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 struct kvm_stats_debugfs_item debugfs_entries[] = {
  38         { "exits",       VCPU_STAT(sum_exits) },
  39         { "mmio",        VCPU_STAT(mmio_exits) },
  40         { "sig",         VCPU_STAT(signal_exits) },
  41         { "sysc",        VCPU_STAT(syscall_exits) },
  42         { "inst_emu",    VCPU_STAT(emulated_inst_exits) },
  43         { "dec",         VCPU_STAT(dec_exits) },
  44         { "ext_intr",    VCPU_STAT(ext_intr_exits) },
  45         { "queue_intr",  VCPU_STAT(queue_intr) },
  46         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  47         { "pf_storage",  VCPU_STAT(pf_storage) },
  48         { "sp_storage",  VCPU_STAT(sp_storage) },
  49         { "pf_instruc",  VCPU_STAT(pf_instruc) },
  50         { "sp_instruc",  VCPU_STAT(sp_instruc) },
  51         { "ld",          VCPU_STAT(ld) },
  52         { "ld_slow",     VCPU_STAT(ld_slow) },
  53         { "st",          VCPU_STAT(st) },
  54         { "st_slow",     VCPU_STAT(st_slow) },
  55         { NULL }
  56 };
  57
  58 void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
  59 {
  60 }
  61
  62 void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
  63 {
  64 }
  65
  66 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  67 {
  68         memcpy(get_paca()->kvm_slb, to_book3s(vcpu)->slb_shadow, sizeof(get_paca()->kvm_slb));
  69         get_paca()->kvm_slb_max = to_book3s(vcpu)->slb_shadow_max;
  70 }
  71
  72 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  73 {
  74         memcpy(to_book3s(vcpu)->slb_shadow, get_paca()->kvm_slb, sizeof(get_paca()->kvm_slb));
  75         to_book3s(vcpu)->slb_shadow_max = get_paca()->kvm_slb_max;
  76 }
  77
  78 #if defined(EXIT_DEBUG)
  79 static u32 kvmppc_get_dec(struct kvm_vcpu *vcpu)
  80 {
  81         u64 jd = mftb() - vcpu->arch.dec_jiffies;
  82         return vcpu->arch.dec - jd;
  83 }
  84 #endif
  85
  86 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
  87 {
  88         ulong old_msr = vcpu->arch.msr;
  89
  90 #ifdef EXIT_DEBUG
  91         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
  92 #endif
  93         msr &= to_book3s(vcpu)->msr_mask;
  94         vcpu->arch.msr = msr;
  95         vcpu->arch.shadow_msr = msr | MSR_USER32;
  96         vcpu->arch.shadow_msr &= ( MSR_VEC | MSR_VSX | MSR_FP | MSR_FE0 |
  97                                    MSR_USER64 | MSR_SE | MSR_BE | MSR_DE |
  98                                    MSR_FE1);
  99
 100         if (msr & (MSR_WE|MSR_POW)) {
 101                 if (!vcpu->arch.pending_exceptions) {
 102                         kvm_vcpu_block(vcpu);
 103                         vcpu->stat.halt_wakeup++;
 104                 }
 105         }
 106
 107         if (((vcpu->arch.msr & (MSR_IR|MSR_DR)) != (old_msr & (MSR_IR|MSR_DR))) ||
 108             (vcpu->arch.msr & MSR_PR) != (old_msr & MSR_PR)) {
 109                 kvmppc_mmu_flush_segments(vcpu);
 110                 kvmppc_mmu_map_segment(vcpu, vcpu->arch.pc);
 111         }
 112 }
 113
 114 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
 115 {
 116         vcpu->arch.srr0 = vcpu->arch.pc;
 117         vcpu->arch.srr1 = vcpu->arch.msr | flags;
 118         vcpu->arch.pc = to_book3s(vcpu)->hior + vec;
 119         vcpu->arch.mmu.reset_msr(vcpu);
 120 }
 121
 122 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
 123 {
 124         unsigned int prio;
 125
 126         switch (vec) {
 127         case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET;         break;
 128         case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK;        break;
 129         case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE;         break;
 130         case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT;         break;
 131         case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE;         break;
 132         case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT;         break;
 133         case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL;             break;
 134         case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT;            break;
 135         case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM;              break;
 136         case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL;           break;
 137         case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER;          break;
 138         case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL;              break;
