arch/powerpc/kvm/book3s_pr.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  *    Paul Mackerras <paulus@samba.org>
   8  *
   9  * Description:
  10  * Functions relating to running KVM on Book 3S processors where
  11  * we don't have access to hypervisor mode, and we run the guest
  12  * in problem state (user mode).
  13  *
  14  * This file is derived from arch/powerpc/kvm/44x.c,
  15  * by Hollis Blanchard <hollisb@us.ibm.com>.
  16  *
  17  * This program is free software; you can redistribute it and/or modify
  18  * it under the terms of the GNU General Public License, version 2, as
  19  * published by the Free Software Foundation.
  20  */
  21
  22 #include <linux/kvm_host.h>
  23 #include <linux/err.h>
  24 #include <linux/slab.h>
  25
  26 #include <asm/reg.h>
  27 #include <asm/cputable.h>
  28 #include <asm/cacheflush.h>
  29 #include <asm/tlbflush.h>
  30 #include <asm/uaccess.h>
  31 #include <asm/io.h>
  32 #include <asm/kvm_ppc.h>
  33 #include <asm/kvm_book3s.h>
  34 #include <asm/mmu_context.h>
  35 #include <linux/gfp.h>
  36 #include <linux/sched.h>
  37 #include <linux/vmalloc.h>
  38 #include <linux/highmem.h>
  39
  40 #include "trace.h"
  41
  42 /* #define EXIT_DEBUG */
  43 /* #define DEBUG_EXT */
  44
  45 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  46                              ulong msr);
  47
  48 /* Some compatibility defines */
  49 #ifdef CONFIG_PPC_BOOK3S_32
  50 #define MSR_USER32 MSR_USER
  51 #define MSR_USER64 MSR_USER
  52 #define HW_PAGE_SIZE PAGE_SIZE
  53 #endif
  54
  55 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  56 {
  57 #ifdef CONFIG_PPC_BOOK3S_64
  58         memcpy(to_svcpu(vcpu)->slb, to_book3s(vcpu)->slb_shadow, sizeof(to_svcpu(vcpu)->slb));
  59         memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
  60                sizeof(get_paca()->shadow_vcpu));
  61         to_svcpu(vcpu)->slb_max = to_book3s(vcpu)->slb_shadow_max;
  62 #endif
  63
  64 #ifdef CONFIG_PPC_BOOK3S_32
  65         current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
  66 #endif
  67 }
  68
  69 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  70 {
  71 #ifdef CONFIG_PPC_BOOK3S_64
  72         memcpy(to_book3s(vcpu)->slb_shadow, to_svcpu(vcpu)->slb, sizeof(to_svcpu(vcpu)->slb));
  73         memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
  74                sizeof(get_paca()->shadow_vcpu));
  75         to_book3s(vcpu)->slb_shadow_max = to_svcpu(vcpu)->slb_max;
  76 #endif
  77
  78         kvmppc_giveup_ext(vcpu, MSR_FP);
  79         kvmppc_giveup_ext(vcpu, MSR_VEC);
  80         kvmppc_giveup_ext(vcpu, MSR_VSX);
  81 }
  82
  83 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
  84 {
  85         ulong smsr = vcpu->arch.shared->msr;
  86
  87         /* Guest MSR values */
  88         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE | MSR_DE;
  89         /* Process MSR values */
  90         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
  91         /* External providers the guest reserved */
  92         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
  93         /* 64-bit Process MSR values */
  94 #ifdef CONFIG_PPC_BOOK3S_64
  95         smsr |= MSR_ISF | MSR_HV;
  96 #endif
  97         vcpu->arch.shadow_msr = smsr;
  98 }
  99
 100 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 101 {
 102         ulong old_msr = vcpu->arch.shared->msr;
 103
 104 #ifdef EXIT_DEBUG
 105         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 106 #endif
 107
 108         msr &= to_book3s(vcpu)->msr_mask;
 109         vcpu->arch.shared->msr = msr;
 110         kvmppc_recalc_shadow_msr(vcpu);
 111
 112         if (msr & MSR_POW) {
 113                 if (!vcpu->arch.pending_exceptions) {
 114                         kvm_vcpu_block(vcpu);
 115                         vcpu->stat.halt_wakeup++;
 116
 117                         /* Unset POW bit after we woke up */
 118                         msr &= ~MSR_POW;
