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 #include <linux/slab.h>
  20 #include "trace.h"
  21
  22 #include <asm/reg.h>
  23 #include <asm/cputable.h>
  24 #include <asm/cacheflush.h>
  25 #include <asm/tlbflush.h>
  26 #include <asm/uaccess.h>
  27 #include <asm/io.h>
  28 #include <asm/kvm_ppc.h>
  29 #include <asm/kvm_book3s.h>
  30 #include <asm/mmu_context.h>
  31 #include <linux/gfp.h>
  32 #include <linux/sched.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/highmem.h>
  35
  36 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  37
  38 /* #define EXIT_DEBUG */
  39 /* #define DEBUG_EXT */
  40
  41 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  42                              ulong msr);
  43
  44 /* Some compatibility defines */
  45 #ifdef CONFIG_PPC_BOOK3S_32
  46 #define MSR_USER32 MSR_USER
  47 #define MSR_USER64 MSR_USER
  48 #define HW_PAGE_SIZE PAGE_SIZE
  49 #endif
  50
  51 struct kvm_stats_debugfs_item debugfs_entries[] = {
  52         { "exits",       VCPU_STAT(sum_exits) },
  53         { "mmio",        VCPU_STAT(mmio_exits) },
  54         { "sig",         VCPU_STAT(signal_exits) },
  55         { "sysc",        VCPU_STAT(syscall_exits) },
  56         { "inst_emu",    VCPU_STAT(emulated_inst_exits) },
  57         { "dec",         VCPU_STAT(dec_exits) },
  58         { "ext_intr",    VCPU_STAT(ext_intr_exits) },
  59         { "queue_intr",  VCPU_STAT(queue_intr) },
  60         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  61         { "pf_storage",  VCPU_STAT(pf_storage) },
  62         { "sp_storage",  VCPU_STAT(sp_storage) },
  63         { "pf_instruc",  VCPU_STAT(pf_instruc) },
  64         { "sp_instruc",  VCPU_STAT(sp_instruc) },
  65         { "ld",          VCPU_STAT(ld) },
  66         { "ld_slow",     VCPU_STAT(ld_slow) },
  67         { "st",          VCPU_STAT(st) },
  68         { "st_slow",     VCPU_STAT(st_slow) },
  69         { NULL }
  70 };
  71
  72 void kvmppc_core_load_host_debugstate(struct kvm_vcpu *vcpu)
  73 {
  74 }
  75
  76 void kvmppc_core_load_guest_debugstate(struct kvm_vcpu *vcpu)
  77 {
  78 }
  79
  80 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  81 {
  82 #ifdef CONFIG_PPC_BOOK3S_64
  83         memcpy(to_svcpu(vcpu)->slb, to_book3s(vcpu)->slb_shadow, sizeof(to_svcpu(vcpu)->slb));
  84         memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
  85                sizeof(get_paca()->shadow_vcpu));
  86         to_svcpu(vcpu)->slb_max = to_book3s(vcpu)->slb_shadow_max;
  87 #endif
  88
  89 #ifdef CONFIG_PPC_BOOK3S_32
  90         current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
  91 #endif
  92 }
  93
  94 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  95 {
  96 #ifdef CONFIG_PPC_BOOK3S_64
  97         memcpy(to_book3s(vcpu)->slb_shadow, to_svcpu(vcpu)->slb, sizeof(to_svcpu(vcpu)->slb));
  98         memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  99                sizeof(get_paca()->shadow_vcpu));
 100         to_book3s(vcpu)->slb_shadow_max = to_svcpu(vcpu)->slb_max;
 101 #endif
 102
 103         kvmppc_giveup_ext(vcpu, MSR_FP);
 104         kvmppc_giveup_ext(vcpu, MSR_VEC);
 105         kvmppc_giveup_ext(vcpu, MSR_VSX);
 106 }
 107
 108 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 109 {
 110         ulong smsr = vcpu->arch.shared->msr;
 111
 112         /* Guest MSR values */
 113         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE;
 114         /* Process MSR values */
 115         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 116         /* External providers the guest reserved */
 117         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
 118         /* 64-bit Process MSR values */
 119 #ifdef CONFIG_PPC_BOOK3S_64
 120         smsr |= MSR_ISF | MSR_HV;
 121 #endif
 122         vcpu->arch.shadow_msr = smsr;
 123 }
 124
 125 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 126 {
 127         ulong old_msr = vcpu->arch.shared->msr;
 128
 129 #ifdef EXIT_DEBUG
 130         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 131 #endif
 132
 133         msr &= to_book3s(vcpu)->msr_mask;
 134         vcpu->arch.shared->msr = msr;
 135         kvmppc_recalc_shadow_msr(vcpu);
 136
 137         if (msr & MSR_POW) {
 138                 if (!vcpu->arch.pending_exceptions) {
 139                         kvm_vcpu_block(vcpu);
 140                         vcpu->stat.halt_wakeup++;
 141
 142                         /* Unset POW bit after we woke up */
 143                         msr &= ~MSR_POW;
 144                         vcpu->arch.shared->msr = msr;
 145                 }
 146         }
 147
 148         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 149                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 150                 kvmppc_mmu_flush_segments(vcpu);
 151                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 152
 153                 /* Preload magic page segment when in kernel mode */
 154                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 155                         struct kvm_vcpu_arch *a = &vcpu->arch;
 156
 157                         if (msr & MSR_DR)
 158                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 159                         else
 160                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 161                 }
 162         }
 163
 164         /* Preload FPU if it's enabled */
 165         if (vcpu->arch.shared->msr & MSR_FP)
 166                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 167 }
 168
 169 void kvmppc_inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 flags)
 170 {
 171         vcpu->arch.shared->srr0 = kvmppc_get_pc(vcpu);
 172         vcpu->arch.shared->srr1 = vcpu->arch.shared->msr | flags;
 173         kvmppc_set_pc(vcpu, to_book3s(vcpu)->hior + vec);
