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 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
 155                                           unsigned int vec)
 156 {
 157         clear_bit(kvmppc_book3s_vec2irqprio(vec),
 158                   &vcpu->arch.pending_exceptions);
 159 }
 160
 161 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
 162 {
 163         vcpu->stat.queue_intr++;
 164
 165         set_bit(kvmppc_book3s_vec2irqprio(vec),
 166                 &vcpu->arch.pending_exceptions);
 167 #ifdef EXIT_DEBUG
 168         printk(KERN_INFO "Queueing interrupt %x\n", vec);
 169 #endif
 170 }
 171
 172
 173 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu)
 174 {
 175         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
 176 }
 177
 178 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 179 {
 180         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 181 }
 182
 183 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 184 {
 185         return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
 186 }
 187
 188 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 189 {
 190         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 191 }
 192
 193 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 194                                 struct kvm_interrupt *irq)
 195 {
 196         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 197 }
 198
 199 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
 200 {
 201         int deliver = 1;
 202         int vec = 0;
 203
 204         switch (priority) {
 205         case BOOK3S_IRQPRIO_DECREMENTER:
 206                 deliver = vcpu->arch.msr & MSR_EE;
 207                 vec = BOOK3S_INTERRUPT_DECREMENTER;
 208                 break;
 209         case BOOK3S_IRQPRIO_EXTERNAL:
 210                 deliver = vcpu->arch.msr & MSR_EE;
 211                 vec = BOOK3S_INTERRUPT_EXTERNAL;
 212                 break;
 213         case BOOK3S_IRQPRIO_SYSTEM_RESET:
 214                 vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
 215                 break;
 216         case BOOK3S_IRQPRIO_MACHINE_CHECK:
 217                 vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
 218                 break;
 219         case BOOK3S_IRQPRIO_DATA_STORAGE:
 220                 vec = BOOK3S_INTERRUPT_DATA_STORAGE;
 221                 break;
 222         case BOOK3S_IRQPRIO_INST_STORAGE:
 223                 vec = BOOK3S_INTERRUPT_INST_STORAGE;
 224                 break;
 225         case BOOK3S_IRQPRIO_DATA_SEGMENT:
 226                 vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
 227                 break;
 228         case BOOK3S_IRQPRIO_INST_SEGMENT:
 229                 vec = BOOK3S_INTERRUPT_INST_SEGMENT;
 230                 break;
 231         case BOOK3S_IRQPRIO_ALIGNMENT:
 232                 vec = BOOK3S_INTERRUPT_ALIGNMENT;
 233                 break;
 234         case BOOK3S_IRQPRIO_PROGRAM:
 235                 vec = BOOK3S_INTERRUPT_PROGRAM;
 236                 break;
 237         case BOOK3S_IRQPRIO_VSX:
 238                 vec = BOOK3S_INTERRUPT_VSX;
 239                 break;
 240         case BOOK3S_IRQPRIO_ALTIVEC:
 241                 vec = BOOK3S_INTERRUPT_ALTIVEC;
 242                 break;
 243         case BOOK3S_IRQPRIO_FP_UNAVAIL:
 244                 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
 245                 break;
 246         case BOOK3S_IRQPRIO_SYSCALL:
 247                 vec = BOOK3S_INTERRUPT_SYSCALL;
 248                 break;
 249         case BOOK3S_IRQPRIO_DEBUG:
 250                 vec = BOOK3S_INTERRUPT_TRACE;
 251                 break;
 252         case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
 253                 vec = BOOK3S_INTERRUPT_PERFMON;
 254                 break;
 255         default:
 256                 deliver = 0;
 257                 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
 258                 break;
 259         }
 260
 261 #if 0
 262         printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
 263 #endif
 264
 265         if (deliver)
 266                 kvmppc_inject_interrupt(vcpu, vec, 0ULL);
 267
 268         return deliver;
 269 }
 270
 271 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 272 {
 273         unsigned long *pending = &vcpu->arch.pending_exceptions;
