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 pfn_t kvmppc_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 423 {
 424         ulong mp_pa = vcpu->arch.magic_page_pa;
 425
 426         /* Magic page override */
 427         if (unlikely(mp_pa) &&
 428             unlikely(((gfn << PAGE_SHIFT) & KVM_PAM) ==
 429                      ((mp_pa & PAGE_MASK) & KVM_PAM))) {
 430                 ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
 431                 pfn_t pfn;
 432
 433                 pfn = (pfn_t)virt_to_phys((void*)shared_page) >> PAGE_SHIFT;
 434                 get_page(pfn_to_page(pfn));
 435                 return pfn;
 436         }
 437
 438         return gfn_to_pfn(vcpu->kvm, gfn);
 439 }
 440
 441 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 442  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 443  * emulate 32 bytes dcbz length.
 444  *
 445  * The Book3s_64 inventors also realized this case and implemented a special bit
 446  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 447  *
 448  * My approach here is to patch the dcbz instruction on executing pages.
 449  */
 450 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 451 {
 452         struct page *hpage;
 453         u64 hpage_offset;
 454         u32 *page;
 455         int i;
 456
 457         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 458         if (is_error_page(hpage)) {
 459                 kvm_release_page_clean(hpage);
 460                 return;
 461         }
 462
 463         hpage_offset = pte->raddr & ~PAGE_MASK;
 464         hpage_offset &= ~0xFFFULL;
 465         hpage_offset /= 4;
 466
 467         get_page(hpage);
 468         page = kmap_atomic(hpage, KM_USER0);
 469
 470         /* patch dcbz into reserved instruction, so we trap */
 471         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 472                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 473                         page[i] &= 0xfffffff7;
 474
 475         kunmap_atomic(page, KM_USER0);
 476         put_page(hpage);
 477 }
 478
 479 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
 480                          struct kvmppc_pte *pte)
 481 {
 482         int relocated = (vcpu->arch.shared->msr & (data ? MSR_DR : MSR_IR));
 483         int r;
 484
 485         if (relocated) {
 486                 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
 487         } else {
 488                 pte->eaddr = eaddr;
 489                 pte->raddr = eaddr & KVM_PAM;
 490                 pte->vpage = VSID_REAL | eaddr >> 12;
 491                 pte->may_read = true;
 492                 pte->may_write = true;
 493                 pte->may_execute = true;
 494                 r = 0;
 495         }
 496
 497         return r;
 498 }
 499
 500 static hva_t kvmppc_bad_hva(void)
 501 {
 502         return PAGE_OFFSET;
 503 }
 504
 505 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
 506                                bool read)
 507 {
 508         hva_t hpage;
 509
 510         if (read && !pte->may_read)
 511                 goto err;
 512
 513         if (!read && !pte->may_write)
 514                 goto err;
 515
 516         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 517         if (kvm_is_error_hva(hpage))
 518                 goto err;
 519
 520         return hpage | (pte->raddr & ~PAGE_MASK);
 521 err:
 522         return kvmppc_bad_hva();
 523 }
 524
 525 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 526               bool data)
 527 {
 528         struct kvmppc_pte pte;
 529
 530         vcpu->stat.st++;
 531
 532         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 533                 return -ENOENT;
 534
 535         *eaddr = pte.raddr;
 536
 537         if (!pte.may_write)
 538                 return -EPERM;
 539
 540         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 541                 return EMULATE_DO_MMIO;
 542
 543         return EMULATE_DONE;
 544 }
 545
 546 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 547                       bool data)
 548 {
 549         struct kvmppc_pte pte;
 550         hva_t hva = *eaddr;
 551
 552         vcpu->stat.ld++;
 553
 554         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 555                 goto nopte;
 556
 557         *eaddr = pte.raddr;
 558
 559         hva = kvmppc_pte_to_hva(vcpu, &pte, true);
 560         if (kvm_is_error_hva(hva))
 561                 goto mmio;
 562
 563         if (copy_from_user(ptr, (void __user *)hva, size)) {
 564                 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
 565                 goto mmio;
 566         }
 567
 568         return EMULATE_DONE;
 569
 570 nopte:
 571         return -ENOENT;
 572 mmio:
 573         return EMULATE_DO_MMIO;
 574 }
 575
 576 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 577 {
 578         ulong mp_pa = vcpu->arch.magic_page_pa;
 579
 580         if (unlikely(mp_pa) &&
 581             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 582                 return 1;
 583         }
 584
 585         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 586 }
 587
 588 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 589                             ulong eaddr, int vec)
 590 {