 174         vcpu->arch.mmu.reset_msr(vcpu);
 175 }
 176
 177 static int kvmppc_book3s_vec2irqprio(unsigned int vec)
 178 {
 179         unsigned int prio;
 180
 181         switch (vec) {
 182         case 0x100: prio = BOOK3S_IRQPRIO_SYSTEM_RESET;         break;
 183         case 0x200: prio = BOOK3S_IRQPRIO_MACHINE_CHECK;        break;
 184         case 0x300: prio = BOOK3S_IRQPRIO_DATA_STORAGE;         break;
 185         case 0x380: prio = BOOK3S_IRQPRIO_DATA_SEGMENT;         break;
 186         case 0x400: prio = BOOK3S_IRQPRIO_INST_STORAGE;         break;
 187         case 0x480: prio = BOOK3S_IRQPRIO_INST_SEGMENT;         break;
 188         case 0x500: prio = BOOK3S_IRQPRIO_EXTERNAL;             break;
 189         case 0x600: prio = BOOK3S_IRQPRIO_ALIGNMENT;            break;
 190         case 0x700: prio = BOOK3S_IRQPRIO_PROGRAM;              break;
 191         case 0x800: prio = BOOK3S_IRQPRIO_FP_UNAVAIL;           break;
 192         case 0x900: prio = BOOK3S_IRQPRIO_DECREMENTER;          break;
 193         case 0xc00: prio = BOOK3S_IRQPRIO_SYSCALL;              break;
 194         case 0xd00: prio = BOOK3S_IRQPRIO_DEBUG;                break;
 195         case 0xf20: prio = BOOK3S_IRQPRIO_ALTIVEC;              break;
 196         case 0xf40: prio = BOOK3S_IRQPRIO_VSX;                  break;
 197         default:    prio = BOOK3S_IRQPRIO_MAX;                  break;
 198         }
 199
 200         return prio;
 201 }
 202
 203 static void kvmppc_book3s_dequeue_irqprio(struct kvm_vcpu *vcpu,
 204                                           unsigned int vec)
 205 {
 206         clear_bit(kvmppc_book3s_vec2irqprio(vec),
 207                   &vcpu->arch.pending_exceptions);
 208
 209         if (!vcpu->arch.pending_exceptions)
 210                 vcpu->arch.shared->int_pending = 0;
 211 }
 212
 213 void kvmppc_book3s_queue_irqprio(struct kvm_vcpu *vcpu, unsigned int vec)
 214 {
 215         vcpu->stat.queue_intr++;
 216
 217         set_bit(kvmppc_book3s_vec2irqprio(vec),
 218                 &vcpu->arch.pending_exceptions);
 219 #ifdef EXIT_DEBUG
 220         printk(KERN_INFO "Queueing interrupt %x\n", vec);
 221 #endif
 222 }
 223
 224
 225 void kvmppc_core_queue_program(struct kvm_vcpu *vcpu, ulong flags)
 226 {
 227         to_book3s(vcpu)->prog_flags = flags;
 228         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_PROGRAM);
 229 }
 230
 231 void kvmppc_core_queue_dec(struct kvm_vcpu *vcpu)
 232 {
 233         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 234 }
 235
 236 int kvmppc_core_pending_dec(struct kvm_vcpu *vcpu)
 237 {
 238         return test_bit(BOOK3S_INTERRUPT_DECREMENTER >> 7, &vcpu->arch.pending_exceptions);
 239 }
 240
 241 void kvmppc_core_dequeue_dec(struct kvm_vcpu *vcpu)
 242 {
 243         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_DECREMENTER);
 244 }
 245
 246 void kvmppc_core_queue_external(struct kvm_vcpu *vcpu,
 247                                 struct kvm_interrupt *irq)
 248 {
 249         kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 250 }
 251
 252 void kvmppc_core_dequeue_external(struct kvm_vcpu *vcpu,
 253                                   struct kvm_interrupt *irq)
 254 {
 255         kvmppc_book3s_dequeue_irqprio(vcpu, BOOK3S_INTERRUPT_EXTERNAL);
 256 }
 257
 258 int kvmppc_book3s_irqprio_deliver(struct kvm_vcpu *vcpu, unsigned int priority)
 259 {
 260         int deliver = 1;
 261         int vec = 0;
 262         ulong flags = 0ULL;
 263         ulong crit_raw = vcpu->arch.shared->critical;
 264         ulong crit_r1 = kvmppc_get_gpr(vcpu, 1);
 265         bool crit;
 266
 267         /* Truncate crit indicators in 32 bit mode */
 268         if (!(vcpu->arch.shared->msr & MSR_SF)) {
 269                 crit_raw &= 0xffffffff;
 270                 crit_r1 &= 0xffffffff;
 271         }
 272
 273         /* Critical section when crit == r1 */
 274         crit = (crit_raw == crit_r1);
 275         /* ... and we're in supervisor mode */
 276         crit = crit && !(vcpu->arch.shared->msr & MSR_PR);
 277
 278         switch (priority) {
 279         case BOOK3S_IRQPRIO_DECREMENTER:
 280                 deliver = (vcpu->arch.shared->msr & MSR_EE) && !crit;
 281                 vec = BOOK3S_INTERRUPT_DECREMENTER;
 282                 break;
 283         case BOOK3S_IRQPRIO_EXTERNAL:
 284                 deliver = (vcpu->arch.shared->msr & MSR_EE) && !crit;
 285                 vec = BOOK3S_INTERRUPT_EXTERNAL;
 286                 break;
 287         case BOOK3S_IRQPRIO_SYSTEM_RESET:
 288                 vec = BOOK3S_INTERRUPT_SYSTEM_RESET;
 289                 break;
 290         case BOOK3S_IRQPRIO_MACHINE_CHECK:
 291                 vec = BOOK3S_INTERRUPT_MACHINE_CHECK;
 292                 break;
 293         case BOOK3S_IRQPRIO_DATA_STORAGE:
 294                 vec = BOOK3S_INTERRUPT_DATA_STORAGE;
 295                 break;
 296         case BOOK3S_IRQPRIO_INST_STORAGE:
 297                 vec = BOOK3S_INTERRUPT_INST_STORAGE;
 298                 break;
 299         case BOOK3S_IRQPRIO_DATA_SEGMENT:
 300                 vec = BOOK3S_INTERRUPT_DATA_SEGMENT;
 301                 break;
 302         case BOOK3S_IRQPRIO_INST_SEGMENT:
 303                 vec = BOOK3S_INTERRUPT_INST_SEGMENT;
 304                 break;
 305         case BOOK3S_IRQPRIO_ALIGNMENT:
 306                 vec = BOOK3S_INTERRUPT_ALIGNMENT;
 307                 break;
 308         case BOOK3S_IRQPRIO_PROGRAM:
 309                 vec = BOOK3S_INTERRUPT_PROGRAM;
 310                 flags = to_book3s(vcpu)->prog_flags;
 311                 break;
 312         case BOOK3S_IRQPRIO_VSX:
 313                 vec = BOOK3S_INTERRUPT_VSX;
 314                 break;
 315         case BOOK3S_IRQPRIO_ALTIVEC:
 316                 vec = BOOK3S_INTERRUPT_ALTIVEC;