 274         unsigned int priority;
 275
 276         /* XXX be more clever here - no need to mftb() on every entry */
 277         /* Issue DEC again if it's still active */
 278 #ifdef AGGRESSIVE_DEC
 279         if (vcpu->arch.msr & MSR_EE)
 280                 if (kvmppc_get_dec(vcpu) & 0x80000000)
 281                         kvmppc_core_queue_dec(vcpu);
 282 #endif
 283
 284 #ifdef EXIT_DEBUG
 285         if (vcpu->arch.pending_exceptions)
 286                 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
 287 #endif
 288         priority = __ffs(*pending);
 289         while (priority <= (sizeof(unsigned int) * 8)) {
 290                 if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
 291                     (priority != BOOK3S_IRQPRIO_DECREMENTER)) {
 292                         /* DEC interrupts get cleared by mtdec */
 293                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 294                         break;
 295                 }
 296
 297                 priority = find_next_bit(pending,
 298                                          BITS_PER_BYTE * sizeof(*pending),
 299                                          priority + 1);
 300         }
 301 }
 302
 303 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 304 {
 305         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 306         vcpu->arch.pvr = pvr;
 307         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 308                 kvmppc_mmu_book3s_64_init(vcpu);
 309                 to_book3s(vcpu)->hior = 0xfff00000;
 310                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 311         } else {
 312                 kvmppc_mmu_book3s_32_init(vcpu);
 313                 to_book3s(vcpu)->hior = 0;
 314                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 315         }
 316
 317         /* If we are in hypervisor level on 970, we can tell the CPU to
 318          * treat DCBZ as 32 bytes store */
 319         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 320         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 321             !strcmp(cur_cpu_spec->platform, "ppc970"))
 322                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 323
 324 }
 325
 326 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 327  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 328  * emulate 32 bytes dcbz length.
 329  *
 330  * The Book3s_64 inventors also realized this case and implemented a special bit
 331  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 332  *
 333  * My approach here is to patch the dcbz instruction on executing pages.
 334  */
 335 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 336 {
 337         bool touched = false;
 338         hva_t hpage;
 339         u32 *page;
 340         int i;
 341
 342         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 343         if (kvm_is_error_hva(hpage))
 344                 return;
 345
 346         hpage |= pte->raddr & ~PAGE_MASK;
 347         hpage &= ~0xFFFULL;
 348
 349         page = vmalloc(HW_PAGE_SIZE);
 350
 351         if (copy_from_user(page, (void __user *)hpage, HW_PAGE_SIZE))
 352                 goto out;
 353
 354         for (i=0; i < HW_PAGE_SIZE / 4; i++)
 355                 if ((page[i] & 0xff0007ff) == INS_DCBZ) {
 356                         page[i] &= 0xfffffff7; // reserved instruction, so we trap
 357                         touched = true;
 358                 }
 359
 360         if (touched)
 361                 copy_to_user((void __user *)hpage, page, HW_PAGE_SIZE);
 362
 363 out:
 364         vfree(page);
 365 }
 366
 367 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
 368                          struct kvmppc_pte *pte)
 369 {
 370         int relocated = (vcpu->arch.msr & (data ? MSR_DR : MSR_IR));
 371         int r;
 372
 373         if (relocated) {
 374                 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
 375         } else {
 376                 pte->eaddr = eaddr;
 377                 pte->raddr = eaddr & 0xffffffff;
 378                 pte->vpage = eaddr >> 12;
 379                 switch (vcpu->arch.msr & (MSR_DR|MSR_IR)) {
 380                 case 0:
 381                         pte->vpage |= VSID_REAL;
 382                 case MSR_DR:
 383                         pte->vpage |= VSID_REAL_DR;
 384                 case MSR_IR:
 385                         pte->vpage |= VSID_REAL_IR;
 386                 }
 387                 pte->may_read = true;
 388                 pte->may_write = true;
 389                 pte->may_execute = true;
 390                 r = 0;
 391         }
 392
 393         return r;
 394 }
 395
 396 static hva_t kvmppc_bad_hva(void)
 397 {
 398         return PAGE_OFFSET;
 399 }
 400
 401 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
 402                                bool read)
 403 {
 404         hva_t hpage;
 405
 406         if (read && !pte->may_read)
 407                 goto err;
 408
 409         if (!read && !pte->may_write)
 410                 goto err;
 411
 412         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 413         if (kvm_is_error_hva(hpage))
 414                 goto err;
 415
 416         return hpage | (pte->raddr & ~PAGE_MASK);
 417 err:
 418         return kvmppc_bad_hva();
 419 }
 420
 421 int kvmppc_st(struct kvm_vcpu *vcpu, ulong eaddr, int size, void *ptr)
 422 {
 423         struct kvmppc_pte pte;
 424         hva_t hva = eaddr;
 425
 426         vcpu->stat.st++;
 427
 428         if (kvmppc_xlate(vcpu, eaddr, false, &pte))
 429                 goto err;
 430
 431         hva = kvmppc_pte_to_hva(vcpu, &pte, false);
 432         if (kvm_is_error_hva(hva))
 433                 goto err;
 434
 435         if (copy_to_user((void __user *)hva, ptr, size)) {
 436                 printk(KERN_INFO "kvmppc_st at 0x%lx failed\n", hva);
 437                 goto err;
 438         }
 439
 440         return 0;
 441
 442 err:
 443         return -ENOENT;
 444 }
 445
 446 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong eaddr, int size, void *ptr,
 447                       bool data)
 448 {
 449         struct kvmppc_pte pte;
 450         hva_t hva = eaddr;
 451
 452         vcpu->stat.ld++;
 453
 454         if (kvmppc_xlate(vcpu, eaddr, data, &pte))
 455                 goto err;
 456
 457         hva = kvmppc_pte_to_hva(vcpu, &pte, true);
 458         if (kvm_is_error_hva(hva))
 459                 goto err;
 460
 461         if (copy_from_user(ptr, (void __user *)hva, size)) {
 462                 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
 463                 goto err;
 464         }
 465
 466         return 0;
 467
 468 err:
 469         return -ENOENT;
 470 }
 471
 472 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 473 {
 474         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 475 }
 476
 477 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 478                             ulong eaddr, int vec)
 479 {
 480         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 481         int r = RESUME_GUEST;
 482         int relocated;
 483         int page_found = 0;
 484         struct kvmppc_pte pte;
 485         bool is_mmio = false;
 486
 487         if ( vec == BOOK3S_INTERRUPT_DATA_STORAGE ) {
 488                 relocated = (vcpu->arch.msr & MSR_DR);
 489         } else {
 490                 relocated = (vcpu->arch.msr & MSR_IR);
 491         }
 492
 493         /* Resolve real address if translation turned on */
 494         if (relocated) {
 495                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 496         } else {
 497                 pte.may_execute = true;
 498                 pte.may_read = true;
 499                 pte.may_write = true;
 500                 pte.raddr = eaddr & 0xffffffff;
 501                 pte.eaddr = eaddr;
 502                 pte.vpage = eaddr >> 12;
 503                 switch (vcpu->arch.msr & (MSR_DR|MSR_IR)) {
 504                 case 0:
 505                         pte.vpage |= VSID_REAL;
 506                 case MSR_DR:
 507                         pte.vpage |= VSID_REAL_DR;
 508                 case MSR_IR:
 509                         pte.vpage |= VSID_REAL_IR;
 510                 }
 511         }
 512
 513         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 514            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 515                 /*
 516                  * If we do the dcbz hack, we have to NX on every execution,
 517                  * so we can patch the executing code. This renders our guest
 518                  * NX-less.
