Pileus Git - ~andy/linux/blob - 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 #include <linux/slab.h>
  20
  21 #include <asm/reg.h>
  22 #include <asm/cputable.h>
  23 #include <asm/cacheflush.h>
  24 #include <asm/tlbflush.h>
  25 #include <asm/uaccess.h>
  26 #include <asm/io.h>
  27 #include <asm/kvm_ppc.h>
  28 #include <asm/kvm_book3s.h>
  29 #include <asm/mmu_context.h>
  30 #include <linux/gfp.h>
  31 #include <linux/sched.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/highmem.h>
  34
  35 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  36
  37 /* #define EXIT_DEBUG */
  38 /* #define EXIT_DEBUG_SIMPLE */
  39 /* #define DEBUG_EXT */
  40
  41 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  42                              ulong msr);
  43
  44 /* Some compatibility defines */
  45 #ifdef CONFIG_PPC_BOOK3S_32
  46 #define MSR_USER32 MSR_USER
  47 #define MSR_USER64 MSR_USER
  48 #define HW_PAGE_SIZE PAGE_SIZE
  49 #endif
  50
  51 struct kvm_stats_debugfs_item debugfs_entries[] = {
  52         { "exits",       VCPU_STAT(sum_exits) },
  53         { "mmio",        VCPU_STAT(mmio_exits) },
  54         { "sig",         VCPU_STAT(signal_exits) },
  55         { "sysc",        VCPU_STAT(syscall_exits) },
  56         { "inst_emu",    VCPU_STAT(emulated_inst_exits) },
  57         { "dec",         VCPU_STAT(dec_exits) },
  58         { "ext_intr",    VCPU_STAT(ext_intr_exits) },
  59         { "queue_intr",  VCPU_STAT(queue_intr) },
  60         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  61         { "pf_storage",  VCPU_STAT(pf_storage) },
  62         { "sp_storage",  VCPU_STAT(sp_storage) },
  63         { "pf_instruc",  VCPU_STAT(pf_instruc) },
  64         { "sp_instruc",  VCPU_STAT(sp_instruc) },
  65         { "ld",          VCPU_STAT(ld) },
  66         { "ld_slow",     VCPU_STAT(ld_slow) },
  67         { "st",          VCPU_STAT(st) },
  68         { "st_slow",     VCPU_STAT(st_slow) },
  69         { NULL }
  70 };
  71
  72 void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
  73 {
  74 }
  75
  76 void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
  77 {
  78 }
  79
  80 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  81 {
  82 #ifdef CONFIG_PPC_BOOK3S_64
  83         memcpy(to_svcpu(vcpu)->slb, to_book3s(vcpu)->slb_shadow, sizeof(to_svcpu(vcpu)->slb));
  84         memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
  85                sizeof(get_paca()->shadow_vcpu));
  86         to_svcpu(vcpu)->slb_max = to_book3s(vcpu)->slb_shadow_max;
  87 #endif
  88
  89 #ifdef CONFIG_PPC_BOOK3S_32
  90         current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
  91 #endif
  92 }
  93
  94 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  95 {
  96 #ifdef CONFIG_PPC_BOOK3S_64
  97         memcpy(to_book3s(vcpu)->slb_shadow, to_svcpu(vcpu)->slb, sizeof(to_svcpu(vcpu)->slb));
  98         memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  99                sizeof(get_paca()->shadow_vcpu));
 100         to_book3s(vcpu)->slb_shadow_max = to_svcpu(vcpu)->slb_max;
 101 #endif
 102
 103         kvmppc_giveup_ext(vcpu, MSR_FP);
 104         kvmppc_giveup_ext(vcpu, MSR_VEC);
 105         kvmppc_giveup_ext(vcpu, MSR_VSX);
 106 }
 107
 108 #if defined(EXIT_DEBUG)
 109 static u32 kvmppc_get_dec(struct kvm_vcpu *vcpu)
 110 {
 111         u64 jd = mftb() - vcpu->arch.dec_jiffies;
 112         return vcpu->arch.dec - jd;
 113 }
 114 #endif
 115
 116 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 117 {
 118         ulong smsr = vcpu->arch.shared->msr;
 119
 120         /* Guest MSR values */
 121         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE;
 122         /* Process MSR values */
 123         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 124         /* External providers the guest reserved */
 125         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
 126         /* 64-bit Process MSR values */
 127 #ifdef CONFIG_PPC_BOOK3S_64
 128         smsr |= MSR_ISF | MSR_HV;
 129 #endif
 130         vcpu->arch.shadow_msr = smsr;
 131 }
 132
 133 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 134 {
 135         ulong old_msr = vcpu->arch.shared->msr;
 136
 137 #ifdef EXIT_DEBUG
 138         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 139 #endif
 140
 141         msr &= to_book3s(vcpu)->msr_mask;
 142         vcpu->arch.shared->msr = msr;
 143         kvmppc_recalc_shadow_msr(vcpu);
 144
 145         if (msr & (MSR_WE|MSR_POW)) {
 146                 if (!vcpu->arch.pending_exceptions) {
 147                         kvm_vcpu_block(vcpu);
 148                         vcpu->stat.halt_wakeup++;
 149                 }
 150         }
 151
 152         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 153                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 154                 kvmppc_mmu_flush_segments(vcpu);
 155                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 156         }
 157
 158         /* Preload FPU if it's enabled */
 159         if (vcpu->arch.shared->msr & MSR_FP)
 160                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 161 }
 162
 163 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
 164 {
 165         vcpu->arch.shared->srr0 = kvmppc_get_pc(vcpu);
 166         vcpu->arch.shared->srr1 = vcpu->arch.shared->msr | flags;
 167         kvmppc_set_pc(vcpu, to_book3s(vcpu)->hior + vec);
 168         vcpu->arch.mmu.reset_msr(vcpu);
 169 }
 170
 171 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
 172 {
 173         unsigned int prio;
 174
 175         switch (vec) {
 176         case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET;         break;
 177         case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK;        break;
 178         case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE;         break;
 179         case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT;         break;
 180         case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE;         break;
 181         case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT;         break;
 182         case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL;             break;
 183         case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT;            break;
 184         case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM;              break;