 591         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 592         int r = RESUME_GUEST;
 593         int relocated;
 594         int page_found = 0;
 595         struct kvmppc_pte pte;
 596         bool is_mmio = false;
 597         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 598         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 599         u64 vsid;
 600
 601         relocated = data ? dr : ir;
 602
 603         /* Resolve real address if translation turned on */
 604         if (relocated) {
 605                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 606         } else {
 607                 pte.may_execute = true;
 608                 pte.may_read = true;
 609                 pte.may_write = true;
 610                 pte.raddr = eaddr & KVM_PAM;
 611                 pte.eaddr = eaddr;
 612                 pte.vpage = eaddr >> 12;
 613         }
 614
 615         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 616         case 0:
 617                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 618                 break;
 619         case MSR_DR:
 620         case MSR_IR:
 621                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 622
 623                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 624                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 625                 else
 626                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 627                 pte.vpage |= vsid;
 628
 629                 if (vsid == -1)
 630                         page_found = -EINVAL;
 631                 break;
 632         }
 633
 634         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 635            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 636                 /*
 637                  * If we do the dcbz hack, we have to NX on every execution,
 638                  * so we can patch the executing code. This renders our guest
 639                  * NX-less.
 640                  */
 641                 pte.may_execute = !data;
 642         }
 643
 644         if (page_found == -ENOENT) {
 645                 /* Page not found in guest PTE entries */
 646                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 647                 vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 648                 vcpu->arch.shared->msr |=
 649                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 650                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 651         } else if (page_found == -EPERM) {
 652                 /* Storage protection */
 653                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 654                 vcpu->arch.shared->dsisr =
 655                         to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
 656                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 657                 vcpu->arch.shared->msr |=
 658                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 659                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 660         } else if (page_found == -EINVAL) {
 661                 /* Page not found in guest SLB */
 662                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 663                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 664         } else if (!is_mmio &&
 665                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 666                 /* The guest's PTE is not mapped yet. Map on the host */
 667                 kvmppc_mmu_map_page(vcpu, &pte);
 668                 if (data)
 669                         vcpu->stat.sp_storage++;
 670                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 671                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 672                         kvmppc_patch_dcbz(vcpu, &pte);
 673         } else {
 674                 /* MMIO */
 675                 vcpu->stat.mmio_exits++;
 676                 vcpu->arch.paddr_accessed = pte.raddr;
 677                 r = kvmppc_emulate_mmio(run, vcpu);
 678                 if ( r == RESUME_HOST_NV )
 679                         r = RESUME_HOST;
 680         }
 681
 682         return r;
 683 }
 684
 685 static inline int get_fpr_index(int i)
 686 {
 687 #ifdef CONFIG_VSX
 688         i *= 2;
 689 #endif
 690         return i;
 691 }
 692
 693 /* Give up external provider (FPU, Altivec, VSX) */
 694 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 695 {
 696         struct thread_struct *t = &current->thread;
 697         u64 *vcpu_fpr = vcpu->arch.fpr;
 698 #ifdef CONFIG_VSX
 699         u64 *vcpu_vsx = vcpu->arch.vsr;
 700 #endif
 701         u64 *thread_fpr = (u64*)t->fpr;
 702         int i;
 703
 704         if (!(vcpu->arch.guest_owned_ext & msr))
 705                 return;
 706
 707 #ifdef DEBUG_EXT
 708         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 709 #endif
 710
 711         switch (msr) {
 712         case MSR_FP:
 713                 giveup_fpu(current);
 714                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 715                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 716
 717                 vcpu->arch.fpscr = t->fpscr.val;
 718                 break;
 719         case MSR_VEC:
 720 #ifdef CONFIG_ALTIVEC
 721                 giveup_altivec(current);