 519                  */
 520                 pte.may_execute = !data;
 521         }
 522
 523         if (page_found == -ENOENT) {
 524                 /* Page not found in guest PTE entries */
 525                 vcpu->arch.dear = vcpu->arch.fault_dear;
 526                 to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr;
 527                 vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x00000000f8000000ULL);
 528                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 529         } else if (page_found == -EPERM) {
 530                 /* Storage protection */
 531                 vcpu->arch.dear = vcpu->arch.fault_dear;
 532                 to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
 533                 to_book3s(vcpu)->dsisr |= DSISR_PROTFAULT;
 534                 vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x00000000f8000000ULL);
 535                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 536         } else if (page_found == -EINVAL) {
 537                 /* Page not found in guest SLB */
 538                 vcpu->arch.dear = vcpu->arch.fault_dear;
 539                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 540         } else if (!is_mmio &&
 541                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 542                 /* The guest's PTE is not mapped yet. Map on the host */
 543                 kvmppc_mmu_map_page(vcpu, &pte);
 544                 if (data)
 545                         vcpu->stat.sp_storage++;
 546                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 547                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 548                         kvmppc_patch_dcbz(vcpu, &pte);
 549         } else {
 550                 /* MMIO */
 551                 vcpu->stat.mmio_exits++;
 552                 vcpu->arch.paddr_accessed = pte.raddr;
 553                 r = kvmppc_emulate_mmio(run, vcpu);
 554                 if ( r == RESUME_HOST_NV )
 555                         r = RESUME_HOST;
 556                 if ( r == RESUME_GUEST_NV )
 557                         r = RESUME_GUEST;
 558         }
 559
 560         return r;
 561 }
 562
 563 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 564                        unsigned int exit_nr)
 565 {
 566         int r = RESUME_HOST;
 567
 568         vcpu->stat.sum_exits++;
 569
 570         run->exit_reason = KVM_EXIT_UNKNOWN;
 571         run->ready_for_interrupt_injection = 1;
 572 #ifdef EXIT_DEBUG
 573         printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | dec=0x%x | msr=0x%lx\n",
 574                 exit_nr, vcpu->arch.pc, vcpu->arch.fault_dear,
 575                 kvmppc_get_dec(vcpu), vcpu->arch.msr);
 576 #elif defined (EXIT_DEBUG_SIMPLE)
 577         if ((exit_nr != 0x900) && (exit_nr != 0x500))
 578                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | msr=0x%lx\n",
 579                         exit_nr, vcpu->arch.pc, vcpu->arch.fault_dear,
 580                         vcpu->arch.msr);
 581 #endif
 582         kvm_resched(vcpu);
 583         switch (exit_nr) {
 584         case BOOK3S_INTERRUPT_INST_STORAGE:
 585                 vcpu->stat.pf_instruc++;
 586                 /* only care about PTEG not found errors, but leave NX alone */
 587                 if (vcpu->arch.shadow_msr & 0x40000000) {
 588                         r = kvmppc_handle_pagefault(run, vcpu, vcpu->arch.pc, exit_nr);
 589                         vcpu->stat.sp_instruc++;
 590                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 591                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 592                         /*
 593                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 594                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 595                          *     that no guest that needs the dcbz hack does NX.
 596                          */
 597                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.pc, ~0xFFFULL);
 598                 } else {
 599                         vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x58000000);
 600                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 601                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.pc, ~0xFFFULL);
 602                         r = RESUME_GUEST;
 603                 }
 604                 break;
 605         case BOOK3S_INTERRUPT_DATA_STORAGE:
 606                 vcpu->stat.pf_storage++;
 607                 /* The only case we need to handle is missing shadow PTEs */
 608                 if (vcpu->arch.fault_dsisr & DSISR_NOHPTE) {
 609                         r = kvmppc_handle_pagefault(run, vcpu, vcpu->arch.fault_dear, exit_nr);
 610                 } else {
 611                         vcpu->arch.dear = vcpu->arch.fault_dear;
 612                         to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr;
 613                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 614                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.dear, ~0xFFFULL);