 317                 break;
 318         case BOOK3S_IRQPRIO_FP_UNAVAIL:
 319                 vec = BOOK3S_INTERRUPT_FP_UNAVAIL;
 320                 break;
 321         case BOOK3S_IRQPRIO_SYSCALL:
 322                 vec = BOOK3S_INTERRUPT_SYSCALL;
 323                 break;
 324         case BOOK3S_IRQPRIO_DEBUG:
 325                 vec = BOOK3S_INTERRUPT_TRACE;
 326                 break;
 327         case BOOK3S_IRQPRIO_PERFORMANCE_MONITOR:
 328                 vec = BOOK3S_INTERRUPT_PERFMON;
 329                 break;
 330         default:
 331                 deliver = 0;
 332                 printk(KERN_ERR "KVM: Unknown interrupt: 0x%x\n", priority);
 333                 break;
 334         }
 335
 336 #if 0
 337         printk(KERN_INFO "Deliver interrupt 0x%x? %x\n", vec, deliver);
 338 #endif
 339
 340         if (deliver)
 341                 kvmppc_inject_interrupt(vcpu, vec, flags);
 342
 343         return deliver;
 344 }
 345
 346 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 347 {
 348         unsigned long *pending = &vcpu->arch.pending_exceptions;
 349         unsigned long old_pending = vcpu->arch.pending_exceptions;
 350         unsigned int priority;
 351
 352 #ifdef EXIT_DEBUG
 353         if (vcpu->arch.pending_exceptions)
 354                 printk(KERN_EMERG "KVM: Check pending: %lx\n", vcpu->arch.pending_exceptions);
 355 #endif
 356         priority = __ffs(*pending);
 357         while (priority < BOOK3S_IRQPRIO_MAX) {
 358                 if (kvmppc_book3s_irqprio_deliver(vcpu, priority) &&
 359                     (priority != BOOK3S_IRQPRIO_DECREMENTER)) {
 360                         /* DEC interrupts get cleared by mtdec */
 361                         clear_bit(priority, &vcpu->arch.pending_exceptions);
 362                         break;
 363                 }
 364
 365                 priority = find_next_bit(pending,
 366                                          BITS_PER_BYTE * sizeof(*pending),
 367                                          priority + 1);
 368         }
 369
 370         /* Tell the guest about our interrupt status */
 371         if (*pending)
 372                 vcpu->arch.shared->int_pending = 1;
 373         else if (old_pending)
 374                 vcpu->arch.shared->int_pending = 0;
 375 }
 376
 377 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 378 {
 379         u32 host_pvr;
 380
 381         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 382         vcpu->arch.pvr = pvr;
 383 #ifdef CONFIG_PPC_BOOK3S_64
 384         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 385                 kvmppc_mmu_book3s_64_init(vcpu);
 386                 to_book3s(vcpu)->hior = 0xfff00000;
 387                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 388         } else
 389 #endif
 390         {
 391                 kvmppc_mmu_book3s_32_init(vcpu);
 392                 to_book3s(vcpu)->hior = 0;
 393                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 394         }
 395
 396         /* If we are in hypervisor level on 970, we can tell the CPU to
 397          * treat DCBZ as 32 bytes store */
 398         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 399         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 400             !strcmp(cur_cpu_spec->platform, "ppc970"))
 401                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 402
 403         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 404            really needs them in a VM on Cell and force disable them. */
 405         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 406                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 407
 408 #ifdef CONFIG_PPC_BOOK3S_32
 409         /* 32 bit Book3S always has 32 byte dcbz */
 410         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 411 #endif
 412
 413         /* On some CPUs we can execute paired single operations natively */
 414         asm ( "mfpvr %0" : "=r"(host_pvr));
 415         switch (host_pvr) {
 416         case 0x00080200:        /* lonestar 2.0 */
 417         case 0x00088202:        /* lonestar 2.2 */
 418         case 0x70000100:        /* gekko 1.0 */
 419         case 0x00080100:        /* gekko 2.0 */
 420         case 0x00083203:        /* gekko 2.3a */
 421         case 0x00083213:        /* gekko 2.3b */
 422         case 0x00083204:        /* gekko 2.4 */
 423         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 424         case 0x00087200:        /* broadway */
 425                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 426                 /* Enable HID2.PSE - in case we need it later */
 427                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 428         }
 429 }
 430
 431 pfn_t kvmppc_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 432 {
 433         ulong mp_pa = vcpu->arch.magic_page_pa;
 434
 435         /* Magic page override */
 436         if (unlikely(mp_pa) &&
 437             unlikely(((gfn << PAGE_SHIFT) & KVM_PAM) ==
 438                      ((mp_pa & PAGE_MASK) & KVM_PAM))) {
 439                 ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
 440                 pfn_t pfn;
 441
 442                 pfn = (pfn_t)virt_to_phys((void*)shared_page) >> PAGE_SHIFT;
 443                 get_page(pfn_to_page(pfn));
 444                 return pfn;
 445         }
 446
 447         return gfn_to_pfn(vcpu->kvm, gfn);
 448 }
 449
 450 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 451  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 452  * emulate 32 bytes dcbz length.
 453  *
 454  * The Book3s_64 inventors also realized this case and implemented a special bit
 455  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 456  *
 457  * My approach here is to patch the dcbz instruction on executing pages.
