503                 switch (vcpu->arch.msr & (MSR_DR|MSR_IR)) {
 504                 case 0:
 505                         pte.vpage |= VSID_REAL;
 506                 case MSR_DR:
 507                         pte.vpage |= VSID_REAL_DR;
 508                 case MSR_IR:
 509                         pte.vpage |= VSID_REAL_IR;
 510                 }
 511         }
 512
 513         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 514            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 515                 /*
 516                  * If we do the dcbz hack, we have to NX on every execution,
 517                  * so we can patch the executing code. This renders our guest
 518                  * NX-less.
 519                  */
 520                 pte.may_execute = !data;
 521         }
 522
 523         if (page_found == -ENOENT) {
 524                 /* Page not found in guest PTE entries */
 525                 vcpu->arch.dear = vcpu->arch.fault_dear;
 526                 to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr;
 527                 vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x00000000f8000000ULL);
 528                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 529         } else if (page_found == -EPERM) {
 530                 /* Storage protection */
 531                 vcpu->arch.dear = vcpu->arch.fault_dear;
 532                 to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
 533                 to_book3s(vcpu)->dsisr |= DSISR_PROTFAULT;
 534                 vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x00000000f8000000ULL);
 535                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 536         } else if (page_found == -EINVAL) {
 537                 /* Page not found in guest SLB */
 538                 vcpu->arch.dear = vcpu->arch.fault_dear;
 539                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 540         } else if (!is_mmio &&
 541                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 542                 /* The guest's PTE is not mapped yet. Map on the host */
 543                 kvmppc_mmu_map_page(vcpu, &pte);
 544                 if (data)
 545                         vcpu->stat.sp_storage++;
 546                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 547                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 548                         kvmppc_patch_dcbz(vcpu, &pte);
 549         } else {
 550                 /* MMIO */
 551                 vcpu->stat.mmio_exits++;
 552                 vcpu->arch.paddr_accessed = pte.raddr;
 553                 r = kvmppc_emulate_mmio(run, vcpu);
 554                 if ( r == RESUME_HOST_NV )
 555                         r = RESUME_HOST;
 556         }
 557
 558         return r;
 559 }
 560
 561 static inline int get_fpr_index(int i)
 562 {
 563 #ifdef CONFIG_VSX
 564         i *= 2;
 565 #endif
 566         return i;
 567 }
 568
 569 /* Give up external provider (FPU, Altivec, VSX) */
 570 static void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 571 {
 572         struct thread_struct *t = &current->thread;
 573         u64 *vcpu_fpr = vcpu->arch.fpr;
 574         u64 *vcpu_vsx = vcpu->arch.vsr;
 575         u64 *thread_fpr = (u64*)t->fpr;
 576         int i;
 577
 578         if (!(vcpu->arch.guest_owned_ext & msr))
 579                 return;
 580
 581 #ifdef DEBUG_EXT
 582         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 583 #endif
 584
 585         switch (msr) {
 586         case MSR_FP:
 587                 giveup_fpu(current);
 588                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 589                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 590
 591                 vcpu->arch.fpscr = t->fpscr.val;
 592                 break;
 593         case MSR_VEC:
 594 #ifdef CONFIG_ALTIVEC
 595                 giveup_altivec(current);
 596                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 597                 vcpu->arch.vscr = t->vscr;
 598 #endif
 599                 break;