 185         case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL;           break;
 186         case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER;          break;
 187         case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL;              break;
 188         case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG;                break;
 189         case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC;              break;
 190         case 0xf40: prio = BOOK3S_IRQPRIO_VSX;                  break;
 191         default:    prio = BOOK3S_IRQPRIO_MAX;                  break;
 192         }
 193
 194         return prio;
 195 }
 196
 197 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
 198                                           unsigned int vec)
 199 {
 200         clear_bit(kvmppc_book3s_vec2irqprio(vec),
 201                   &vcpu->arch.pending_exceptions);
 202 }
 203
 204 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
 205 {
 206         vcpu->stat.queue_intr++;
 207
 208         set_bit(kvmppc_book3s_vec2irqprio(vec),
 209                 &vcpu->arch.pending_exceptions);
 210 #ifdef EXIT_DEBUG
 211         printk(KERN_INFO "Queueing interrupt %x\n", vec);
 212 #endif
 213 }
 214
 215
 216 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong flags)
 217 {
 218         to_book3s(vcpu)->prog_flags = flags;
 219         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
 220 }
 221
 222 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 223 {
 224         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 225 }
 226
 227 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 228 {
 229         return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
 230 }
 231
 232 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 233 {
 234         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 235 }
 236
 237 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 238                                 struct kvm_interrupt *irq)
 239 {
 240         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 241 }
 242
 243 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
 244                                   struct kvm_interrupt *irq)
 245 {
 246         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 247 }
 248
 249 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
 250 {
 251         int deliver = 1;
 252         int vec = 0;
 253         ulong flags = 0ULL;
 254         ulong crit_raw = vcpu->arch.shared->critical;
 255         ulong crit_r1 = kvmppc_get_gpr(vcpu, 1);
 256         bool crit;
 257
 258         /* Truncate crit indicators in 32 bit mode */
 259         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 260                 crit_raw &= 0xffffffff;
 261                 crit_r1 &= 0xffffffff;
 262         }
 263
 264         /* Critical section when crit == r1 */
 265         crit = (crit_raw == crit_r1);
 266         /* ... and we're in supervisor mode */
 267         crit = crit && !(vcpu->arch.shared->msr & MSR_PR);
 268
 269         switch (priority) {
 270         case BOOK3S_IRQPRIO_DECREMENTER:
 271                 deliver = (vcpu->arch.shared->msr & MSR_EE) && !crit;
 272                 vec = BOOK3S_INTERRUPT_DECREMENTER;
 273                 break;
 274         case BOOK3S_IRQPRIO_EXTERNAL:
 275                 deliver = (vcpu->arch.shared->msr & MSR_EE) && !crit;
 276                 vec = BOOK3S_INTERRUPT_EXTERNAL;
 277                 break;
 278         case BOOK3S_IRQPRIO_SYSTEM_RESET:
 279                 vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
 280                 break;
 281         case BOOK3S_IRQPRIO_MACHINE_CHECK:
 282                 vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
 283                 break;
 284         case BOOK3S_IRQPRIO_DATA_STORAGE:
 285                 vec = BOOK3S_INTERRUPT_DATA_STORAGE;
 286                 break;
 287         case BOOK3S_IRQPRIO_INST_STORAGE:
 288                 vec = BOOK3S_INTERRUPT_INST_STORAGE;
 289                 break;
 290         case BOOK3S_IRQPRIO_DATA_SEGMENT:
 291                 vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
 292                 break;
 293         case BOOK3S_IRQPRIO_INST_SEGMENT:
 294                 vec = BOOK3S_INTERRUPT_INST_SEGMENT;
 295                 break;
 296         case BOOK3S_IRQPRIO_ALIGNMENT:
 297                 vec = BOOK3S_INTERRUPT_ALIGNMENT;
 298                 break;
 299         case BOOK3S_IRQPRIO_PROGRAM:
 300                 vec = BOOK3S_INTERRUPT_PROGRAM;
 301                 flags = to_book3s(vcpu)->prog_flags;
 302                 break;
 303         case BOOK3S_IRQPRIO_VSX:
 304                 vec = BOOK3S_INTERRUPT_VSX;
 305                 break;
 306         case BOOK3S_IRQPRIO_ALTIVEC:
 307                 vec = BOOK3S_INTERRUPT_ALTIVEC;
 308                 break;
 309         case BOOK3S_IRQPRIO_FP_UNAVAIL:
 310                 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
 311                 break;
 312         case BOOK3S_IRQPRIO_SYSCALL:
 313                 vec = BOOK3S_INTERRUPT_SYSCALL;
 314                 break;
 315         case BOOK3S_IRQPRIO_DEBUG:
 316                 vec = BOOK3S_INTERRUPT_TRACE;
 317                 break;
 318         case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
 319                 vec = BOOK3S_INTERRUPT_PERFMON;
 320                 break;
 321         default:
 322                 deliver = 0;
 323                 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
 324                 break;
 325         }
 326
 327 #if 0
 328         printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
 329 #endif
 330
 331         if (deliver)
 332                 kvmppc_inject_interrupt(vcpu, vec, flags);
 333
 334         return deliver;
 335 }
 336
 337 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 338 {
 339         unsigned long *pending = &vcpu->arch.pending_exceptions;
 340         unsigned long old_pending = vcpu->arch.pending_exceptions;
 341         unsigned int priority;
 342
 343 #ifdef EXIT_DEBUG
 344         if (vcpu->arch.pending_exceptions)
 345                 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
 346 #endif
 347         priority = __ffs(*pending);
 348         while (priority < BOOK3S_IRQPRIO_MAX) {
 349                 if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