 139         case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG;                break;
 140         case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC;              break;
 141         case 0xf40: prio = BOOK3S_IRQPRIO_VSX;                  break;
 142         default:    prio = BOOK3S_IRQPRIO_MAX;                  break;
 143         }
 144
 145         return prio;
 146 }
 147
 148 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
 149                                           unsigned int vec)
 150 {
 151         clear_bit(kvmppc_book3s_vec2irqprio(vec),
 152                   &vcpu->arch.pending_exceptions);
 153 }
 154
 155 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
 156 {
 157         vcpu->stat.queue_intr++;
 158
 159         set_bit(kvmppc_book3s_vec2irqprio(vec),
 160                 &vcpu->arch.pending_exceptions);
 161 #ifdef EXIT_DEBUG
 162         printk(KERN_INFO "Queueing interrupt %x\n", vec);
 163 #endif
 164 }
 165
 166
 167 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu)
 168 {
 169         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
 170 }
 171
 172 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 173 {
 174         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 175 }
 176
 177 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 178 {
 179         return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
 180 }
 181
 182 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 183 {
 184         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 185 }
 186
 187 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 188                                 struct kvm_interrupt *irq)
 189 {
 190         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 191 }
 192
 193 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
 194 {
 195         int deliver = 1;
 196         int vec = 0;
 197
 198         switch (priority) {
 199         case BOOK3S_IRQPRIO_DECREMENTER:
 200                 deliver = vcpu->arch.msr & MSR_EE;
 201                 vec = BOOK3S_INTERRUPT_DECREMENTER;
 202                 break;
 203         case BOOK3S_IRQPRIO_EXTERNAL:
 204                 deliver = vcpu->arch.msr & MSR_EE;
 205                 vec = BOOK3S_INTERRUPT_EXTERNAL;
 206                 break;
 207         case BOOK3S_IRQPRIO_SYSTEM_RESET:
 208                 vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
 209                 break;
 210         case BOOK3S_IRQPRIO_MACHINE_CHECK:
 211                 vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
 212                 break;
 213         case BOOK3S_IRQPRIO_DATA_STORAGE:
 214                 vec = BOOK3S_INTERRUPT_DATA_STORAGE;
 215                 break;
 216         case BOOK3S_IRQPRIO_INST_STORAGE:
 217                 vec = BOOK3S_INTERRUPT_INST_STORAGE;
 218                 break;
 219         case BOOK3S_IRQPRIO_DATA_SEGMENT:
 220                 vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
 221                 break;
 222         case BOOK3S_IRQPRIO_INST_SEGMENT:
 223                 vec = BOOK3S_INTERRUPT_INST_SEGMENT;
 224                 break;
 225         case BOOK3S_IRQPRIO_ALIGNMENT:
 226                 vec = BOOK3S_INTERRUPT_ALIGNMENT;
 227                 break;
 228         case BOOK3S_IRQPRIO_PROGRAM:
 229                 vec = BOOK3S_INTERRUPT_PROGRAM;
 230                 break;
 231         case BOOK3S_IRQPRIO_VSX:
 232                 vec = BOOK3S_INTERRUPT_VSX;
 233                 break;
 234         case BOOK3S_IRQPRIO_ALTIVEC:
 235                 vec = BOOK3S_INTERRUPT_ALTIVEC;
 236                 break;
 237         case BOOK3S_IRQPRIO_FP_UNAVAIL:
 238                 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
 239                 break;
 240         case BOOK3S_IRQPRIO_SYSCALL:
 241                 vec = BOOK3S_INTERRUPT_SYSCALL;
 242                 break;
 243         case BOOK3S_IRQPRIO_DEBUG:
 244                 vec = BOOK3S_INTERRUPT_TRACE;
 245                 break;
 246         case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
 247                 vec = BOOK3S_INTERRUPT_PERFMON;
 248                 break;
 249         default:
 250                 deliver = 0;
 251                 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
 252                 break;
 253         }
 254
 255 #if 0
 256         printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
 257 #endif
 258
 259         if (deliver)
 260                 kvmppc_inject_interrupt(vcpu, vec, 0ULL);
 261
 262         return deliver;
 263 }
 264
 265 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 266 {