 722                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 723                 vcpu->arch.vscr = t->vscr;
 724 #endif
 725                 break;
 726         case MSR_VSX:
 727 #ifdef CONFIG_VSX
 728                 __giveup_vsx(current);
 729                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 730                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 731 #endif
 732                 break;
 733         default:
 734                 BUG();
 735         }
 736
 737         vcpu->arch.guest_owned_ext &= ~msr;
 738         current->thread.regs->msr &= ~msr;
 739         kvmppc_recalc_shadow_msr(vcpu);
 740 }
 741
 742 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 743 {
 744         ulong srr0 = kvmppc_get_pc(vcpu);
 745         u32 last_inst = kvmppc_get_last_inst(vcpu);
 746         int ret;
 747
 748         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 749         if (ret == -ENOENT) {
 750                 ulong msr = vcpu->arch.shared->msr;
 751
 752                 msr = kvmppc_set_field(msr, 33, 33, 1);
 753                 msr = kvmppc_set_field(msr, 34, 36, 0);
 754                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 755                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 756                 return EMULATE_AGAIN;
 757         }
 758
 759         return EMULATE_DONE;
 760 }
 761
 762 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 763 {
 764
 765         /* Need to do paired single emulation? */
 766         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 767                 return EMULATE_DONE;
 768
 769         /* Read out the instruction */
 770         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 771                 /* Need to emulate */
 772                 return EMULATE_FAIL;
 773
 774         return EMULATE_AGAIN;
 775 }
 776
 777 /* Handle external providers (FPU, Altivec, VSX) */
 778 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 779                              ulong msr)
 780 {
 781         struct thread_struct *t = &current->thread;
 782         u64 *vcpu_fpr = vcpu->arch.fpr;
 783 #ifdef CONFIG_VSX
 784         u64 *vcpu_vsx = vcpu->arch.vsr;
 785 #endif
 786         u64 *thread_fpr = (u64*)t->fpr;
 787         int i;
 788
 789         /* When we have paired singles, we emulate in software */
 790         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 791                 return RESUME_GUEST;
 792
 793         if (!(vcpu->arch.shared->msr & msr)) {
 794                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 795                 return RESUME_GUEST;
 796         }
 797
 798         /* We already own the ext */
 799         if (vcpu->arch.guest_owned_ext & msr) {
 800                 return RESUME_GUEST;
 801         }
 802
 803 #ifdef DEBUG_EXT
 804         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 805 #endif
 806
 807         current->thread.regs->msr |= msr;
 808
 809         switch (msr) {
 810         case MSR_FP:
 811                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 812                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 813
 814                 t->fpscr.val = vcpu->arch.fpscr;
 815                 t->fpexc_mode = 0;
 816                 kvmppc_load_up_fpu();
 817                 break;
 818         case MSR_VEC:
 819 #ifdef CONFIG_ALTIVEC
 820                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 821                 t->vscr = vcpu->arch.vscr;
 822                 t->vrsave = -1;
 823                 kvmppc_load_up_altivec();
 824 #endif
 825                 break;
 826         case MSR_VSX:
 827 #ifdef CONFIG_VSX
 828                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 829                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 830                 kvmppc_load_up_vsx();
 831 #endif
 832                 break;
 833         default:
 834                 BUG();
 835         }
 836
 837         vcpu->arch.guest_owned_ext |= msr;
 838
 839         kvmppc_recalc_shadow_msr(vcpu);
 840
 841         return RESUME_GUEST;
 842 }
 843
 844 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 845                        unsigned int exit_nr)
 846 {
 847         int r = RESUME_HOST;
 848
 849         vcpu->stat.sum_exits++;
 850
 851         run->exit_reason = KVM_EXIT_UNKNOWN;
 852         run->ready_for_interrupt_injection = 1;
 853 #ifdef EXIT_DEBUG
 854         printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | dec=0x%x | msr=0x%lx\n",
 855                 exit_nr, kvmppc_get_pc(vcpu), kvmppc_get_fault_dar(vcpu),
 856                 kvmppc_get_dec(vcpu), to_svcpu(vcpu)->shadow_srr1);
 857 #elif defined (EXIT_DEBUG_SIMPLE)
 858         if ((exit_nr != 0x900) && (exit_nr != 0x500))
 859                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | msr=0x%lx\n",
 860                         exit_nr, kvmppc_get_pc(vcpu), kvmppc_get_fault_dar(vcpu),
 861                         vcpu->arch.shared->msr);
 862 #endif
 863         kvm_resched(vcpu);
 864         switch (exit_nr) {
 865         case BOOK3S_INTERRUPT_INST_STORAGE:
 866                 vcpu->stat.pf_instruc++;
 867
 868 #ifdef CONFIG_PPC_BOOK3S_32
 869                 /* We set segments as unused segments when invalidating them. So
