SAFE public projects git trees. - safe/jmp/linux-2.6/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
  20 #include <asm/reg.h>
  21 #include <asm/cputable.h>
  22 #include <asm/cacheflush.h>
  23 #include <asm/tlbflush.h>
  24 #include <asm/uaccess.h>
  25 #include <asm/io.h>
  26 #include <asm/kvm_ppc.h>
  27 #include <asm/kvm_book3s.h>
  28 #include <asm/mmu_context.h>
  29 #include <linux/gfp.h>
  30 #include <linux/sched.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/highmem.h>
  33
  34 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  35
  36 /* #define EXIT_DEBUG */
  37 /* #define EXIT_DEBUG_SIMPLE */
  38 /* #define DEBUG_EXT */
  39
  40 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  41                              ulong msr);
  42
  43 /* Some compatibility defines */
  44 #ifdef CONFIG_PPC_BOOK3S_32
  45 #define MSR_USER32 MSR_USER
  46 #define MSR_USER64 MSR_USER
  47 #define HW_PAGE_SIZE PAGE_SIZE
  48 #endif
  49
  50 struct kvm_stats_debugfs_item debugfs_entries[] = {
  51         { "exits",       VCPU_STAT(sum_exits) },
  52         { "mmio",        VCPU_STAT(mmio_exits) },
  53         { "sig",         VCPU_STAT(signal_exits) },
  54         { "sysc",        VCPU_STAT(syscall_exits) },
  55         { "inst_emu",    VCPU_STAT(emulated_inst_exits) },
  56         { "dec",         VCPU_STAT(dec_exits) },
  57         { "ext_intr",    VCPU_STAT(ext_intr_exits) },
  58         { "queue_intr",  VCPU_STAT(queue_intr) },
  59         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  60         { "pf_storage",  VCPU_STAT(pf_storage) },
  61         { "sp_storage",  VCPU_STAT(sp_storage) },
  62         { "pf_instruc",  VCPU_STAT(pf_instruc) },
  63         { "sp_instruc",  VCPU_STAT(sp_instruc) },
  64         { "ld",          VCPU_STAT(ld) },
  65         { "ld_slow",     VCPU_STAT(ld_slow) },
  66         { "st",          VCPU_STAT(st) },
  67         { "st_slow",     VCPU_STAT(st_slow) },
  68         { NULL }
  69 };
  70
  71 void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
  72 {
  73 }
  74
  75 void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
  76 {
  77 }
  78
  79 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  80 {
  81 #ifdef CONFIG_PPC_BOOK3S_64
  82         memcpy(to_svcpu(vcpu)->slb, to_book3s(vcpu)->slb_shadow, sizeof(to_svcpu(vcpu)->slb));
  83         memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
  84                sizeof(get_paca()->shadow_vcpu));
  85         to_svcpu(vcpu)->slb_max = to_book3s(vcpu)->slb_shadow_max;
  86 #endif
  87
  88 #ifdef CONFIG_PPC_BOOK3S_32
  89         current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
  90 #endif
  91 }
  92
  93 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  94 {
  95 #ifdef CONFIG_PPC_BOOK3S_64
  96         memcpy(to_book3s(vcpu)->slb_shadow, to_svcpu(vcpu)->slb, sizeof(to_svcpu(vcpu)->slb));
  97         memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  98                sizeof(get_paca()->shadow_vcpu));
  99         to_book3s(vcpu)->slb_shadow_max = to_svcpu(vcpu)->slb_max;
 100 #endif
 101
 102         kvmppc_giveup_ext(vcpu, MSR_FP);
 103         kvmppc_giveup_ext(vcpu, MSR_VEC);
 104         kvmppc_giveup_ext(vcpu, MSR_VSX);
 105 }
 106
 107 #if defined(EXIT_DEBUG)
 108 static u32 kvmppc_get_dec(struct kvm_vcpu *vcpu)
 109 {
 110         u64 jd = mftb() - vcpu->arch.dec_jiffies;
 111         return vcpu->arch.dec - jd;
 112 }
 113 #endif
 114
 115 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 116 {
 117         vcpu->arch.shadow_msr = vcpu->arch.msr;
 118         /* Guest MSR values */
 119         vcpu->arch.shadow_msr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE |
 120                                  MSR_BE | MSR_DE;
 121         /* Process MSR values */
 122         vcpu->arch.shadow_msr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR |
 123                                  MSR_EE;
 124         /* External providers the guest reserved */
 125         vcpu->arch.shadow_msr |= (vcpu->arch.msr & vcpu->arch.guest_owned_ext);
 126         /* 64-bit Process MSR values */
 127 #ifdef CONFIG_PPC_BOOK3S_64
 128         vcpu->arch.shadow_msr |= MSR_ISF | MSR_HV;
 129 #endif
 130 }
 131
 132 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 133 {
 134         ulong old_msr = vcpu->arch.msr;
 135
 136 #ifdef EXIT_DEBUG
 137         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 138 #endif
 139
 140         msr &= to_book3s(vcpu)->msr_mask;
 141         vcpu->arch.msr = msr;
 142         kvmppc_recalc_shadow_msr(vcpu);
 143
 144         if (msr & (MSR_WE|MSR_POW)) {
 145                 if (!vcpu->arch.pending_exceptions) {
 146                         kvm_vcpu_block(vcpu);
 147                         vcpu->stat.halt_wakeup++;
 148                 }
 149         }
 150
 151         if ((vcpu->arch.msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 152                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 153                 kvmppc_mmu_flush_segments(vcpu);
 154                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 155         }
 156
 157         /* Preload FPU if it's enabled */
 158         if (vcpu->arch.msr & MSR_FP)
 159                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 160 }
 161
 162 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
 163 {
 164         vcpu->arch.srr0 = kvmppc_get_pc(vcpu);
 165         vcpu->arch.srr1 = vcpu->arch.msr | flags;
 166         kvmppc_set_pc(vcpu, to_book3s(vcpu)->hior + vec);
 167         vcpu->arch.mmu.reset_msr(vcpu);
 168 }
 169
 170 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
 171 {
 172         unsigned int prio;
 173
 174         switch (vec) {
 175         case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET;         break;
 176         case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK;        break;
 177         case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE;         break;
 178         case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT;         break;
 179         case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE;         break;
 180         case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT;         break;
 181         case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL;             break;
 182         case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT;            break;
 183         case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM;              break;
 184         case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL;           break;
 185         case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER;          break;
 186         case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL;              break;
 187         case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG;                break;
 188         case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC;              break;
 189         case 0xf40: prio = BOOK3S_IRQPRIO_VSX;                  break;
 190         default:    prio = BOOK3S_IRQPRIO_MAX;                  break;
 191         }
 192
 193         return prio;
 194 }
 195
