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/sched.h>
  30 #include <linux/vmalloc.h>
  31
  32 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  33
  34 /* #define EXIT_DEBUG */
  35 /* #define EXIT_DEBUG_SIMPLE */
  36
  37 /* Without AGGRESSIVE_DEC we only fire off a DEC interrupt when DEC turns 0.
  38  * When set, we retrigger a DEC interrupt after that if DEC <= 0.
  39  * PPC32 Linux runs faster without AGGRESSIVE_DEC, PPC64 Linux requires it. */
  40
  41 /* #define AGGRESSIVE_DEC */
  42
  43 struct kvm_stats_debugfs_item debugfs_entries[] = {
  44         { "exits",       VCPU_STAT(sum_exits) },
  45         { "mmio",        VCPU_STAT(mmio_exits) },
  46         { "sig",         VCPU_STAT(signal_exits) },
  47         { "sysc",        VCPU_STAT(syscall_exits) },
  48         { "inst_emu",    VCPU_STAT(emulated_inst_exits) },
  49         { "dec",         VCPU_STAT(dec_exits) },
  50         { "ext_intr",    VCPU_STAT(ext_intr_exits) },
  51         { "queue_intr",  VCPU_STAT(queue_intr) },
  52         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  53         { "pf_storage",  VCPU_STAT(pf_storage) },
  54         { "sp_storage",  VCPU_STAT(sp_storage) },
  55         { "pf_instruc",  VCPU_STAT(pf_instruc) },
  56         { "sp_instruc",  VCPU_STAT(sp_instruc) },
  57         { "ld",          VCPU_STAT(ld) },
  58         { "ld_slow",     VCPU_STAT(ld_slow) },
  59         { "st",          VCPU_STAT(st) },
  60         { "st_slow",     VCPU_STAT(st_slow) },
  61         { NULL }
  62 };
  63
  64 void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
  65 {
  66 }
  67
  68 void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
  69 {
  70 }
  71
  72 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  73 {
  74         memcpy(get_paca()->kvm_slb, to_book3s(vcpu)->slb_shadow, sizeof(get_paca()->kvm_slb));
  75         get_paca()->kvm_slb_max = to_book3s(vcpu)->slb_shadow_max;
  76 }
  77
  78 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  79 {
  80         memcpy(to_book3s(vcpu)->slb_shadow, get_paca()->kvm_slb, sizeof(get_paca()->kvm_slb));
  81         to_book3s(vcpu)->slb_shadow_max = get_paca()->kvm_slb_max;
  82 }
  83
  84 #if defined(AGGRESSIVE_DEC) || defined(EXIT_DEBUG)
  85 static u32 kvmppc_get_dec(struct kvm_vcpu *vcpu)
  86 {
  87         u64 jd = mftb() - vcpu->arch.dec_jiffies;
  88         return vcpu->arch.dec - jd;
  89 }
  90 #endif
  91
  92 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
  93 {
  94         ulong old_msr = vcpu->arch.msr;
  95
  96 #ifdef EXIT_DEBUG
  97         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
  98 #endif
  99         msr &= to_book3s(vcpu)->msr_mask;
 100         vcpu->arch.msr = msr;
 101         vcpu->arch.shadow_msr = msr | MSR_USER32;
 102         vcpu->arch.shadow_msr &= ( MSR_VEC | MSR_VSX | MSR_FP | MSR_FE0 |
 103                                    MSR_USER64 | MSR_SE | MSR_BE | MSR_DE |
 104                                    MSR_FE1);
 105
 106         if (msr & (MSR_WE|MSR_POW)) {
 107                 if (!vcpu->arch.pending_exceptions) {
 108                         kvm_vcpu_block(vcpu);
 109                         vcpu->stat.halt_wakeup++;
 110                 }
 111         }
 112
 113         if (((vcpu->arch.msr & (MSR_IR|MSR_DR)) != (old_msr & (MSR_IR|MSR_DR))) ||
 114             (vcpu->arch.msr & MSR_PR) != (old_msr & MSR_PR)) {
 115                 kvmppc_mmu_flush_segments(vcpu);
 116                 kvmppc_mmu_map_segment(vcpu, vcpu->arch.pc);
 117         }
 118 }
 119
 120 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
 121 {
 122         vcpu->arch.srr0 = vcpu->arch.pc;
 123         vcpu->arch.srr1 = vcpu->arch.msr | flags;
 124         vcpu->arch.pc = to_book3s(vcpu)->hior + vec;
 125         vcpu->arch.mmu.reset_msr(vcpu);
 126 }
 127
 128 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
 129 {
 130         unsigned int prio;
 131
 132         switch (vec) {
 133         case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET;         break;
 134         case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK;        break;
 135         case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE;         break;
 136         case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT;         break;
 137         case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE;         break;
 138         case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT;         break;