 350                     (priority != BOOK3S_IRQPRIO_DECREMENTER)) {
 351                         /* DEC interrupts get cleared by mtdec */
 352                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 353                         break;
 354                 }
 355
 356                 priority = find_next_bit(pending,
 357                                          BITS_PER_BYTE * sizeof(*pending),
 358                                          priority + 1);
 359         }
 360
 361         /* Tell the guest about our interrupt status */
 362         if (*pending)
 363                 vcpu->arch.shared->int_pending = 1;
 364         else if (old_pending)
 365                 vcpu->arch.shared->int_pending = 0;
 366 }
 367
 368 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 369 {
 370         u32 host_pvr;
 371
 372         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 373         vcpu->arch.pvr = pvr;
 374 #ifdef CONFIG_PPC_BOOK3S_64
 375         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 376                 kvmppc_mmu_book3s_64_init(vcpu);
 377                 to_book3s(vcpu)->hior = 0xfff00000;
 378                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 379         } else
 380 #endif
 381         {
 382                 kvmppc_mmu_book3s_32_init(vcpu);
 383                 to_book3s(vcpu)->hior = 0;
 384                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 385         }
 386
 387         /* If we are in hypervisor level on 970, we can tell the CPU to
 388          * treat DCBZ as 32 bytes store */
 389         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 390         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 391             !strcmp(cur_cpu_spec->platform, "ppc970"))
 392                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 393
 394         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 395            really needs them in a VM on Cell and force disable them. */
 396         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 397                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 398
 399 #ifdef CONFIG_PPC_BOOK3S_32
 400         /* 32 bit Book3S always has 32 byte dcbz */
 401         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 402 #endif
 403
 404         /* On some CPUs we can execute paired single operations natively */
 405         asm ( "mfpvr %0" : "=r"(host_pvr));
 406         switch (host_pvr) {
 407         case 0x00080200:        /* lonestar 2.0 */
 408         case 0x00088202:        /* lonestar 2.2 */
 409         case 0x70000100:        /* gekko 1.0 */
 410         case 0x00080100:        /* gekko 2.0 */
 411         case 0x00083203:        /* gekko 2.3a */
 412         case 0x00083213:        /* gekko 2.3b */
 413         case 0x00083204:        /* gekko 2.4 */
 414         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 415         case 0x00087200:        /* broadway */
 416                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 417                 /* Enable HID2.PSE - in case we need it later */
 418                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 419         }
 420 }
 421
 422 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 423  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 424  * emulate 32 bytes dcbz length.
 425  *
 426  * The Book3s_64 inventors also realized this case and implemented a special bit
 427  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 428  *
 429  * My approach here is to patch the dcbz instruction on executing pages.
 430  */
 431 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 432 {
 433         struct page *hpage;
 434         u64 hpage_offset;
 435         u32 *page;
 436         int i;
 437
 438         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 439         if (is_error_page(hpage))
 440                 return;
 441
 442         hpage_offset = pte->raddr & ~PAGE_MASK;
 443         hpage_offset &= ~0xFFFULL;
 444         hpage_offset /= 4;
 445
 446         get_page(hpage);
 447         page = kmap_atomic(hpage, KM_USER0);
 448
 449         /* patch dcbz into reserved instruction, so we trap */
 450         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 451                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 452                         page[i] &= 0xfffffff7;
 453
 454         kunmap_atomic(page, KM_USER0);
 455         put_page(hpage);
 456 }
 457
 458 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
 459                          struct kvmppc_pte *pte)
 460 {
 461         int relocated = (vcpu->arch.shared->msr & (data ? MSR_DR : MSR_IR));
 462         int r;
 463
 464         if (relocated) {
 465                 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
 466         } else {
 467                 pte->eaddr = eaddr;
 468                 pte->raddr = eaddr & 0xffffffff;
 469                 pte->vpage = VSID_REAL | eaddr >> 12;
 470                 pte->may_read = true;
 471                 pte->may_write = true;
 472                 pte->may_execute = true;
 473                 r = 0;
 474         }
 475
 476         return r;
 477 }
 478
 479 static hva_t kvmppc_bad_hva(void)
 480 {
 481         return PAGE_OFFSET;
 482 }
 483
 484 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
 485                                bool read)
 486 {
 487         hva_t hpage;
 488
 489         if (read && !pte->may_read)
 490                 goto err;
 491
 492         if (!read && !pte->may_write)
 493                 goto err;
 494
 495         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 496         if (kvm_is_error_hva(hpage))
 497                 goto err;
 498
 499         return hpage | (pte->raddr & ~PAGE_MASK);
 500 err:
 501         return kvmppc_bad_hva();
 502 }
 503
 504 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 505               bool data)
 506 {
 507         struct kvmppc_pte pte;
 508
 509         vcpu->stat.st++;
 510
 511         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 512                 return -ENOENT;
 513
 514         *eaddr = pte.raddr;
 515
 516         if (!pte.may_write)
 517                 return -EPERM;
 518
 519         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 520                 return EMULATE_DO_MMIO;
 521
 522         return EMULATE_DONE;
 523 }
 524
