Pileus Git - ~andy/linux/blob - arch/powerpc/kvm/book3s.c

   1 /*
   2  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   3  *
   4  * Authors:
   5  *    Alexander Graf <agraf@suse.de>
   6  *    Kevin Wolf <mail@kevin-wolf.de>
   7  *
   8  * Description:
   9  * This file is derived from arch/powerpc/kvm/44x.c,
  10  * by Hollis Blanchard <hollisb@us.ibm.com>.
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License, version 2, as
  14  * published by the Free Software Foundation.
  15  */
  16
  17 #include <linux/kvm_host.h>
  18 #include <linux/err.h>
  19 #include <linux/slab.h>
  20
  21 #include <asm/reg.h>
  22 #include <asm/cputable.h>
  23 #include <asm/cacheflush.h>
  24 #include <asm/tlbflush.h>
  25 #include <asm/uaccess.h>
  26 #include <asm/io.h>
  27 #include <asm/kvm_ppc.h>
  28 #include <asm/kvm_book3s.h>
  29 #include <asm/mmu_context.h>
  30 #include <linux/gfp.h>
  31 #include <linux/sched.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/highmem.h>
  34
  35 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  36
  37 /* #define EXIT_DEBUG */
  38 /* #define EXIT_DEBUG_SIMPLE */
  39 /* #define DEBUG_EXT */
  40
  41 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  42                              ulong msr);
  43
  44 /* Some compatibility defines */
  45 #ifdef CONFIG_PPC_BOOK3S_32
  46 #define MSR_USER32 MSR_USER
  47 #define MSR_USER64 MSR_USER
  48 #define HW_PAGE_SIZE PAGE_SIZE
  49 #endif
  50
  51 struct kvm_stats_debugfs_item debugfs_entries[] = {
  52         { "exits",       VCPU_STAT(sum_exits) },
  53         { "mmio",        VCPU_STAT(mmio_exits) },
  54         { "sig",         VCPU_STAT(signal_exits) },
  55         { "sysc",        VCPU_STAT(syscall_exits) },
  56         { "inst_emu",    VCPU_STAT(emulated_inst_exits) },
  57         { "dec",         VCPU_STAT(dec_exits) },
  58         { "ext_intr",    VCPU_STAT(ext_intr_exits) },
  59         { "queue_intr",  VCPU_STAT(queue_intr) },
  60         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  61         { "pf_storage",  VCPU_STAT(pf_storage) },
  62         { "sp_storage",  VCPU_STAT(sp_storage) },
  63         { "pf_instruc",  VCPU_STAT(pf_instruc) },
  64         { "sp_instruc",  VCPU_STAT(sp_instruc) },
  65         { "ld",          VCPU_STAT(ld) },
  66         { "ld_slow",     VCPU_STAT(ld_slow) },
  67         { "st",          VCPU_STAT(st) },
  68         { "st_slow",     VCPU_STAT(st_slow) },
  69         { NULL }
  70 };
  71
  72 void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
  73 {
  74 }
  75
  76 void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
  77 {
  78 }
  79
  80 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  81 {
  82 #ifdef CONFIG_PPC_BOOK3S_64
  83         memcpy(to_svcpu(vcpu)->slb, to_book3s(vcpu)->slb_shadow, sizeof(to_svcpu(vcpu)->slb));
  84         memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
  85                sizeof(get_paca()->shadow_vcpu));
  86         to_svcpu(vcpu)->slb_max = to_book3s(vcpu)->slb_shadow_max;
  87 #endif
  88
  89 #ifdef CONFIG_PPC_BOOK3S_32
  90         current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
  91 #endif
  92 }
  93
  94 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  95 {
  96 #ifdef CONFIG_PPC_BOOK3S_64
  97         memcpy(to_book3s(vcpu)->slb_shadow, to_svcpu(vcpu)->slb, sizeof(to_svcpu(vcpu)->slb));
  98         memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  99                sizeof(get_paca()->shadow_vcpu));
 100         to_book3s(vcpu)->slb_shadow_max = to_svcpu(vcpu)->slb_max;
 101 #endif
 102
 103         kvmppc_giveup_ext(vcpu, MSR_FP);
 104         kvmppc_giveup_ext(vcpu, MSR_VEC);
 105         kvmppc_giveup_ext(vcpu, MSR_VSX);
 106 }
 107
 108 #if defined(EXIT_DEBUG)
 109 static u32 kvmppc_get_dec(struct kvm_vcpu *vcpu)
 110 {
 111         u64 jd = mftb() - vcpu->arch.dec_jiffies;
 112         return vcpu->arch.dec - jd;
 113 }
 114 #endif
 115
 116 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 117 {
 118         ulong smsr = vcpu->arch.shared->msr;
 119
 120         /* Guest MSR values */
 121         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE;
 122         /* Process MSR values */
 123         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 124         /* External providers the guest reserved */
 125         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
 126         /* 64-bit Process MSR values */
 127 #ifdef CONFIG_PPC_BOOK3S_64
 128         smsr |= MSR_ISF | MSR_HV;
 129 #endif
 130         vcpu->arch.shadow_msr = smsr;
 131 }
 132
 133 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 134 {
 135         ulong old_msr = vcpu->arch.shared->msr;
 136
 137 #ifdef EXIT_DEBUG
 138         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 139 #endif
 140
 141         msr &= to_book3s(vcpu)->msr_mask;
 142         vcpu->arch.shared->msr = msr;
 143         kvmppc_recalc_shadow_msr(vcpu);
 144
 145         if (msr & (MSR_WE|MSR_POW)) {
 146                 if (!vcpu->arch.pending_exceptions) {
 147                         kvm_vcpu_block(vcpu);
 148                         vcpu->stat.halt_wakeup++;
 149                 }
 150         }
 151
 152         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 153                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 154                 kvmppc_mmu_flush_segments(vcpu);
 155                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 156         }
 157
 158         /* Preload FPU if it's enabled */
 159         if (vcpu->arch.shared->msr & MSR_FP)
 160                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 161 }
 162
 163 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
 164 {
 165         vcpu->arch.shared->srr0 = kvmppc_get_pc(vcpu);
 166         vcpu->arch.shared->srr1 = vcpu->arch.shared->msr | flags;
 167         kvmppc_set_pc(vcpu, to_book3s(vcpu)->hior + vec);
 168         vcpu->arch.mmu.reset_msr(vcpu);
 169 }
 170
 171 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
 172 {
 173         unsigned int prio;
 174
 175         switch (vec) {
 176         case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET;         break;
 177         case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK;        break;
 178         case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE;         break;
 179         case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT;         break;
