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         memcpy(&get_paca()->shadow_vcpu, &to_book3s(vcpu)->shadow_vcpu,
  70                sizeof(get_paca()->shadow_vcpu));
  71         get_paca()->kvm_slb_max = to_book3s(vcpu)->slb_shadow_max;
  72 }
  73
  74 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  75 {
  76         memcpy(to_book3s(vcpu)->slb_shadow, get_paca()->kvm_slb, sizeof(get_paca()->kvm_slb));
  77         memcpy(&to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  78                sizeof(get_paca()->shadow_vcpu));
  79         to_book3s(vcpu)->slb_shadow_max = get_paca()->kvm_slb_max;
  80 }
  81
  82 #if defined(EXIT_DEBUG)
  83 static u32 kvmppc_get_dec(struct kvm_vcpu *vcpu)
  84 {
  85         u64 jd = mftb() - vcpu->arch.dec_jiffies;
  86         return vcpu->arch.dec - jd;
  87 }
  88 #endif
  89
  90 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
  91 {
  92         ulong old_msr = vcpu->arch.msr;
  93
  94 #ifdef EXIT_DEBUG
  95         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
  96 #endif
  97         msr &= to_book3s(vcpu)->msr_mask;
  98         vcpu->arch.msr = msr;
  99         vcpu->arch.shadow_msr = msr | MSR_USER32;
 100         vcpu->arch.shadow_msr &= ( MSR_VEC | MSR_VSX | MSR_FP | MSR_FE0 |
 101                                    MSR_USER64 | MSR_SE | MSR_BE | MSR_DE |
 102                                    MSR_FE1);
 103
 104         if (msr & (MSR_WE|MSR_POW)) {
 105                 if (!vcpu->arch.pending_exceptions) {
 106                         kvm_vcpu_block(vcpu);
 107                         vcpu->stat.halt_wakeup++;
 108                 }
 109         }
 110
 111         if (((vcpu->arch.msr & (MSR_IR|MSR_DR)) != (old_msr & (MSR_IR|MSR_DR))) ||
 112             (vcpu->arch.msr & MSR_PR) != (old_msr & MSR_PR)) {
 113                 kvmppc_mmu_flush_segments(vcpu);
 114                 kvmppc_mmu_map_segment(vcpu, vcpu->arch.pc);
 115         }
 116 }
 117
 118 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
 119 {
 120         vcpu->arch.srr0 = vcpu->arch.pc;
 121         vcpu->arch.srr1 = vcpu->arch.msr | flags;
 122         vcpu->arch.pc = to_book3s(vcpu)->hior + vec;
 123         vcpu->arch.mmu.reset_msr(vcpu);
 124 }
 125
 126 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
 127 {
 128         unsigned int prio;
 129
 130         switch (vec) {
 131         case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET;         break;
 132         case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK;        break;
 133         case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE;         break;
 134         case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT;         break;
 135         case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE;         break;
 136         case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT;         break;
 137         case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL;             break;
 138         case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT;            break;
 139         case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM;              break;
 140         case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL;           break;
 141         case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER;          break;
 142         case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL;              break;
 143         case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG;                break;
 144         case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC;              break;
 145         case 0xf40: prio = BOOK3S_IRQPRIO_VSX;                  break;
 146         default:    prio = BOOK3S_IRQPRIO_MAX;                  break;
 147         }
 148
 149         return prio;
 150 }
 151
 152 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
 153                                           unsigned int vec)
 154 {
 155         clear_bit(kvmppc_book3s_vec2irqprio(vec),
 156                   &vcpu->arch.pending_exceptions);
 157 }
 158
 159 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
 160 {
 161         vcpu->stat.queue_intr++;
 162
 163         set_bit(kvmppc_book3s_vec2irqprio(vec),
 164                 &vcpu->arch.pending_exceptions);
 165 #ifdef EXIT_DEBUG
 166         printk(KERN_INFO "Queueing interrupt %x\n", vec);
 167 #endif
 168 }
 169
 170
 171 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong flags)
 172 {
 173         to_book3s(vcpu)->prog_flags = flags;
 174         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