 525 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 526                       bool data)
 527 {
 528         struct kvmppc_pte pte;
 529         hva_t hva = *eaddr;
 530
 531         vcpu->stat.ld++;
 532
 533         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 534                 goto nopte;
 535
 536         *eaddr = pte.raddr;
 537
 538         hva = kvmppc_pte_to_hva(vcpu, &pte, true);
 539         if (kvm_is_error_hva(hva))
 540                 goto mmio;
 541
 542         if (copy_from_user(ptr, (void __user *)hva, size)) {
 543                 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
 544                 goto mmio;
 545         }
 546
 547         return EMULATE_DONE;
 548
 549 nopte:
 550         return -ENOENT;
 551 mmio:
 552         return EMULATE_DO_MMIO;
 553 }
 554
 555 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 556 {
 557         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 558 }
 559
 560 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 561                             ulong eaddr, int vec)
 562 {
 563         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 564         int r = RESUME_GUEST;
 565         int relocated;
 566         int page_found = 0;
 567         struct kvmppc_pte pte;
 568         bool is_mmio = false;
 569         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 570         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 571         u64 vsid;
 572
 573         relocated = data ? dr : ir;
 574
 575         /* Resolve real address if translation turned on */
 576         if (relocated) {
 577                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 578         } else {
 579                 pte.may_execute = true;
 580                 pte.may_read = true;
 581                 pte.may_write = true;
 582                 pte.raddr = eaddr & 0xffffffff;
 583                 pte.eaddr = eaddr;
 584                 pte.vpage = eaddr >> 12;
 585         }
 586
 587         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 588         case 0:
 589                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 590                 break;
 591         case MSR_DR:
 592         case MSR_IR:
 593                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 594
 595                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 596                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 597                 else
 598                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 599                 pte.vpage |= vsid;
 600
 601                 if (vsid == -1)
 602                         page_found = -EINVAL;
 603                 break;
 604         }
 605
 606         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 607            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 608                 /*
 609                  * If we do the dcbz hack, we have to NX on every execution,
 610                  * so we can patch the executing code. This renders our guest
 611                  * NX-less.
 612                  */
 613                 pte.may_execute = !data;
 614         }
 615
 616         if (page_found == -ENOENT) {
 617                 /* Page not found in guest PTE entries */
 618                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 619                 vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 620                 vcpu->arch.shared->msr |=
 621                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 622                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 623         } else if (page_found == -EPERM) {
 624                 /* Storage protection */
 625                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 626                 vcpu->arch.shared->dsisr =
 627                         to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
 628                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 629                 vcpu->arch.shared->msr |=
 630                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 631                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 632         } else if (page_found == -EINVAL) {
 633                 /* Page not found in guest SLB */
 634                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 635                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 636         } else if (!is_mmio &&
 637                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 638                 /* The guest's PTE is not mapped yet. Map on the host */
 639                 kvmppc_mmu_map_page(vcpu, &pte);
 640                 if (data)
 641                         vcpu->stat.sp_storage++;
 642                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 643                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 644                         kvmppc_patch_dcbz(vcpu, &pte);
 645         } else {
 646                 /* MMIO */
 647                 vcpu->stat.mmio_exits++;
 648                 vcpu->arch.paddr_accessed = pte.raddr;
 649                 r = kvmppc_emulate_mmio(run, vcpu);
 650                 if ( r == RESUME_HOST_NV )
 651                         r = RESUME_HOST;
 652         }
 653
 654         return r;
 655 }
 656
 657 static inline int get_fpr_index(int i)
 658 {
 659 #ifdef CONFIG_VSX
 660         i *= 2;
 661 #endif
 662         return i;
 663 }
 664
 665 /* Give up external provider (FPU, Altivec, VSX) */
 666 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 667 {
 668         struct thread_struct *t = &current->thread;
 669         u64 *vcpu_fpr = vcpu->arch.fpr;
 670 #ifdef CONFIG_VSX
 671         u64 *vcpu_vsx = vcpu->arch.vsr;
 672 #endif
 673         u64 *thread_fpr = (u64*)t->fpr;
 674         int i;
 675
 676         if (!(vcpu->arch.guest_owned_ext & msr))
 677                 return;
 678
 679 #ifdef DEBUG_EXT
 680         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 681 #endif
 682
 683         switch (msr) {
 684         case MSR_FP:
 685                 giveup_fpu(current);
 686                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 687                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 688
 689                 vcpu->arch.fpscr = t->fpscr.val;
 690                 break;
 691         case MSR_VEC:
 692 #ifdef CONFIG_ALTIVEC
 693                 giveup_altivec(current);