 119                         vcpu->arch.shared->msr = msr;
 120                 }
 121         }
 122
 123         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 124                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 125                 kvmppc_mmu_flush_segments(vcpu);
 126                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 127
 128                 /* Preload magic page segment when in kernel mode */
 129                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 130                         struct kvm_vcpu_arch *a = &vcpu->arch;
 131
 132                         if (msr & MSR_DR)
 133                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 134                         else
 135                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 136                 }
 137         }
 138
 139         /* Preload FPU if it's enabled */
 140         if (vcpu->arch.shared->msr & MSR_FP)
 141                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 142 }
 143
 144 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 145 {
 146         u32 host_pvr;
 147
 148         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 149         vcpu->arch.pvr = pvr;
 150 #ifdef CONFIG_PPC_BOOK3S_64
 151         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 152                 kvmppc_mmu_book3s_64_init(vcpu);
 153                 if (!to_book3s(vcpu)->hior_sregs)
 154                         to_book3s(vcpu)->hior = 0xfff00000;
 155                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 156         } else
 157 #endif
 158         {
 159                 kvmppc_mmu_book3s_32_init(vcpu);
 160                 if (!to_book3s(vcpu)->hior_sregs)
 161                         to_book3s(vcpu)->hior = 0;
 162                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 163         }
 164
 165         /* If we are in hypervisor level on 970, we can tell the CPU to
 166          * treat DCBZ as 32 bytes store */
 167         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 168         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 169             !strcmp(cur_cpu_spec->platform, "ppc970"))
 170                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 171
 172         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 173            really needs them in a VM on Cell and force disable them. */
 174         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 175                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 176
 177 #ifdef CONFIG_PPC_BOOK3S_32
 178         /* 32 bit Book3S always has 32 byte dcbz */
 179         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 180 #endif
 181
 182         /* On some CPUs we can execute paired single operations natively */
 183         asm ( "mfpvr %0" : "=r"(host_pvr));
 184         switch (host_pvr) {
 185         case 0x00080200:        /* lonestar 2.0 */
 186         case 0x00088202:        /* lonestar 2.2 */
 187         case 0x70000100:        /* gekko 1.0 */
 188         case 0x00080100:        /* gekko 2.0 */
 189         case 0x00083203:        /* gekko 2.3a */
 190         case 0x00083213:        /* gekko 2.3b */
 191         case 0x00083204:        /* gekko 2.4 */
 192         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 193         case 0x00087200:        /* broadway */
 194                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 195                 /* Enable HID2.PSE - in case we need it later */
 196                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 197         }
 198 }
 199
 200 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 201  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 202  * emulate 32 bytes dcbz length.
 203  *
 204  * The Book3s_64 inventors also realized this case and implemented a special bit
 205  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 206  *
 207  * My approach here is to patch the dcbz instruction on executing pages.