 175 }
 176
 177 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 178 {
 179         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 180 }
 181
 182 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 183 {
 184         return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
 185 }
 186
 187 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 188 {
 189         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 190 }
 191
 192 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 193                                 struct kvm_interrupt *irq)
 194 {
 195         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 196 }
 197
 198 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
 199 {
 200         int deliver = 1;
 201         int vec = 0;
 202         ulong flags = 0ULL;
 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                 flags = to_book3s(vcpu)->prog_flags;
 237                 break;
 238         case BOOK3S_IRQPRIO_VSX:
 239                 vec = BOOK3S_INTERRUPT_VSX;
 240                 break;
 241         case BOOK3S_IRQPRIO_ALTIVEC:
 242                 vec = BOOK3S_INTERRUPT_ALTIVEC;
 243                 break;
 244         case BOOK3S_IRQPRIO_FP_UNAVAIL:
 245                 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
 246                 break;
 247         case BOOK3S_IRQPRIO_SYSCALL:
 248                 vec = BOOK3S_INTERRUPT_SYSCALL;
 249                 break;
 250         case BOOK3S_IRQPRIO_DEBUG:
 251                 vec = BOOK3S_INTERRUPT_TRACE;
 252                 break;
 253         case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
 254                 vec = BOOK3S_INTERRUPT_PERFMON;
 255                 break;
 256         default:
 257                 deliver = 0;
 258                 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
 259                 break;
 260         }
 261
 262 #if 0
 263         printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
 264 #endif
 265
 266         if (deliver)
 267                 kvmppc_inject_interrupt(vcpu, vec, flags);
 268
 269         return deliver;
 270 }
 271
 272 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 273 {
 274         unsigned long *pending = &vcpu->arch.pending_exceptions;
 275         unsigned int priority;
 276
 277 #ifdef EXIT_DEBUG
 278         if (vcpu->arch.pending_exceptions)
 279                 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
 280 #endif
 281         priority = __ffs(*pending);
 282         while (priority <= (sizeof(unsigned int) * 8)) {
 283                 if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
 284                     (priority != BOOK3S_IRQPRIO_DECREMENTER)) {
 285                         /* DEC interrupts get cleared by mtdec */
 286                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 287                         break;
 288                 }
 289
 290                 priority = find_next_bit(pending,
 291                                          BITS_PER_BYTE * sizeof(*pending),
 292                                          priority + 1);
 293         }
 294 }
 295
 296 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 297 {
 298         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 299         vcpu->arch.pvr = pvr;
 300         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 301                 kvmppc_mmu_book3s_64_init(vcpu);
 302                 to_book3s(vcpu)->hior = 0xfff00000;
 303                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 304         } else {
 305                 kvmppc_mmu_book3s_32_init(vcpu);
 306                 to_book3s(vcpu)->hior = 0;
 307                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 308         }
 309
 310         /* If we are in hypervisor level on 970, we can tell the CPU to
 311          * treat DCBZ as 32 bytes store */
 312         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 313         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 314             !strcmp(cur_cpu_spec->platform, "ppc970"))
 315                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 316
 317 }
 318
 319 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 320  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 321  * emulate 32 bytes dcbz length.
 322  *
 323  * The Book3s_64 inventors also realized this case and implemented a special bit
 324  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 325  *
 326  * My approach here is to patch the dcbz instruction on executing pages.