 694                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 695                 vcpu->arch.vscr = t->vscr;
 696 #endif
 697                 break;
 698         case MSR_VSX:
 699 #ifdef CONFIG_VSX
 700                 __giveup_vsx(current);
 701                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 702                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 703 #endif
 704                 break;
 705         default:
 706                 BUG();
 707         }
 708
 709         vcpu->arch.guest_owned_ext &= ~msr;
 710         current->thread.regs->msr &= ~msr;
 711         kvmppc_recalc_shadow_msr(vcpu);
 712 }
 713
 714 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 715 {
 716         ulong srr0 = kvmppc_get_pc(vcpu);
 717         u32 last_inst = kvmppc_get_last_inst(vcpu);
 718         int ret;
 719
 720         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 721         if (ret == -ENOENT) {
 722                 ulong msr = vcpu->arch.shared->msr;
 723
 724                 msr = kvmppc_set_field(msr, 33, 33, 1);
 725                 msr = kvmppc_set_field(msr, 34, 36, 0);
 726                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 727                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 728                 return EMULATE_AGAIN;
 729         }
 730
 731         return EMULATE_DONE;
 732 }
 733
 734 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 735 {
 736
 737         /* Need to do paired single emulation? */
 738         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 739                 return EMULATE_DONE;
 740
 741         /* Read out the instruction */
 742         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 743                 /* Need to emulate */
 744                 return EMULATE_FAIL;
 745
 746         return EMULATE_AGAIN;
 747 }
 748
 749 /* Handle external providers (FPU, Altivec, VSX) */
 750 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 751                              ulong msr)
 752 {
 753         struct thread_struct *t = &current->thread;
 754         u64 *vcpu_fpr = vcpu->arch.fpr;
 755 #ifdef CONFIG_VSX
 756         u64 *vcpu_vsx = vcpu->arch.vsr;
 757 #endif
 758         u64 *thread_fpr = (u64*)t->fpr;
 759         int i;
 760
 761         /* When we have paired singles, we emulate in software */
 762         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 763                 return RESUME_GUEST;
 764
 765         if (!(vcpu->arch.shared->msr & msr)) {
 766                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 767                 return RESUME_GUEST;
 768         }
 769
 770         /* We already own the ext */
 771         if (vcpu->arch.guest_owned_ext & msr) {
 772                 return RESUME_GUEST;
 773         }
 774
 775 #ifdef DEBUG_EXT
 776         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 777 #endif
 778
 779         current->thread.regs->msr |= msr;
 780
 781         switch (msr) {
 782         case MSR_FP:
 783                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 784                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 785
 786                 t->fpscr.val = vcpu->arch.fpscr;
 787                 t->fpexc_mode = 0;
 788                 kvmppc_load_up_fpu();
 789                 break;
 790         case MSR_VEC:
 791 #ifdef CONFIG_ALTIVEC
 792                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 793                 t->vscr = vcpu->arch.vscr;
 794                 t->vrsave = -1;
 795                 kvmppc_load_up_altivec();
 796 #endif
 797                 break;
 798         case MSR_VSX:
 799 #ifdef CONFIG_VSX
 800                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 801                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 802                 kvmppc_load_up_vsx();
 803 #endif
 804                 break;
 805         default:
 806                 BUG();
 807         }
 808
 809         vcpu->arch.guest_owned_ext |= msr;
 810
 811         kvmppc_recalc_shadow_msr(vcpu);
 812
 813         return RESUME_GUEST;
 814 }
 815
 816 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 817                        unsigned int exit_nr)
 818 {
 819         int r = RESUME_HOST;
 820
 821         vcpu->stat.sum_exits++;
 822
 823         run->exit_reason = KVM_EXIT_UNKNOWN;
 824         run->ready_for_interrupt_injection = 1;
 825 #ifdef EXIT_DEBUG
 826         printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | dec=0x%x | msr=0x%lx\n",
 827                 exit_nr, kvmppc_get_pc(vcpu), kvmppc_get_fault_dar(vcpu),
 828                 kvmppc_get_dec(vcpu), to_svcpu(vcpu)->shadow_srr1);
 829 #elif defined (EXIT_DEBUG_SIMPLE)
 830         if ((exit_nr != 0x900) && (exit_nr != 0x500))
 831                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | msr=0x%lx\n",
 832                         exit_nr, kvmppc_get_pc(vcpu), kvmppc_get_fault_dar(vcpu),
 833                         vcpu->arch.shared->msr);
 834 #endif
 835         kvm_resched(vcpu);
 836         switch (exit_nr) {
 837         case BOOK3S_INTERRUPT_INST_STORAGE:
 838                 vcpu->stat.pf_instruc++;
 839
 840 #ifdef CONFIG_PPC_BOOK3S_32
 841                 /* We set segments as unused segments when invalidating them. So
 842                  * treat the respective fault as segment fault. */
 843                 if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
 844                     == SR_INVALID) {
 845                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 846                         r = RESUME_GUEST;
 847                         break;
 848                 }
 849 #endif
 850
 851                 /* only care about PTEG not found errors, but leave NX alone */
 852                 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
 853                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 854                         vcpu->stat.sp_instruc++;
 855                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 856                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 857                         /*
 858                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 859                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 860                          *     that no guest that needs the dcbz hack does NX.