 458  */
 459 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 460 {
 461         struct page *hpage;
 462         u64 hpage_offset;
 463         u32 *page;
 464         int i;
 465
 466         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 467         if (is_error_page(hpage)) {
 468                 kvm_release_page_clean(hpage);
 469                 return;
 470         }
 471
 472         hpage_offset = pte->raddr & ~PAGE_MASK;
 473         hpage_offset &= ~0xFFFULL;
 474         hpage_offset /= 4;
 475
 476         get_page(hpage);
 477         page = kmap_atomic(hpage, KM_USER0);
 478
 479         /* patch dcbz into reserved instruction, so we trap */
 480         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 481                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 482                         page[i] &= 0xfffffff7;
 483
 484         kunmap_atomic(page, KM_USER0);
 485         put_page(hpage);
 486 }
 487
 488 static int kvmppc_xlate(struct kvm_vcpu *vcpu, ulong eaddr, bool data,
 489                          struct kvmppc_pte *pte)
 490 {
 491         int relocated = (vcpu->arch.shared->msr & (data ? MSR_DR : MSR_IR));
 492         int r;
 493
 494         if (relocated) {
 495                 r = vcpu->arch.mmu.xlate(vcpu, eaddr, pte, data);
 496         } else {
 497                 pte->eaddr = eaddr;
 498                 pte->raddr = eaddr & KVM_PAM;
 499                 pte->vpage = VSID_REAL | eaddr >> 12;
 500                 pte->may_read = true;
 501                 pte->may_write = true;
 502                 pte->may_execute = true;
 503                 r = 0;
 504         }
 505
 506         return r;
 507 }
 508
 509 static hva_t kvmppc_bad_hva(void)
 510 {
 511         return PAGE_OFFSET;
 512 }
 513
 514 static hva_t kvmppc_pte_to_hva(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte,
 515                                bool read)
 516 {
 517         hva_t hpage;
 518
 519         if (read && !pte->may_read)
 520                 goto err;
 521
 522         if (!read && !pte->may_write)
 523                 goto err;
 524
 525         hpage = gfn_to_hva(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 526         if (kvm_is_error_hva(hpage))
 527                 goto err;
 528
 529         return hpage | (pte->raddr & ~PAGE_MASK);
 530 err:
 531         return kvmppc_bad_hva();
 532 }
 533
 534 int kvmppc_st(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 535               bool data)
 536 {
 537         struct kvmppc_pte pte;
 538
 539         vcpu->stat.st++;
 540
 541         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 542                 return -ENOENT;
 543
 544         *eaddr = pte.raddr;
 545
 546         if (!pte.may_write)
 547                 return -EPERM;
 548
 549         if (kvm_write_guest(vcpu->kvm, pte.raddr, ptr, size))
 550                 return EMULATE_DO_MMIO;
 551
 552         return EMULATE_DONE;
 553 }
 554
 555 int kvmppc_ld(struct kvm_vcpu *vcpu, ulong *eaddr, int size, void *ptr,
 556                       bool data)
 557 {
 558         struct kvmppc_pte pte;
 559         hva_t hva = *eaddr;
 560
 561         vcpu->stat.ld++;
 562
 563         if (kvmppc_xlate(vcpu, *eaddr, data, &pte))
 564                 goto nopte;
 565
 566         *eaddr = pte.raddr;
 567
 568         hva = kvmppc_pte_to_hva(vcpu, &pte, true);
 569         if (kvm_is_error_hva(hva))
 570                 goto mmio;
 571
 572         if (copy_from_user(ptr, (void __user *)hva, size)) {
 573                 printk(KERN_INFO "kvmppc_ld at 0x%lx failed\n", hva);
 574                 goto mmio;
 575         }
 576
 577         return EMULATE_DONE;
 578
 579 nopte:
 580         return -ENOENT;
 581 mmio:
 582         return EMULATE_DO_MMIO;
 583 }
 584
 585 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 586 {
 587         ulong mp_pa = vcpu->arch.magic_page_pa;
 588
 589         if (unlikely(mp_pa) &&
 590             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 591                 return 1;
 592         }
 593
 594         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 595 }
 596
 597 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 598                             ulong eaddr, int vec)
 599 {
 600         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 601         int r = RESUME_GUEST;
 602         int relocated;
 603         int page_found = 0;
 604         struct kvmppc_pte pte;
 605         bool is_mmio = false;
 606         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 607         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 608         u64 vsid;
 609
 610         relocated = data ? dr : ir;
 611
 612         /* Resolve real address if translation turned on */
 613         if (relocated) {
 614                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 615         } else {
 616                 pte.may_execute = true;
 617                 pte.may_read = true;
 618                 pte.may_write = true;
 619                 pte.raddr = eaddr & KVM_PAM;
 620                 pte.eaddr = eaddr;
 621                 pte.vpage = eaddr >> 12;
 622         }
 623
 624         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 625         case 0:
 626                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 627                 break;
 628         case MSR_DR:
 629         case MSR_IR:
 630                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 631
 632                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 633                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 634                 else
 635                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 636                 pte.vpage |= vsid;
 637
 638                 if (vsid == -1)
 639                         page_found = -EINVAL;
 640                 break;
 641         }
 642
 643         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 644            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 645                 /*
 646                  * If we do the dcbz hack, we have to NX on every execution,
 647                  * so we can patch the executing code. This renders our guest
 648                  * NX-less.