 615                         r = RESUME_GUEST;
 616                 }
 617                 break;
 618         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 619                 if (kvmppc_mmu_map_segment(vcpu, vcpu->arch.fault_dear) < 0) {
 620                         vcpu->arch.dear = vcpu->arch.fault_dear;
 621                         kvmppc_book3s_queue_irqprio(vcpu,
 622                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 623                 }
 624                 r = RESUME_GUEST;
 625                 break;
 626         case BOOK3S_INTERRUPT_INST_SEGMENT:
 627                 if (kvmppc_mmu_map_segment(vcpu, vcpu->arch.pc) < 0) {
 628                         kvmppc_book3s_queue_irqprio(vcpu,
 629                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 630                 }
 631                 r = RESUME_GUEST;
 632                 break;
 633         /* We're good on these - the host merely wanted to get our attention */
 634         case BOOK3S_INTERRUPT_DECREMENTER:
 635                 vcpu->stat.dec_exits++;
 636                 r = RESUME_GUEST;
 637                 break;
 638         case BOOK3S_INTERRUPT_EXTERNAL:
 639                 vcpu->stat.ext_intr_exits++;
 640                 r = RESUME_GUEST;
 641                 break;
 642         case BOOK3S_INTERRUPT_PROGRAM:
 643         {
 644                 enum emulation_result er;
 645
 646                 if (vcpu->arch.msr & MSR_PR) {
 647 #ifdef EXIT_DEBUG
 648                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", vcpu->arch.pc, vcpu->arch.last_inst);
 649 #endif
 650                         if ((vcpu->arch.last_inst & 0xff0007ff) !=
 651                             (INS_DCBZ & 0xfffffff7)) {
 652                                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 653                                 r = RESUME_GUEST;
 654                                 break;
 655                         }
 656                 }
 657
 658                 vcpu->stat.emulated_inst_exits++;
 659                 er = kvmppc_emulate_instruction(run, vcpu);
 660                 switch (er) {
 661                 case EMULATE_DONE:
 662                         r = RESUME_GUEST;
 663                         break;
 664                 case EMULATE_FAIL:
 665                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 666                                __func__, vcpu->arch.pc, vcpu->arch.last_inst);
 667                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 668                         r = RESUME_GUEST;
 669                         break;
 670                 default:
 671                         BUG();
 672                 }
 673                 break;
 674         }
 675         case BOOK3S_INTERRUPT_SYSCALL:
 676 #ifdef EXIT_DEBUG
 677                 printk(KERN_INFO "Syscall Nr %d\n", (int)vcpu->arch.gpr[0]);
 678 #endif
 679                 vcpu->stat.syscall_exits++;
 680                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 681                 r = RESUME_GUEST;
 682                 break;
 683         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 684         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 685         case BOOK3S_INTERRUPT_TRACE:
 686         case BOOK3S_INTERRUPT_ALTIVEC:
 687         case BOOK3S_INTERRUPT_VSX:
 688                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 689                 r = RESUME_GUEST;
 690                 break;
 691         default:
 692                 /* Ugh - bork here! What did we get? */
 693                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n", exit_nr, vcpu->arch.pc, vcpu->arch.shadow_msr);
 694                 r = RESUME_HOST;
 695                 BUG();
 696                 break;
 697         }
 698
 699
 700         if (!(r & RESUME_HOST)) {
 701                 /* To avoid clobbering exit_reason, only check for signals if
 702                  * we aren't already exiting to userspace for some other
 703                  * reason. */
 704                 if (signal_pending(current)) {
 705 #ifdef EXIT_DEBUG
 706                         printk(KERN_EMERG "KVM: Going back to host\n");
 707 #endif
 708                         vcpu->stat.signal_exits++;
 709                         run->exit_reason = KVM_EXIT_INTR;
 710                         r = -EINTR;
 711                 } else {
 712                         /* In case an interrupt came in that was triggered
 713                          * from userspace (like DEC), we need to check what
 714                          * to inject now! */
 715                         kvmppc_core_deliver_interrupts(vcpu);
 716                 }
 717         }
 718
 719 #ifdef EXIT_DEBUG
 720         printk(KERN_EMERG "KVM exit: vcpu=0x%p pc=0x%lx r=0x%x\n", vcpu, vcpu->arch.pc, r);
 721 #endif
 722
 723         return r;
 724 }
 725
 726 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 727 {
 728         return 0;