 861                          */
 862                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 863                         r = RESUME_GUEST;
 864                 } else {
 865                         vcpu->arch.shared->msr |=
 866                                 to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
 867                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 868                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 869                         r = RESUME_GUEST;
 870                 }
 871                 break;
 872         case BOOK3S_INTERRUPT_DATA_STORAGE:
 873         {
 874                 ulong dar = kvmppc_get_fault_dar(vcpu);
 875                 vcpu->stat.pf_storage++;
 876
 877 #ifdef CONFIG_PPC_BOOK3S_32
 878                 /* We set segments as unused segments when invalidating them. So
 879                  * treat the respective fault as segment fault. */
 880                 if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 881                         kvmppc_mmu_map_segment(vcpu, dar);
 882                         r = RESUME_GUEST;
 883                         break;
 884                 }
 885 #endif
 886
 887                 /* The only case we need to handle is missing shadow PTEs */
 888                 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
 889                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 890                 } else {
 891                         vcpu->arch.shared->dar = dar;
 892                         vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 893                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 894                         kvmppc_mmu_pte_flush(vcpu, dar, ~0xFFFUL);
 895                         r = RESUME_GUEST;
 896                 }
 897                 break;
 898         }
 899         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 900                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 901                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 902                         kvmppc_book3s_queue_irqprio(vcpu,
 903                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 904                 }
 905                 r = RESUME_GUEST;
 906                 break;
 907         case BOOK3S_INTERRUPT_INST_SEGMENT:
 908                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 909                         kvmppc_book3s_queue_irqprio(vcpu,
 910                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 911                 }
 912                 r = RESUME_GUEST;
 913                 break;
 914         /* We're good on these - the host merely wanted to get our attention */
 915         case BOOK3S_INTERRUPT_DECREMENTER:
 916                 vcpu->stat.dec_exits++;
 917                 r = RESUME_GUEST;
 918                 break;
 919         case BOOK3S_INTERRUPT_EXTERNAL:
 920                 vcpu->stat.ext_intr_exits++;
 921                 r = RESUME_GUEST;
 922                 break;
 923         case BOOK3S_INTERRUPT_PERFMON:
 924                 r = RESUME_GUEST;
 925                 break;
 926         case BOOK3S_INTERRUPT_PROGRAM:
 927         {
 928                 enum emulation_result er;
 929                 ulong flags;
 930
 931 program_interrupt:
 932                 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
 933
 934                 if (vcpu->arch.shared->msr & MSR_PR) {
 935 #ifdef EXIT_DEBUG
 936                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 937 #endif
 938                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 939                             (INS_DCBZ & 0xfffffff7)) {
 940                                 kvmppc_core_queue_program(vcpu, flags);
 941                                 r = RESUME_GUEST;
 942                                 break;
 943                         }
 944                 }
 945
 946                 vcpu->stat.emulated_inst_exits++;
 947                 er = kvmppc_emulate_instruction(run, vcpu);
 948                 switch (er) {
 949                 case EMULATE_DONE:
 950                         r = RESUME_GUEST_NV;
 951                         break;
 952                 case EMULATE_AGAIN:
 953                         r = RESUME_GUEST;
 954                         break;
 955                 case EMULATE_FAIL:
 956                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 957                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 958                         kvmppc_core_queue_program(vcpu, flags);
 959                         r = RESUME_GUEST;
 960                         break;
 961                 case EMULATE_DO_MMIO:
 962                         run->exit_reason = KVM_EXIT_MMIO;
 963                         r = RESUME_HOST_NV;
 964                         break;
 965                 default:
 966                         BUG();
 967                 }
 968                 break;
 969         }
 970         case BOOK3S_INTERRUPT_SYSCALL:
 971                 if (vcpu->arch.osi_enabled &&
 972                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 973                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 974                         /* MOL hypercalls */
 975                         u64 *gprs = run->osi.gprs;
 976                         int i;
 977
 978                         run->exit_reason = KVM_EXIT_OSI;
 979                         for (i = 0; i < 32; i++)
 980                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 981                         vcpu->arch.osi_needed = 1;
 982                         r = RESUME_HOST_NV;
 983                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 984                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 985                         /* KVM PV hypercalls */
 986                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 987                         r = RESUME_GUEST;
 988                 } else {
 989                         /* Guest syscalls */
 990                         vcpu->stat.syscall_exits++;
 991                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 992                         r = RESUME_GUEST;
 993                 }
 994                 break;
 995         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 996         case BOOK3S_INTERRUPT_ALTIVEC:
 997         case BOOK3S_INTERRUPT_VSX:
 998         {
 999                 int ext_msr = 0;
1000
1001                 switch (exit_nr) {
1002                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
1003                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
1004                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
1005                 }
1006
1007                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
1008                 case EMULATE_DONE:
1009                         /* everything ok - let's enable the ext */
1010                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1011                         break;
1012                 case EMULATE_FAIL:
1013                         /* we need to emulate this instruction */
1014                         goto program_interrupt;
1015                         break;
1016                 default:
1017                         /* nothing to worry about - go again */
1018                         break;
1019                 }
1020                 break;
1021         }
1022         case BOOK3S_INTERRUPT_ALIGNMENT:
1023                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