 180         case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE;         break;
 181         case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT;         break;
 182         case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL;             break;
 183         case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT;            break;
 184         case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM;              break;
 185         case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL;           break;
 186         case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER;          break;
 187         case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL;              break;
 188         case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG;                break;
 189         case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC;              break;
 190         case 0xf40: prio = BOOK3S_IRQPRIO_VSX;                  break;
 191         default:    prio = BOOK3S_IRQPRIO_MAX;                  break;
 192         }
 193
 194         return prio;
 195 }
 196
 197 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
 198                                           unsigned int vec)
 199 {
 200         clear_bit(kvmppc_book3s_vec2irqprio(vec),
 201                   &vcpu->arch.pending_exceptions);
 202 }
 203
 204 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
 205 {
 206         vcpu->stat.queue_intr++;
 207
 208         set_bit(kvmppc_book3s_vec2irqprio(vec),
 209                 &vcpu->arch.pending_exceptions);
 210 #ifdef EXIT_DEBUG
 211         printk(KERN_INFO "Queueing interrupt %x\n", vec);
 212 #endif
 213 }
 214
 215
 216 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong flags)
 217 {
 218         to_book3s(vcpu)->prog_flags = flags;
 219         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
 220 }
 221
 222 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 223 {
 224         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 225 }
 226
 227 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 228 {
 229         return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
 230 }
 231
 232 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 233 {
 234         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 235 }
 236
 237 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 238                                 struct kvm_interrupt *irq)
 239 {
 240         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 241 }
 242
 243 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
 244                                   struct kvm_interrupt *irq)
 245 {
 246         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 247 }
 248
 249 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
 250 {
 251         int deliver = 1;
 252         int vec = 0;
 253         ulong flags = 0ULL;
 254         ulong crit_raw = vcpu->arch.shared->critical;
 255         ulong crit_r1 = kvmppc_get_gpr(vcpu, 1);
 256         bool crit;
 257
 258         /* Truncate crit indicators in 32 bit mode */
 259         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 260                 crit_raw &= 0xffffffff;
 261                 crit_r1 &= 0xffffffff;
 262         }
 263
 264         /* Critical section when crit == r1 */
 265         crit = (crit_raw == crit_r1);
 266         /* ... and we're in supervisor mode */
 267         crit = crit && !(vcpu->arch.shared->msr & MSR_PR);
 268
 269         switch (priority) {
 270         case BOOK3S_IRQPRIO_DECREMENTER:
 271                 deliver = (vcpu->arch.shared->msr & MSR_EE) && !crit;
 272                 vec = BOOK3S_INTERRUPT_DECREMENTER;
 273                 break;
 274         case BOOK3S_IRQPRIO_EXTERNAL:
 275                 deliver = (vcpu->arch.shared->msr & MSR_EE) && !crit;
 276                 vec = BOOK3S_INTERRUPT_EXTERNAL;
 277                 break;
 278         case BOOK3S_IRQPRIO_SYSTEM_RESET:
 279                 vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
 280                 break;
 281         case BOOK3S_IRQPRIO_MACHINE_CHECK:
 282                 vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
 283                 break;
 284         case BOOK3S_IRQPRIO_DATA_STORAGE:
 285                 vec = BOOK3S_INTERRUPT_DATA_STORAGE;
 286                 break;
 287         case BOOK3S_IRQPRIO_INST_STORAGE:
 288                 vec = BOOK3S_INTERRUPT_INST_STORAGE;
 289                 break;
 290         case BOOK3S_IRQPRIO_DATA_SEGMENT:
 291                 vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
 292                 break;
 293         case BOOK3S_IRQPRIO_INST_SEGMENT:
 294                 vec = BOOK3S_INTERRUPT_INST_SEGMENT;
 295                 break;
 296         case BOOK3S_IRQPRIO_ALIGNMENT:
 297                 vec = BOOK3S_INTERRUPT_ALIGNMENT;
 298                 break;
 299         case BOOK3S_IRQPRIO_PROGRAM:
 300                 vec = BOOK3S_INTERRUPT_PROGRAM;
 301                 flags = to_book3s(vcpu)->prog_flags;
 302                 break;
 303         case BOOK3S_IRQPRIO_VSX:
 304                 vec = BOOK3S_INTERRUPT_VSX;
 305                 break;
 306         case BOOK3S_IRQPRIO_ALTIVEC:
 307                 vec = BOOK3S_INTERRUPT_ALTIVEC;
 308                 break;
 309         case BOOK3S_IRQPRIO_FP_UNAVAIL:
 310                 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
 311                 break;
 312         case BOOK3S_IRQPRIO_SYSCALL:
 313                 vec = BOOK3S_INTERRUPT_SYSCALL;
 314                 break;
 315         case BOOK3S_IRQPRIO_DEBUG:
 316                 vec = BOOK3S_INTERRUPT_TRACE;
 317                 break;
 318         case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
 319                 vec = BOOK3S_INTERRUPT_PERFMON;
 320                 break;
 321         default:
 322                 deliver = 0;
 323                 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
 324                 break;
 325         }
 326
 327 #if 0
 328         printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
 329 #endif
 330
 331         if (deliver)
 332                 kvmppc_inject_interrupt(vcpu, vec, flags);
 333
 334         return deliver;
 335 }
 336
 337 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 338 {
 339         unsigned long *pending = &vcpu->arch.pending_exceptions;
 340         unsigned int priority;