 139         case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL;             break;
 140         case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT;            break;
 141         case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM;              break;
 142         case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL;           break;
 143         case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER;          break;
 144         case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL;              break;
 145         case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG;                break;
 146         case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC;              break;
 147         case 0xf40: prio = BOOK3S_IRQPRIO_VSX;                  break;
 148         default:    prio = BOOK3S_IRQPRIO_MAX;                  break;
 149         }
 150
 151         return prio;
 152 }
 153
 154 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
 155 {
 156         vcpu->stat.queue_intr++;
 157
 158         set_bit(kvmppc_book3s_vec2irqprio(vec),
 159                 &vcpu->arch.pending_exceptions);
 160 #ifdef EXIT_DEBUG
 161         printk(KERN_INFO "Queueing interrupt %x\n", vec);
 162 #endif
 163 }
 164
 165
 166 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu)
 167 {
 168         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
 169 }
 170
 171 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 172 {
 173         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 174 }
 175
 176 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 177 {
 178         return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
 179 }
 180
 181 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 182                                 struct kvm_interrupt *irq)
 183 {
 184         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 185 }
 186
 187 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
 188 {
 189         int deliver = 1;
 190         int vec = 0;
 191
 192         switch (priority) {
 193         case BOOK3S_IRQPRIO_DECREMENTER:
 194                 deliver = vcpu->arch.msr & MSR_EE;
 195                 vec = BOOK3S_INTERRUPT_DECREMENTER;
 196                 break;
 197         case BOOK3S_IRQPRIO_EXTERNAL:
 198                 deliver = vcpu->arch.msr & MSR_EE;
 199                 vec = BOOK3S_INTERRUPT_EXTERNAL;
 200                 break;
 201         case BOOK3S_IRQPRIO_SYSTEM_RESET:
 202                 vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
 203                 break;
 204         case BOOK3S_IRQPRIO_MACHINE_CHECK:
 205                 vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
 206                 break;
 207         case BOOK3S_IRQPRIO_DATA_STORAGE:
 208                 vec = BOOK3S_INTERRUPT_DATA_STORAGE;
 209                 break;
 210         case BOOK3S_IRQPRIO_INST_STORAGE:
 211                 vec = BOOK3S_INTERRUPT_INST_STORAGE;
 212                 break;
 213         case BOOK3S_IRQPRIO_DATA_SEGMENT:
 214                 vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
 215                 break;
 216         case BOOK3S_IRQPRIO_INST_SEGMENT:
 217                 vec = BOOK3S_INTERRUPT_INST_SEGMENT;
 218                 break;
 219         case BOOK3S_IRQPRIO_ALIGNMENT:
 220                 vec = BOOK3S_INTERRUPT_ALIGNMENT;
 221                 break;
 222         case BOOK3S_IRQPRIO_PROGRAM:
 223                 vec = BOOK3S_INTERRUPT_PROGRAM;
 224                 break;
 225         case BOOK3S_IRQPRIO_VSX:
 226                 vec = BOOK3S_INTERRUPT_VSX;
 227                 break;
 228         case BOOK3S_IRQPRIO_ALTIVEC:
 229                 vec = BOOK3S_INTERRUPT_ALTIVEC;
 230                 break;
 231         case BOOK3S_IRQPRIO_FP_UNAVAIL:
 232                 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
 233                 break;
 234         case BOOK3S_IRQPRIO_SYSCALL:
 235                 vec = BOOK3S_INTERRUPT_SYSCALL;
 236                 break;
 237         case BOOK3S_IRQPRIO_DEBUG:
 238                 vec = BOOK3S_INTERRUPT_TRACE;
 239                 break;
 240         case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
 241                 vec = BOOK3S_INTERRUPT_PERFMON;
 242                 break;
 243         default:
 244                 deliver = 0;
 245                 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