 870                  * treat the respective fault as segment fault. */
 871                 if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
 872                     == SR_INVALID) {
 873                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 874                         r = RESUME_GUEST;
 875                         break;
 876                 }
 877 #endif
 878
 879                 /* only care about PTEG not found errors, but leave NX alone */
 880                 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
 881                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 882                         vcpu->stat.sp_instruc++;
 883                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 884                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 885                         /*
 886                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 887                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 888                          *     that no guest that needs the dcbz hack does NX.
 889                          */
 890                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 891                         r = RESUME_GUEST;
 892                 } else {
 893                         vcpu->arch.shared->msr |=
 894                                 to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
 895                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 896                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 897                         r = RESUME_GUEST;
 898                 }
 899                 break;
 900         case BOOK3S_INTERRUPT_DATA_STORAGE:
 901         {
 902                 ulong dar = kvmppc_get_fault_dar(vcpu);
 903                 vcpu->stat.pf_storage++;
 904
 905 #ifdef CONFIG_PPC_BOOK3S_32
 906                 /* We set segments as unused segments when invalidating them. So
 907                  * treat the respective fault as segment fault. */
 908                 if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 909                         kvmppc_mmu_map_segment(vcpu, dar);
 910                         r = RESUME_GUEST;
 911                         break;
 912                 }
 913 #endif
 914
 915                 /* The only case we need to handle is missing shadow PTEs */
 916                 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
 917                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 918                 } else {
 919                         vcpu->arch.shared->dar = dar;
 920                         vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 921                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 922                         kvmppc_mmu_pte_flush(vcpu, dar, ~0xFFFUL);
 923                         r = RESUME_GUEST;
 924                 }
 925                 break;
 926         }
 927         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 928                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 929                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 930                         kvmppc_book3s_queue_irqprio(vcpu,
 931                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 932                 }
 933                 r = RESUME_GUEST;
 934                 break;
 935         case BOOK3S_INTERRUPT_INST_SEGMENT:
 936                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 937                         kvmppc_book3s_queue_irqprio(vcpu,
 938                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 939                 }
 940                 r = RESUME_GUEST;
 941                 break;
 942         /* We're good on these - the host merely wanted to get our attention */
 943         case BOOK3S_INTERRUPT_DECREMENTER:
 944                 vcpu->stat.dec_exits++;
 945                 r = RESUME_GUEST;
 946                 break;
 947         case BOOK3S_INTERRUPT_EXTERNAL:
 948                 vcpu->stat.ext_intr_exits++;
 949                 r = RESUME_GUEST;
 950                 break;
 951         case BOOK3S_INTERRUPT_PERFMON:
 952                 r = RESUME_GUEST;
 953                 break;
 954         case BOOK3S_INTERRUPT_PROGRAM:
 955         {
 956                 enum emulation_result er;
 957                 ulong flags;
 958
 959 program_interrupt:
 960                 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
 961
 962                 if (vcpu->arch.shared->msr & MSR_PR) {
 963 #ifdef EXIT_DEBUG
 964                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 965 #endif
 966                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 967                             (INS_DCBZ & 0xfffffff7)) {
 968                                 kvmppc_core_queue_program(vcpu, flags);
 969                                 r = RESUME_GUEST;
 970                                 break;
 971                         }
 972                 }
 973
 974                 vcpu->stat.emulated_inst_exits++;
 975                 er = kvmppc_emulate_instruction(run, vcpu);
 976                 switch (er) {
 977                 case EMULATE_DONE:
 978                         r = RESUME_GUEST_NV;
 979                         break;
 980                 case EMULATE_AGAIN:
 981                         r = RESUME_GUEST;
 982                         break;
 983                 case EMULATE_FAIL:
 984                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 985                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 986                         kvmppc_core_queue_program(vcpu, flags);
 987                         r = RESUME_GUEST;
 988                         break;
 989                 case EMULATE_DO_MMIO:
 990                         run->exit_reason = KVM_EXIT_MMIO;
 991                         r = RESUME_HOST_NV;
 992                         break;
 993                 default:
 994                         BUG();
 995                 }
 996                 break;
 997         }
 998         case BOOK3S_INTERRUPT_SYSCALL:
 999                 if (vcpu->arch.osi_enabled &&
1000                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1001                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1002                         /* MOL hypercalls */
1003                         u64 *gprs = run->osi.gprs;
1004                         int i;
1005
1006                         run->exit_reason = KVM_EXIT_OSI;
1007                         for (i = 0; i < 32; i++)
1008                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
1009                         vcpu->arch.osi_needed = 1;
1010                         r = RESUME_HOST_NV;
1011                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
1012                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1013                         /* KVM PV hypercalls */
1014                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1015                         r = RESUME_GUEST;
1016                 } else {
1017                         /* Guest syscalls */
1018                         vcpu->stat.syscall_exits++;
1019                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1020                         r = RESUME_GUEST;
1021                 }
1022                 break;
1023         case BOOK3S_INTERRUPT_FP_UNAVAIL:
1024         case BOOK3S_INTERRUPT_ALTIVEC:
1025         case BOOK3S_INTERRUPT_VSX:
1026         {
1027                 int ext_msr = 0;
1028
1029                 switch (exit_nr) {
1030                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
1031                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
1032                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
1033                 }
1034
1035                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
1036                 case EMULATE_DONE:
1037                         /* everything ok - let's enable the ext */
1038                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1039                         break;
1040                 case EMULATE_FAIL:
1041                         /* we need to emulate this instruction */
1042                         goto program_interrupt;
1043                         break;
1044                 default:
1045                         /* nothing to worry about - go again */
1046                         break;
1047                 }
1048                 break;
1049         }
1050         case BOOK3S_INTERRUPT_ALIGNMENT:
1051                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
1052                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
1053                                 kvmppc_get_last_inst(vcpu));
1054                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
1055                                 kvmppc_get_last_inst(vcpu));
1056                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1057                 }
1058                 r = RESUME_GUEST;
1059                 break;
1060         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1061         case BOOK3S_INTERRUPT_TRACE:
1062                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1063                 r = RESUME_GUEST;
1064                 break;
1065         default:
1066                 /* Ugh - bork here! What did we get? */
1067                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1068                         exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
1069                 r = RESUME_HOST;
1070                 BUG();
1071                 break;
1072         }
1073
1074
1075         if (!(r & RESUME_HOST)) {
1076                 /* To avoid clobbering exit_reason, only check for signals if
1077                  * we aren't already exiting to userspace for some other
1078                  * reason. */
1079                 if (signal_pending(current)) {
1080 #ifdef EXIT_DEBUG
1081                         printk(KERN_EMERG "KVM: Going back to host\n");
1082 #endif
1083                         vcpu->stat.signal_exits++;
1084                         run->exit_reason = KVM_EXIT_INTR;
1085                         r = -EINTR;
1086                 } else {
1087                         /* In case an interrupt came in that was triggered
1088                          * from userspace (like DEC), we need to check what
1089                          * to inject now! */
1090                         kvmppc_core_deliver_interrupts(vcpu);
1091                 }
1092         }
1093
1094 #ifdef EXIT_DEBUG