 649                  */
 650                 pte.may_execute = !data;
 651         }
 652
 653         if (page_found == -ENOENT) {
 654                 /* Page not found in guest PTE entries */
 655                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 656                 vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 657                 vcpu->arch.shared->msr |=
 658                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 659                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 660         } else if (page_found == -EPERM) {
 661                 /* Storage protection */
 662                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 663                 vcpu->arch.shared->dsisr =
 664                         to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
 665                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 666                 vcpu->arch.shared->msr |=
 667                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 668                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 669         } else if (page_found == -EINVAL) {
 670                 /* Page not found in guest SLB */
 671                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 672                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 673         } else if (!is_mmio &&
 674                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 675                 /* The guest's PTE is not mapped yet. Map on the host */
 676                 kvmppc_mmu_map_page(vcpu, &pte);
 677                 if (data)
 678                         vcpu->stat.sp_storage++;
 679                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 680                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 681                         kvmppc_patch_dcbz(vcpu, &pte);
 682         } else {
 683                 /* MMIO */
 684                 vcpu->stat.mmio_exits++;
 685                 vcpu->arch.paddr_accessed = pte.raddr;
 686                 r = kvmppc_emulate_mmio(run, vcpu);
 687                 if ( r == RESUME_HOST_NV )
 688                         r = RESUME_HOST;
 689         }
 690
 691         return r;
 692 }
 693
 694 static inline int get_fpr_index(int i)
 695 {
 696 #ifdef CONFIG_VSX
 697         i *= 2;
 698 #endif
 699         return i;
 700 }
 701
 702 /* Give up external provider (FPU, Altivec, VSX) */
 703 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 704 {
 705         struct thread_struct *t = &current->thread;
 706         u64 *vcpu_fpr = vcpu->arch.fpr;
 707 #ifdef CONFIG_VSX
 708         u64 *vcpu_vsx = vcpu->arch.vsr;
 709 #endif
 710         u64 *thread_fpr = (u64*)t->fpr;
 711         int i;
 712
 713         if (!(vcpu->arch.guest_owned_ext & msr))
 714                 return;
 715
 716 #ifdef DEBUG_EXT
 717         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 718 #endif
 719
 720         switch (msr) {
 721         case MSR_FP:
 722                 giveup_fpu(current);
 723                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 724                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 725
 726                 vcpu->arch.fpscr = t->fpscr.val;
 727                 break;
 728         case MSR_VEC:
 729 #ifdef CONFIG_ALTIVEC
 730                 giveup_altivec(current);
 731                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 732                 vcpu->arch.vscr = t->vscr;
 733 #endif
 734                 break;
 735         case MSR_VSX:
 736 #ifdef CONFIG_VSX
 737                 __giveup_vsx(current);
 738                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 739                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 740 #endif
 741                 break;
 742         default:
 743                 BUG();
 744         }
 745
 746         vcpu->arch.guest_owned_ext &= ~msr;
 747         current->thread.regs->msr &= ~msr;
 748         kvmppc_recalc_shadow_msr(vcpu);
 749 }
 750
 751 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 752 {
 753         ulong srr0 = kvmppc_get_pc(vcpu);
 754         u32 last_inst = kvmppc_get_last_inst(vcpu);
 755         int ret;
 756
 757         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 758         if (ret == -ENOENT) {
 759                 ulong msr = vcpu->arch.shared->msr;
 760
 761                 msr = kvmppc_set_field(msr, 33, 33, 1);
 762                 msr = kvmppc_set_field(msr, 34, 36, 0);
 763                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 764                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 765                 return EMULATE_AGAIN;
 766         }
 767
 768         return EMULATE_DONE;
 769 }
 770
 771 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 772 {
 773
 774         /* Need to do paired single emulation? */
 775         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 776                 return EMULATE_DONE;
 777
 778         /* Read out the instruction */
 779         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 780                 /* Need to emulate */
 781                 return EMULATE_FAIL;
 782
 783         return EMULATE_AGAIN;
 784 }
 785
 786 /* Handle external providers (FPU, Altivec, VSX) */
 787 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 788                              ulong msr)
 789 {
 790         struct thread_struct *t = &current->thread;
 791         u64 *vcpu_fpr = vcpu->arch.fpr;
 792 #ifdef CONFIG_VSX
 793         u64 *vcpu_vsx = vcpu->arch.vsr;
 794 #endif
 795         u64 *thread_fpr = (u64*)t->fpr;
 796         int i;
 797
 798         /* When we have paired singles, we emulate in software */
 799         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 800                 return RESUME_GUEST;
 801
 802         if (!(vcpu->arch.shared->msr & msr)) {
 803                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 804                 return RESUME_GUEST;
 805         }
 806
 807         /* We already own the ext */
 808         if (vcpu->arch.guest_owned_ext & msr) {
 809                 return RESUME_GUEST;
 810         }
 811
 812 #ifdef DEBUG_EXT
 813         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 814 #endif
 815
 816         current->thread.regs->msr |= msr;
 817
 818         switch (msr) {
 819         case MSR_FP:
 820                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 821                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 822
 823                 t->fpscr.val = vcpu->arch.fpscr;
 824                 t->fpexc_mode = 0;
 825                 kvmppc_load_up_fpu();
 826                 break;
 827         case MSR_VEC:
 828 #ifdef CONFIG_ALTIVEC
 829                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 830                 t->vscr = vcpu->arch.vscr;
 831                 t->vrsave = -1;
 832                 kvmppc_load_up_altivec();
 833 #endif
 834                 break;
 835         case MSR_VSX:
 836 #ifdef CONFIG_VSX
 837                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 838                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 839                 kvmppc_load_up_vsx();
 840 #endif
 841                 break;
 842         default:
 843                 BUG();
 844         }
 845
 846         vcpu->arch.guest_owned_ext |= msr;
 847
 848         kvmppc_recalc_shadow_msr(vcpu);
 849
 850         return RESUME_GUEST;
 851 }
 852
 853 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 854                        unsigned int exit_nr)
 855 {
 856         int r = RESUME_HOST;
 857
 858         vcpu->stat.sum_exits++;
 859
 860         run->exit_reason = KVM_EXIT_UNKNOWN;
 861         run->ready_for_interrupt_injection = 1;
 862
 863         trace_kvm_book3s_exit(exit_nr, vcpu);
 864         kvm_resched(vcpu);
 865         switch (exit_nr) {
 866         case BOOK3S_INTERRUPT_INST_STORAGE:
 867                 vcpu->stat.pf_instruc++;
 868
 869 #ifdef CONFIG_PPC_BOOK3S_32
 870                 /* We set segments as unused segments when invalidating them. So
 871                  * treat the respective fault as segment fault. */
 872                 if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
 873                     == SR_INVALID) {
 874                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 875                         r = RESUME_GUEST;
 876                         break;
 877                 }
 878 #endif
 879
 880                 /* only care about PTEG not found errors, but leave NX alone */
 881                 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
 882                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 883                         vcpu->stat.sp_instruc++;
 884                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 885                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 886                         /*
 887                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 888                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 889                          *     that no guest that needs the dcbz hack does NX.