 246                 break;
 247         }
 248
 249 #if 0
 250         printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
 251 #endif
 252
 253         if (deliver)
 254                 kvmppc_inject_interrupt(vcpu, vec, 0ULL);
 255
 256         return deliver;
 257 }
 258
 259 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 260 {
 261         unsigned long *pending = &vcpu->arch.pending_exceptions;
 262         unsigned int priority;
 263
 264         /* XXX be more clever here - no need to mftb() on every entry */
 265         /* Issue DEC again if it's still active */
 266 #ifdef AGGRESSIVE_DEC
 267         if (vcpu->arch.msr & MSR_EE)
 268                 if (kvmppc_get_dec(vcpu) & 0x80000000)
 269                         kvmppc_core_queue_dec(vcpu);
 270 #endif
 271
 272 #ifdef EXIT_DEBUG
 273         if (vcpu->arch.pending_exceptions)
 274                 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
 275 #endif
 276         priority = __ffs(*pending);
 277         while (priority <= (sizeof(unsigned int) * 8)) {
 278                 if (kvmppc_book3s_irqprio_deliver(vcpu, priority)) {
 279                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 280                         break;
 281                 }
 282
 283                 priority = find_next_bit(pending,
 284                                          BITS_PER_BYTE * sizeof(*pending),
 285                                          priority + 1);
 286         }
 287 }
 288
 289 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 290 {
 291         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 292         vcpu->arch.pvr = pvr;
 293         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 294                 kvmppc_mmu_book3s_64_init(vcpu);
 295                 to_book3s(vcpu)->hior = 0xfff00000;
 296                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 297         } else {
 298                 kvmppc_mmu_book3s_32_init(vcpu);
 299                 to_book3s(vcpu)->hior = 0;
 300                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 301         }
 302
 303         /* If we are in hypervisor level on 970, we can tell the CPU to
 304          * treat DCBZ as 32 bytes store */
 305         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 306         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 307             !strcmp(cur_cpu_spec->platform, "ppc970"))
 308                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 309
 310 }
 311
 312 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 313  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 314  * emulate 32 bytes dcbz length.
 315  *
 316  * The Book3s_64 inventors also realized this case and implemented a special bit
 317  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 318  *
 319  * My approach here is to patch the dcbz instruction on executing pages.
 320  */
 321 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 322 {
 323         bool touched = false;
 324         hva_t hpage;
 325         u32 *page;
 326         int i;
 327
 328         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 329         if (kvm_is_error_hva(hpage))
 330                 return;
 331
 332         hpage |= pte->raddr & ~PAGE_MASK;
 333         hpage &= ~0xFFFULL;
 334
 335         page = vmalloc(HW_PAGE_SIZE);
 336
 337         if (copy_from_user(page, (void __user *)hpage, HW_PAGE_SIZE))
 338                 goto out;
 339
 340         for (i=0; i < HW_PAGE_SIZE / 4; i++)
 341                 if ((page[i] & 0xff0007ff) == INS_DCBZ) {
 342                         page[i] &= 0xfffffff7; // reserved instruction, so we trap
 343                         touched = true;
 344                 }
 345
 346         if (touched)
 347                 copy_to_user((void __user *)hpage, page, HW_PAGE_SIZE);
 348
 349 out:
 350         vfree(page);
 351 }
 352
 353 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
 354                          struct kvmppc_pte *pte)
 355 {
 356         int relocated = (vcpu->arch.msr & (data ? MSR_DR : MSR_IR));
 357         int r;
 358
 359         if (relocated) {
 360                 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
 361         } else {
 362                 pte->eaddr = eaddr;
 363                 pte->raddr = eaddr & 0xffffffff;
 364                 pte->vpage = eaddr >> 12;
 365                 switch (vcpu->arch.msr & (MSR_DR|MSR_IR)) {