 196 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
 197                                           unsigned int vec)
 198 {
 199         clear_bit(kvmppc_book3s_vec2irqprio(vec),
 200                   &vcpu->arch.pending_exceptions);
 201 }
 202
 203 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
 204 {
 205         vcpu->stat.queue_intr++;
 206
 207         set_bit(kvmppc_book3s_vec2irqprio(vec),
 208                 &vcpu->arch.pending_exceptions);
 209 #ifdef EXIT_DEBUG
 210         printk(KERN_INFO "Queueing interrupt %x\n", vec);
 211 #endif
 212 }
 213
 214
 215 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong flags)
 216 {
 217         to_book3s(vcpu)->prog_flags = flags;
 218         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
 219 }
 220
 221 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 222 {
 223         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 224 }
 225
 226 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 227 {
 228         return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
 229 }
 230
 231 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 232 {
 233         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 234 }
 235
 236 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 237                                 struct kvm_interrupt *irq)
 238 {
 239         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 240 }
 241
 242 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
 243                                   struct kvm_interrupt *irq)
 244 {
 245         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 246 }
 247
 248 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
 249 {
 250         int deliver = 1;
 251         int vec = 0;
 252         ulong flags = 0ULL;
 253
 254         switch (priority) {
 255         case BOOK3S_IRQPRIO_DECREMENTER:
 256                 deliver = vcpu->arch.msr & MSR_EE;
 257                 vec = BOOK3S_INTERRUPT_DECREMENTER;
 258                 break;
 259         case BOOK3S_IRQPRIO_EXTERNAL:
 260                 deliver = vcpu->arch.msr & MSR_EE;
 261                 vec = BOOK3S_INTERRUPT_EXTERNAL;
 262                 break;
 263         case BOOK3S_IRQPRIO_SYSTEM_RESET:
 264                 vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
 265                 break;
 266         case BOOK3S_IRQPRIO_MACHINE_CHECK:
 267                 vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
 268                 break;
 269         case BOOK3S_IRQPRIO_DATA_STORAGE:
 270                 vec = BOOK3S_INTERRUPT_DATA_STORAGE;
 271                 break;
 272         case BOOK3S_IRQPRIO_INST_STORAGE:
 273                 vec = BOOK3S_INTERRUPT_INST_STORAGE;
 274                 break;
 275         case BOOK3S_IRQPRIO_DATA_SEGMENT:
 276                 vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
 277                 break;
 278         case BOOK3S_IRQPRIO_INST_SEGMENT:
 279                 vec = BOOK3S_INTERRUPT_INST_SEGMENT;
 280                 break;
 281         case BOOK3S_IRQPRIO_ALIGNMENT:
 282                 vec = BOOK3S_INTERRUPT_ALIGNMENT;
 283                 break;
 284         case BOOK3S_IRQPRIO_PROGRAM:
 285                 vec = BOOK3S_INTERRUPT_PROGRAM;
 286                 flags = to_book3s(vcpu)->prog_flags;
 287                 break;
 288         case BOOK3S_IRQPRIO_VSX:
 289                 vec = BOOK3S_INTERRUPT_VSX;
 290                 break;
 291         case BOOK3S_IRQPRIO_ALTIVEC:
 292                 vec = BOOK3S_INTERRUPT_ALTIVEC;
 293                 break;
 294         case BOOK3S_IRQPRIO_FP_UNAVAIL:
 295                 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
 296                 break;
 297         case BOOK3S_IRQPRIO_SYSCALL:
 298                 vec = BOOK3S_INTERRUPT_SYSCALL;
 299                 break;
 300         case BOOK3S_IRQPRIO_DEBUG:
 301                 vec = BOOK3S_INTERRUPT_TRACE;
 302                 break;
 303         case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
 304                 vec = BOOK3S_INTERRUPT_PERFMON;
 305                 break;
 306         default:
 307                 deliver = 0;
 308                 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
 309                 break;
 310         }
 311
 312 #if 0
 313         printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
 314 #endif
 315
 316         if (deliver)
 317                 kvmppc_inject_interrupt(vcpu, vec, flags);
 318
 319         return deliver;
 320 }
 321
 322 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 323 {
 324         unsigned long *pending = &vcpu->arch.pending_exceptions;
 325         unsigned int priority;
 326
 327 #ifdef EXIT_DEBUG
 328         if (vcpu->arch.pending_exceptions)
 329                 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
 330 #endif
 331         priority = __ffs(*pending);
 332         while (priority < BOOK3S_IRQPRIO_MAX) {
 333                 if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
 334                     (priority != BOOK3S_IRQPRIO_DECREMENTER)) {
 335                         /* DEC interrupts get cleared by mtdec */
 336                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 337                         break;
 338                 }
 339
 340                 priority = find_next_bit(pending,
 341                                          BITS_PER_BYTE * sizeof(*pending),
 342                                          priority + 1);
 343         }
 344 }
 345
 346 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 347 {
 348         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 349         vcpu->arch.pvr = pvr;
 350 #ifdef CONFIG_PPC_BOOK3S_64
 351         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 352                 kvmppc_mmu_book3s_64_init(vcpu);
 353                 to_book3s(vcpu)->hior = 0xfff00000;
 354                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 355         } else
 356 #endif
 357         {
 358                 kvmppc_mmu_book3s_32_init(vcpu);
 359                 to_book3s(vcpu)->hior = 0;
 360                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 361         }
 362
 363         /* If we are in hypervisor level on 970, we can tell the CPU to
 364          * treat DCBZ as 32 bytes store */
 365         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 366         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 367             !strcmp(cur_cpu_spec->platform, "ppc970"))
 368                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 369
 370         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 371            really needs them in a VM on Cell and force disable them. */
 372         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 373                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 374
 375 #ifdef CONFIG_PPC_BOOK3S_32
 376         /* 32 bit Book3S always has 32 byte dcbz */
 377         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 378 #endif
 379 }
 380
 381 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 382  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 383  * emulate 32 bytes dcbz length.