 600         case MSR_VSX:
 601 #ifdef CONFIG_VSX
 602                 __giveup_vsx(current);
 603                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 604                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 605 #endif
 606                 break;
 607         default:
 608                 BUG();
 609         }
 610
 611         vcpu->arch.guest_owned_ext &= ~msr;
 612         current->thread.regs->msr &= ~msr;
 613         kvmppc_set_msr(vcpu, vcpu->arch.msr);
 614 }
 615
 616 /* Handle external providers (FPU, Altivec, VSX) */
 617 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 618                              ulong msr)
 619 {
 620         struct thread_struct *t = &current->thread;
 621         u64 *vcpu_fpr = vcpu->arch.fpr;
 622         u64 *vcpu_vsx = vcpu->arch.vsr;
 623         u64 *thread_fpr = (u64*)t->fpr;
 624         int i;
 625
 626         if (!(vcpu->arch.msr & msr)) {
 627                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 628                 return RESUME_GUEST;
 629         }
 630
 631 #ifdef DEBUG_EXT
 632         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 633 #endif
 634
 635         current->thread.regs->msr |= msr;
 636
 637         switch (msr) {
 638         case MSR_FP:
 639                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 640                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 641
 642                 t->fpscr.val = vcpu->arch.fpscr;
 643                 t->fpexc_mode = 0;
 644                 kvmppc_load_up_fpu();
 645                 break;
 646         case MSR_VEC:
 647 #ifdef CONFIG_ALTIVEC
 648                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 649                 t->vscr = vcpu->arch.vscr;
 650                 t->vrsave = -1;
 651                 kvmppc_load_up_altivec();
 652 #endif
 653                 break;
 654         case MSR_VSX:
 655 #ifdef CONFIG_VSX
 656                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 657                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 658                 kvmppc_load_up_vsx();
 659 #endif
 660                 break;
 661         default:
 662                 BUG();
 663         }
 664
 665         vcpu->arch.guest_owned_ext |= msr;
 666
 667         kvmppc_set_msr(vcpu, vcpu->arch.msr);
 668
 669         return RESUME_GUEST;
 670 }
 671
 672 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 673                        unsigned int exit_nr)
 674 {
 675         int r = RESUME_HOST;
 676
 677         vcpu->stat.sum_exits++;
 678
 679         run->exit_reason = KVM_EXIT_UNKNOWN;
 680         run->ready_for_interrupt_injection = 1;
 681 #ifdef EXIT_DEBUG
 682         printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | dec=0x%x | msr=0x%lx\n",
 683                 exit_nr, vcpu->arch.pc, vcpu->arch.fault_dear,
 684                 kvmppc_get_dec(vcpu), vcpu->arch.msr);
 685 #elif defined (EXIT_DEBUG_SIMPLE)
 686         if ((exit_nr != 0x900) && (exit_nr != 0x500))
 687                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | msr=0x%lx\n",
 688                         exit_nr, vcpu->arch.pc, vcpu->arch.fault_dear,
 689                         vcpu->arch.msr);
 690 #endif
 691         kvm_resched(vcpu);
 692         switch (exit_nr) {
 693         case BOOK3S_INTERRUPT_INST_STORAGE:
 694                 vcpu->stat.pf_instruc++;
 695                 /* only care about PTEG not found errors, but leave NX alone */
 696                 if (vcpu->arch.shadow_msr & 0x40000000) {
 697                         r = kvmppc_handle_pagefault(run, vcpu, vcpu->arch.pc, exit_nr);
 698                         vcpu->stat.sp_instruc++;
 699                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 700                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 701                         /*
 702                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 703                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 704                          *     that no guest that needs the dcbz hack does NX.