1095         printk(KERN_EMERG "KVM exit: vcpu=0x%p pc=0x%lx r=0x%x\n", vcpu, kvmppc_get_pc(vcpu), r);
1096 #endif
1097
1098         return r;
1099 }
1100
1101 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1102 {
1103         return 0;
1104 }
1105
1106 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1107 {
1108         int i;
1109
1110         regs->pc = kvmppc_get_pc(vcpu);
1111         regs->cr = kvmppc_get_cr(vcpu);
1112         regs->ctr = kvmppc_get_ctr(vcpu);
1113         regs->lr = kvmppc_get_lr(vcpu);
1114         regs->xer = kvmppc_get_xer(vcpu);
1115         regs->msr = vcpu->arch.shared->msr;
1116         regs->srr0 = vcpu->arch.shared->srr0;
1117         regs->srr1 = vcpu->arch.shared->srr1;
1118         regs->pid = vcpu->arch.pid;
1119         regs->sprg0 = vcpu->arch.shared->sprg0;
1120         regs->sprg1 = vcpu->arch.shared->sprg1;
1121         regs->sprg2 = vcpu->arch.shared->sprg2;
1122         regs->sprg3 = vcpu->arch.shared->sprg3;
1123         regs->sprg5 = vcpu->arch.sprg4;
1124         regs->sprg6 = vcpu->arch.sprg5;
1125         regs->sprg7 = vcpu->arch.sprg6;
1126
1127         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1128                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
1129
1130         return 0;
1131 }
1132
1133 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1134 {
1135         int i;
1136
1137         kvmppc_set_pc(vcpu, regs->pc);
1138         kvmppc_set_cr(vcpu, regs->cr);
1139         kvmppc_set_ctr(vcpu, regs->ctr);
1140         kvmppc_set_lr(vcpu, regs->lr);
1141         kvmppc_set_xer(vcpu, regs->xer);
1142         kvmppc_set_msr(vcpu, regs->msr);
1143         vcpu->arch.shared->srr0 = regs->srr0;
1144         vcpu->arch.shared->srr1 = regs->srr1;
1145         vcpu->arch.shared->sprg0 = regs->sprg0;
1146         vcpu->arch.shared->sprg1 = regs->sprg1;
1147         vcpu->arch.shared->sprg2 = regs->sprg2;
1148         vcpu->arch.shared->sprg3 = regs->sprg3;
1149         vcpu->arch.sprg5 = regs->sprg4;
1150         vcpu->arch.sprg6 = regs->sprg5;
1151         vcpu->arch.sprg7 = regs->sprg6;
1152
1153         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1154                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
1155
1156         return 0;
1157 }
1158
1159 int kvm_arch_vcpu_ioctl_get_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         sregs->pvr = vcpu->arch.pvr;
1166
1167         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1168         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1169                 for (i = 0; i < 64; i++) {
1170                         sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
1171                         sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
1172                 }
1173         } else {
1174                 for (i = 0; i < 16; i++) {
1175                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
1176                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
1177                 }
1178                 for (i = 0; i < 8; i++) {
1179                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1180                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1181                 }
1182         }
1183
1184         return 0;
1185 }
1186
1187 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1188                                   struct kvm_sregs *sregs)
1189 {
1190         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1191         int i;
1192
1193         kvmppc_set_pvr(vcpu, sregs->pvr);
1194
1195         vcpu3s->sdr1 = sregs->u.s.sdr1;
1196         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1197                 for (i = 0; i < 64; i++) {
1198                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1199                                                     sregs->u.s.ppc64.slb[i].slbe);
1200                 }
1201         } else {
1202                 for (i = 0; i < 16; i++) {
1203                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1204                 }
1205                 for (i = 0; i < 8; i++) {
1206                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1207                                        (u32)sregs->u.s.ppc32.ibat[i]);
1208                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1209                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1210                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1211                                        (u32)sregs->u.s.ppc32.dbat[i]);
1212                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1213                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1214                 }
1215         }
1216
1217         /* Flush the MMU after messing with the segments */
1218         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1219
1220         return 0;
1221 }
1222
1223 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1224 {
1225         return -ENOTSUPP;
1226 }
1227
1228 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1229 {
1230         return -ENOTSUPP;
1231 }
1232
1233 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1234                                   struct kvm_translation *tr)
1235 {
1236         return 0;
1237 }
1238
1239 /*
1240  * Get (and clear) the dirty memory log for a memory slot.