 341
 342 #ifdef EXIT_DEBUG
 343         if (vcpu->arch.pending_exceptions)
 344                 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
 345 #endif
 346         priority = __ffs(*pending);
 347         while (priority < BOOK3S_IRQPRIO_MAX) {
 348                 if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
 349                     (priority != BOOK3S_IRQPRIO_DECREMENTER)) {
 350                         /* DEC interrupts get cleared by mtdec */
 351                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 352                         break;
 353                 }
 354
 355                 priority = find_next_bit(pending,
 356                                          BITS_PER_BYTE * sizeof(*pending),
 357                                          priority + 1);
 358         }
 359 }
 360
 361 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 362 {
 363         u32 host_pvr;
 364
 365         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 366         vcpu->arch.pvr = pvr;
 367 #ifdef CONFIG_PPC_BOOK3S_64
 368         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 369                 kvmppc_mmu_book3s_64_init(vcpu);
 370                 to_book3s(vcpu)->hior = 0xfff00000;
 371                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 372         } else
 373 #endif
 374         {
 375                 kvmppc_mmu_book3s_32_init(vcpu);
 376                 to_book3s(vcpu)->hior = 0;
 377                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 378         }
 379
 380         /* If we are in hypervisor level on 970, we can tell the CPU to
 381          * treat DCBZ as 32 bytes store */
 382         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 383         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 384             !strcmp(cur_cpu_spec->platform, "ppc970"))
 385                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 386
 387         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 388            really needs them in a VM on Cell and force disable them. */
 389         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 390                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 391
 392 #ifdef CONFIG_PPC_BOOK3S_32
 393         /* 32 bit Book3S always has 32 byte dcbz */
 394         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 395 #endif
 396
 397         /* On some CPUs we can execute paired single operations natively */
 398         asm ( "mfpvr %0" : "=r"(host_pvr));
 399         switch (host_pvr) {
 400         case 0x00080200:        /* lonestar 2.0 */
 401         case 0x00088202:        /* lonestar 2.2 */
 402         case 0x70000100:        /* gekko 1.0 */
 403         case 0x00080100:        /* gekko 2.0 */
 404         case 0x00083203:        /* gekko 2.3a */
 405         case 0x00083213:        /* gekko 2.3b */
 406         case 0x00083204:        /* gekko 2.4 */
 407         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 408         case 0x00087200:        /* broadway */
 409                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 410                 /* Enable HID2.PSE - in case we need it later */
 411                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 412         }
 413 }
 414
 415 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 416  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 417  * emulate 32 bytes dcbz length.
 418  *
 419  * The Book3s_64 inventors also realized this case and implemented a special bit
 420  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 421  *
 422  * My approach here is to patch the dcbz instruction on executing pages.
 423  */
 424 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 425 {
 426         struct page *hpage;
 427         u64 hpage_offset;
 428         u32 *page;
 429         int i;
 430
 431         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 432         if (is_error_page(hpage))
 433                 return;
 434
 435         hpage_offset = pte->raddr & ~PAGE_MASK;
 436         hpage_offset &= ~0xFFFULL;
 437         hpage_offset /= 4;
 438
 439         get_page(hpage);
 440         page = kmap_atomic(hpage, KM_USER0);
 441
 442         /* patch dcbz into reserved instruction, so we trap */
 443         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 444                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 445                         page[i] &= 0xfffffff7;
 446
 447         kunmap_atomic(page, KM_USER0);
 448         put_page(hpage);
 449 }
 450
 451 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
 452                          struct kvmppc_pte *pte)
 453 {
 454         int relocated = (vcpu->arch.shared->msr & (data ? MSR_DR : MSR_IR));
 455         int r;
 456
 457         if (relocated) {
 458                 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
 459         } else {
 460                 pte->eaddr = eaddr;
 461                 pte->raddr = eaddr & 0xffffffff;
 462                 pte->vpage = VSID_REAL | eaddr >> 12;
 463                 pte->may_read = true;
 464                 pte->may_write = true;
 465                 pte->may_execute = true;
 466                 r = 0;
 467         }
 468
 469         return r;
 470 }
 471
 472 static hva_t kvmppc_bad_hva(void)
 473 {
 474         return PAGE_OFFSET;
 475 }
 476
 477 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
 478                                bool read)
 479 {
 480         hva_t hpage;
 481
 482         if (read && !pte->may_read)
 483                 goto err;
 484
 485         if (!read && !pte->may_write)
 486                 goto err;
 487
 488         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 489         if (kvm_is_error_hva(hpage))
 490                 goto err;
 491
 492         return hpage | (pte->raddr & ~PAGE_MASK);
 493 err:
 494         return kvmppc_bad_hva();
 495 }
 496
 497 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 498               bool data)
 499 {
 500         struct kvmppc_pte pte;
 501
 502         vcpu->stat.st++;
 503
 504         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 505                 return -ENOENT;
 506
 507         *eaddr = pte.raddr;
 508
 509         if (!pte.may_write)
 510                 return -EPERM;
 511
 512         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 513                 return EMULATE_DO_MMIO;