 384  *
 385  * The Book3s_64 inventors also realized this case and implemented a special bit
 386  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 387  *
 388  * My approach here is to patch the dcbz instruction on executing pages.
 389  */
 390 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 391 {
 392         struct page *hpage;
 393         u64 hpage_offset;
 394         u32 *page;
 395         int i;
 396
 397         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 398         if (is_error_page(hpage))
 399                 return;
 400
 401         hpage_offset = pte->raddr & ~PAGE_MASK;
 402         hpage_offset &= ~0xFFFULL;
 403         hpage_offset /= 4;
 404
 405         get_page(hpage);
 406         page = kmap_atomic(hpage, KM_USER0);
 407
 408         /* patch dcbz into reserved instruction, so we trap */
 409         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 410                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 411                         page[i] &= 0xfffffff7;
 412
 413         kunmap_atomic(page, KM_USER0);
 414         put_page(hpage);
 415 }
 416
 417 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
 418                          struct kvmppc_pte *pte)
 419 {
 420         int relocated = (vcpu->arch.msr & (data ? MSR_DR : MSR_IR));
 421         int r;
 422
 423         if (relocated) {
 424                 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
 425         } else {
 426                 pte->eaddr = eaddr;
 427                 pte->raddr = eaddr & 0xffffffff;
 428                 pte->vpage = VSID_REAL | eaddr >> 12;
 429                 pte->may_read = true;
 430                 pte->may_write = true;
 431                 pte->may_execute = true;
 432                 r = 0;
 433         }
 434
 435         return r;
 436 }
 437
 438 static hva_t kvmppc_bad_hva(void)
 439 {
 440         return PAGE_OFFSET;
 441 }
 442
 443 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
 444                                bool read)
 445 {
 446         hva_t hpage;
 447
 448         if (read && !pte->may_read)
 449                 goto err;
 450
 451         if (!read && !pte->may_write)
 452                 goto err;
 453
 454         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 455         if (kvm_is_error_hva(hpage))
 456                 goto err;
 457
 458         return hpage | (pte->raddr & ~PAGE_MASK);
 459 err:
 460         return kvmppc_bad_hva();
 461 }
 462
 463 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 464               bool data)
 465 {
 466         struct kvmppc_pte pte;
 467
 468         vcpu->stat.st++;
 469
 470         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 471                 return -ENOENT;
 472
 473         *eaddr = pte.raddr;
 474
 475         if (!pte.may_write)
 476                 return -EPERM;
 477
 478         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 479                 return EMULATE_DO_MMIO;
 480
 481         return EMULATE_DONE;
 482 }
 483
 484 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 485                       bool data)
 486 {
 487         struct kvmppc_pte pte;
 488         hva_t hva = *eaddr;
 489
 490         vcpu->stat.ld++;
 491
 492         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 493                 goto nopte;
 494
 495         *eaddr = pte.raddr;
 496
 497         hva = kvmppc_pte_to_hva(vcpu, &pte, true);
 498         if (kvm_is_error_hva(hva))
 499                 goto mmio;
 500
 501         if (copy_from_user(ptr, (void __user *)hva, size)) {
 502                 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
 503                 goto mmio;
 504         }
 505
 506         return EMULATE_DONE;
 507
 508 nopte:
 509         return -ENOENT;
 510 mmio:
 511         return EMULATE_DO_MMIO;
 512 }
 513
 514 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 515 {
 516         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 517 }
 518
 519 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 520                             ulong eaddr, int vec)
 521 {
 522         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 523         int r = RESUME_GUEST;
 524         int relocated;
 525         int page_found = 0;
 526         struct kvmppc_pte pte;
 527         bool is_mmio = false;
 528         bool dr = (vcpu->arch.msr & MSR_DR) ? true : false;
 529         bool ir = (vcpu->arch.msr & MSR_IR) ? true : false;
 530         u64 vsid;
 531
 532         relocated = data ? dr : ir;
 533
 534         /* Resolve real address if translation turned on */
 535         if (relocated) {
 536                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 537         } else {
 538                 pte.may_execute = true;
 539                 pte.may_read = true;
 540                 pte.may_write = true;
 541                 pte.raddr = eaddr & 0xffffffff;
 542                 pte.eaddr = eaddr;
 543                 pte.vpage = eaddr >> 12;
 544         }
 545
 546         switch (vcpu->arch.msr & (MSR_DR|MSR_IR)) {
 547         case 0:
 548                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 549                 break;
 550         case MSR_DR:
 551         case MSR_IR:
 552                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 553
 554                 if ((vcpu->arch.msr & (MSR_DR|MSR_IR)) == MSR_DR)
 555                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 556                 else
 557                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 558                 pte.vpage |= vsid;
 559
 560                 if (vsid == -1)
 561                         page_found = -EINVAL;
 562                 break;
 563         }
 564
 565         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 566            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 567                 /*
 568                  * If we do the dcbz hack, we have to NX on every execution,
 569                  * so we can patch the executing code. This renders our guest
 570                  * NX-less.