 366                 case 0:
 367                         pte->vpage |= VSID_REAL;
 368                 case MSR_DR:
 369                         pte->vpage |= VSID_REAL_DR;
 370                 case MSR_IR:
 371                         pte->vpage |= VSID_REAL_IR;
 372                 }
 373                 pte->may_read = true;
 374                 pte->may_write = true;
 375                 pte->may_execute = true;
 376                 r = 0;
 377         }
 378
 379         return r;
 380 }
 381
 382 static hva_t kvmppc_bad_hva(void)
 383 {
 384         return PAGE_OFFSET;
 385 }
 386
 387 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
 388                                bool read)
 389 {
 390         hva_t hpage;
 391
 392         if (read && !pte->may_read)
 393                 goto err;
 394
 395         if (!read && !pte->may_write)
 396                 goto err;
 397
 398         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 399         if (kvm_is_error_hva(hpage))
 400                 goto err;
 401
 402         return hpage | (pte->raddr & ~PAGE_MASK);
 403 err:
 404         return kvmppc_bad_hva();
 405 }
 406
 407 int kvmppc_st(struct kvm_vcpu *vcpu, ulong eaddr, int size, void *ptr)
 408 {
 409         struct kvmppc_pte pte;
 410         hva_t hva = eaddr;
 411
 412         vcpu->stat.st++;
 413
 414         if (kvmppc_xlate(vcpu, eaddr, false, &pte))
 415                 goto err;
 416
 417         hva = kvmppc_pte_to_hva(vcpu, &pte, false);
 418         if (kvm_is_error_hva(hva))
 419                 goto err;
 420
 421         if (copy_to_user((void __user *)hva, ptr, size)) {
 422                 printk(KERN_INFO "kvmppc_st at 0x%lx failed\n", hva);
 423                 goto err;
 424         }
 425
 426         return 0;
 427
 428 err:
 429         return -ENOENT;
 430 }
 431
 432 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong eaddr, int size, void *ptr,
 433                       bool data)
 434 {
 435         struct kvmppc_pte pte;
 436         hva_t hva = eaddr;
 437
 438         vcpu->stat.ld++;
 439
 440         if (kvmppc_xlate(vcpu, eaddr, data, &pte))
 441                 goto err;
 442
 443         hva = kvmppc_pte_to_hva(vcpu, &pte, true);
 444         if (kvm_is_error_hva(hva))
 445                 goto err;
 446
 447         if (copy_from_user(ptr, (void __user *)hva, size)) {
 448                 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
 449                 goto err;
 450         }
 451
 452         return 0;
 453
 454 err:
 455         return -ENOENT;
 456 }
 457
 458 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 459 {
 460         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 461 }
 462
 463 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 464                             ulong eaddr, int vec)
 465 {
 466         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 467         int r = RESUME_GUEST;
 468         int relocated;
 469         int page_found = 0;
 470         struct kvmppc_pte pte;
 471         bool is_mmio = false;
 472
 473         if ( vec == BOOK3S_INTERRUPT_DATA_STORAGE ) {
 474                 relocated = (vcpu->arch.msr & MSR_DR);
 475         } else {
 476                 relocated = (vcpu->arch.msr & MSR_IR);
 477         }
 478
 479         /* Resolve real address if translation turned on */
 480         if (relocated) {
 481                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 482         } else {
 483                 pte.may_execute = true;
 484                 pte.may_read = true;
 485                 pte.may_write = true;
 486                 pte.raddr = eaddr & 0xffffffff;
 487                 pte.eaddr = eaddr;
 488                 pte.vpage = eaddr >> 12;
 489                 switch (vcpu->arch.msr & (MSR_DR|MSR_IR)) {
 490                 case 0:
 491                         pte.vpage |= VSID_REAL;
 492                 case MSR_DR:
 493                         pte.vpage |= VSID_REAL_DR;
 494                 case MSR_IR:
 495                         pte.vpage |= VSID_REAL_IR;
 496                 }
 497         }
 498
 499         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 500            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 501                 /*
 502                  * If we do the dcbz hack, we have to NX on every execution,
 503                  * so we can patch the executing code. This renders our guest
 504                  * NX-less.