 514
 515         return EMULATE_DONE;
 516 }
 517
 518 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 519                       bool data)
 520 {
 521         struct kvmppc_pte pte;
 522         hva_t hva = *eaddr;
 523
 524         vcpu->stat.ld++;
 525
 526         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 527                 goto nopte;
 528
 529         *eaddr = pte.raddr;
 530
 531         hva = kvmppc_pte_to_hva(vcpu, &pte, true);
 532         if (kvm_is_error_hva(hva))
 533                 goto mmio;
 534
 535         if (copy_from_user(ptr, (void __user *)hva, size)) {
 536                 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
 537                 goto mmio;
 538         }
 539
 540         return EMULATE_DONE;
 541
 542 nopte:
 543         return -ENOENT;
 544 mmio:
 545         return EMULATE_DO_MMIO;
 546 }
 547
 548 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 549 {
 550         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 551 }
 552
 553 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 554                             ulong eaddr, int vec)
 555 {
 556         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 557         int r = RESUME_GUEST;
 558         int relocated;
 559         int page_found = 0;
 560         struct kvmppc_pte pte;
 561         bool is_mmio = false;
 562         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 563         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 564         u64 vsid;
 565
 566         relocated = data ? dr : ir;
 567
 568         /* Resolve real address if translation turned on */
 569         if (relocated) {
 570                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 571         } else {
 572                 pte.may_execute = true;
 573                 pte.may_read = true;
 574                 pte.may_write = true;
 575                 pte.raddr = eaddr & 0xffffffff;
 576                 pte.eaddr = eaddr;
 577                 pte.vpage = eaddr >> 12;
 578         }
 579
 580         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 581         case 0:
 582                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 583                 break;
 584         case MSR_DR:
 585         case MSR_IR:
 586                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 587
 588                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 589                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 590                 else
 591                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 592                 pte.vpage |= vsid;
 593
 594                 if (vsid == -1)
 595                         page_found = -EINVAL;
 596                 break;
 597         }
 598
 599         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 600            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 601                 /*
 602                  * If we do the dcbz hack, we have to NX on every execution,
 603                  * so we can patch the executing code. This renders our guest
 604                  * NX-less.
 605                  */
 606                 pte.may_execute = !data;
 607         }
 608
 609         if (page_found == -ENOENT) {
 610                 /* Page not found in guest PTE entries */
 611                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 612                 vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 613                 vcpu->arch.shared->msr |=
 614                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 615                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 616         } else if (page_found == -EPERM) {
 617                 /* Storage protection */
 618                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 619                 vcpu->arch.shared->dsisr =
 620                         to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
 621                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 622                 vcpu->arch.shared->msr |=
 623                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 624                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 625         } else if (page_found == -EINVAL) {
 626                 /* Page not found in guest SLB */
 627                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 628                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 629         } else if (!is_mmio &&
 630                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 631                 /* The guest's PTE is not mapped yet. Map on the host */
 632                 kvmppc_mmu_map_page(vcpu, &pte);
 633                 if (data)
 634                         vcpu->stat.sp_storage++;
 635                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 636                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 637                         kvmppc_patch_dcbz(vcpu, &pte);
 638         } else {
 639                 /* MMIO */
 640                 vcpu->stat.mmio_exits++;
 641                 vcpu->arch.paddr_accessed = pte.raddr;
 642                 r = kvmppc_emulate_mmio(run, vcpu);
 643                 if ( r == RESUME_HOST_NV )
 644                         r = RESUME_HOST;
 645         }
 646
 647         return r;
 648 }
 649
 650 static inline int get_fpr_index(int i)
 651 {
 652 #ifdef CONFIG_VSX
 653         i *= 2;
 654 #endif
 655         return i;
 656 }
 657
 658 /* Give up external provider (FPU, Altivec, VSX) */
 659 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 660 {
 661         struct thread_struct *t = &current->thread;
 662         u64 *vcpu_fpr = vcpu->arch.fpr;
 663 #ifdef CONFIG_VSX
 664         u64 *vcpu_vsx = vcpu->arch.vsr;
 665 #endif
 666         u64 *thread_fpr = (u64*)t->fpr;
 667         int i;
 668
 669         if (!(vcpu->arch.guest_owned_ext & msr))
 670                 return;
 671
 672 #ifdef DEBUG_EXT
 673         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 674 #endif
 675
 676         switch (msr) {
 677         case MSR_FP:
 678                 giveup_fpu(current);
 679                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 680                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 681
 682                 vcpu->arch.fpscr = t->fpscr.val;
 683                 break;
 684         case MSR_VEC:
 685 #ifdef CONFIG_ALTIVEC