 705                          */
 706                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.pc, ~0xFFFULL);
 707                 } else {
 708                         vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x58000000);
 709                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 710                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.pc, ~0xFFFULL);
 711                         r = RESUME_GUEST;
 712                 }
 713                 break;
 714         case BOOK3S_INTERRUPT_DATA_STORAGE:
 715                 vcpu->stat.pf_storage++;
 716                 /* The only case we need to handle is missing shadow PTEs */
 717                 if (vcpu->arch.fault_dsisr & DSISR_NOHPTE) {
 718                         r = kvmppc_handle_pagefault(run, vcpu, vcpu->arch.fault_dear, exit_nr);
 719                 } else {
 720                         vcpu->arch.dear = vcpu->arch.fault_dear;
 721                         to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr;
 722                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 723                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.dear, ~0xFFFULL);
 724                         r = RESUME_GUEST;
 725                 }
 726                 break;
 727         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 728                 if (kvmppc_mmu_map_segment(vcpu, vcpu->arch.fault_dear) < 0) {
 729                         vcpu->arch.dear = vcpu->arch.fault_dear;
 730                         kvmppc_book3s_queue_irqprio(vcpu,
 731                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 732                 }
 733                 r = RESUME_GUEST;
 734                 break;
 735         case BOOK3S_INTERRUPT_INST_SEGMENT:
 736                 if (kvmppc_mmu_map_segment(vcpu, vcpu->arch.pc) < 0) {
 737                         kvmppc_book3s_queue_irqprio(vcpu,
 738                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 739                 }
 740                 r = RESUME_GUEST;
 741                 break;
 742         /* We're good on these - the host merely wanted to get our attention */
 743         case BOOK3S_INTERRUPT_DECREMENTER:
 744                 vcpu->stat.dec_exits++;
 745                 r = RESUME_GUEST;
 746                 break;
 747         case BOOK3S_INTERRUPT_EXTERNAL:
 748                 vcpu->stat.ext_intr_exits++;
 749                 r = RESUME_GUEST;
 750                 break;
 751         case BOOK3S_INTERRUPT_PROGRAM:
 752         {
 753                 enum emulation_result er;
 754                 ulong flags;
 755
 756                 flags = (vcpu->arch.shadow_msr & 0x1f0000ull);
 757
 758                 if (vcpu->arch.msr & MSR_PR) {
 759 #ifdef EXIT_DEBUG
 760                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", vcpu->arch.pc, vcpu->arch.last_inst);
 761 #endif
 762                         if ((vcpu->arch.last_inst & 0xff0007ff) !=
 763                             (INS_DCBZ & 0xfffffff7)) {
 764                                 kvmppc_core_queue_program(vcpu, flags);
 765                                 r = RESUME_GUEST;
 766                                 break;
 767                         }
 768                 }
 769
 770                 vcpu->stat.emulated_inst_exits++;
 771                 er = kvmppc_emulate_instruction(run, vcpu);
 772                 switch (er) {
 773                 case EMULATE_DONE:
 774                         r = RESUME_GUEST_NV;
 775                         break;
 776                 case EMULATE_FAIL:
 777                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 778                                __func__, vcpu->arch.pc, vcpu->arch.last_inst);
 779                         kvmppc_core_queue_program(vcpu, flags);
 780                         r = RESUME_GUEST;
 781                         break;
 782                 default:
 783                         BUG();
 784                 }
 785                 break;
 786         }
 787         case BOOK3S_INTERRUPT_SYSCALL:
 788 #ifdef EXIT_DEBUG
 789                 printk(KERN_INFO "Syscall Nr %d\n", (int)kvmppc_get_gpr(vcpu, 0));
 790 #endif
 791                 vcpu->stat.syscall_exits++;
 792                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 793                 r = RESUME_GUEST;
 794                 break;
 795         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 796                 r = kvmppc_handle_ext(vcpu, exit_nr, MSR_FP);
 797                 break;
 798         case BOOK3S_INTERRUPT_ALTIVEC:
 799                 r = kvmppc_handle_ext(vcpu, exit_nr, MSR_VEC);
 800                 break;
 801         case BOOK3S_INTERRUPT_VSX:
 802                 r = kvmppc_handle_ext(vcpu, exit_nr, MSR_VSX);
 803                 break;
 804         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 805         case BOOK3S_INTERRUPT_TRACE:
 806                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 807                 r = RESUME_GUEST;
 808                 break;
 809         default:
 810                 /* Ugh - bork here! What did we get? */
 811                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n", exit_nr, vcpu->arch.pc, vcpu->arch.shadow_msr);
 812                 r = RESUME_HOST;
 813                 BUG();
 814                 break;
 815         }
 816
 817
 818         if (!(r & RESUME_HOST)) {
 819                 /* To avoid clobbering exit_reason, only check for signals if
 820                  * we aren't already exiting to userspace for some other
 821                  * reason. */
 822                 if (signal_pending(current)) {
 823 #ifdef EXIT_DEBUG
 824                         printk(KERN_EMERG "KVM: Going back to host\n");
 825 #endif
 826                         vcpu->stat.signal_exits++;