 571                  */
 572                 pte.may_execute = !data;
 573         }
 574
 575         if (page_found == -ENOENT) {
 576                 /* Page not found in guest PTE entries */
 577                 vcpu->arch.dear = kvmppc_get_fault_dar(vcpu);
 578                 to_book3s(vcpu)->dsisr = to_svcpu(vcpu)->fault_dsisr;
 579                 vcpu->arch.msr |= (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 580                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 581         } else if (page_found == -EPERM) {
 582                 /* Storage protection */
 583                 vcpu->arch.dear = kvmppc_get_fault_dar(vcpu);
 584                 to_book3s(vcpu)->dsisr = to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
 585                 to_book3s(vcpu)->dsisr |= DSISR_PROTFAULT;
 586                 vcpu->arch.msr |= (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 587                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 588         } else if (page_found == -EINVAL) {
 589                 /* Page not found in guest SLB */
 590                 vcpu->arch.dear = kvmppc_get_fault_dar(vcpu);
 591                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 592         } else if (!is_mmio &&
 593                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 594                 /* The guest's PTE is not mapped yet. Map on the host */
 595                 kvmppc_mmu_map_page(vcpu, &pte);
 596                 if (data)
 597                         vcpu->stat.sp_storage++;
 598                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 599                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 600                         kvmppc_patch_dcbz(vcpu, &pte);
 601         } else {
 602                 /* MMIO */
 603                 vcpu->stat.mmio_exits++;
 604                 vcpu->arch.paddr_accessed = pte.raddr;
 605                 r = kvmppc_emulate_mmio(run, vcpu);
 606                 if ( r == RESUME_HOST_NV )
 607                         r = RESUME_HOST;
 608         }
 609
 610         return r;
 611 }
 612
 613 static inline int get_fpr_index(int i)
 614 {
 615 #ifdef CONFIG_VSX
 616         i *= 2;
 617 #endif
 618         return i;
 619 }
 620
 621 /* Give up external provider (FPU, Altivec, VSX) */
 622 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 623 {
 624         struct thread_struct *t = &current->thread;
 625         u64 *vcpu_fpr = vcpu->arch.fpr;
 626 #ifdef CONFIG_VSX
 627         u64 *vcpu_vsx = vcpu->arch.vsr;
 628 #endif
 629         u64 *thread_fpr = (u64*)t->fpr;
 630         int i;
 631
 632         if (!(vcpu->arch.guest_owned_ext & msr))
 633                 return;
 634
 635 #ifdef DEBUG_EXT
 636         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 637 #endif
 638
 639         switch (msr) {
 640         case MSR_FP:
 641                 giveup_fpu(current);
 642                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 643                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 644
 645                 vcpu->arch.fpscr = t->fpscr.val;
 646                 break;
 647         case MSR_VEC:
 648 #ifdef CONFIG_ALTIVEC
 649                 giveup_altivec(current);
 650                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 651                 vcpu->arch.vscr = t->vscr;
 652 #endif
 653                 break;
 654         case MSR_VSX:
 655 #ifdef CONFIG_VSX
 656                 __giveup_vsx(current);
 657                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 658                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 659 #endif
 660                 break;
 661         default:
 662                 BUG();
 663         }
 664
 665         vcpu->arch.guest_owned_ext &= ~msr;
 666         current->thread.regs->msr &= ~msr;
 667         kvmppc_recalc_shadow_msr(vcpu);
 668 }
 669
 670 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 671 {
 672         ulong srr0 = kvmppc_get_pc(vcpu);
 673         u32 last_inst = kvmppc_get_last_inst(vcpu);
 674         int ret;
 675
 676         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 677         if (ret == -ENOENT) {
 678                 vcpu->arch.msr = kvmppc_set_field(vcpu->arch.msr, 33, 33, 1);
 679                 vcpu->arch.msr = kvmppc_set_field(vcpu->arch.msr, 34, 36, 0);
 680                 vcpu->arch.msr = kvmppc_set_field(vcpu->arch.msr, 42, 47, 0);
 681                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 682                 return EMULATE_AGAIN;
 683         }
 684
 685         return EMULATE_DONE;
 686 }
 687
 688 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 689 {
 690
 691         /* Need to do paired single emulation? */
 692         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 693                 return EMULATE_DONE;
 694
 695         /* Read out the instruction */
 696         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 697                 /* Need to emulate */
 698                 return EMULATE_FAIL;
 699
 700         return EMULATE_AGAIN;
 701 }
 702
 703 /* Handle external providers (FPU, Altivec, VSX) */
 704 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 705                              ulong msr)
 706 {
 707         struct thread_struct *t = &current->thread;
 708         u64 *vcpu_fpr = vcpu->arch.fpr;
 709 #ifdef CONFIG_VSX
 710         u64 *vcpu_vsx = vcpu->arch.vsr;
 711 #endif
 712         u64 *thread_fpr = (u64*)t->fpr;
 713         int i;
 714
 715         /* When we have paired singles, we emulate in software */
 716         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 717                 return RESUME_GUEST;
 718
 719         if (!(vcpu->arch.msr & msr)) {
 720                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 721                 return RESUME_GUEST;
 722         }
 723
 724         /* We already own the ext */
 725         if (vcpu->arch.guest_owned_ext & msr) {
 726                 return RESUME_GUEST;
 727         }
 728
 729 #ifdef DEBUG_EXT
 730         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 731 #endif
 732
 733         current->thread.regs->msr |= msr;
 734
 735         switch (msr) {
 736         case MSR_FP:
 737                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 738                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 739
 740                 t->fpscr.val = vcpu->arch.fpscr;
 741                 t->fpexc_mode = 0;
 742                 kvmppc_load_up_fpu();
 743                 break;
 744         case MSR_VEC:
 745 #ifdef CONFIG_ALTIVEC
 746                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 747                 t->vscr = vcpu->arch.vscr;
 748                 t->vrsave = -1;
 749                 kvmppc_load_up_altivec();
 750 #endif
 751                 break;
 752         case MSR_VSX:
 753 #ifdef CONFIG_VSX
 754                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 755                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 756                 kvmppc_load_up_vsx();
 757 #endif
 758                 break;
 759         default:
 760                 BUG();
 761         }
 762
 763         vcpu->arch.guest_owned_ext |= msr;
 764
 765         kvmppc_recalc_shadow_msr(vcpu);
 766
 767         return RESUME_GUEST;
 768 }
 769
 770 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 771                        unsigned int exit_nr)
 772 {
 773         int r = RESUME_HOST;
 774
 775         vcpu->stat.sum_exits++;
 776
 777         run->exit_reason = KVM_EXIT_UNKNOWN;
 778         run->ready_for_interrupt_injection = 1;
 779 #ifdef EXIT_DEBUG
 780         printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | dec=0x%x | msr=0x%lx\n",
 781                 exit_nr, kvmppc_get_pc(vcpu), kvmppc_get_fault_dar(vcpu),
 782                 kvmppc_get_dec(vcpu), to_svcpu(vcpu)->shadow_srr1);
 783 #elif defined (EXIT_DEBUG_SIMPLE)
 784         if ((exit_nr != 0x900) && (exit_nr != 0x500))
 785                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | msr=0x%lx\n",
 786                         exit_nr, kvmppc_get_pc(vcpu), kvmppc_get_fault_dar(vcpu),
 787                         vcpu->arch.msr);
 788 #endif
 789         kvm_resched(vcpu);
 790         switch (exit_nr) {
 791         case BOOK3S_INTERRUPT_INST_STORAGE:
 792                 vcpu->stat.pf_instruc++;
 793
 794 #ifdef CONFIG_PPC_BOOK3S_32
 795                 /* We set segments as unused segments when invalidating them. So
 796                  * treat the respective fault as segment fault. */
 797                 if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
 798                     == SR_INVALID) {
 799                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 800                         r = RESUME_GUEST;
 801                         break;
 802                 }
 803 #endif
 804
 805                 /* only care about PTEG not found errors, but leave NX alone */
 806                 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
 807                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 808                         vcpu->stat.sp_instruc++;
 809                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 810                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 811                         /*
 812                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 813                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 814                          *     that no guest that needs the dcbz hack does NX.
 815                          */
 816                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 817                         r = RESUME_GUEST;
 818                 } else {
 819                         vcpu->arch.msr |= to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
 820                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 821                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 822                         r = RESUME_GUEST;
 823                 }
 824                 break;
 825         case BOOK3S_INTERRUPT_DATA_STORAGE:
 826         {
 827                 ulong dar = kvmppc_get_fault_dar(vcpu);
 828                 vcpu->stat.pf_storage++;
 829
 830 #ifdef CONFIG_PPC_BOOK3S_32
 831                 /* We set segments as unused segments when invalidating them. So
 832                  * treat the respective fault as segment fault. */
 833                 if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 834                         kvmppc_mmu_map_segment(vcpu, dar);
 835                         r = RESUME_GUEST;
 836                         break;
 837                 }
 838 #endif
 839
 840                 /* The only case we need to handle is missing shadow PTEs */
 841                 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
 842                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 843                 } else {
 844                         vcpu->arch.dear = dar;
 845                         to_book3s(vcpu)->dsisr = to_svcpu(vcpu)->fault_dsisr;
 846                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 847                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.dear, ~0xFFFUL);
 848                         r = RESUME_GUEST;
 849                 }
 850                 break;
 851         }
 852         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 853                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 854                         vcpu->arch.dear = kvmppc_get_fault_dar(vcpu);
 855                         kvmppc_book3s_queue_irqprio(vcpu,
 856                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 857                 }
 858                 r = RESUME_GUEST;
 859                 break;
 860         case BOOK3S_INTERRUPT_INST_SEGMENT:
 861                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 862                         kvmppc_book3s_queue_irqprio(vcpu,
 863                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 864                 }
 865                 r = RESUME_GUEST;
 866                 break;
 867         /* We're good on these - the host merely wanted to get our attention */
 868         case BOOK3S_INTERRUPT_DECREMENTER:
 869                 vcpu->stat.dec_exits++;
 870                 r = RESUME_GUEST;
 871                 break;
 872         case BOOK3S_INTERRUPT_EXTERNAL:
 873                 vcpu->stat.ext_intr_exits++;
 874                 r = RESUME_GUEST;