 686                 giveup_altivec(current);
 687                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 688                 vcpu->arch.vscr = t->vscr;
 689 #endif
 690                 break;
 691         case MSR_VSX:
 692 #ifdef CONFIG_VSX
 693                 __giveup_vsx(current);
 694                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 695                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 696 #endif
 697                 break;
 698         default:
 699                 BUG();
 700         }
 701
 702         vcpu->arch.guest_owned_ext &= ~msr;
 703         current->thread.regs->msr &= ~msr;
 704         kvmppc_recalc_shadow_msr(vcpu);
 705 }
 706
 707 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 708 {
 709         ulong srr0 = kvmppc_get_pc(vcpu);
 710         u32 last_inst = kvmppc_get_last_inst(vcpu);
 711         int ret;
 712
 713         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 714         if (ret == -ENOENT) {
 715                 ulong msr = vcpu->arch.shared->msr;
 716
 717                 msr = kvmppc_set_field(msr, 33, 33, 1);
 718                 msr = kvmppc_set_field(msr, 34, 36, 0);
 719                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 720                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 721                 return EMULATE_AGAIN;
 722         }
 723
 724         return EMULATE_DONE;
 725 }
 726
 727 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 728 {
 729
 730         /* Need to do paired single emulation? */
 731         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 732                 return EMULATE_DONE;
 733
 734         /* Read out the instruction */
 735         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 736                 /* Need to emulate */
 737                 return EMULATE_FAIL;
 738
 739         return EMULATE_AGAIN;
 740 }
 741
 742 /* Handle external providers (FPU, Altivec, VSX) */
 743 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 744                              ulong msr)
 745 {
 746         struct thread_struct *t = &current->thread;
 747         u64 *vcpu_fpr = vcpu->arch.fpr;
 748 #ifdef CONFIG_VSX
 749         u64 *vcpu_vsx = vcpu->arch.vsr;
 750 #endif
 751         u64 *thread_fpr = (u64*)t->fpr;
 752         int i;
 753
 754         /* When we have paired singles, we emulate in software */
 755         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 756                 return RESUME_GUEST;
 757
 758         if (!(vcpu->arch.shared->msr & msr)) {
 759                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 760                 return RESUME_GUEST;
 761         }
 762
 763         /* We already own the ext */
 764         if (vcpu->arch.guest_owned_ext & msr) {
 765                 return RESUME_GUEST;
 766         }
 767
 768 #ifdef DEBUG_EXT
 769         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 770 #endif
 771
 772         current->thread.regs->msr |= msr;
 773
 774         switch (msr) {
 775         case MSR_FP:
 776                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 777                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 778
 779                 t->fpscr.val = vcpu->arch.fpscr;
 780                 t->fpexc_mode = 0;
 781                 kvmppc_load_up_fpu();
 782                 break;
 783         case MSR_VEC:
 784 #ifdef CONFIG_ALTIVEC
 785                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 786                 t->vscr = vcpu->arch.vscr;
 787                 t->vrsave = -1;
 788                 kvmppc_load_up_altivec();
 789 #endif
 790                 break;
 791         case MSR_VSX:
 792 #ifdef CONFIG_VSX
 793                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 794                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 795                 kvmppc_load_up_vsx();
 796 #endif
 797                 break;
 798         default:
 799                 BUG();
 800         }
 801
 802         vcpu->arch.guest_owned_ext |= msr;
 803
 804         kvmppc_recalc_shadow_msr(vcpu);
 805
 806         return RESUME_GUEST;
 807 }
 808
 809 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 810                        unsigned int exit_nr)
 811 {
 812         int r = RESUME_HOST;
 813
 814         vcpu->stat.sum_exits++;
 815
 816         run->exit_reason = KVM_EXIT_UNKNOWN;
 817         run->ready_for_interrupt_injection = 1;
 818 #ifdef EXIT_DEBUG
 819         printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | dec=0x%x | msr=0x%lx\n",
 820                 exit_nr, kvmppc_get_pc(vcpu), kvmppc_get_fault_dar(vcpu),
 821                 kvmppc_get_dec(vcpu), to_svcpu(vcpu)->shadow_srr1);
 822 #elif defined (EXIT_DEBUG_SIMPLE)
 823         if ((exit_nr != 0x900) && (exit_nr != 0x500))
 824                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | msr=0x%lx\n",
 825                         exit_nr, kvmppc_get_pc(vcpu), kvmppc_get_fault_dar(vcpu),
 826                         vcpu->arch.shared->msr);
 827 #endif
 828         kvm_resched(vcpu);
 829         switch (exit_nr) {
 830         case BOOK3S_INTERRUPT_INST_STORAGE:
 831                 vcpu->stat.pf_instruc++;
 832
 833 #ifdef CONFIG_PPC_BOOK3S_32
 834                 /* We set segments as unused segments when invalidating them. So
 835                  * treat the respective fault as segment fault. */
 836                 if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
 837                     == SR_INVALID) {
 838                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 839                         r = RESUME_GUEST;
 840                         break;
 841                 }
 842 #endif
 843
 844                 /* only care about PTEG not found errors, but leave NX alone */
 845                 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
 846                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 847                         vcpu->stat.sp_instruc++;
 848                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 849                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 850                         /*
 851                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 852                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 853                          *     that no guest that needs the dcbz hack does NX.
