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