1024                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
1025                                 kvmppc_get_last_inst(vcpu));
1026                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
1027                                 kvmppc_get_last_inst(vcpu));
1028                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1029                 }
1030                 r = RESUME_GUEST;
1031                 break;
1032         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1033         case BOOK3S_INTERRUPT_TRACE:
1034                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1035                 r = RESUME_GUEST;
1036                 break;
1037         default:
1038                 /* Ugh - bork here! What did we get? */
1039                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1040                         exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
1041                 r = RESUME_HOST;
1042                 BUG();
1043                 break;
1044         }
1045
1046
1047         if (!(r & RESUME_HOST)) {
1048                 /* To avoid clobbering exit_reason, only check for signals if
1049                  * we aren't already exiting to userspace for some other
1050                  * reason. */
1051                 if (signal_pending(current)) {
1052 #ifdef EXIT_DEBUG
1053                         printk(KERN_EMERG "KVM: Going back to host\n");
1054 #endif
1055                         vcpu->stat.signal_exits++;
1056                         run->exit_reason = KVM_EXIT_INTR;
1057                         r = -EINTR;
1058                 } else {
1059                         /* In case an interrupt came in that was triggered
1060                          * from userspace (like DEC), we need to check what
1061                          * to inject now! */
1062                         kvmppc_core_deliver_interrupts(vcpu);
1063                 }
1064         }
1065
1066 #ifdef EXIT_DEBUG
1067         printk(KERN_EMERG "KVM exit: vcpu=0x%p pc=0x%lx r=0x%x\n", vcpu, kvmppc_get_pc(vcpu), r);
1068 #endif
1069
1070         return r;
1071 }
1072
1073 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1074 {
1075         return 0;
1076 }
1077
1078 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1079 {
1080         int i;
1081
1082         regs->pc = kvmppc_get_pc(vcpu);
1083         regs->cr = kvmppc_get_cr(vcpu);
1084         regs->ctr = kvmppc_get_ctr(vcpu);
1085         regs->lr = kvmppc_get_lr(vcpu);
1086         regs->xer = kvmppc_get_xer(vcpu);
1087         regs->msr = vcpu->arch.shared->msr;
1088         regs->srr0 = vcpu->arch.shared->srr0;
1089         regs->srr1 = vcpu->arch.shared->srr1;
1090         regs->pid = vcpu->arch.pid;
1091         regs->sprg0 = vcpu->arch.shared->sprg0;
1092         regs->sprg1 = vcpu->arch.shared->sprg1;
1093         regs->sprg2 = vcpu->arch.shared->sprg2;
1094         regs->sprg3 = vcpu->arch.shared->sprg3;
1095         regs->sprg5 = vcpu->arch.sprg4;
1096         regs->sprg6 = vcpu->arch.sprg5;
1097         regs->sprg7 = vcpu->arch.sprg6;
1098
1099         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1100                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
1101
1102         return 0;
1103 }
1104
1105 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1106 {
1107         int i;
1108
1109         kvmppc_set_pc(vcpu, regs->pc);
1110         kvmppc_set_cr(vcpu, regs->cr);
1111         kvmppc_set_ctr(vcpu, regs->ctr);
1112         kvmppc_set_lr(vcpu, regs->lr);
1113         kvmppc_set_xer(vcpu, regs->xer);
1114         kvmppc_set_msr(vcpu, regs->msr);
1115         vcpu->arch.shared->srr0 = regs->srr0;
1116         vcpu->arch.shared->srr1 = regs->srr1;
1117         vcpu->arch.shared->sprg0 = regs->sprg0;
1118         vcpu->arch.shared->sprg1 = regs->sprg1;
1119         vcpu->arch.shared->sprg2 = regs->sprg2;
1120         vcpu->arch.shared->sprg3 = regs->sprg3;
1121         vcpu->arch.sprg5 = regs->sprg4;
1122         vcpu->arch.sprg6 = regs->sprg5;
1123         vcpu->arch.sprg7 = regs->sprg6;
1124
1125         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1126                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
1127
1128         return 0;
1129 }
1130
1131 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1132                                   struct kvm_sregs *sregs)
1133 {
1134         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1135         int i;
1136
1137         sregs->pvr = vcpu->arch.pvr;
1138
1139         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1140         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1141                 for (i = 0; i < 64; i++) {
1142                         sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
1143                         sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
1144                 }
1145         } else {
1146                 for (i = 0; i < 16; i++) {
1147                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
1148                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
1149                 }
1150                 for (i = 0; i < 8; i++) {
1151                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1152                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1153                 }
1154         }
1155
1156         return 0;
1157 }
1158
1159 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1160                                   struct kvm_sregs *sregs)
1161 {
1162         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1163         int i;
1164
1165         kvmppc_set_pvr(vcpu, sregs->pvr);
1166
1167         vcpu3s->sdr1 = sregs->u.s.sdr1;
1168         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1169                 for (i = 0; i < 64; i++) {
1170                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1171                                                     sregs->u.s.ppc64.slb[i].slbe);
1172                 }
1173         } else {
1174                 for (i = 0; i < 16; i++) {
1175                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1176                 }
1177                 for (i = 0; i < 8; i++) {
1178                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1179                                        (u32)sregs->u.s.ppc32.ibat[i]);
1180                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1181                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1182                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1183                                        (u32)sregs->u.s.ppc32.dbat[i]);
1184                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1185                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1186                 }
1187         }
1188
1189         /* Flush the MMU after messing with the segments */
1190         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1191
1192         return 0;
1193 }
1194
1195 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1196 {
1197         return -ENOTSUPP;
1198 }
1199
1200 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1201 {
1202         return -ENOTSUPP;
1203 }
1204
1205 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1206                                   struct kvm_translation *tr)
1207 {
1208         return 0;
1209 }
1210
1211 /*
1212  * Get (and clear) the dirty memory log for a memory slot.
1213  */
1214 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1215                                       struct kvm_dirty_log *log)