 267         unsigned long *pending = &vcpu->arch.pending_exceptions;
 268         unsigned int priority;
 269
 270 #ifdef EXIT_DEBUG
 271         if (vcpu->arch.pending_exceptions)
 272                 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
 273 #endif
 274         priority = __ffs(*pending);
 275         while (priority <= (sizeof(unsigned int) * 8)) {
 276                 if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
 277                     (priority != BOOK3S_IRQPRIO_DECREMENTER)) {
 278                         /* DEC interrupts get cleared by mtdec */
 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         }
 543
 544         return r;
 545 }
 546
 547 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 548                        unsigned int exit_nr)
 549 {
 550         int r = RESUME_HOST;
 551
 552         vcpu->stat.sum_exits++;
 553
 554         run->exit_reason = KVM_EXIT_UNKNOWN;
 555         run->ready_for_interrupt_injection = 1;
 556 #ifdef EXIT_DEBUG
 557         printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | dec=0x%x | msr=0x%lx\n",
 558                 exit_nr, vcpu->arch.pc, vcpu->arch.fault_dear,
 559                 kvmppc_get_dec(vcpu), vcpu->arch.msr);
 560 #elif defined (EXIT_DEBUG_SIMPLE)
 561         if ((exit_nr != 0x900) && (exit_nr != 0x500))
 562                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | msr=0x%lx\n",
 563                         exit_nr, vcpu->arch.pc, vcpu->arch.fault_dear,
 564                         vcpu->arch.msr);
 565 #endif
 566         kvm_resched(vcpu);
 567         switch (exit_nr) {
 568         case BOOK3S_INTERRUPT_INST_STORAGE:
 569                 vcpu->stat.pf_instruc++;
 570                 /* only care about PTEG not found errors, but leave NX alone */
 571                 if (vcpu->arch.shadow_msr & 0x40000000) {
 572                         r = kvmppc_handle_pagefault(run, vcpu, vcpu->arch.pc, exit_nr);
 573                         vcpu->stat.sp_instruc++;
 574                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 575                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 576                         /*
 577                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 578                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 579                          *     that no guest that needs the dcbz hack does NX.
 580                          */
 581                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.pc, ~0xFFFULL);
 582                 } else {
 583                         vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x58000000);
 584                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 585                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.pc, ~0xFFFULL);
 586                         r = RESUME_GUEST;
 587                 }
 588                 break;
 589         case BOOK3S_INTERRUPT_DATA_STORAGE:
 590                 vcpu->stat.pf_storage++;
 591                 /* The only case we need to handle is missing shadow PTEs */
 592                 if (vcpu->arch.fault_dsisr & DSISR_NOHPTE) {
 593                         r = kvmppc_handle_pagefault(run, vcpu, vcpu->arch.fault_dear, exit_nr);
 594                 } else {
 595                         vcpu->arch.dear = vcpu->arch.fault_dear;
 596                         to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr;
 597                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 598                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.dear, ~0xFFFULL);
 599                         r = RESUME_GUEST;
 600                 }
 601                 break;
 602         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 603                 if (kvmppc_mmu_map_segment(vcpu, vcpu->arch.fault_dear) < 0) {
 604                         vcpu->arch.dear = vcpu->arch.fault_dear;
 605                         kvmppc_book3s_queue_irqprio(vcpu,
 606                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 607                 }
 608                 r = RESUME_GUEST;
 609                 break;
 610         case BOOK3S_INTERRUPT_INST_SEGMENT:
 611                 if (kvmppc_mmu_map_segment(vcpu, vcpu->arch.pc) < 0) {
 612                         kvmppc_book3s_queue_irqprio(vcpu,
 613                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 614                 }
 615                 r = RESUME_GUEST;
 616                 break;
 617         /* We're good on these - the host merely wanted to get our attention */