 327  */
 328 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 329 {
 330         bool touched = false;
 331         hva_t hpage;
 332         u32 *page;
 333         int i;
 334
 335         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 336         if (kvm_is_error_hva(hpage))
 337                 return;
 338
 339         hpage |= pte->raddr & ~PAGE_MASK;
 340         hpage &= ~0xFFFULL;
 341
 342         page = vmalloc(HW_PAGE_SIZE);
 343
 344         if (copy_from_user(page, (void __user *)hpage, HW_PAGE_SIZE))
 345                 goto out;
 346
 347         for (i=0; i < HW_PAGE_SIZE / 4; i++)
 348                 if ((page[i] & 0xff0007ff) == INS_DCBZ) {
 349                         page[i] &= 0xfffffff7; // reserved instruction, so we trap
 350                         touched = true;
 351                 }
 352
 353         if (touched)
 354                 copy_to_user((void __user *)hpage, page, HW_PAGE_SIZE);
 355
 356 out:
 357         vfree(page);
 358 }
 359
 360 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
 361                          struct kvmppc_pte *pte)
 362 {
 363         int relocated = (vcpu->arch.msr & (data ? MSR_DR : MSR_IR));
 364         int r;
 365
 366         if (relocated) {
 367                 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
 368         } else {
 369                 pte->eaddr = eaddr;
 370                 pte->raddr = eaddr & 0xffffffff;
 371                 pte->vpage = eaddr >> 12;
 372                 switch (vcpu->arch.msr & (MSR_DR|MSR_IR)) {
 373                 case 0:
 374                         pte->vpage |= VSID_REAL;
 375                 case MSR_DR:
 376                         pte->vpage |= VSID_REAL_DR;
 377                 case MSR_IR:
 378                         pte->vpage |= VSID_REAL_IR;
 379                 }
 380                 pte->may_read = true;
 381                 pte->may_write = true;
 382                 pte->may_execute = true;
 383                 r = 0;
 384         }
 385
 386         return r;
 387 }
 388
 389 static hva_t kvmppc_bad_hva(void)
 390 {
 391         return PAGE_OFFSET;
 392 }
 393
 394 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
 395                                bool read)
 396 {
 397         hva_t hpage;
 398
 399         if (read && !pte->may_read)
 400                 goto err;
 401
 402         if (!read && !pte->may_write)
 403                 goto err;
 404
 405         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 406         if (kvm_is_error_hva(hpage))
 407                 goto err;
 408
 409         return hpage | (pte->raddr & ~PAGE_MASK);
 410 err:
 411         return kvmppc_bad_hva();
 412 }
 413
 414 int kvmppc_st(struct kvm_vcpu *vcpu, ulong eaddr, int size, void *ptr)
 415 {
 416         struct kvmppc_pte pte;
 417         hva_t hva = eaddr;
 418
 419         vcpu->stat.st++;
 420
 421         if (kvmppc_xlate(vcpu, eaddr, false, &pte))
 422                 goto err;
 423
 424         hva = kvmppc_pte_to_hva(vcpu, &pte, false);
 425         if (kvm_is_error_hva(hva))
 426                 goto err;
 427
 428         if (copy_to_user((void __user *)hva, ptr, size)) {
 429                 printk(KERN_INFO "kvmppc_st at 0x%lx failed\n", hva);
 430                 goto err;
 431         }
 432
 433         return 0;
 434
 435 err:
 436         return -ENOENT;
 437 }
 438
 439 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong eaddr, int size, void *ptr,
 440                       bool data)
 441 {
 442         struct kvmppc_pte pte;
 443         hva_t hva = eaddr;
 444
 445         vcpu->stat.ld++;
 446
 447         if (kvmppc_xlate(vcpu, eaddr, data, &pte))
 448                 goto err;
 449
 450         hva = kvmppc_pte_to_hva(vcpu, &pte, true);
 451         if (kvm_is_error_hva(hva))
 452                 goto err;
 453
 454         if (copy_from_user(ptr, (void __user *)hva, size)) {
 455                 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
 456                 goto err;
 457         }
 458
 459         return 0;
 460
 461 err:
 462         return -ENOENT;
 463 }
 464
 465 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 466 {
 467         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 468 }
 469
 470 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 471                             ulong eaddr, int vec)
 472 {
 473         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 474         int r = RESUME_GUEST;
 475         int relocated;
 476         int page_found = 0;
 477         struct kvmppc_pte pte;
 478         bool is_mmio = false;
 479
 480         if ( vec == BOOK3S_INTERRUPT_DATA_STORAGE ) {
 481                 relocated = (vcpu->arch.msr & MSR_DR);
 482         } else {
 483                 relocated = (vcpu->arch.msr & MSR_IR);
 484         }
 485
 486         /* Resolve real address if translation turned on */
 487         if (relocated) {
 488                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 489         } else {
 490                 pte.may_execute = true;
 491                 pte.may_read = true;
 492                 pte.may_write = true;
 493                 pte.raddr = eaddr & 0xffffffff;
 494                 pte.eaddr = eaddr;
 495                 pte.vpage = eaddr >> 12;
 496                 switch (vcpu->arch.msr & (MSR_DR|MSR_IR)) {
 497                 case 0:
 498                         pte.vpage |= VSID_REAL;
 499                 case MSR_DR:
 500                         pte.vpage |= VSID_REAL_DR;
 501                 case MSR_IR:
 502                         pte.vpage |= VSID_REAL_IR;
 503                 }
 504         }
 505
 506         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 507            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 508                 /*
 509                  * If we do the dcbz hack, we have to NX on every execution,
 510                  * so we can patch the executing code. This renders our guest
 511                  * NX-less.