 854                          */
 855                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 856                         r = RESUME_GUEST;
 857                 } else {
 858                         vcpu->arch.shared->msr |=
 859                                 to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
 860                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 861                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 862                         r = RESUME_GUEST;
 863                 }
 864                 break;
 865         case BOOK3S_INTERRUPT_DATA_STORAGE:
 866         {
 867                 ulong dar = kvmppc_get_fault_dar(vcpu);
 868                 vcpu->stat.pf_storage++;
 869
 870 #ifdef CONFIG_PPC_BOOK3S_32
 871                 /* We set segments as unused segments when invalidating them. So
 872                  * treat the respective fault as segment fault. */
 873                 if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 874                         kvmppc_mmu_map_segment(vcpu, dar);
 875                         r = RESUME_GUEST;
 876                         break;
 877                 }
 878 #endif
 879
 880                 /* The only case we need to handle is missing shadow PTEs */
 881                 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
 882                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 883                 } else {
 884                         vcpu->arch.shared->dar = dar;
 885                         vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 886                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 887                         kvmppc_mmu_pte_flush(vcpu, dar, ~0xFFFUL);
 888                         r = RESUME_GUEST;
 889                 }
 890                 break;
 891         }
 892         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 893                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 894                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 895                         kvmppc_book3s_queue_irqprio(vcpu,
 896                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 897                 }
 898                 r = RESUME_GUEST;
 899                 break;
 900         case BOOK3S_INTERRUPT_INST_SEGMENT:
 901                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 902                         kvmppc_book3s_queue_irqprio(vcpu,
 903                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 904                 }
 905                 r = RESUME_GUEST;
 906                 break;
 907         /* We're good on these - the host merely wanted to get our attention */
 908         case BOOK3S_INTERRUPT_DECREMENTER:
 909                 vcpu->stat.dec_exits++;
 910                 r = RESUME_GUEST;
 911                 break;
 912         case BOOK3S_INTERRUPT_EXTERNAL:
 913                 vcpu->stat.ext_intr_exits++;
 914                 r = RESUME_GUEST;
 915                 break;
 916         case BOOK3S_INTERRUPT_PERFMON:
 917                 r = RESUME_GUEST;
 918                 break;
 919         case BOOK3S_INTERRUPT_PROGRAM:
 920         {
 921                 enum emulation_result er;
 922                 ulong flags;
 923
 924 program_interrupt:
 925                 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
 926
 927                 if (vcpu->arch.shared->msr & MSR_PR) {
 928 #ifdef EXIT_DEBUG
 929                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 930 #endif
 931                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 932                             (INS_DCBZ & 0xfffffff7)) {
 933                                 kvmppc_core_queue_program(vcpu, flags);
 934                                 r = RESUME_GUEST;
 935                                 break;
 936                         }
 937                 }
 938
 939                 vcpu->stat.emulated_inst_exits++;
 940                 er = kvmppc_emulate_instruction(run, vcpu);
 941                 switch (er) {
 942                 case EMULATE_DONE:
 943                         r = RESUME_GUEST_NV;
 944                         break;
 945                 case EMULATE_AGAIN:
 946                         r = RESUME_GUEST;
 947                         break;
 948                 case EMULATE_FAIL:
 949                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 950                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 951                         kvmppc_core_queue_program(vcpu, flags);
 952                         r = RESUME_GUEST;
 953                         break;
 954                 case EMULATE_DO_MMIO:
 955                         run->exit_reason = KVM_EXIT_MMIO;
 956                         r = RESUME_HOST_NV;
 957                         break;
 958                 default:
 959                         BUG();
 960                 }
 961                 break;
 962         }
 963         case BOOK3S_INTERRUPT_SYSCALL:
 964                 if (vcpu->arch.osi_enabled &&
 965                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 966                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 967                         /* MOL hypercalls */
 968                         u64 *gprs = run->osi.gprs;
 969                         int i;
 970
 971                         run->exit_reason = KVM_EXIT_OSI;
 972                         for (i = 0; i < 32; i++)
 973                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 974                         vcpu->arch.osi_needed = 1;
 975                         r = RESUME_HOST_NV;
 976                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 977                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 978                         /* KVM PV hypercalls */
 979                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 980                         r = RESUME_GUEST;
 981                 } else {
 982                         /* Guest syscalls */
 983                         vcpu->stat.syscall_exits++;
 984                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 985                         r = RESUME_GUEST;
 986                 }
 987                 break;
 988         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 989         case BOOK3S_INTERRUPT_ALTIVEC:
 990         case BOOK3S_INTERRUPT_VSX:
 991         {
 992                 int ext_msr = 0;
 993
 994                 switch (exit_nr) {
 995                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 996                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 997                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 998                 }
 999
1000                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
1001                 case EMULATE_DONE:
1002                         /* everything ok - let's enable the ext */
1003                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1004                         break;
1005                 case EMULATE_FAIL:
1006                         /* we need to emulate this instruction */
1007                         goto program_interrupt;
1008                         break;
1009                 default:
1010                         /* nothing to worry about - go again */
1011                         break;
1012                 }
1013                 break;
1014         }
1015         case BOOK3S_INTERRUPT_ALIGNMENT:
1016                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