 618         case BOOK3S_INTERRUPT_DECREMENTER:
 619                 vcpu->stat.dec_exits++;
 620                 r = RESUME_GUEST;
 621                 break;
 622         case BOOK3S_INTERRUPT_EXTERNAL:
 623                 vcpu->stat.ext_intr_exits++;
 624                 r = RESUME_GUEST;
 625                 break;
 626         case BOOK3S_INTERRUPT_PROGRAM:
 627         {
 628                 enum emulation_result er;
 629
 630                 if (vcpu->arch.msr & MSR_PR) {
 631 #ifdef EXIT_DEBUG
 632                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", vcpu->arch.pc, vcpu->arch.last_inst);
 633 #endif
 634                         if ((vcpu->arch.last_inst & 0xff0007ff) !=
 635                             (INS_DCBZ & 0xfffffff7)) {
 636                                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 637                                 r = RESUME_GUEST;
 638                                 break;
 639                         }
 640                 }
 641
 642                 vcpu->stat.emulated_inst_exits++;
 643                 er = kvmppc_emulate_instruction(run, vcpu);
 644                 switch (er) {
 645                 case EMULATE_DONE:
 646                         r = RESUME_GUEST_NV;
 647                         break;
 648                 case EMULATE_FAIL:
 649                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 650                                __func__, vcpu->arch.pc, vcpu->arch.last_inst);
 651                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 652                         r = RESUME_GUEST;
 653                         break;
 654                 default:
 655                         BUG();
 656                 }
 657                 break;
 658         }
 659         case BOOK3S_INTERRUPT_SYSCALL:
 660 #ifdef EXIT_DEBUG
 661                 printk(KERN_INFO "Syscall Nr %d\n", (int)kvmppc_get_gpr(vcpu, 0));
 662 #endif
 663                 vcpu->stat.syscall_exits++;
 664                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 665                 r = RESUME_GUEST;
 666                 break;
 667         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 668         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 669         case BOOK3S_INTERRUPT_TRACE:
 670         case BOOK3S_INTERRUPT_ALTIVEC:
 671         case BOOK3S_INTERRUPT_VSX:
 672                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 673                 r = RESUME_GUEST;
 674                 break;
 675         default:
 676                 /* Ugh - bork here! What did we get? */
 677                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n", exit_nr, vcpu->arch.pc, vcpu->arch.shadow_msr);
 678                 r = RESUME_HOST;
 679                 BUG();
 680                 break;
 681         }
 682
 683
 684         if (!(r & RESUME_HOST)) {
 685                 /* To avoid clobbering exit_reason, only check for signals if
 686                  * we aren't already exiting to userspace for some other
 687                  * reason. */
 688                 if (signal_pending(current)) {
 689 #ifdef EXIT_DEBUG
 690                         printk(KERN_EMERG "KVM: Going back to host\n");
 691 #endif
 692                         vcpu->stat.signal_exits++;
 693                         run->exit_reason = KVM_EXIT_INTR;
 694                         r = -EINTR;
 695                 } else {
 696                         /* In case an interrupt came in that was triggered
 697                          * from userspace (like DEC), we need to check what
 698                          * to inject now! */
 699                         kvmppc_core_deliver_interrupts(vcpu);
 700                 }
 701         }
 702
 703 #ifdef EXIT_DEBUG
 704         printk(KERN_EMERG "KVM exit: vcpu=0x%p pc=0x%lx r=0x%x\n", vcpu, vcpu->arch.pc, r);
 705 #endif
 706
 707         return r;
 708 }
 709
 710 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 711 {
 712         return 0;
 713 }
 714
 715 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 716 {
 717         int i;
 718
 719         regs->pc = vcpu->arch.pc;
 720         regs->cr = kvmppc_get_cr(vcpu);
 721         regs->ctr = vcpu->arch.ctr;
 722         regs->lr = vcpu->arch.lr;
 723         regs->xer = kvmppc_get_xer(vcpu);
 724         regs->msr = vcpu->arch.msr;
 725         regs->srr0 = vcpu->arch.srr0;
 726         regs->srr1 = vcpu->arch.srr1;
 727         regs->pid = vcpu->arch.pid;
 728         regs->sprg0 = vcpu->arch.sprg0;