 512                  */
 513                 pte.may_execute = !data;
 514         }
 515
 516         if (page_found == -ENOENT) {
 517                 /* Page not found in guest PTE entries */
 518                 vcpu->arch.dear = vcpu->arch.fault_dear;
 519                 to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr;
 520                 vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x00000000f8000000ULL);
 521                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 522         } else if (page_found == -EPERM) {
 523                 /* Storage protection */
 524                 vcpu->arch.dear = vcpu->arch.fault_dear;
 525                 to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
 526                 to_book3s(vcpu)->dsisr |= DSISR_PROTFAULT;
 527                 vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x00000000f8000000ULL);
 528                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 529         } else if (page_found == -EINVAL) {
 530                 /* Page not found in guest SLB */
 531                 vcpu->arch.dear = vcpu->arch.fault_dear;
 532                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 533         } else if (!is_mmio &&
 534                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 535                 /* The guest's PTE is not mapped yet. Map on the host */
 536                 kvmppc_mmu_map_page(vcpu, &pte);
 537                 if (data)
 538                         vcpu->stat.sp_storage++;
 539                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 540                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 541                         kvmppc_patch_dcbz(vcpu, &pte);
 542         } else {
 543                 /* MMIO */
 544                 vcpu->stat.mmio_exits++;
 545                 vcpu->arch.paddr_accessed = pte.raddr;
 546                 r = kvmppc_emulate_mmio(run, vcpu);
 547                 if ( r == RESUME_HOST_NV )
 548                         r = RESUME_HOST;
 549         }
 550
 551         return r;
 552 }
 553
 554 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 555                        unsigned int exit_nr)
 556 {
 557         int r = RESUME_HOST;
 558
 559         vcpu->stat.sum_exits++;
 560
 561         run->exit_reason = KVM_EXIT_UNKNOWN;
 562         run->ready_for_interrupt_injection = 1;
 563 #ifdef EXIT_DEBUG
 564         printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | dec=0x%x | msr=0x%lx\n",
 565                 exit_nr, vcpu->arch.pc, vcpu->arch.fault_dear,
 566                 kvmppc_get_dec(vcpu), vcpu->arch.msr);
 567 #elif defined (EXIT_DEBUG_SIMPLE)
 568         if ((exit_nr != 0x900) && (exit_nr != 0x500))
 569                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | dar=0x%lx | msr=0x%lx\n",
 570                         exit_nr, vcpu->arch.pc, vcpu->arch.fault_dear,
 571                         vcpu->arch.msr);
 572 #endif
 573         kvm_resched(vcpu);
 574         switch (exit_nr) {
 575         case BOOK3S_INTERRUPT_INST_STORAGE:
 576                 vcpu->stat.pf_instruc++;
 577                 /* only care about PTEG not found errors, but leave NX alone */
 578                 if (vcpu->arch.shadow_msr & 0x40000000) {
 579                         r = kvmppc_handle_pagefault(run, vcpu, vcpu->arch.pc, exit_nr);
 580                         vcpu->stat.sp_instruc++;
 581                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 582                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 583                         /*
 584                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 585                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 586                          *     that no guest that needs the dcbz hack does NX.