 729 }
 730
 731 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 732 {
 733         int i;
 734
 735         regs->pc = vcpu->arch.pc;
 736         regs->cr = vcpu->arch.cr;
 737         regs->ctr = vcpu->arch.ctr;
 738         regs->lr = vcpu->arch.lr;
 739         regs->xer = vcpu->arch.xer;
 740         regs->msr = vcpu->arch.msr;
 741         regs->srr0 = vcpu->arch.srr0;
 742         regs->srr1 = vcpu->arch.srr1;
 743         regs->pid = vcpu->arch.pid;
 744         regs->sprg0 = vcpu->arch.sprg0;
 745         regs->sprg1 = vcpu->arch.sprg1;
 746         regs->sprg2 = vcpu->arch.sprg2;
 747         regs->sprg3 = vcpu->arch.sprg3;
 748         regs->sprg5 = vcpu->arch.sprg4;
 749         regs->sprg6 = vcpu->arch.sprg5;
 750         regs->sprg7 = vcpu->arch.sprg6;
 751
 752         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 753                 regs->gpr[i] = vcpu->arch.gpr[i];
 754
 755         return 0;
 756 }
 757
 758 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 759 {
 760         int i;
 761
 762         vcpu->arch.pc = regs->pc;
 763         vcpu->arch.cr = regs->cr;
 764         vcpu->arch.ctr = regs->ctr;
 765         vcpu->arch.lr = regs->lr;
 766         vcpu->arch.xer = regs->xer;
 767         kvmppc_set_msr(vcpu, regs->msr);
 768         vcpu->arch.srr0 = regs->srr0;
 769         vcpu->arch.srr1 = regs->srr1;
 770         vcpu->arch.sprg0 = regs->sprg0;
 771         vcpu->arch.sprg1 = regs->sprg1;
 772         vcpu->arch.sprg2 = regs->sprg2;
 773         vcpu->arch.sprg3 = regs->sprg3;
 774         vcpu->arch.sprg5 = regs->sprg4;
 775         vcpu->arch.sprg6 = regs->sprg5;
 776         vcpu->arch.sprg7 = regs->sprg6;
 777
 778         for (i = 0; i < ARRAY_SIZE(vcpu->arch.gpr); i++)
 779                 vcpu->arch.gpr[i] = regs->gpr[i];
 780
 781         return 0;
 782 }
 783
 784 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 785                                   struct kvm_sregs *sregs)
 786 {
 787         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 788         int i;
 789
 790         sregs->pvr = vcpu->arch.pvr;
 791
 792         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 793         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 794                 for (i = 0; i < 64; i++) {
 795                         sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
 796                         sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
 797                 }
 798         } else {
 799                 for (i = 0; i < 16; i++) {
 800                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
 801                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
 802                 }
 803                 for (i = 0; i < 8; i++) {
 804                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 805                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 806                 }
 807         }
 808         return 0;
 809 }
 810
 811 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 812                                   struct kvm_sregs *sregs)
 813 {
 814         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 815         int i;
 816
 817         kvmppc_set_pvr(vcpu, sregs->pvr);
 818
 819         vcpu3s->sdr1 = sregs->u.s.sdr1;
 820         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 821                 for (i = 0; i < 64; i++) {
 822                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 823                                                     sregs->u.s.ppc64.slb[i].slbe);
 824                 }
 825         } else {
 826                 for (i = 0; i < 16; i++) {
 827                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 828                 }
 829                 for (i = 0; i < 8; i++) {
 830                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 831                                        (u32)sregs->u.s.ppc32.ibat[i]);
 832                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 833                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 834                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 835                                        (u32)sregs->u.s.ppc32.dbat[i]);
 836                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 837                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 838                 }
 839         }
 840
 841         /* Flush the MMU after messing with the segments */
 842         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 843         return 0;
 844 }
 845
 846 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 847 {
 848         return -ENOTSUPP;
 849 }
 850
 851 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 852 {
 853         return -ENOTSUPP;
 854 }
 855
 856 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 857                                   struct kvm_translation *tr)
 858 {
 859         return 0;
 860 }
 861
 862 /*
 863  * Get (and clear) the dirty memory log for a memory slot.