 827                         run->exit_reason = KVM_EXIT_INTR;
 828                         r = -EINTR;
 829                 } else {
 830                         /* In case an interrupt came in that was triggered
 831                          * from userspace (like DEC), we need to check what
 832                          * to inject now! */
 833                         kvmppc_core_deliver_interrupts(vcpu);
 834                 }
 835         }
 836
 837 #ifdef EXIT_DEBUG
 838         printk(KERN_EMERG "KVM exit: vcpu=0x%p pc=0x%lx r=0x%x\n", vcpu, vcpu->arch.pc, r);
 839 #endif
 840
 841         return r;
 842 }
 843
 844 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 845 {
 846         return 0;
 847 }
 848
 849 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 850 {
 851         int i;
 852
 853         regs->pc = vcpu->arch.pc;
 854         regs->cr = kvmppc_get_cr(vcpu);
 855         regs->ctr = vcpu->arch.ctr;
 856         regs->lr = vcpu->arch.lr;
 857         regs->xer = kvmppc_get_xer(vcpu);
 858         regs->msr = vcpu->arch.msr;
 859         regs->srr0 = vcpu->arch.srr0;
 860         regs->srr1 = vcpu->arch.srr1;
 861         regs->pid = vcpu->arch.pid;
 862         regs->sprg0 = vcpu->arch.sprg0;
 863         regs->sprg1 = vcpu->arch.sprg1;
 864         regs->sprg2 = vcpu->arch.sprg2;
 865         regs->sprg3 = vcpu->arch.sprg3;
 866         regs->sprg5 = vcpu->arch.sprg4;
 867         regs->sprg6 = vcpu->arch.sprg5;
 868         regs->sprg7 = vcpu->arch.sprg6;
 869
 870         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 871                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
 872
 873         return 0;
 874 }
 875
 876 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 877 {
 878         int i;
 879
 880         vcpu->arch.pc = regs->pc;
 881         kvmppc_set_cr(vcpu, regs->cr);
 882         vcpu->arch.ctr = regs->ctr;
 883         vcpu->arch.lr = regs->lr;
 884         kvmppc_set_xer(vcpu, regs->xer);
 885         kvmppc_set_msr(vcpu, regs->msr);
 886         vcpu->arch.srr0 = regs->srr0;
 887         vcpu->arch.srr1 = regs->srr1;
 888         vcpu->arch.sprg0 = regs->sprg0;
 889         vcpu->arch.sprg1 = regs->sprg1;
 890         vcpu->arch.sprg2 = regs->sprg2;
 891         vcpu->arch.sprg3 = regs->sprg3;
 892         vcpu->arch.sprg5 = regs->sprg4;
 893         vcpu->arch.sprg6 = regs->sprg5;
 894         vcpu->arch.sprg7 = regs->sprg6;
 895
 896         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 897                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
 898
 899         return 0;
 900 }
 901
 902 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 903                                   struct kvm_sregs *sregs)
 904 {
 905         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 906         int i;
 907
 908         sregs->pvr = vcpu->arch.pvr;
 909
 910         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 911         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 912                 for (i = 0; i < 64; i++) {
 913                         sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
 914                         sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
 915                 }
 916         } else {
 917                 for (i = 0; i < 16; i++) {
 918                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
 919                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
 920                 }
 921                 for (i = 0; i < 8; i++) {
 922                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 923                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 924                 }
 925         }
 926         return 0;
 927 }
 928
 929 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 930                                   struct kvm_sregs *sregs)
 931 {
 932         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 933         int i;
 934
 935         kvmppc_set_pvr(vcpu, sregs->pvr);
 936
 937         vcpu3s->sdr1 = sregs->u.s.sdr1;
 938         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 939                 for (i = 0; i < 64; i++) {
 940                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 941                                                     sregs->u.s.ppc64.slb[i].slbe);
 942                 }
 943         } else {
 944                 for (i = 0; i < 16; i++) {
 945                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 946                 }
 947                 for (i = 0; i < 8; i++) {
 948                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 949                                        (u32)sregs->u.s.ppc32.ibat[i]);
 950                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 951                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 952                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 953                                        (u32)sregs->u.s.ppc32.dbat[i]);
 954                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 955                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 956                 }
 957         }
 958
 959         /* Flush the MMU after messing with the segments */
 960         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 961         return 0;
 962 }
 963
 964 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 965 {
 966         return -ENOTSUPP;
 967 }
 968
 969 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 970 {
 971         return -ENOTSUPP;
 972 }
 973
 974 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 975                                   struct kvm_translation *tr)
 976 {
 977         return 0;
 978 }
 979
 980 /*
 981  * Get (and clear) the dirty memory log for a memory slot.