 505                  */
 506                 pte.may_execute = !data;
 507         }
 508
 509         if (page_found == -ENOENT) {
 510                 /* Page not found in guest PTE entries */
 511                 vcpu->arch.dear = vcpu->arch.fault_dear;
 512                 to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr;
 513                 vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x00000000f8000000ULL);
 514                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 515         } else if (page_found == -EPERM) {
 516                 /* Storage protection */
 517                 vcpu->arch.dear = vcpu->arch.fault_dear;
 518                 to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
 519                 to_book3s(vcpu)->dsisr |= DSISR_PROTFAULT;
 520                 vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x00000000f8000000ULL);
 521                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 522         } else if (page_found == -EINVAL) {
 523                 /* Page not found in guest SLB */
 524                 vcpu->arch.dear = vcpu->arch.fault_dear;
 525                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 526         } else if (!is_mmio &&
 527                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 528                 /* The guest's PTE is not mapped yet. Map on the host */
 529                 kvmppc_mmu_map_page(vcpu, &pte);
 530                 if (data)
 531                         vcpu->stat.sp_storage++;
 532                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 533                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 534                         kvmppc_patch_dcbz(vcpu, &pte);
 535         } else {
 536                 /* MMIO */
 537                 vcpu->stat.mmio_exits++;
 538                 vcpu->arch.paddr_accessed = pte.raddr;
 539                 r = kvmppc_emulate_mmio(run, vcpu);
 540                 if ( r == RESUME_HOST_NV )
 541                         r = RESUME_HOST;
 542                 if ( r == RESUME_GUEST_NV )
 543                         r = RESUME_GUEST;
 544         }
 545
 546         return r;
 547 }
 548
 549 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 550                        unsigned int exit_nr)
 551 {
 552         int r = RESUME_HOST;
 553
 554         vcpu->stat.sum_exits++;
 555
 556         run->exit_reason = KVM_EXIT_UNKNOWN;
 557         run->ready_for_interrupt_injection = 1;
 558 #ifdef EXIT_DEBUG
 559         printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | dec=0x%x | msr=0x%lx\n",
 560                 exit_nr, vcpu->arch.pc, vcpu->arch.fault_dear,
 561                 kvmppc_get_dec(vcpu), vcpu->arch.msr);
 562 #elif defined (EXIT_DEBUG_SIMPLE)
 563         if ((exit_nr != 0x900) && (exit_nr != 0x500))
 564                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | msr=0x%lx\n",
 565                         exit_nr, vcpu->arch.pc, vcpu->arch.fault_dear,
 566                         vcpu->arch.msr);
 567 #endif
 568         kvm_resched(vcpu);
 569         switch (exit_nr) {
 570         case BOOK3S_INTERRUPT_INST_STORAGE:
 571                 vcpu->stat.pf_instruc++;
 572                 /* only care about PTEG not found errors, but leave NX alone */
 573                 if (vcpu->arch.shadow_msr & 0x40000000) {
 574                         r = kvmppc_handle_pagefault(run, vcpu, vcpu->arch.pc, exit_nr);
 575                         vcpu->stat.sp_instruc++;
 576                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 577                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 578                         /*
 579                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 580                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 581                          *     that no guest that needs the dcbz hack does NX.