 587                          */
 588                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.pc, ~0xFFFULL);
 589                 } else {
 590                         vcpu->arch.msr |= (vcpu->arch.shadow_msr & 0x58000000);
 591                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 592                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.pc, ~0xFFFULL);
 593                         r = RESUME_GUEST;
 594                 }
 595                 break;
 596         case BOOK3S_INTERRUPT_DATA_STORAGE:
 597                 vcpu->stat.pf_storage++;
 598                 /* The only case we need to handle is missing shadow PTEs */
 599                 if (vcpu->arch.fault_dsisr & DSISR_NOHPTE) {
 600                         r = kvmppc_handle_pagefault(run, vcpu, vcpu->arch.fault_dear, exit_nr);
 601                 } else {
 602                         vcpu->arch.dear = vcpu->arch.fault_dear;
 603                         to_book3s(vcpu)->dsisr = vcpu->arch.fault_dsisr;
 604                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 605                         kvmppc_mmu_pte_flush(vcpu, vcpu->arch.dear, ~0xFFFULL);
 606                         r = RESUME_GUEST;
 607                 }
 608                 break;
 609         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 610                 if (kvmppc_mmu_map_segment(vcpu, vcpu->arch.fault_dear) < 0) {
 611                         vcpu->arch.dear = vcpu->arch.fault_dear;
 612                         kvmppc_book3s_queue_irqprio(vcpu,
 613                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 614                 }
 615                 r = RESUME_GUEST;
 616                 break;
 617         case BOOK3S_INTERRUPT_INST_SEGMENT:
 618                 if (kvmppc_mmu_map_segment(vcpu, vcpu->arch.pc) < 0) {
 619                         kvmppc_book3s_queue_irqprio(vcpu,
 620                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 621                 }
 622                 r = RESUME_GUEST;
 623                 break;
 624         /* We're good on these - the host merely wanted to get our attention */
 625         case BOOK3S_INTERRUPT_DECREMENTER:
 626                 vcpu->stat.dec_exits++;
 627                 r = RESUME_GUEST;
 628                 break;
 629         case BOOK3S_INTERRUPT_EXTERNAL:
 630                 vcpu->stat.ext_intr_exits++;
 631                 r = RESUME_GUEST;
 632                 break;
 633         case BOOK3S_INTERRUPT_PROGRAM:
 634         {
 635                 enum emulation_result er;
 636
 637                 if (vcpu->arch.msr & MSR_PR) {
 638 #ifdef EXIT_DEBUG
 639                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", vcpu->arch.pc, vcpu->arch.last_inst);
 640 #endif
 641                         if ((vcpu->arch.last_inst & 0xff0007ff) !=
 642                             (INS_DCBZ & 0xfffffff7)) {
 643                                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 644                                 r = RESUME_GUEST;
 645                                 break;
 646                         }
 647                 }
 648
 649                 vcpu->stat.emulated_inst_exits++;
 650                 er = kvmppc_emulate_instruction(run, vcpu);
 651                 switch (er) {
 652                 case EMULATE_DONE:
 653                         r = RESUME_GUEST_NV;
 654                         break;
 655                 case EMULATE_FAIL:
 656                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 657                                __func__, vcpu->arch.pc, vcpu->arch.last_inst);
 658                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 659                         r = RESUME_GUEST;
 660                         break;
 661                 default:
 662                         BUG();
 663                 }
 664                 break;
 665         }
 666         case BOOK3S_INTERRUPT_SYSCALL:
 667 #ifdef EXIT_DEBUG
 668                 printk(KERN_INFO "Syscall Nr %d\n", (int)kvmppc_get_gpr(vcpu, 0));
 669 #endif
 670                 vcpu->stat.syscall_exits++;
 671                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 672                 r = RESUME_GUEST;
 673                 break;
 674         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 675         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 676         case BOOK3S_INTERRUPT_TRACE:
 677         case BOOK3S_INTERRUPT_ALTIVEC:
 678         case BOOK3S_INTERRUPT_VSX:
 679                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 680                 r = RESUME_GUEST;
 681                 break;
 682         default:
 683                 /* Ugh - bork here! What did we get? */
 684                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n", exit_nr, vcpu->arch.pc, vcpu->arch.shadow_msr);
 685                 r = RESUME_HOST;
 686                 BUG();
 687                 break;
 688         }
 689
 690
 691         if (!(r & RESUME_HOST)) {
 692                 /* To avoid clobbering exit_reason, only check for signals if
 693                  * we aren't already exiting to userspace for some other
 694                  * reason. */
 695                 if (signal_pending(current)) {
 696 #ifdef EXIT_DEBUG
 697                         printk(KERN_EMERG "KVM: Going back to host\n");
 698 #endif
 699                         vcpu->stat.signal_exits++;
 700                         run->exit_reason = KVM_EXIT_INTR;
 701                         r = -EINTR;
 702                 } else {
 703                         /* In case an interrupt came in that was triggered