 864  */
 865 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 866                                       struct kvm_dirty_log *log)
 867 {
 868         struct kvm_memory_slot *memslot;
 869         struct kvm_vcpu *vcpu;
 870         ulong ga, ga_end;
 871         int is_dirty = 0;
 872         int r, n;
 873
 874         down_write(&kvm->slots_lock);
 875
 876         r = kvm_get_dirty_log(kvm, log, &is_dirty);
 877         if (r)
 878                 goto out;
 879
 880         /* If nothing is dirty, don't bother messing with page tables. */
 881         if (is_dirty) {
 882                 memslot = &kvm->memslots[log->slot];
 883
 884                 ga = memslot->base_gfn << PAGE_SHIFT;
 885                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
 886
 887                 kvm_for_each_vcpu(n, vcpu, kvm)
 888                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
 889
 890                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 891                 memset(memslot->dirty_bitmap, 0, n);
 892         }
 893
 894         r = 0;
 895 out:
 896         up_write(&kvm->slots_lock);
 897         return r;
 898 }
 899
 900 int kvmppc_core_check_processor_compat(void)
 901 {
 902         return 0;
 903 }
 904
 905 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 906 {
 907         struct kvmppc_vcpu_book3s *vcpu_book3s;
 908         struct kvm_vcpu *vcpu;
 909         int err;
 910
 911         vcpu_book3s = (struct kvmppc_vcpu_book3s *)__get_free_pages( GFP_KERNEL | __GFP_ZERO,
 912                         get_order(sizeof(struct kvmppc_vcpu_book3s)));
 913         if (!vcpu_book3s) {
 914                 err = -ENOMEM;
 915                 goto out;
 916         }
 917
 918         vcpu = &vcpu_book3s->vcpu;
 919         err = kvm_vcpu_init(vcpu, kvm, id);
 920         if (err)
 921                 goto free_vcpu;
 922
 923         vcpu->arch.host_retip = kvm_return_point;
 924         vcpu->arch.host_msr = mfmsr();
 925         /* default to book3s_64 (970fx) */
 926         vcpu->arch.pvr = 0x3C0301;
 927         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 928         vcpu_book3s->slb_nr = 64;
 929
 930         /* remember where some real-mode handlers are */
 931         vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
 932         vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
 933         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
 934
 935         vcpu->arch.shadow_msr = MSR_USER64;
 936
 937         err = __init_new_context();
 938         if (err < 0)
 939                 goto free_vcpu;
 940         vcpu_book3s->context_id = err;
 941
 942         vcpu_book3s->vsid_max = ((vcpu_book3s->context_id + 1) << USER_ESID_BITS) - 1;
 943         vcpu_book3s->vsid_first = vcpu_book3s->context_id << USER_ESID_BITS;
 944         vcpu_book3s->vsid_next = vcpu_book3s->vsid_first;
 945
 946         return vcpu;
 947
 948 free_vcpu:
 949         free_pages((long)vcpu_book3s, get_order(sizeof(struct kvmppc_vcpu_book3s)));
 950 out:
 951         return ERR_PTR(err);
 952 }
 953
 954 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 955 {
 956         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
 957
 958         __destroy_context(vcpu_book3s->context_id);
 959         kvm_vcpu_uninit(vcpu);
 960         free_pages((long)vcpu_book3s, get_order(sizeof(struct kvmppc_vcpu_book3s)));
 961 }
 962
 963 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
 964 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 965 {
 966         int ret;
 967
 968         /* No need to go into the guest when all we do is going out */
 969         if (signal_pending(current)) {
 970                 kvm_run->exit_reason = KVM_EXIT_INTR;
 971                 return -EINTR;
 972         }
 973
 974         /* XXX we get called with irq disabled - change that! */
 975         local_irq_enable();
 976
 977         ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
 978
 979         local_irq_disable();
 980
 981         return ret;
 982 }
 983
 984 static int kvmppc_book3s_init(void)
 985 {
 986         return kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), THIS_MODULE);
 987 }
 988
 989 static void kvmppc_book3s_exit(void)
 990 {
 991         kvm_exit();
 992 }
 993
 994 module_init(kvmppc_book3s_init);
 995 module_exit(kvmppc_book3s_exit);