 982  */
 983 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 984                                       struct kvm_dirty_log *log)
 985 {
 986         struct kvm_memory_slot *memslot;
 987         struct kvm_vcpu *vcpu;
 988         ulong ga, ga_end;
 989         int is_dirty = 0;
 990         int r, n;
 991
 992         mutex_lock(&kvm->slots_lock);
 993
 994         r = kvm_get_dirty_log(kvm, log, &is_dirty);
 995         if (r)
 996                 goto out;
 997
 998         /* If nothing is dirty, don't bother messing with page tables. */
 999         if (is_dirty) {
1000                 memslot = &kvm->memslots->memslots[log->slot];
1001
1002                 ga = memslot->base_gfn << PAGE_SHIFT;
1003                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1004
1005                 kvm_for_each_vcpu(n, vcpu, kvm)
1006                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1007
1008                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1009                 memset(memslot->dirty_bitmap, 0, n);
1010         }
1011
1012         r = 0;
1013 out:
1014         mutex_unlock(&kvm->slots_lock);
1015         return r;
1016 }
1017
1018 int kvmppc_core_check_processor_compat(void)
1019 {
1020         return 0;
1021 }
1022
1023 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1024 {
1025         struct kvmppc_vcpu_book3s *vcpu_book3s;
1026         struct kvm_vcpu *vcpu;
1027         int err;
1028
1029         vcpu_book3s = (struct kvmppc_vcpu_book3s *)__get_free_pages( GFP_KERNEL | __GFP_ZERO,
1030                         get_order(sizeof(struct kvmppc_vcpu_book3s)));
1031         if (!vcpu_book3s) {
1032                 err = -ENOMEM;
1033                 goto out;
1034         }
1035
1036         vcpu = &vcpu_book3s->vcpu;
1037         err = kvm_vcpu_init(vcpu, kvm, id);
1038         if (err)
1039                 goto free_vcpu;
1040
1041         vcpu->arch.host_retip = kvm_return_point;
1042         vcpu->arch.host_msr = mfmsr();
1043         /* default to book3s_64 (970fx) */
1044         vcpu->arch.pvr = 0x3C0301;
1045         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1046         vcpu_book3s->slb_nr = 64;
1047
1048         /* remember where some real-mode handlers are */
1049         vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
1050         vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
1051         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
1052         vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
1053
1054         vcpu->arch.shadow_msr = MSR_USER64;
1055
1056         err = __init_new_context();
1057         if (err < 0)
1058                 goto free_vcpu;
1059         vcpu_book3s->context_id = err;
1060
1061         vcpu_book3s->vsid_max = ((vcpu_book3s->context_id + 1) << USER_ESID_BITS) - 1;
1062         vcpu_book3s->vsid_first = vcpu_book3s->context_id << USER_ESID_BITS;
1063         vcpu_book3s->vsid_next = vcpu_book3s->vsid_first;
1064
1065         return vcpu;
1066
1067 free_vcpu:
1068         free_pages((long)vcpu_book3s, get_order(sizeof(struct kvmppc_vcpu_book3s)));
1069 out:
1070         return ERR_PTR(err);
1071 }
1072
1073 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1074 {
1075         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1076
1077         __destroy_context(vcpu_book3s->context_id);
1078         kvm_vcpu_uninit(vcpu);