1017                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
1018                                 kvmppc_get_last_inst(vcpu));
1019                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
1020                                 kvmppc_get_last_inst(vcpu));
1021                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1022                 }
1023                 r = RESUME_GUEST;
1024                 break;
1025         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1026         case BOOK3S_INTERRUPT_TRACE:
1027                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1028                 r = RESUME_GUEST;
1029                 break;
1030         default:
1031                 /* Ugh - bork here! What did we get? */
1032                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1033                         exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
1034                 r = RESUME_HOST;
1035                 BUG();
1036                 break;
1037         }
1038
1039
1040         if (!(r & RESUME_HOST)) {
1041                 /* To avoid clobbering exit_reason, only check for signals if
1042                  * we aren't already exiting to userspace for some other
1043                  * reason. */
1044                 if (signal_pending(current)) {
1045 #ifdef EXIT_DEBUG
1046                         printk(KERN_EMERG "KVM: Going back to host\n");
1047 #endif
1048                         vcpu->stat.signal_exits++;
1049                         run->exit_reason = KVM_EXIT_INTR;
1050                         r = -EINTR;
1051                 } else {
1052                         /* In case an interrupt came in that was triggered
1053                          * from userspace (like DEC), we need to check what
1054                          * to inject now! */
1055                         kvmppc_core_deliver_interrupts(vcpu);
1056                 }
1057         }
1058
1059 #ifdef EXIT_DEBUG
1060         printk(KERN_EMERG "KVM exit: vcpu=0x%p pc=0x%lx r=0x%x\n", vcpu, kvmppc_get_pc(vcpu), r);
1061 #endif
1062
1063         return r;
1064 }
1065
1066 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1067 {
1068         return 0;
1069 }
1070
1071 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1072 {
1073         int i;
1074
1075         regs->pc = kvmppc_get_pc(vcpu);
1076         regs->cr = kvmppc_get_cr(vcpu);
1077         regs->ctr = kvmppc_get_ctr(vcpu);
1078         regs->lr = kvmppc_get_lr(vcpu);
1079         regs->xer = kvmppc_get_xer(vcpu);
1080         regs->msr = vcpu->arch.shared->msr;
1081         regs->srr0 = vcpu->arch.shared->srr0;
1082         regs->srr1 = vcpu->arch.shared->srr1;
1083         regs->pid = vcpu->arch.pid;
1084         regs->sprg0 = vcpu->arch.shared->sprg0;
1085         regs->sprg1 = vcpu->arch.shared->sprg1;
1086         regs->sprg2 = vcpu->arch.shared->sprg2;
1087         regs->sprg3 = vcpu->arch.shared->sprg3;
1088         regs->sprg5 = vcpu->arch.sprg4;
1089         regs->sprg6 = vcpu->arch.sprg5;
1090         regs->sprg7 = vcpu->arch.sprg6;
1091
1092         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1093                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
1094
1095         return 0;
1096 }
1097
1098 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1099 {
1100         int i;
1101
1102         kvmppc_set_pc(vcpu, regs->pc);
1103         kvmppc_set_cr(vcpu, regs->cr);
1104         kvmppc_set_ctr(vcpu, regs->ctr);
1105         kvmppc_set_lr(vcpu, regs->lr);
1106         kvmppc_set_xer(vcpu, regs->xer);
1107         kvmppc_set_msr(vcpu, regs->msr);
1108         vcpu->arch.shared->srr0 = regs->srr0;
1109         vcpu->arch.shared->srr1 = regs->srr1;
1110         vcpu->arch.shared->sprg0 = regs->sprg0;
1111         vcpu->arch.shared->sprg1 = regs->sprg1;
1112         vcpu->arch.shared->sprg2 = regs->sprg2;
1113         vcpu->arch.shared->sprg3 = regs->sprg3;
1114         vcpu->arch.sprg5 = regs->sprg4;
1115         vcpu->arch.sprg6 = regs->sprg5;
1116         vcpu->arch.sprg7 = regs->sprg6;
1117
1118         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1119                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
1120
1121         return 0;
1122 }
1123
1124 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1125                                   struct kvm_sregs *sregs)
1126 {
1127         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1128         int i;
1129
1130         sregs->pvr = vcpu->arch.pvr;
1131
1132         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1133         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1134                 for (i = 0; i < 64; i++) {
1135                         sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
1136                         sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
1137                 }
1138         } else {
1139                 for (i = 0; i < 16; i++) {
1140                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
1141                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
1142                 }
1143                 for (i = 0; i < 8; i++) {
1144                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1145                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1146                 }
1147         }
1148
1149         return 0;
1150 }
1151
1152 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1153                                   struct kvm_sregs *sregs)
1154 {
1155         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1156         int i;
1157
1158         kvmppc_set_pvr(vcpu, sregs->pvr);
1159
1160         vcpu3s->sdr1 = sregs->u.s.sdr1;
1161         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1162                 for (i = 0; i < 64; i++) {
1163                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1164                                                     sregs->u.s.ppc64.slb[i].slbe);
1165                 }
1166         } else {
1167                 for (i = 0; i < 16; i++) {
1168                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1169                 }
1170                 for (i = 0; i < 8; i++) {
1171                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1172                                        (u32)sregs->u.s.ppc32.ibat[i]);
1173                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1174                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1175                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1176                                        (u32)sregs->u.s.ppc32.dbat[i]);
1177                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1178                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1179                 }
1180         }
1181
1182         /* Flush the MMU after messing with the segments */
1183         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1184
1185         return 0;
1186 }
1187
1188 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1189 {
1190         return -ENOTSUPP;
1191 }
1192
1193 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1194 {
1195         return -ENOTSUPP;
1196 }
1197
1198 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1199                                   struct kvm_translation *tr)
1200 {
1201         return 0;
1202 }
1203
1204 /*
1205  * Get (and clear) the dirty memory log for a memory slot.
1206  */
1207 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1208                                       struct kvm_dirty_log *log)