 582                          */
 583                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.pc, ~0xFFFULL);
 584                 } else {
 585                         vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x58000000);
 586                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 587                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.pc, ~0xFFFULL);
 588                         r = RESUME_GUEST;
 589                 }
 590                 break;
 591         case BOOK3S_INTERRUPT_DATA_STORAGE:
 592                 vcpu->stat.pf_storage++;
 593                 /* The only case we need to handle is missing shadow PTEs */
 594                 if (vcpu->arch.fault_dsisr & DSISR_NOHPTE) {
 595                         r = kvmppc_handle_pagefault(run, vcpu, vcpu->arch.fault_dear, exit_nr);
 596                 } else {
 597                         vcpu->arch.dear = vcpu->arch.fault_dear;
 598                         to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr;
 599                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 600                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.dear, ~0xFFFULL);
 601                         r = RESUME_GUEST;
 602                 }
 603                 break;
 604         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 605                 if (kvmppc_mmu_map_segment(vcpu, vcpu->arch.fault_dear) < 0) {
 606                         vcpu->arch.dear = vcpu->arch.fault_dear;
 607                         kvmppc_book3s_queue_irqprio(vcpu,
 608                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 609                 }
 610                 r = RESUME_GUEST;
 611                 break;
 612         case BOOK3S_INTERRUPT_INST_SEGMENT:
 613                 if (kvmppc_mmu_map_segment(vcpu, vcpu->arch.pc) < 0) {
 614                         kvmppc_book3s_queue_irqprio(vcpu,
 615                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 616                 }
 617                 r = RESUME_GUEST;
 618                 break;
 619         /* We're good on these - the host merely wanted to get our attention */
 620         case BOOK3S_INTERRUPT_DECREMENTER:
 621                 vcpu->stat.dec_exits++;
 622                 r = RESUME_GUEST;
 623                 break;
 624         case BOOK3S_INTERRUPT_EXTERNAL:
 625                 vcpu->stat.ext_intr_exits++;
 626                 r = RESUME_GUEST;
 627                 break;
 628         case BOOK3S_INTERRUPT_PROGRAM:
 629         {
 630                 enum emulation_result er;
 631
 632                 if (vcpu->arch.msr & MSR_PR) {
 633 #ifdef EXIT_DEBUG
 634                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", vcpu->arch.pc, vcpu->arch.last_inst);
 635 #endif
 636                         if ((vcpu->arch.last_inst & 0xff0007ff) !=
 637                             (INS_DCBZ & 0xfffffff7)) {
 638                                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 639                                 r = RESUME_GUEST;
 640                                 break;
 641                         }
 642                 }
 643
 644                 vcpu->stat.emulated_inst_exits++;
 645                 er = kvmppc_emulate_instruction(run, vcpu);
 646                 switch (er) {
 647                 case EMULATE_DONE:
 648                         r = RESUME_GUEST;
 649                         break;
 650                 case EMULATE_FAIL:
 651                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 652                                __func__, vcpu->arch.pc, vcpu->arch.last_inst);
 653                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 654                         r = RESUME_GUEST;
 655                         break;
 656                 default:
 657                         BUG();
 658                 }
 659                 break;
 660         }
 661         case BOOK3S_INTERRUPT_SYSCALL:
 662 #ifdef EXIT_DEBUG
 663                 printk(KERN_INFO "Syscall Nr %d\n", (int)vcpu->arch.gpr[0]);
 664 #endif
 665                 vcpu->stat.syscall_exits++;
 666                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 667                 r = RESUME_GUEST;
 668                 break;
 669         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 670         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 671         case BOOK3S_INTERRUPT_TRACE:
 672         case BOOK3S_INTERRUPT_ALTIVEC:
 673         case BOOK3S_INTERRUPT_VSX:
 674                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 675                 r = RESUME_GUEST;
 676                 break;
 677         default:
 678                 /* Ugh - bork here! What did we get? */
 679                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n", exit_nr, vcpu->arch.pc, vcpu->arch.shadow_msr);
 680                 r = RESUME_HOST;
 681                 BUG();
 682                 break;
 683         }
 684
 685
 686         if (!(r & RESUME_HOST)) {
 687                 /* To avoid clobbering exit_reason, only check for signals if
 688                  * we aren't already exiting to userspace for some other
 689                  * reason. */
 690                 if (signal_pending(current)) {
 691 #ifdef EXIT_DEBUG
 692                         printk(KERN_EMERG "KVM: Going back to host\n");
 693 #endif
 694                         vcpu->stat.signal_exits++;
 695                         run->exit_reason = KVM_EXIT_INTR;
 696                         r = -EINTR;
 697                 } else {