1241  */
1242 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1243                                       struct kvm_dirty_log *log)
1244 {
1245         struct kvm_memory_slot *memslot;
1246         struct kvm_vcpu *vcpu;
1247         ulong ga, ga_end;
1248         int is_dirty = 0;
1249         int r;
1250         unsigned long n;
1251
1252         mutex_lock(&kvm->slots_lock);
1253
1254         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1255         if (r)
1256                 goto out;
1257
1258         /* If nothing is dirty, don't bother messing with page tables. */
1259         if (is_dirty) {
1260                 memslot = &kvm->memslots->memslots[log->slot];
1261
1262                 ga = memslot->base_gfn << PAGE_SHIFT;
1263                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1264
1265                 kvm_for_each_vcpu(n, vcpu, kvm)
1266                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1267
1268                 n = kvm_dirty_bitmap_bytes(memslot);
1269                 memset(memslot->dirty_bitmap, 0, n);
1270         }
1271
1272         r = 0;
1273 out:
1274         mutex_unlock(&kvm->slots_lock);
1275         return r;
1276 }
1277
1278 int kvmppc_core_check_processor_compat(void)
1279 {
1280         return 0;
1281 }
1282
1283 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1284 {
1285         struct kvmppc_vcpu_book3s *vcpu_book3s;
1286         struct kvm_vcpu *vcpu;
1287         int err = -ENOMEM;
1288         unsigned long p;
1289
1290         vcpu_book3s = vmalloc(sizeof(struct kvmppc_vcpu_book3s));
1291         if (!vcpu_book3s)
1292                 goto out;
1293
1294         memset(vcpu_book3s, 0, sizeof(struct kvmppc_vcpu_book3s));
1295
1296         vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
1297                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
1298         if (!vcpu_book3s->shadow_vcpu)
1299                 goto free_vcpu;
1300
1301         vcpu = &vcpu_book3s->vcpu;
1302         err = kvm_vcpu_init(vcpu, kvm, id);
1303         if (err)
1304                 goto free_shadow_vcpu;
1305
1306         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1307         /* the real shared page fills the last 4k of our page */
1308         vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
1309         if (!p)
1310                 goto uninit_vcpu;
1311
1312         vcpu->arch.host_retip = kvm_return_point;
1313         vcpu->arch.host_msr = mfmsr();
1314 #ifdef CONFIG_PPC_BOOK3S_64
1315         /* default to book3s_64 (970fx) */
1316         vcpu->arch.pvr = 0x3C0301;
1317 #else
1318         /* default to book3s_32 (750) */
1319         vcpu->arch.pvr = 0x84202;
1320 #endif
1321         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1322         vcpu_book3s->slb_nr = 64;
1323
1324         /* remember where some real-mode handlers are */
1325         vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
1326         vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
1327         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
1328 #ifdef CONFIG_PPC_BOOK3S_64
1329         vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
1330 #else
1331         vcpu->arch.rmcall = (ulong)kvmppc_rmcall;
1332 #endif
1333
1334         vcpu->arch.shadow_msr = MSR_USER64;
1335
1336         err = kvmppc_mmu_init(vcpu);
1337         if (err < 0)
1338                 goto uninit_vcpu;
1339
1340         return vcpu;
1341
1342 uninit_vcpu:
1343         kvm_vcpu_uninit(vcpu);
1344 free_shadow_vcpu:
1345         kfree(vcpu_book3s->shadow_vcpu);
1346 free_vcpu:
1347         vfree(vcpu_book3s);
1348 out:
1349         return ERR_PTR(err);
1350 }
1351
1352 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1353 {
1354         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1355
1356         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1357         kvm_vcpu_uninit(vcpu);