1216 {
1217         struct kvm_memory_slot *memslot;
1218         struct kvm_vcpu *vcpu;
1219         ulong ga, ga_end;
1220         int is_dirty = 0;
1221         int r;
1222         unsigned long n;
1223
1224         mutex_lock(&kvm->slots_lock);
1225
1226         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1227         if (r)
1228                 goto out;
1229
1230         /* If nothing is dirty, don't bother messing with page tables. */
1231         if (is_dirty) {
1232                 memslot = &kvm->memslots->memslots[log->slot];
1233
1234                 ga = memslot->base_gfn << PAGE_SHIFT;
1235                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1236
1237                 kvm_for_each_vcpu(n, vcpu, kvm)
1238                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1239
1240                 n = kvm_dirty_bitmap_bytes(memslot);
1241                 memset(memslot->dirty_bitmap, 0, n);
1242         }
1243
1244         r = 0;
1245 out:
1246         mutex_unlock(&kvm->slots_lock);
1247         return r;
1248 }
1249
1250 int kvmppc_core_check_processor_compat(void)
1251 {
1252         return 0;
1253 }
1254
1255 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1256 {
1257         struct kvmppc_vcpu_book3s *vcpu_book3s;
1258         struct kvm_vcpu *vcpu;
1259         int err = -ENOMEM;
1260
1261         vcpu_book3s = vmalloc(sizeof(struct kvmppc_vcpu_book3s));
1262         if (!vcpu_book3s)
1263                 goto out;
1264
1265         memset(vcpu_book3s, 0, sizeof(struct kvmppc_vcpu_book3s));
1266
1267         vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
1268                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
1269         if (!vcpu_book3s->shadow_vcpu)
1270                 goto free_vcpu;
1271
1272         vcpu = &vcpu_book3s->vcpu;
1273         err = kvm_vcpu_init(vcpu, kvm, id);
1274         if (err)
1275                 goto free_shadow_vcpu;
1276
1277         vcpu->arch.shared = (void*)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1278         if (!vcpu->arch.shared)
1279                 goto uninit_vcpu;
1280
1281         vcpu->arch.host_retip = kvm_return_point;
1282         vcpu->arch.host_msr = mfmsr();
1283 #ifdef CONFIG_PPC_BOOK3S_64
1284         /* default to book3s_64 (970fx) */
1285         vcpu->arch.pvr = 0x3C0301;
1286 #else
1287         /* default to book3s_32 (750) */
1288         vcpu->arch.pvr = 0x84202;
1289 #endif
1290         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1291         vcpu_book3s->slb_nr = 64;
1292
1293         /* remember where some real-mode handlers are */
1294         vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
1295         vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
1296         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
1297 #ifdef CONFIG_PPC_BOOK3S_64
1298         vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
1299 #else
1300         vcpu->arch.rmcall = (ulong)kvmppc_rmcall;
1301 #endif
1302
1303         vcpu->arch.shadow_msr = MSR_USER64;
1304
1305         err = kvmppc_mmu_init(vcpu);
1306         if (err < 0)
1307                 goto uninit_vcpu;
1308
1309         return vcpu;
1310
1311 uninit_vcpu:
1312         kvm_vcpu_uninit(vcpu);
1313 free_shadow_vcpu:
1314         kfree(vcpu_book3s->shadow_vcpu);
1315 free_vcpu:
1316         vfree(vcpu_book3s);
1317 out:
1318         return ERR_PTR(err);
1319 }
1320
1321 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1322 {
1323         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1324
1325         free_page((unsigned long)vcpu->arch.shared);
1326         kvm_vcpu_uninit(vcpu);
1327         kfree(vcpu_book3s->shadow_vcpu);
1328         vfree(vcpu_book3s);
1329 }
1330
1331 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1332 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1333 {
1334         int ret;
1335         double fpr[32][TS_FPRWIDTH];
1336         unsigned int fpscr;
1337         int fpexc_mode;
1338 #ifdef CONFIG_ALTIVEC
1339         vector128 vr[32];
1340         vector128 vscr;
1341         unsigned long uninitialized_var(vrsave);
1342         int used_vr;
1343 #endif
1344 #ifdef CONFIG_VSX
1345         int used_vsr;
1346 #endif
1347         ulong ext_msr;
1348
1349         /* No need to go into the guest when all we do is going out */
1350         if (signal_pending(current)) {
1351                 kvm_run->exit_reason = KVM_EXIT_INTR;
1352                 return -EINTR;
1353         }
1354
1355         /* Save FPU state in stack */
1356         if (current->thread.regs->msr & MSR_FP)
1357                 giveup_fpu(current);
1358         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1359         fpscr = current->thread.fpscr.val;
1360         fpexc_mode = current->thread.fpexc_mode;
1361
1362 #ifdef CONFIG_ALTIVEC
1363         /* Save Altivec state in stack */
1364         used_vr = current->thread.used_vr;
1365         if (used_vr) {
1366                 if (current->thread.regs->msr & MSR_VEC)
1367                         giveup_altivec(current);
1368                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1369                 vscr = current->thread.vscr;
1370                 vrsave = current->thread.vrsave;
1371         }
1372 #endif
1373
1374 #ifdef CONFIG_VSX
1375         /* Save VSX state in stack */
1376         used_vsr = current->thread.used_vsr;
1377         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1378                         __giveup_vsx(current);
1379 #endif
1380
1381         /* Remember the MSR with disabled extensions */
1382         ext_msr = current->thread.regs->msr;
1383
1384         /* XXX we get called with irq disabled - change that! */
1385         local_irq_enable();
1386
1387         /* Preload FPU if it's enabled */
1388         if (vcpu->arch.shared->msr & MSR_FP)
1389                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1390
1391         ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
1392
1393         local_irq_disable();
1394
1395         current->thread.regs->msr = ext_msr;
1396
1397         /* Make sure we save the guest FPU/Altivec/VSX state */
1398         kvmppc_giveup_ext(vcpu, MSR_FP);
1399         kvmppc_giveup_ext(vcpu, MSR_VEC);
1400         kvmppc_giveup_ext(vcpu, MSR_VSX);
1401
1402         /* Restore FPU state from stack */
1403         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1404         current->thread.fpscr.val = fpscr;
1405         current->thread.fpexc_mode = fpexc_mode;
1406
1407 #ifdef CONFIG_ALTIVEC
1408         /* Restore Altivec state from stack */
1409         if (used_vr && current->thread.used_vr) {
1410                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1411                 current->thread.vscr = vscr;
1412                 current->thread.vrsave = vrsave;
1413         }
1414         current->thread.used_vr = used_vr;
1415 #endif
1416
1417 #ifdef CONFIG_VSX
1418         current->thread.used_vsr = used_vsr;
1419 #endif
1420
1421         return ret;
1422 }
1423
1424 static int kvmppc_book3s_init(void)
1425 {
1426         int r;
1427
1428         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1429                      THIS_MODULE);
1430
1431         if (r)
1432                 return r;
1433
1434         r = kvmppc_mmu_hpte_sysinit();
1435
1436         return r;
1437 }
1438
1439 static void kvmppc_book3s_exit(void)
1440 {
1441         kvmppc_mmu_hpte_sysexit();
1442         kvm_exit();
1443 }
1444
1445 module_init(kvmppc_book3s_init);
1446 module_exit(kvmppc_book3s_exit);