 698                         /* In case an interrupt came in that was triggered
 699                          * from userspace (like DEC), we need to check what
 700                          * to inject now! */
 701                         kvmppc_core_deliver_interrupts(vcpu);
 702                 }
 703         }
 704
 705 #ifdef EXIT_DEBUG
 706         printk(KERN_EMERG "KVM exit: vcpu=0x%p pc=0x%lx r=0x%x\n", vcpu, vcpu->arch.pc, r);
 707 #endif
 708
 709         return r;
 710 }
 711
 712 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 713 {
 714         return 0;
 715 }
 716
 717 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 718 {
 719         int i;
 720
 721         regs->pc = vcpu->arch.pc;
 722         regs->cr = vcpu->arch.cr;
 723         regs->ctr = vcpu->arch.ctr;
 724         regs->lr = vcpu->arch.lr;
 725         regs->xer = vcpu->arch.xer;
 726         regs->msr = vcpu->arch.msr;
 727         regs->srr0 = vcpu->arch.srr0;
 728         regs->srr1 = vcpu->arch.srr1;
 729         regs->pid = vcpu->arch.pid;
 730         regs->sprg0 = vcpu->arch.sprg0;
 731         regs->sprg1 = vcpu->arch.sprg1;
 732         regs->sprg2 = vcpu->arch.sprg2;
 733         regs->sprg3 = vcpu->arch.sprg3;
 734         regs->sprg5 = vcpu->arch.sprg4;
 735         regs->sprg6 = vcpu->arch.sprg5;
 736         regs->sprg7 = vcpu->arch.sprg6;
 737
 738         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 739                 regs->gpr[i] = vcpu->arch.gpr[i];
 740
 741         return 0;
 742 }
 743
 744 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 745 {
 746         int i;
 747
 748         vcpu->arch.pc = regs->pc;
 749         vcpu->arch.cr = regs->cr;
 750         vcpu->arch.ctr = regs->ctr;
 751         vcpu->arch.lr = regs->lr;
 752         vcpu->arch.xer = regs->xer;
 753         kvmppc_set_msr(vcpu, regs->msr);
 754         vcpu->arch.srr0 = regs->srr0;
 755         vcpu->arch.srr1 = regs->srr1;
 756         vcpu->arch.sprg0 = regs->sprg0;
 757         vcpu->arch.sprg1 = regs->sprg1;
 758         vcpu->arch.sprg2 = regs->sprg2;
 759         vcpu->arch.sprg3 = regs->sprg3;
 760         vcpu->arch.sprg5 = regs->sprg4;
 761         vcpu->arch.sprg6 = regs->sprg5;
 762         vcpu->arch.sprg7 = regs->sprg6;
 763
 764         for (i = 0; i < ARRAY_SIZE(vcpu->arch.gpr); i++)
 765                 vcpu->arch.gpr[i] = regs->gpr[i];
 766
 767         return 0;
 768 }
 769
 770 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 771                                   struct kvm_sregs *sregs)
 772 {
 773         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 774         int i;
 775
 776         sregs->pvr = vcpu->arch.pvr;
 777
 778         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 779         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 780                 for (i = 0; i < 64; i++) {
 781                         sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
 782                         sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
 783                 }
 784         } else {
 785                 for (i = 0; i < 16; i++) {
 786                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
 787                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
 788                 }
 789                 for (i = 0; i < 8; i++) {
 790                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 791                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 792                 }
 793         }
 794         return 0;
 795 }
 796
 797 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 798                                   struct kvm_sregs *sregs)
 799 {
 800         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 801         int i;
 802
 803         kvmppc_set_pvr(vcpu, sregs->pvr);
 804
 805         vcpu3s->sdr1 = sregs->u.s.sdr1;
 806         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 807                 for (i = 0; i < 64; i++) {
 808                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 809                                                     sregs->u.s.ppc64.slb[i].slbe);
 810                 }
 811         } else {
 812                 for (i = 0; i < 16; i++) {
 813                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 814                 }
 815                 for (i = 0; i < 8; i++) {
 816                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 817                                        (u32)sregs->u.s.ppc32.ibat[i]);
 818                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 819                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 820                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 821                                        (u32)sregs->u.s.ppc32.dbat[i]);
 822                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 823                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 824                 }
 825         }
 826
 827         /* Flush the MMU after messing with the segments */
 828         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 829         return 0;
 830 }
 831
 832 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 833 {
 834         return -ENOTSUPP;
 835 }
 836
 837 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 838 {
 839         return -ENOTSUPP;
 840 }
 841
 842 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 843                                   struct kvm_translation *tr)
 844 {
 845         return 0;
 846 }
 847
 848 /*
 849  * Get (and clear) the dirty memory log for a memory slot.