1358         kfree(vcpu_book3s->shadow_vcpu);
1359         vfree(vcpu_book3s);
1360 }
1361
1362 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1363 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1364 {
1365         int ret;
1366         double fpr[32][TS_FPRWIDTH];
1367         unsigned int fpscr;
1368         int fpexc_mode;
1369 #ifdef CONFIG_ALTIVEC
1370         vector128 vr[32];
1371         vector128 vscr;
1372         unsigned long uninitialized_var(vrsave);
1373         int used_vr;
1374 #endif
1375 #ifdef CONFIG_VSX
1376         int used_vsr;
1377 #endif
1378         ulong ext_msr;
1379
1380         /* No need to go into the guest when all we do is going out */
1381         if (signal_pending(current)) {
1382                 kvm_run->exit_reason = KVM_EXIT_INTR;
1383                 return -EINTR;
1384         }
1385
1386         /* Save FPU state in stack */
1387         if (current->thread.regs->msr & MSR_FP)
1388                 giveup_fpu(current);
1389         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1390         fpscr = current->thread.fpscr.val;
1391         fpexc_mode = current->thread.fpexc_mode;
1392
1393 #ifdef CONFIG_ALTIVEC
1394         /* Save Altivec state in stack */
1395         used_vr = current->thread.used_vr;
1396         if (used_vr) {
1397                 if (current->thread.regs->msr & MSR_VEC)
1398                         giveup_altivec(current);
1399                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1400                 vscr = current->thread.vscr;
1401                 vrsave = current->thread.vrsave;
1402         }
1403 #endif
1404
1405 #ifdef CONFIG_VSX
1406         /* Save VSX state in stack */
1407         used_vsr = current->thread.used_vsr;
1408         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1409                         __giveup_vsx(current);
1410 #endif
1411
1412         /* Remember the MSR with disabled extensions */
1413         ext_msr = current->thread.regs->msr;
1414
1415         /* XXX we get called with irq disabled - change that! */
1416         local_irq_enable();
1417
1418         /* Preload FPU if it's enabled */
1419         if (vcpu->arch.shared->msr & MSR_FP)
1420                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1421
1422         ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
1423
1424         local_irq_disable();
1425
1426         current->thread.regs->msr = ext_msr;
1427
1428         /* Make sure we save the guest FPU/Altivec/VSX state */
1429         kvmppc_giveup_ext(vcpu, MSR_FP);
1430         kvmppc_giveup_ext(vcpu, MSR_VEC);
1431         kvmppc_giveup_ext(vcpu, MSR_VSX);
1432
1433         /* Restore FPU state from stack */
1434         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1435         current->thread.fpscr.val = fpscr;
1436         current->thread.fpexc_mode = fpexc_mode;
1437
1438 #ifdef CONFIG_ALTIVEC
1439         /* Restore Altivec state from stack */
1440         if (used_vr && current->thread.used_vr) {
1441                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1442                 current->thread.vscr = vscr;
1443                 current->thread.vrsave = vrsave;
1444         }
1445         current->thread.used_vr = used_vr;
1446 #endif
1447
1448 #ifdef CONFIG_VSX
1449         current->thread.used_vsr = used_vsr;
1450 #endif
1451
1452         return ret;
1453 }
1454
1455 static int kvmppc_book3s_init(void)
1456 {
1457         int r;
1458
1459         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1460                      THIS_MODULE);
1461
1462         if (r)
1463                 return r;
1464
1465         r = kvmppc_mmu_hpte_sysinit();
1466
1467         return r;
1468 }
1469
1470 static void kvmppc_book3s_exit(void)
1471 {
1472         kvmppc_mmu_hpte_sysexit();
1473         kvm_exit();
1474 }
1475
1476 module_init(kvmppc_book3s_init);
1477 module_exit(kvmppc_book3s_exit);