 704                          * from userspace (like DEC), we need to check what
 705                          * to inject now! */
 706                         kvmppc_core_deliver_interrupts(vcpu);
 707                 }
 708         }
 709
 710 #ifdef EXIT_DEBUG
 711         printk(KERN_EMERG "KVM exit: vcpu=0x%p pc=0x%lx r=0x%x\n", vcpu, vcpu->arch.pc, r);
 712 #endif
 713
 714         return r;
 715 }
 716
 717 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 718 {
 719         return 0;
 720 }
 721
 722 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 723 {
 724         int i;
 725
 726         regs->pc = vcpu->arch.pc;
 727         regs->cr = kvmppc_get_cr(vcpu);
 728         regs->ctr = vcpu->arch.ctr;
 729         regs->lr = vcpu->arch.lr;
 730         regs->xer = kvmppc_get_xer(vcpu);
 731         regs->msr = vcpu->arch.msr;
 732         regs->srr0 = vcpu->arch.srr0;
 733         regs->srr1 = vcpu->arch.srr1;
 734         regs->pid = vcpu->arch.pid;
 735         regs->sprg0 = vcpu->arch.sprg0;
 736         regs->sprg1 = vcpu->arch.sprg1;
 737         regs->sprg2 = vcpu->arch.sprg2;
 738         regs->sprg3 = vcpu->arch.sprg3;
 739         regs->sprg5 = vcpu->arch.sprg4;
 740         regs->sprg6 = vcpu->arch.sprg5;
 741         regs->sprg7 = vcpu->arch.sprg6;
 742
 743         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 744                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
 745
 746         return 0;
 747 }
 748
 749 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 750 {
 751         int i;
 752
 753         vcpu->arch.pc = regs->pc;
 754         kvmppc_set_cr(vcpu, regs->cr);
 755         vcpu->arch.ctr = regs->ctr;
 756         vcpu->arch.lr = regs->lr;
 757         kvmppc_set_xer(vcpu, regs->xer);
 758         kvmppc_set_msr(vcpu, regs->msr);
 759         vcpu->arch.srr0 = regs->srr0;
 760         vcpu->arch.srr1 = regs->srr1;
 761         vcpu->arch.sprg0 = regs->sprg0;
 762         vcpu->arch.sprg1 = regs->sprg1;
 763         vcpu->arch.sprg2 = regs->sprg2;
 764         vcpu->arch.sprg3 = regs->sprg3;
 765         vcpu->arch.sprg5 = regs->sprg4;
 766         vcpu->arch.sprg6 = regs->sprg5;
 767         vcpu->arch.sprg7 = regs->sprg6;
 768
 769         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
 770                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
 771
 772         return 0;
 773 }
 774
 775 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 776                                   struct kvm_sregs *sregs)
 777 {
 778         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 779         int i;
 780
 781         sregs->pvr = vcpu->arch.pvr;
 782
 783         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 784         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 785                 for (i = 0; i < 64; i++) {
 786                         sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
 787                         sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
 788                 }
 789         } else {
 790                 for (i = 0; i < 16; i++) {
 791                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
 792                         sregs->u.s.ppc32.sr[i] = vcpu3s->sr[i].raw;
 793                 }
 794                 for (i = 0; i < 8; i++) {
 795                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 796                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 797                 }
 798         }
 799         return 0;
 800 }
 801
 802 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 803                                   struct kvm_sregs *sregs)
 804 {
 805         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 806         int i;
 807
 808         kvmppc_set_pvr(vcpu, sregs->pvr);
 809
 810         vcpu3s->sdr1 = sregs->u.s.sdr1;
 811         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 812                 for (i = 0; i < 64; i++) {
 813                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 814                                                     sregs->u.s.ppc64.slb[i].slbe);
 815                 }
 816         } else {
 817                 for (i = 0; i < 16; i++) {
 818                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 819                 }
 820                 for (i = 0; i < 8; i++) {
 821                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 822                                        (u32)sregs->u.s.ppc32.ibat[i]);
 823                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 824                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 825                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 826                                        (u32)sregs->u.s.ppc32.dbat[i]);
 827                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 828                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 829                 }
 830         }
 831
 832         /* Flush the MMU after messing with the segments */
 833         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 834         return 0;
 835 }
 836
 837 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 838 {
 839         return -ENOTSUPP;
 840 }
 841
 842 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 843 {
 844         return -ENOTSUPP;
 845 }
 846
 847 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 848                                   struct kvm_translation *tr)
 849 {
 850         return 0;
 851 }
 852
 853 /*
 854  * Get (and clear) the dirty memory log for a memory slot.