 890                          */
 891                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 892                         r = RESUME_GUEST;
 893                 } else {
 894                         vcpu->arch.shared->msr |=
 895                                 to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
 896                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 897                         r = RESUME_GUEST;
 898                 }
 899                 break;
 900         case BOOK3S_INTERRUPT_DATA_STORAGE:
 901         {
 902                 ulong dar = kvmppc_get_fault_dar(vcpu);
 903                 vcpu->stat.pf_storage++;
 904
 905 #ifdef CONFIG_PPC_BOOK3S_32
 906                 /* We set segments as unused segments when invalidating them. So
 907                  * treat the respective fault as segment fault. */
 908                 if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 909                         kvmppc_mmu_map_segment(vcpu, dar);
 910                         r = RESUME_GUEST;
 911                         break;
 912                 }
 913 #endif
 914
 915                 /* The only case we need to handle is missing shadow PTEs */
 916                 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
 917                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 918                 } else {
 919                         vcpu->arch.shared->dar = dar;
 920                         vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 921                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 922                         r = RESUME_GUEST;
 923                 }
 924                 break;
 925         }
 926         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 927                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 928                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 929                         kvmppc_book3s_queue_irqprio(vcpu,
 930                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 931                 }
 932                 r = RESUME_GUEST;
 933                 break;
 934         case BOOK3S_INTERRUPT_INST_SEGMENT:
 935                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 936                         kvmppc_book3s_queue_irqprio(vcpu,
 937                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 938                 }
 939                 r = RESUME_GUEST;
 940                 break;
 941         /* We're good on these - the host merely wanted to get our attention */
 942         case BOOK3S_INTERRUPT_DECREMENTER:
 943                 vcpu->stat.dec_exits++;
 944                 r = RESUME_GUEST;
 945                 break;
 946         case BOOK3S_INTERRUPT_EXTERNAL:
 947                 vcpu->stat.ext_intr_exits++;
 948                 r = RESUME_GUEST;
 949                 break;
 950         case BOOK3S_INTERRUPT_PERFMON:
 951                 r = RESUME_GUEST;
 952                 break;
 953         case BOOK3S_INTERRUPT_PROGRAM:
 954         {
 955                 enum emulation_result er;
 956                 ulong flags;
 957
 958 program_interrupt:
 959                 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
 960
 961                 if (vcpu->arch.shared->msr & MSR_PR) {
 962 #ifdef EXIT_DEBUG
 963                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 964 #endif
 965                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 966                             (INS_DCBZ & 0xfffffff7)) {
 967                                 kvmppc_core_queue_program(vcpu, flags);
 968                                 r = RESUME_GUEST;
 969                                 break;
 970                         }
 971                 }
 972
 973                 vcpu->stat.emulated_inst_exits++;
 974                 er = kvmppc_emulate_instruction(run, vcpu);
 975                 switch (er) {
 976                 case EMULATE_DONE:
 977                         r = RESUME_GUEST_NV;
 978                         break;
 979                 case EMULATE_AGAIN:
 980                         r = RESUME_GUEST;
 981                         break;
 982                 case EMULATE_FAIL:
 983                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 984                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 985                         kvmppc_core_queue_program(vcpu, flags);
 986                         r = RESUME_GUEST;
 987                         break;
 988                 case EMULATE_DO_MMIO:
 989                         run->exit_reason = KVM_EXIT_MMIO;
 990                         r = RESUME_HOST_NV;
 991                         break;
 992                 default:
 993                         BUG();
 994                 }
 995                 break;
 996         }
 997         case BOOK3S_INTERRUPT_SYSCALL:
 998                 if (vcpu->arch.osi_enabled &&
 999                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
1000                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
1001                         /* MOL hypercalls */
1002                         u64 *gprs = run->osi.gprs;
1003                         int i;
1004
1005                         run->exit_reason = KVM_EXIT_OSI;
1006                         for (i = 0; i < 32; i++)
1007                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
1008                         vcpu->arch.osi_needed = 1;
1009                         r = RESUME_HOST_NV;
1010                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
1011                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
1012                         /* KVM PV hypercalls */
1013                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
1014                         r = RESUME_GUEST;
1015                 } else {
1016                         /* Guest syscalls */
1017                         vcpu->stat.syscall_exits++;
1018                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1019                         r = RESUME_GUEST;
1020                 }
1021                 break;
1022         case BOOK3S_INTERRUPT_FP_UNAVAIL:
1023         case BOOK3S_INTERRUPT_ALTIVEC:
1024         case BOOK3S_INTERRUPT_VSX:
1025         {
1026                 int ext_msr = 0;
1027
1028                 switch (exit_nr) {
1029                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
1030                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
1031                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
1032                 }
1033
1034                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
1035                 case EMULATE_DONE:
1036                         /* everything ok - let's enable the ext */
1037                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
1038                         break;
1039                 case EMULATE_FAIL:
1040                         /* we need to emulate this instruction */
1041                         goto program_interrupt;
1042                         break;
1043                 default:
1044                         /* nothing to worry about - go again */
1045                         break;
1046                 }
1047                 break;
1048         }
1049         case BOOK3S_INTERRUPT_ALIGNMENT:
1050                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
1051                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
1052                                 kvmppc_get_last_inst(vcpu));
1053                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
1054                                 kvmppc_get_last_inst(vcpu));
1055                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1056                 }
1057                 r = RESUME_GUEST;
1058                 break;
1059         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1060         case BOOK3S_INTERRUPT_TRACE:
1061                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
1062                 r = RESUME_GUEST;
1063                 break;
1064         default:
1065                 /* Ugh - bork here! What did we get? */
1066                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
1067                         exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
1068                 r = RESUME_HOST;
1069                 BUG();
1070                 break;
1071         }
1072
1073
1074         if (!(r & RESUME_HOST)) {
1075                 /* To avoid clobbering exit_reason, only check for signals if
1076                  * we aren't already exiting to userspace for some other
1077                  * reason. */
1078                 if (signal_pending(current)) {
1079 #ifdef EXIT_DEBUG
1080                         printk(KERN_EMERG "KVM: Going back to host\n");
1081 #endif
1082                         vcpu->stat.signal_exits++;
1083                         run->exit_reason = KVM_EXIT_INTR;
1084                         r = -EINTR;
1085                 } else {
1086                         /* In case an interrupt came in that was triggered
1087                          * from userspace (like DEC), we need to check what
1088                          * to inject now! */
1089                         kvmppc_core_deliver_interrupts(vcpu);
1090                 }
1091         }
1092
1093         trace_kvm_book3s_reenter(r, vcpu);
1094
1095         return r;
1096 }
1097
1098 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1099 {
1100         return 0;
1101 }
1102
1103 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1104 {
1105         int i;
1106
1107         regs->pc = kvmppc_get_pc(vcpu);
1108         regs->cr = kvmppc_get_cr(vcpu);
1109         regs->ctr = kvmppc_get_ctr(vcpu);
1110         regs->lr = kvmppc_get_lr(vcpu);
1111         regs->xer = kvmppc_get_xer(vcpu);
1112         regs->msr = vcpu->arch.shared->msr;
1113         regs->srr0 = vcpu->arch.shared->srr0;
1114         regs->srr1 = vcpu->arch.shared->srr1;
1115         regs->pid = vcpu->arch.pid;
1116         regs->sprg0 = vcpu->arch.shared->sprg0;
1117         regs->sprg1 = vcpu->arch.shared->sprg1;
1118         regs->sprg2 = vcpu->arch.shared->sprg2;
1119         regs->sprg3 = vcpu->arch.shared->sprg3;
1120         regs->sprg5 = vcpu->arch.sprg4;
1121         regs->sprg6 = vcpu->arch.sprg5;
1122         regs->sprg7 = vcpu->arch.sprg6;
1123
1124         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1125                 regs->gpr[i] = kvmppc_get_gpr(vcpu, i);
1126
1127         return 0;
1128 }
1129
1130 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1131 {
1132         int i;
1133
1134         kvmppc_set_pc(vcpu, regs->pc);
1135         kvmppc_set_cr(vcpu, regs->cr);
1136         kvmppc_set_ctr(vcpu, regs->ctr);
1137         kvmppc_set_lr(vcpu, regs->lr);
1138         kvmppc_set_xer(vcpu, regs->xer);
1139         kvmppc_set_msr(vcpu, regs->msr);
1140         vcpu->arch.shared->srr0 = regs->srr0;
1141         vcpu->arch.shared->srr1 = regs->srr1;
1142         vcpu->arch.shared->sprg0 = regs->sprg0;
1143         vcpu->arch.shared->sprg1 = regs->sprg1;
1144         vcpu->arch.shared->sprg2 = regs->sprg2;
1145         vcpu->arch.shared->sprg3 = regs->sprg3;
1146         vcpu->arch.sprg5 = regs->sprg4;
1147         vcpu->arch.sprg6 = regs->sprg5;
1148         vcpu->arch.sprg7 = regs->sprg6;
1149
1150         for (i = 0; i < ARRAY_SIZE(regs->gpr); i++)
1151                 kvmppc_set_gpr(vcpu, i, regs->gpr[i]);
1152
1153         return 0;
1154 }
1155
1156 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1157                                   struct kvm_sregs *sregs)
1158 {
1159         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1160         int i;
1161
1162         sregs->pvr = vcpu->arch.pvr;
1163
1164         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
1165         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1166                 for (i = 0; i < 64; i++) {
1167                         sregs->u.s.ppc64.slb[i].slbe = vcpu3s->slb[i].orige | i;
1168                         sregs->u.s.ppc64.slb[i].slbv = vcpu3s->slb[i].origv;
1169                 }
1170         } else {
1171                 for (i = 0; i < 16; i++)
1172                         sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
1173
1174                 for (i = 0; i < 8; i++) {
1175                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
1176                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
1177                 }
1178         }
1179
1180         return 0;
1181 }
1182
1183 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1184                                   struct kvm_sregs *sregs)
1185 {
1186         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1187         int i;
1188
1189         kvmppc_set_pvr(vcpu, sregs->pvr);
1190
1191         vcpu3s->sdr1 = sregs->u.s.sdr1;
1192         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1193                 for (i = 0; i < 64; i++) {
1194                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1195                                                     sregs->u.s.ppc64.slb[i].slbe);
1196                 }
1197         } else {
1198                 for (i = 0; i < 16; i++) {
1199                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1200                 }
1201                 for (i = 0; i < 8; i++) {
1202                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1203                                        (u32)sregs->u.s.ppc32.ibat[i]);
1204                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1205                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1206                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1207                                        (u32)sregs->u.s.ppc32.dbat[i]);
1208                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1209                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1210                 }
1211         }
1212
1213         /* Flush the MMU after messing with the segments */
1214         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1215
1216         return 0;
1217 }
1218
1219 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1220 {
1221         return -ENOTSUPP;
1222 }
1223
1224 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1225 {
1226         return -ENOTSUPP;
1227 }
1228
1229 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1230                                   struct kvm_translation *tr)
1231 {
1232         return 0;
1233 }
1234
1235 /*
1236  * Get (and clear) the dirty memory log for a memory slot.