1079         free_pages((long)vcpu_book3s, get_order(sizeof(struct kvmppc_vcpu_book3s)));
1080 }
1081
1082 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1083 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1084 {
1085         int ret;
1086         struct thread_struct ext_bkp;
1087         bool save_vec = current->thread.used_vr;
1088         bool save_vsx = current->thread.used_vsr;
1089         ulong ext_msr;
1090
1091         /* No need to go into the guest when all we do is going out */
1092         if (signal_pending(current)) {
1093                 kvm_run->exit_reason = KVM_EXIT_INTR;
1094                 return -EINTR;
1095         }
1096
1097         /* Save FPU state in stack */
1098         if (current->thread.regs->msr & MSR_FP)
1099                 giveup_fpu(current);
1100         memcpy(ext_bkp.fpr, current->thread.fpr, sizeof(current->thread.fpr));
1101         ext_bkp.fpscr = current->thread.fpscr;
1102         ext_bkp.fpexc_mode = current->thread.fpexc_mode;
1103
1104 #ifdef CONFIG_ALTIVEC
1105         /* Save Altivec state in stack */
1106         if (save_vec) {
1107                 if (current->thread.regs->msr & MSR_VEC)
1108                         giveup_altivec(current);
1109                 memcpy(ext_bkp.vr, current->thread.vr, sizeof(ext_bkp.vr));
1110                 ext_bkp.vscr = current->thread.vscr;
1111                 ext_bkp.vrsave = current->thread.vrsave;
1112         }
1113         ext_bkp.used_vr = current->thread.used_vr;
1114 #endif
1115
1116 #ifdef CONFIG_VSX
1117         /* Save VSX state in stack */
1118         if (save_vsx && (current->thread.regs->msr & MSR_VSX))
1119                         __giveup_vsx(current);
1120         ext_bkp.used_vsr = current->thread.used_vsr;
1121 #endif
1122
1123         /* Remember the MSR with disabled extensions */
1124         ext_msr = current->thread.regs->msr;
1125
1126         /* XXX we get called with irq disabled - change that! */
1127         local_irq_enable();
1128
1129         ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
1130
1131         local_irq_disable();
1132
1133         current->thread.regs->msr = ext_msr;
1134
1135         /* Make sure we save the guest FPU/Altivec/VSX state */
1136         kvmppc_giveup_ext(vcpu, MSR_FP);
1137         kvmppc_giveup_ext(vcpu, MSR_VEC);
1138         kvmppc_giveup_ext(vcpu, MSR_VSX);
1139
1140         /* Restore FPU state from stack */
1141         memcpy(current->thread.fpr, ext_bkp.fpr, sizeof(ext_bkp.fpr));
1142         current->thread.fpscr = ext_bkp.fpscr;
1143         current->thread.fpexc_mode = ext_bkp.fpexc_mode;
1144
1145 #ifdef CONFIG_ALTIVEC
1146         /* Restore Altivec state from stack */
1147         if (save_vec && current->thread.used_vr) {
1148                 memcpy(current->thread.vr, ext_bkp.vr, sizeof(ext_bkp.vr));
1149                 current->thread.vscr = ext_bkp.vscr;
1150                 current->thread.vrsave= ext_bkp.vrsave;
1151         }
1152         current->thread.used_vr = ext_bkp.used_vr;
1153 #endif
1154
1155 #ifdef CONFIG_VSX
1156         current->thread.used_vsr = ext_bkp.used_vsr;
1157 #endif
1158
1159         return ret;
1160 }
1161
1162 static int kvmppc_book3s_init(void)
1163 {
1164         return kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), THIS_MODULE);
1165 }
1166
1167 static void kvmppc_book3s_exit(void)
1168 {
1169         kvm_exit();
1170 }
1171
1172 module_init(kvmppc_book3s_init);
1173 module_exit(kvmppc_book3s_exit);