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