 850  */
 851 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 852                                       struct kvm_dirty_log *log)
 853 {
 854         struct kvm_memory_slot *memslot;
 855         struct kvm_vcpu *vcpu;
 856         ulong ga, ga_end;
 857         int is_dirty = 0;
 858         int r, n;
 859
 860         down_write(&kvm->slots_lock);
 861
 862         r = kvm_get_dirty_log(kvm, log, &is_dirty);
 863         if (r)
 864                 goto out;
 865
 866         /* If nothing is dirty, don't bother messing with page tables. */
 867         if (is_dirty) {
 868                 memslot = &kvm->memslots[log->slot];
 869
 870                 ga = memslot->base_gfn << PAGE_SHIFT;
 871                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
 872
 873                 kvm_for_each_vcpu(n, vcpu, kvm)
 874                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
 875
 876                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 877                 memset(memslot->dirty_bitmap, 0, n);
 878         }
 879
 880         r = 0;
 881 out:
 882         up_write(&kvm->slots_lock);
 883         return r;
 884 }
 885
 886 int kvmppc_core_check_processor_compat(void)
 887 {
 888         return 0;
 889 }
 890
 891 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 892 {
 893         struct kvmppc_vcpu_book3s *vcpu_book3s;
 894         struct kvm_vcpu *vcpu;
 895         int err;
 896
 897         vcpu_book3s = (struct kvmppc_vcpu_book3s *)__get_free_pages( GFP_KERNEL | __GFP_ZERO,
 898                         get_order(sizeof(struct kvmppc_vcpu_book3s)));
 899         if (!vcpu_book3s) {
 900                 err = -ENOMEM;
 901                 goto out;
 902         }
 903
 904         vcpu = &vcpu_book3s->vcpu;
 905         err = kvm_vcpu_init(vcpu, kvm, id);
 906         if (err)
 907                 goto free_vcpu;
 908
 909         vcpu->arch.host_retip = kvm_return_point;
 910         vcpu->arch.host_msr = mfmsr();
 911         /* default to book3s_64 (970fx) */
 912         vcpu->arch.pvr = 0x3C0301;
 913         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 914         vcpu_book3s->slb_nr = 64;
 915
 916         /* remember where some real-mode handlers are */
 917         vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
 918         vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
 919         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
 920
 921         vcpu->arch.shadow_msr = MSR_USER64;
 922
 923         err = __init_new_context();
 924         if (err < 0)
 925                 goto free_vcpu;
 926         vcpu_book3s->context_id = err;
 927
 928         vcpu_book3s->vsid_max = ((vcpu_book3s->context_id + 1) << USER_ESID_BITS) - 1;
 929         vcpu_book3s->vsid_first = vcpu_book3s->context_id << USER_ESID_BITS;
 930         vcpu_book3s->vsid_next = vcpu_book3s->vsid_first;
 931
 932         return vcpu;
 933
 934 free_vcpu:
 935         free_pages((long)vcpu_book3s, get_order(sizeof(struct kvmppc_vcpu_book3s)));
 936 out:
 937         return ERR_PTR(err);
 938 }
 939
 940 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 941 {
 942         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
 943
 944         __destroy_context(vcpu_book3s->context_id);
 945         kvm_vcpu_uninit(vcpu);
 946         free_pages((long)vcpu_book3s, get_order(sizeof(struct kvmppc_vcpu_book3s)));
 947 }
 948
 949 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
 950 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 951 {
 952         int ret;
 953
 954         /* No need to go into the guest when all we do is going out */
 955         if (signal_pending(current)) {
 956                 kvm_run->exit_reason = KVM_EXIT_INTR;
 957                 return -EINTR;
 958         }
 959
 960         /* XXX we get called with irq disabled - change that! */
 961         local_irq_enable();
 962
 963         ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
 964
 965         local_irq_disable();
 966
 967         return ret;
 968 }
 969
 970 static int kvmppc_book3s_init(void)
 971 {
 972         return kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), THIS_MODULE);
 973 }
 974
 975 static void kvmppc_book3s_exit(void)
 976 {
 977         kvm_exit();
 978 }
 979
 980 module_init(kvmppc_book3s_init);
 981 module_exit(kvmppc_book3s_exit);