 729         regs->sprg1 = vcpu->arch.sprg1;
 730         regs->sprg2 = vcpu->arch.sprg2;
 731         regs->sprg3 = vcpu->arch.sprg3;
 732         regs->sprg5 = vcpu->arch.sprg4;
 733         regs->sprg6 = vcpu->arch.sprg5;
 734         regs->sprg7 = vcpu->arch.sprg6;
 735
 736         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 737                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
 738
 739         return 0;
 740 }
 741
 742 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 743 {
 744         int i;
 745
 746         vcpu->arch.pc = regs->pc;
 747         kvmppc_set_cr(vcpu, regs->cr);
 748         vcpu->arch.ctr = regs->ctr;
 749         vcpu->arch.lr = regs->lr;
 750         kvmppc_set_xer(vcpu, regs->xer);
 751         kvmppc_set_msr(vcpu, regs->msr);
 752         vcpu->arch.srr0 = regs->srr0;
 753         vcpu->arch.srr1 = regs->srr1;
 754         vcpu->arch.sprg0 = regs->sprg0;
 755         vcpu->arch.sprg1 = regs->sprg1;
 756         vcpu->arch.sprg2 = regs->sprg2;
 757         vcpu->arch.sprg3 = regs->sprg3;
 758         vcpu->arch.sprg5 = regs->sprg4;
 759         vcpu->arch.sprg6 = regs->sprg5;
 760         vcpu->arch.sprg7 = regs->sprg6;
 761
 762         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 763                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
 764
 765         return 0;
 766 }
 767
 768 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 769                                   struct kvm_sregs *sregs)
 770 {
 771         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 772         int i;
 773
 774         sregs->pvr = vcpu->arch.pvr;
 775
 776         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 777         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 778                 for (i = 0; i < 64; i++) {
 779                         sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
 780                         sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
 781                 }
 782         } else {
 783                 for (i = 0; i < 16; i++) {
 784                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
 785                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
 786                 }
 787                 for (i = 0; i < 8; i++) {
 788                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 789                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 790                 }
 791         }
 792         return 0;
 793 }
 794
 795 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 796                                   struct kvm_sregs *sregs)
 797 {
 798         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 799         int i;
 800
 801         kvmppc_set_pvr(vcpu, sregs->pvr);
 802
 803         vcpu3s->sdr1 = sregs->u.s.sdr1;
 804         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 805                 for (i = 0; i < 64; i++) {
 806                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 807                                                     sregs->u.s.ppc64.slb[i].slbe);
 808                 }
 809         } else {
 810                 for (i = 0; i < 16; i++) {
 811                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 812                 }
 813                 for (i = 0; i < 8; i++) {
 814                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 815                                        (u32)sregs->u.s.ppc32.ibat[i]);
 816                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 817                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 818                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 819                                        (u32)sregs->u.s.ppc32.dbat[i]);
 820                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 821                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 822                 }
 823         }
 824
 825         /* Flush the MMU after messing with the segments */
 826         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 827         return 0;
 828 }
 829
 830 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 831 {
 832         return -ENOTSUPP;
 833 }
 834
 835 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 836 {
 837         return -ENOTSUPP;
 838 }
 839
 840 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 841                                   struct kvm_translation *tr)
 842 {
 843         return 0;
 844 }
 845
 846 /*
 847  * Get (and clear) the dirty memory log for a memory slot.