 875                 break;
 876         case BOOK3S_INTERRUPT_PERFMON:
 877                 r = RESUME_GUEST;
 878                 break;
 879         case BOOK3S_INTERRUPT_PROGRAM:
 880         {
 881                 enum emulation_result er;
 882                 ulong flags;
 883
 884 program_interrupt:
 885                 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
 886
 887                 if (vcpu->arch.msr & MSR_PR) {
 888 #ifdef EXIT_DEBUG
 889                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 890 #endif
 891                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 892                             (INS_DCBZ & 0xfffffff7)) {
 893                                 kvmppc_core_queue_program(vcpu, flags);
 894                                 r = RESUME_GUEST;
 895                                 break;
 896                         }
 897                 }
 898
 899                 vcpu->stat.emulated_inst_exits++;
 900                 er = kvmppc_emulate_instruction(run, vcpu);
 901                 switch (er) {
 902                 case EMULATE_DONE:
 903                         r = RESUME_GUEST_NV;
 904                         break;
 905                 case EMULATE_AGAIN:
 906                         r = RESUME_GUEST;
 907                         break;
 908                 case EMULATE_FAIL:
 909                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 910                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 911                         kvmppc_core_queue_program(vcpu, flags);
 912                         r = RESUME_GUEST;
 913                         break;
 914                 case EMULATE_DO_MMIO:
 915                         run->exit_reason = KVM_EXIT_MMIO;
 916                         r = RESUME_HOST_NV;
 917                         break;
 918                 default:
 919                         BUG();
 920                 }
 921                 break;
 922         }
 923         case BOOK3S_INTERRUPT_SYSCALL:
 924                 // XXX make user settable
 925                 if (vcpu->arch.osi_enabled &&
 926                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 927                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 928                         u64 *gprs = run->osi.gprs;
 929                         int i;
 930
 931                         run->exit_reason = KVM_EXIT_OSI;
 932                         for (i = 0; i < 32; i++)
 933                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 934                         vcpu->arch.osi_needed = 1;
 935                         r = RESUME_HOST_NV;
 936
 937                 } else {
 938                         vcpu->stat.syscall_exits++;
 939                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 940                         r = RESUME_GUEST;
 941                 }
 942                 break;
 943         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 944         case BOOK3S_INTERRUPT_ALTIVEC:
 945         case BOOK3S_INTERRUPT_VSX:
 946         {
 947                 int ext_msr = 0;
 948
 949                 switch (exit_nr) {
 950                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 951                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 952                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 953                 }
 954
 955                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
 956                 case EMULATE_DONE:
 957                         /* everything ok - let's enable the ext */
 958                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
 959                         break;
 960                 case EMULATE_FAIL:
 961                         /* we need to emulate this instruction */
 962                         goto program_interrupt;
 963                         break;
 964                 default:
 965                         /* nothing to worry about - go again */
 966                         break;
 967                 }
 968                 break;
 969         }
 970         case BOOK3S_INTERRUPT_ALIGNMENT:
 971                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
 972                         to_book3s(vcpu)->dsisr = kvmppc_alignment_dsisr(vcpu,
 973                                 kvmppc_get_last_inst(vcpu));
 974                         vcpu->arch.dear = kvmppc_alignment_dar(vcpu,
 975                                 kvmppc_get_last_inst(vcpu));
 976                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 977                 }
 978                 r = RESUME_GUEST;
 979                 break;
 980         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 981         case BOOK3S_INTERRUPT_TRACE:
 982                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 983                 r = RESUME_GUEST;
 984                 break;
 985         default:
 986                 /* Ugh - bork here! What did we get? */
 987                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
 988                         exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
 989                 r = RESUME_HOST;
 990                 BUG();
 991                 break;
 992         }
 993
 994
 995         if (!(r & RESUME_HOST)) {
 996                 /* To avoid clobbering exit_reason, only check for signals if
 997                  * we aren't already exiting to userspace for some other
 998                  * reason. */
 999                 if (signal_pending(current)) {
1000 #ifdef EXIT_DEBUG
1001                         printk(KERN_EMERG "KVM: Going back to host\n");
1002 #endif
1003                         vcpu->stat.signal_exits++;
1004                         run->exit_reason = KVM_EXIT_INTR;
1005                         r = -EINTR;
1006                 } else {
1007                         /* In case an interrupt came in that was triggered
1008                          * from userspace (like DEC), we need to check what
1009                          * to inject now! */
1010                         kvmppc_core_deliver_interrupts(vcpu);
1011                 }
1012         }
1013
1014 #ifdef EXIT_DEBUG
1015         printk(KERN_EMERG "KVM exit: vcpu=0x%p pc=0x%lx r=0x%x\n", vcpu, kvmppc_get_pc(vcpu), r);
1016 #endif