1209 {
1210         struct kvm_memory_slot *memslot;
1211         struct kvm_vcpu *vcpu;
1212         ulong ga, ga_end;
1213         int is_dirty = 0;
1214         int r;
1215         unsigned long n;
1216
1217         mutex_lock(&kvm->slots_lock);
1218
1219         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1220         if (r)
1221                 goto out;
1222
1223         /* If nothing is dirty, don't bother messing with page tables. */
1224         if (is_dirty) {
1225                 memslot = &kvm->memslots->memslots[log->slot];
1226
1227                 ga = memslot->base_gfn << PAGE_SHIFT;
1228                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1229
1230                 kvm_for_each_vcpu(n, vcpu, kvm)
1231                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1232
1233                 n = kvm_dirty_bitmap_bytes(memslot);
1234                 memset(memslot->dirty_bitmap, 0, n);
1235         }
1236
1237         r = 0;
1238 out:
1239         mutex_unlock(&kvm->slots_lock);
1240         return r;
1241 }
1242
1243 int kvmppc_core_check_processor_compat(void)
1244 {
1245         return 0;
1246 }
1247
1248 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1249 {
1250         struct kvmppc_vcpu_book3s *vcpu_book3s;
1251         struct kvm_vcpu *vcpu;
1252         int err = -ENOMEM;
1253
1254         vcpu_book3s = vmalloc(sizeof(struct kvmppc_vcpu_book3s));
1255         if (!vcpu_book3s)
1256                 goto out;
1257
1258         memset(vcpu_book3s, 0, sizeof(struct kvmppc_vcpu_book3s));
1259
1260         vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
1261                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
1262         if (!vcpu_book3s->shadow_vcpu)
1263                 goto free_vcpu;
1264
1265         vcpu = &vcpu_book3s->vcpu;
1266         err = kvm_vcpu_init(vcpu, kvm, id);
1267         if (err)
1268                 goto free_shadow_vcpu;
1269
1270         vcpu->arch.shared = (void*)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1271         if (!vcpu->arch.shared)
1272                 goto uninit_vcpu;
1273
1274         vcpu->arch.host_retip = kvm_return_point;
1275         vcpu->arch.host_msr = mfmsr();
1276 #ifdef CONFIG_PPC_BOOK3S_64
1277         /* default to book3s_64 (970fx) */
1278         vcpu->arch.pvr = 0x3C0301;
1279 #else
1280         /* default to book3s_32 (750) */
1281         vcpu->arch.pvr = 0x84202;
1282 #endif
1283         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1284         vcpu_book3s->slb_nr = 64;
1285
1286         /* remember where some real-mode handlers are */
1287         vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
1288         vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
1289         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
1290 #ifdef CONFIG_PPC_BOOK3S_64
1291         vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
1292 #else
1293         vcpu->arch.rmcall = (ulong)kvmppc_rmcall;
1294 #endif
1295
1296         vcpu->arch.shadow_msr = MSR_USER64;
1297
1298         err = kvmppc_mmu_init(vcpu);
1299         if (err < 0)
1300                 goto uninit_vcpu;
1301
1302         return vcpu;
1303
1304 uninit_vcpu:
1305         kvm_vcpu_uninit(vcpu);
1306 free_shadow_vcpu:
1307         kfree(vcpu_book3s->shadow_vcpu);
1308 free_vcpu:
1309         vfree(vcpu_book3s);
1310 out:
1311         return ERR_PTR(err);
1312 }
1313
1314 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1315 {
1316         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1317
1318         free_page((unsigned long)vcpu->arch.shared);
1319         kvm_vcpu_uninit(vcpu);
1320         kfree(vcpu_book3s->shadow_vcpu);
1321         vfree(vcpu_book3s);
1322 }
1323
1324 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1325 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1326 {
1327         int ret;
1328         double fpr[32][TS_FPRWIDTH];
1329         unsigned int fpscr;
1330         int fpexc_mode;
1331 #ifdef CONFIG_ALTIVEC
1332         vector128 vr[32];
1333         vector128 vscr;
1334         unsigned long uninitialized_var(vrsave);
1335         int used_vr;
1336 #endif
1337 #ifdef CONFIG_VSX
1338         int used_vsr;
1339 #endif
1340         ulong ext_msr;
1341
1342         /* No need to go into the guest when all we do is going out */
1343         if (signal_pending(current)) {
1344                 kvm_run->exit_reason = KVM_EXIT_INTR;
1345                 return -EINTR;
1346         }
1347
1348         /* Save FPU state in stack */
1349         if (current->thread.regs->msr & MSR_FP)
1350                 giveup_fpu(current);
1351         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1352         fpscr = current->thread.fpscr.val;
1353         fpexc_mode = current->thread.fpexc_mode;
1354
1355 #ifdef CONFIG_ALTIVEC
1356         /* Save Altivec state in stack */
1357         used_vr = current->thread.used_vr;
1358         if (used_vr) {
1359                 if (current->thread.regs->msr & MSR_VEC)
1360                         giveup_altivec(current);
1361                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1362                 vscr = current->thread.vscr;
1363                 vrsave = current->thread.vrsave;
1364         }
1365 #endif
1366
1367 #ifdef CONFIG_VSX
1368         /* Save VSX state in stack */
1369         used_vsr = current->thread.used_vsr;
1370         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1371                         __giveup_vsx(current);
1372 #endif
1373
1374         /* Remember the MSR with disabled extensions */
1375         ext_msr = current->thread.regs->msr;
1376
1377         /* XXX we get called with irq disabled - change that! */
1378         local_irq_enable();
1379
1380         /* Preload FPU if it's enabled */
1381         if (vcpu->arch.shared->msr & MSR_FP)
1382                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1383
1384         ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
1385
1386         local_irq_disable();
1387
1388         current->thread.regs->msr = ext_msr;
1389
1390         /* Make sure we save the guest FPU/Altivec/VSX state */
1391         kvmppc_giveup_ext(vcpu, MSR_FP);
1392         kvmppc_giveup_ext(vcpu, MSR_VEC);
1393         kvmppc_giveup_ext(vcpu, MSR_VSX);
1394
1395         /* Restore FPU state from stack */
1396         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1397         current->thread.fpscr.val = fpscr;
1398         current->thread.fpexc_mode = fpexc_mode;
1399
1400 #ifdef CONFIG_ALTIVEC
1401         /* Restore Altivec state from stack */
1402         if (used_vr && current->thread.used_vr) {
1403                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1404                 current->thread.vscr = vscr;
1405                 current->thread.vrsave = vrsave;
1406         }
1407         current->thread.used_vr = used_vr;
1408 #endif
1409
1410 #ifdef CONFIG_VSX
1411         current->thread.used_vsr = used_vsr;
1412 #endif
1413
1414         return ret;
1415 }
1416
1417 static int kvmppc_book3s_init(void)
1418 {
1419         int r;
1420
1421         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1422                      THIS_MODULE);
1423
1424         if (r)
1425                 return r;
1426
1427         r = kvmppc_mmu_hpte_sysinit();
1428
1429         return r;
1430 }
1431
1432 static void kvmppc_book3s_exit(void)
1433 {
1434         kvmppc_mmu_hpte_sysexit();
1435         kvm_exit();
1436 }
1437
1438 module_init(kvmppc_book3s_init);
1439 module_exit(kvmppc_book3s_exit);