1237  */
1238 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1239                                       struct kvm_dirty_log *log)
1240 {
1241         struct kvm_memory_slot *memslot;
1242         struct kvm_vcpu *vcpu;
1243         ulong ga, ga_end;
1244         int is_dirty = 0;
1245         int r;
1246         unsigned long n;
1247
1248         mutex_lock(&kvm->slots_lock);
1249
1250         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1251         if (r)
1252                 goto out;
1253
1254         /* If nothing is dirty, don't bother messing with page tables. */
1255         if (is_dirty) {
1256                 memslot = &kvm->memslots->memslots[log->slot];
1257
1258                 ga = memslot->base_gfn << PAGE_SHIFT;
1259                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1260
1261                 kvm_for_each_vcpu(n, vcpu, kvm)
1262                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1263
1264                 n = kvm_dirty_bitmap_bytes(memslot);
1265                 memset(memslot->dirty_bitmap, 0, n);
1266         }
1267
1268         r = 0;
1269 out:
1270         mutex_unlock(&kvm->slots_lock);
1271         return r;
1272 }
1273
1274 int kvmppc_core_check_processor_compat(void)
1275 {
1276         return 0;
1277 }
1278
1279 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1280 {
1281         struct kvmppc_vcpu_book3s *vcpu_book3s;
1282         struct kvm_vcpu *vcpu;
1283         int err = -ENOMEM;
1284         unsigned long p;
1285
1286         vcpu_book3s = vmalloc(sizeof(struct kvmppc_vcpu_book3s));
1287         if (!vcpu_book3s)
1288                 goto out;
1289
1290         memset(vcpu_book3s, 0, sizeof(struct kvmppc_vcpu_book3s));
1291
1292         vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
1293                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
1294         if (!vcpu_book3s->shadow_vcpu)
1295                 goto free_vcpu;
1296
1297         vcpu = &vcpu_book3s->vcpu;
1298         err = kvm_vcpu_init(vcpu, kvm, id);
1299         if (err)
1300                 goto free_shadow_vcpu;
1301
1302         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1303         /* the real shared page fills the last 4k of our page */
1304         vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
1305         if (!p)
1306                 goto uninit_vcpu;
1307
1308         vcpu->arch.host_retip = kvm_return_point;
1309         vcpu->arch.host_msr = mfmsr();
1310 #ifdef CONFIG_PPC_BOOK3S_64
1311         /* default to book3s_64 (970fx) */
1312         vcpu->arch.pvr = 0x3C0301;
1313 #else
1314         /* default to book3s_32 (750) */
1315         vcpu->arch.pvr = 0x84202;
1316 #endif
1317         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1318         vcpu_book3s->slb_nr = 64;
1319
1320         /* remember where some real-mode handlers are */
1321         vcpu->arch.trampoline_lowmem = kvmppc_trampoline_lowmem;
1322         vcpu->arch.trampoline_enter = kvmppc_trampoline_enter;
1323         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
1324 #ifdef CONFIG_PPC_BOOK3S_64
1325         vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
1326 #else
1327         vcpu->arch.rmcall = (ulong)kvmppc_rmcall;
1328 #endif
1329
1330         vcpu->arch.shadow_msr = MSR_USER64;
1331
1332         err = kvmppc_mmu_init(vcpu);
1333         if (err < 0)
1334                 goto uninit_vcpu;
1335
1336         return vcpu;
1337
1338 uninit_vcpu:
1339         kvm_vcpu_uninit(vcpu);
1340 free_shadow_vcpu:
1341         kfree(vcpu_book3s->shadow_vcpu);
1342 free_vcpu:
1343         vfree(vcpu_book3s);
1344 out:
1345         return ERR_PTR(err);
1346 }
1347
1348 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1349 {
1350         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1351
1352         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1353         kvm_vcpu_uninit(vcpu);
1354         kfree(vcpu_book3s->shadow_vcpu);
1355         vfree(vcpu_book3s);
1356 }
1357
1358 extern int __kvmppc_vcpu_entry(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
1359 int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1360 {
1361         int ret;
1362         double fpr[32][TS_FPRWIDTH];
1363         unsigned int fpscr;
1364         int fpexc_mode;
1365 #ifdef CONFIG_ALTIVEC
1366         vector128 vr[32];
1367         vector128 vscr;
1368         unsigned long uninitialized_var(vrsave);
1369         int used_vr;
1370 #endif
1371 #ifdef CONFIG_VSX
1372         int used_vsr;
1373 #endif
1374         ulong ext_msr;
1375
1376         /* No need to go into the guest when all we do is going out */
1377         if (signal_pending(current)) {
1378                 kvm_run->exit_reason = KVM_EXIT_INTR;
1379                 return -EINTR;
1380         }
1381
1382         /* Save FPU state in stack */
1383         if (current->thread.regs->msr & MSR_FP)
1384                 giveup_fpu(current);
1385         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1386         fpscr = current->thread.fpscr.val;
1387         fpexc_mode = current->thread.fpexc_mode;
1388
1389 #ifdef CONFIG_ALTIVEC
1390         /* Save Altivec state in stack */
1391         used_vr = current->thread.used_vr;
1392         if (used_vr) {
1393                 if (current->thread.regs->msr & MSR_VEC)
1394                         giveup_altivec(current);
1395                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1396                 vscr = current->thread.vscr;
1397                 vrsave = current->thread.vrsave;
1398         }
1399 #endif
1400
1401 #ifdef CONFIG_VSX
1402         /* Save VSX state in stack */
1403         used_vsr = current->thread.used_vsr;
1404         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1405                         __giveup_vsx(current);
1406 #endif
1407
1408         /* Remember the MSR with disabled extensions */
1409         ext_msr = current->thread.regs->msr;
1410
1411         /* XXX we get called with irq disabled - change that! */
1412         local_irq_enable();
1413
1414         /* Preload FPU if it's enabled */
1415         if (vcpu->arch.shared->msr & MSR_FP)
1416                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1417
1418         ret = __kvmppc_vcpu_entry(kvm_run, vcpu);
1419
1420         local_irq_disable();
1421
1422         current->thread.regs->msr = ext_msr;
1423
1424         /* Make sure we save the guest FPU/Altivec/VSX state */
1425         kvmppc_giveup_ext(vcpu, MSR_FP);
1426         kvmppc_giveup_ext(vcpu, MSR_VEC);
1427         kvmppc_giveup_ext(vcpu, MSR_VSX);
1428
1429         /* Restore FPU state from stack */
1430         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1431         current->thread.fpscr.val = fpscr;
1432         current->thread.fpexc_mode = fpexc_mode;
1433
1434 #ifdef CONFIG_ALTIVEC
1435         /* Restore Altivec state from stack */
1436         if (used_vr && current->thread.used_vr) {
1437                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1438                 current->thread.vscr = vscr;
1439                 current->thread.vrsave = vrsave;
1440         }
1441         current->thread.used_vr = used_vr;
1442 #endif
1443
1444 #ifdef CONFIG_VSX
1445         current->thread.used_vsr = used_vsr;
1446 #endif
1447
1448         return ret;
1449 }
1450
1451 static int kvmppc_book3s_init(void)
1452 {
1453         int r;
1454
1455         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1456                      THIS_MODULE);
1457
1458         if (r)
1459                 return r;
1460
1461         r = kvmppc_mmu_hpte_sysinit();
1462
1463         return r;
1464 }
1465
1466 static void kvmppc_book3s_exit(void)
1467 {
1468         kvmppc_mmu_hpte_sysexit();
1469         kvm_exit();
1470 }
1471
1472 module_init(kvmppc_book3s_init);
1473 module_exit(kvmppc_book3s_exit);