1017
1018         return r;
1019 }
1020
1021 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1022 {
1023         return 0;
1024 }
1025
1026 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1027 {
1028         int i;
1029
1030         vcpu_load(vcpu);
1031
1032         regs->pc = kvmppc_get_pc(vcpu);
1033         regs->cr = kvmppc_get_cr(vcpu);
1034         regs->ctr = kvmppc_get_ctr(vcpu);
1035         regs->lr = kvmppc_get_lr(vcpu);
1036         regs->xer = kvmppc_get_xer(vcpu);
1037         regs->msr = vcpu->arch.msr;
1038         regs->srr0 = vcpu->arch.srr0;
1039         regs->srr1 = vcpu->arch.srr1;
1040         regs->pid = vcpu->arch.pid;
1041         regs->sprg0 = vcpu->arch.sprg0;
1042         regs->sprg1 = vcpu->arch.sprg1;
1043         regs->sprg2 = vcpu->arch.sprg2;
1044         regs->sprg3 = vcpu->arch.sprg3;
1045         regs->sprg5 = vcpu->arch.sprg4;
1046         regs->sprg6 = vcpu->arch.sprg5;
1047         regs->sprg7 = vcpu->arch.sprg6;
1048
1049         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1050                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
1051
1052         vcpu_put(vcpu);
1053
1054         return 0;
1055 }
1056
1057 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1058 {
1059         int i;
1060
1061         vcpu_load(vcpu);
1062
1063         kvmppc_set_pc(vcpu, regs->pc);
1064         kvmppc_set_cr(vcpu, regs->cr);
1065         kvmppc_set_ctr(vcpu, regs->ctr);
1066         kvmppc_set_lr(vcpu, regs->lr);
1067         kvmppc_set_xer(vcpu, regs->xer);
1068         kvmppc_set_msr(vcpu, regs->msr);
1069         vcpu->arch.srr0 = regs->srr0;
1070         vcpu->arch.srr1 = regs->srr1;
1071         vcpu->arch.sprg0 = regs->sprg0;
1072         vcpu->arch.sprg1 = regs->sprg1;
1073         vcpu->arch.sprg2 = regs->sprg2;
1074         vcpu->arch.sprg3 = regs->sprg3;
1075         vcpu->arch.sprg5 = regs->sprg4;
1076         vcpu->arch.sprg6 = regs->sprg5;
1077         vcpu->arch.sprg7 = regs->sprg6;
1078
1079         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1080                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
1081
1082         vcpu_put(vcpu);
1083
1084         return 0;
1085 }
1086
1087 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1088                                   struct kvm_sregs *sregs)
1089 {
1090         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1091         int i;
1092
1093         sregs->pvr = vcpu->arch.pvr;
1094
1095         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1096         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1097                 for (i = 0; i < 64; i++) {
1098                         sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
1099                         sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
1100                 }
1101         } else {
1102                 for (i = 0; i < 16; i++) {
1103                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
1104                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
1105                 }
1106                 for (i = 0; i < 8; i++) {
1107                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1108                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1109                 }
1110         }
1111         return 0;
1112 }
1113
1114 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1115                                   struct kvm_sregs *sregs)
1116 {
1117         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1118         int i;
1119
1120         kvmppc_set_pvr(vcpu, sregs->pvr);
1121
1122         vcpu3s->sdr1 = sregs->u.s.sdr1;
1123         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1124                 for (i = 0; i < 64; i++) {
1125                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1126                                                     sregs->u.s.ppc64.slb[i].slbe);
1127                 }
1128         } else {
1129                 for (i = 0; i < 16; i++) {
1130                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1131                 }
1132                 for (i = 0; i < 8; i++) {
1133                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1134                                        (u32)sregs->u.s.ppc32.ibat[i]);
1135                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1136                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1137                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1138                                        (u32)sregs->u.s.ppc32.dbat[i]);
1139                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1140                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1141                 }
1142         }
1143
1144         /* Flush the MMU after messing with the segments */
1145         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1146         return 0;
1147 }
1148
1149 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1150 {
1151         return -ENOTSUPP;
1152 }
1153
1154 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1155 {
1156         return -ENOTSUPP;
1157 }
1158
1159 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1160                                   struct kvm_translation *tr)
1161 {
1162         return 0;
1163 }
1164
1165 /*
1166  * Get (and clear) the dirty memory log for a memory slot.
1167  */
1168 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1169                                       struct kvm_dirty_log *log)
1170 {
1171         struct kvm_memory_slot *memslot;
1172         struct kvm_vcpu *vcpu;
1173         ulong ga, ga_end;
1174         int is_dirty = 0;
1175         int r;
1176         unsigned long n;
1177
1178         mutex_lock(&kvm->slots_lock);
1179
1180         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1181         if (r)
1182                 goto out;
1183
1184         /* If nothing is dirty, don't bother messing with page tables. */
1185         if (is_dirty) {
1186                 memslot = &kvm->memslots->memslots[log->slot];
1187
1188                 ga = memslot->base_gfn << PAGE_SHIFT;
1189                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1190
1191                 kvm_for_each_vcpu(n, vcpu, kvm)