 848  */
 849 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 850                                       struct kvm_dirty_log *log)
 851 {
 852         struct kvm_memory_slot *memslot;
 853         struct kvm_vcpu *vcpu;
 854         ulong ga, ga_end;
 855         int is_dirty = 0;
 856         int r, n;
 857
 858         mutex_lock(&kvm->slots_lock);
 859
 860         r = kvm_get_dirty_log(kvm, log, &is_dirty);
 861         if (r)
 862                 goto out;
 863
 864         /* If nothing is dirty, don't bother messing with page tables. */
 865         if (is_dirty) {
 866                 memslot = &kvm->memslots->memslots[log->slot];
 867
 868                 ga = memslot->base_gfn << PAGE_SHIFT;
 869                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
 870
 871                 kvm_for_each_vcpu(n, vcpu, kvm)
 872                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
 873
 874                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 875                 memset(memslot->dirty_bitmap, 0, n);
 876         }
 877
 878         r = 0;
 879 out:
 880         mutex_unlock(&kvm->slots_lock);
 881         return r;
 882 }
 883
 884 int kvmppc_core_check_processor_compat(void)
 885 {
 886         return 0;
 887 }
 888
 889 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 890 {
 891         struct kvmppc_vcpu_book3s *vcpu_book3s;
 892         struct kvm_vcpu *vcpu;
 893         int err;
 894
 895         vcpu_book3s = (struct kvmppc_vcpu_book3s *)__get_free_pages( GFP_KERNEL | __GFP_ZERO,
 896                         get_order(sizeof(struct kvmppc_vcpu_book3s)));
 897         if (!vcpu_book3s) {
 898                 err = -ENOMEM;
 899                 goto out;
 900         }
 901
 902         vcpu = &vcpu_book3s->vcpu;
 903         err = kvm_vcpu_init(vcpu, kvm, id);
 904         if (err)
 905                 goto free_vcpu;
 906
 907         vcpu->arch.host_retip = kvm_return_point;
 908         vcpu->arch.host_msr = mfmsr();
 909         /* default to book3s_64 (970fx) */
 910         vcpu->arch.pvr = 0x3C0301;
 911         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 912         vcpu_book3s->slb_nr = 64;
 913
 914         /* remember where some real-mode handlers are */
 915         vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
 916         vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
 917         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
 918
 919         vcpu->arch.shadow_msr = MSR_USER64;
 920
 921         err = __init_new_context();
 922         if (err < 0)
 923                 goto free_vcpu;
 924         vcpu_book3s->context_id = err;
 925
 926         vcpu_book3s->vsid_max = ((vcpu_book3s->context_id + 1) << USER_ESID_BITS) - 1;
 927         vcpu_book3s->vsid_first = vcpu_book3s->context_id << USER_ESID_BITS;
 928         vcpu_book3s->vsid_next = vcpu_book3s->vsid_first;
 929
 930         return vcpu;
 931
 932 free_vcpu:
 933         free_pages((long)vcpu_book3s, get_order(sizeof(struct kvmppc_vcpu_book3s)));
 934 out:
 935         return ERR_PTR(err);
 936 }
 937
 938 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 939 {
 940         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
 941
 942         __destroy_context(vcpu_book3s->context_id);
 943         kvm_vcpu_uninit(vcpu);
 944         free_pages((long)vcpu_book3s, get_order(sizeof(struct kvmppc_vcpu_book3s)));
 945 }
 946
 947 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
 948 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 949 {
 950         int ret;
 951
 952         /* No need to go into the guest when all we do is going out */
 953         if (signal_pending(current)) {
 954                 kvm_run->exit_reason = KVM_EXIT_INTR;
 955                 return -EINTR;
 956         }
 957
 958         /* XXX we get called with irq disabled - change that! */
 959         local_irq_enable();
 960
 961         ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
 962
 963         local_irq_disable();
 964
 965         return ret;
 966 }
 967
 968 static int kvmppc_book3s_init(void)
 969 {
 970         return kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), THIS_MODULE);
 971 }
 972
 973 static void kvmppc_book3s_exit(void)
 974 {
 975         kvm_exit();
 976 }
 977
 978 module_init(kvmppc_book3s_init);
 979 module_exit(kvmppc_book3s_exit);