 855  */
 856 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 857                                       struct kvm_dirty_log *log)
 858 {
 859         struct kvm_memory_slot *memslot;
 860         struct kvm_vcpu *vcpu;
 861         ulong ga, ga_end;
 862         int is_dirty = 0;
 863         int r, n;
 864
 865         mutex_lock(&kvm->slots_lock);
 866
 867         r = kvm_get_dirty_log(kvm, log, &is_dirty);
 868         if (r)
 869                 goto out;
 870
 871         /* If nothing is dirty, don't bother messing with page tables. */
 872         if (is_dirty) {
 873                 memslot = &kvm->memslots->memslots[log->slot];
 874
 875                 ga = memslot->base_gfn << PAGE_SHIFT;
 876                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
 877
 878                 kvm_for_each_vcpu(n, vcpu, kvm)
 879                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
 880
 881                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 882                 memset(memslot->dirty_bitmap, 0, n);
 883         }
 884
 885         r = 0;
 886 out:
 887         mutex_unlock(&kvm->slots_lock);
 888         return r;
 889 }
 890
 891 int kvmppc_core_check_processor_compat(void)
 892 {
 893         return 0;
 894 }
 895
 896 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 897 {
 898         struct kvmppc_vcpu_book3s *vcpu_book3s;
 899         struct kvm_vcpu *vcpu;
 900         int err;
 901
 902         vcpu_book3s = (struct kvmppc_vcpu_book3s *)__get_free_pages( GFP_KERNEL | __GFP_ZERO,
 903                         get_order(sizeof(struct kvmppc_vcpu_book3s)));
 904         if (!vcpu_book3s) {
 905                 err = -ENOMEM;
 906                 goto out;
 907         }
 908
 909         vcpu = &vcpu_book3s->vcpu;
 910         err = kvm_vcpu_init(vcpu, kvm, id);
 911         if (err)
 912                 goto free_vcpu;
 913
 914         vcpu->arch.host_retip = kvm_return_point;
 915         vcpu->arch.host_msr = mfmsr();
 916         /* default to book3s_64 (970fx) */
 917         vcpu->arch.pvr = 0x3C0301;
 918         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 919         vcpu_book3s->slb_nr = 64;
 920
 921         /* remember where some real-mode handlers are */
 922         vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
 923         vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
 924         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
 925         vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
 926
 927         vcpu->arch.shadow_msr = MSR_USER64;
 928
 929         err = __init_new_context();
 930         if (err < 0)
 931                 goto free_vcpu;
 932         vcpu_book3s->context_id = err;
 933
 934         vcpu_book3s->vsid_max = ((vcpu_book3s->context_id + 1) << USER_ESID_BITS) - 1;
 935         vcpu_book3s->vsid_first = vcpu_book3s->context_id << USER_ESID_BITS;
 936         vcpu_book3s->vsid_next = vcpu_book3s->vsid_first;
 937
 938         return vcpu;
 939
 940 free_vcpu:
 941         free_pages((long)vcpu_book3s, get_order(sizeof(struct kvmppc_vcpu_book3s)));
 942 out:
 943         return ERR_PTR(err);
 944 }
 945
 946 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 947 {
 948         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
 949
 950         __destroy_context(vcpu_book3s->context_id);
 951         kvm_vcpu_uninit(vcpu);
 952         free_pages((long)vcpu_book3s, get_order(sizeof(struct kvmppc_vcpu_book3s)));
 953 }
 954
 955 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
 956 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 957 {
 958         int ret;
 959
 960         /* No need to go into the guest when all we do is going out */
 961         if (signal_pending(current)) {
 962                 kvm_run->exit_reason = KVM_EXIT_INTR;
 963                 return -EINTR;
 964         }
 965
 966         /* XXX we get called with irq disabled - change that! */
 967         local_irq_enable();
 968
 969         ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
 970
 971         local_irq_disable();
 972
 973         return ret;
 974 }
 975
 976 static int kvmppc_book3s_init(void)
 977 {
 978         return kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), THIS_MODULE);
 979 }
 980
 981 static void kvmppc_book3s_exit(void)
 982 {
 983         kvm_exit();
 984 }
 985
 986 module_init(kvmppc_book3s_init);
 987 module_exit(kvmppc_book3s_exit);