1192                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1193
1194                 n = kvm_dirty_bitmap_bytes(memslot);
1195                 memset(memslot->dirty_bitmap, 0, n);
1196         }
1197
1198         r = 0;
1199 out:
1200         mutex_unlock(&kvm->slots_lock);
1201         return r;
1202 }
1203
1204 int kvmppc_core_check_processor_compat(void)
1205 {
1206         return 0;
1207 }
1208
1209 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1210 {
1211         struct kvmppc_vcpu_book3s *vcpu_book3s;
1212         struct kvm_vcpu *vcpu;
1213         int err = -ENOMEM;
1214
1215         vcpu_book3s = vmalloc(sizeof(struct kvmppc_vcpu_book3s));
1216         if (!vcpu_book3s)
1217                 goto out;
1218
1219         memset(vcpu_book3s, 0, sizeof(struct kvmppc_vcpu_book3s));
1220
1221         vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
1222                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
1223         if (!vcpu_book3s->shadow_vcpu)
1224                 goto free_vcpu;
1225
1226         vcpu = &vcpu_book3s->vcpu;
1227         err = kvm_vcpu_init(vcpu, kvm, id);
1228         if (err)
1229                 goto free_shadow_vcpu;
1230
1231         vcpu->arch.host_retip = kvm_return_point;
1232         vcpu->arch.host_msr = mfmsr();
1233 #ifdef CONFIG_PPC_BOOK3S_64
1234         /* default to book3s_64 (970fx) */
1235         vcpu->arch.pvr = 0x3C0301;
1236 #else
1237         /* default to book3s_32 (750) */
1238         vcpu->arch.pvr = 0x84202;
1239 #endif
1240         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1241         vcpu_book3s->slb_nr = 64;
1242
1243         /* remember where some real-mode handlers are */
1244         vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
1245         vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
1246         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
1247 #ifdef CONFIG_PPC_BOOK3S_64
1248         vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
1249 #else
1250         vcpu->arch.rmcall = (ulong)kvmppc_rmcall;
1251 #endif
1252
1253         vcpu->arch.shadow_msr = MSR_USER64;
1254
1255         err = kvmppc_mmu_init(vcpu);
1256         if (err < 0)
1257                 goto free_shadow_vcpu;
1258
1259         return vcpu;
1260
1261 free_shadow_vcpu:
1262         kfree(vcpu_book3s->shadow_vcpu);
1263 free_vcpu:
1264         vfree(vcpu_book3s);
1265 out:
1266         return ERR_PTR(err);
1267 }
1268
1269 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1270 {
1271         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1272
1273         kvm_vcpu_uninit(vcpu);
1274         kfree(vcpu_book3s->shadow_vcpu);
1275         vfree(vcpu_book3s);
1276 }
1277
1278 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1279 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1280 {
1281         int ret;
1282         struct thread_struct ext_bkp;
1283 #ifdef CONFIG_ALTIVEC
1284         bool save_vec = current->thread.used_vr;
1285 #endif
1286 #ifdef CONFIG_VSX
1287         bool save_vsx = current->thread.used_vsr;
1288 #endif
1289         ulong ext_msr;
1290
1291         /* No need to go into the guest when all we do is going out */
1292         if (signal_pending(current)) {
1293                 kvm_run->exit_reason = KVM_EXIT_INTR;
1294                 return -EINTR;
1295         }
1296
1297         /* Save FPU state in stack */
1298         if (current->thread.regs->msr & MSR_FP)
1299                 giveup_fpu(current);
1300         memcpy(ext_bkp.fpr, current->thread.fpr, sizeof(current->thread.fpr));
1301         ext_bkp.fpscr = current->thread.fpscr;
1302         ext_bkp.fpexc_mode = current->thread.fpexc_mode;
1303
1304 #ifdef CONFIG_ALTIVEC
1305         /* Save Altivec state in stack */
1306         if (save_vec) {
1307                 if (current->thread.regs->msr & MSR_VEC)
1308                         giveup_altivec(current);
1309                 memcpy(ext_bkp.vr, current->thread.vr, sizeof(ext_bkp.vr));
1310                 ext_bkp.vscr = current->thread.vscr;
1311                 ext_bkp.vrsave = current->thread.vrsave;
1312         }
1313         ext_bkp.used_vr = current->thread.used_vr;
1314 #endif
1315
1316 #ifdef CONFIG_VSX
1317         /* Save VSX state in stack */
1318         if (save_vsx && (current->thread.regs->msr & MSR_VSX))
1319                         __giveup_vsx(current);
1320         ext_bkp.used_vsr = current->thread.used_vsr;
1321 #endif
1322
1323         /* Remember the MSR with disabled extensions */
1324         ext_msr = current->thread.regs->msr;
1325
1326         /* XXX we get called with irq disabled - change that! */
1327         local_irq_enable();
1328
1329         /* Preload FPU if it's enabled */
1330         if (vcpu->arch.msr & MSR_FP)
1331                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1332
1333         ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
1334
1335         local_irq_disable();
1336
1337         current->thread.regs->msr = ext_msr;
1338
1339         /* Make sure we save the guest FPU/Altivec/VSX state */
1340         kvmppc_giveup_ext(vcpu, MSR_FP);
1341         kvmppc_giveup_ext(vcpu, MSR_VEC);
1342         kvmppc_giveup_ext(vcpu, MSR_VSX);
1343
1344         /* Restore FPU state from stack */
1345         memcpy(current->thread.fpr, ext_bkp.fpr, sizeof(ext_bkp.fpr));
1346         current->thread.fpscr = ext_bkp.fpscr;
1347         current->thread.fpexc_mode = ext_bkp.fpexc_mode;
1348
1349 #ifdef CONFIG_ALTIVEC
1350         /* Restore Altivec state from stack */
1351         if (save_vec && current->thread.used_vr) {
1352                 memcpy(current->thread.vr, ext_bkp.vr, sizeof(ext_bkp.vr));
1353                 current->thread.vscr = ext_bkp.vscr;
1354                 current->thread.vrsave= ext_bkp.vrsave;
1355         }
1356         current->thread.used_vr = ext_bkp.used_vr;
1357 #endif
1358
1359 #ifdef CONFIG_VSX
1360         current->thread.used_vsr = ext_bkp.used_vsr;
1361 #endif
1362
1363         return ret;
1364 }
1365
1366 static int kvmppc_book3s_init(void)
1367 {
1368         return kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), THIS_MODULE);
1369 }
1370
1371 static void kvmppc_book3s_exit(void)
1372 {
1373         kvm_exit();
1374 }
1375
1376 module_init(kvmppc_book3s_init);
1377 module_exit(kvmppc_book3s_exit);