 208  */
 209 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 210 {
 211         struct page *hpage;
 212         u64 hpage_offset;
 213         u32 *page;
 214         int i;
 215
 216         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 217         if (is_error_page(hpage)) {
 218                 kvm_release_page_clean(hpage);
 219                 return;
 220         }
 221
 222         hpage_offset = pte->raddr & ~PAGE_MASK;
 223         hpage_offset &= ~0xFFFULL;
 224         hpage_offset /= 4;
 225
 226         get_page(hpage);
 227         page = kmap_atomic(hpage, KM_USER0);
 228
 229         /* patch dcbz into reserved instruction, so we trap */
 230         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 231                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 232                         page[i] &= 0xfffffff7;
 233
 234         kunmap_atomic(page, KM_USER0);
 235         put_page(hpage);
 236 }
 237
 238 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 239 {
 240         ulong mp_pa = vcpu->arch.magic_page_pa;
 241
 242         if (unlikely(mp_pa) &&
 243             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 244                 return 1;
 245         }
 246
 247         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 248 }
 249
 250 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 251                             ulong eaddr, int vec)
 252 {
 253         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 254         int r = RESUME_GUEST;
 255         int relocated;
 256         int page_found = 0;
 257         struct kvmppc_pte pte;
 258         bool is_mmio = false;
 259         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 260         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 261         u64 vsid;
 262
 263         relocated = data ? dr : ir;
 264
 265         /* Resolve real address if translation turned on */
 266         if (relocated) {
 267                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 268         } else {
 269                 pte.may_execute = true;
 270                 pte.may_read = true;
 271                 pte.may_write = true;
 272                 pte.raddr = eaddr & KVM_PAM;
 273                 pte.eaddr = eaddr;
 274                 pte.vpage = eaddr >> 12;
 275         }
 276
 277         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 278         case 0:
 279                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 280                 break;
 281         case MSR_DR:
 282         case MSR_IR:
 283                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 284
 285                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 286                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 287                 else
 288                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 289                 pte.vpage |= vsid;
 290
 291                 if (vsid == -1)
 292                         page_found = -EINVAL;
 293                 break;
 294         }
 295
 296         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 297            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 298                 /*
 299                  * If we do the dcbz hack, we have to NX on every execution,
 300                  * so we can patch the executing code. This renders our guest
 301                  * NX-less.
 302                  */
 303                 pte.may_execute = !data;
 304         }
 305
 306         if (page_found == -ENOENT) {
 307                 /* Page not found in guest PTE entries */
 308                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 309                 vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 310                 vcpu->arch.shared->msr |=
 311                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 312                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 313         } else if (page_found == -EPERM) {
 314                 /* Storage protection */
 315                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 316                 vcpu->arch.shared->dsisr =
 317                         to_svcpu(vcpu)->fault_dsisr & ~DSISR_NOHPTE;
 318                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 319                 vcpu->arch.shared->msr |=
 320                         (to_svcpu(vcpu)->shadow_srr1 & 0x00000000f8000000ULL);
 321                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 322         } else if (page_found == -EINVAL) {
 323                 /* Page not found in guest SLB */
 324                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 325                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 326         } else if (!is_mmio &&
 327                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 328                 /* The guest's PTE is not mapped yet. Map on the host */
 329                 kvmppc_mmu_map_page(vcpu, &pte);
 330                 if (data)
 331                         vcpu->stat.sp_storage++;
 332                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 333                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 334                         kvmppc_patch_dcbz(vcpu, &pte);
 335         } else {
 336                 /* MMIO */
 337                 vcpu->stat.mmio_exits++;
 338                 vcpu->arch.paddr_accessed = pte.raddr;
 339                 r = kvmppc_emulate_mmio(run, vcpu);
 340                 if ( r == RESUME_HOST_NV )
 341                         r = RESUME_HOST;
 342         }
 343
 344         return r;
 345 }
 346
 347 static inline int get_fpr_index(int i)
 348 {
 349 #ifdef CONFIG_VSX
 350         i *= 2;
 351 #endif
 352         return i;
 353 }
 354
 355 /* Give up external provider (FPU, Altivec, VSX) */
 356 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 357 {
 358         struct thread_struct *t = &current->thread;
 359         u64 *vcpu_fpr = vcpu->arch.fpr;
 360 #ifdef CONFIG_VSX
 361         u64 *vcpu_vsx = vcpu->arch.vsr;
 362 #endif
 363         u64 *thread_fpr = (u64*)t->fpr;
 364         int i;
 365
 366         if (!(vcpu->arch.guest_owned_ext & msr))
 367                 return;
 368
 369 #ifdef DEBUG_EXT
 370         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 371 #endif
 372
 373         switch (msr) {
 374         case MSR_FP:
 375                 giveup_fpu(current);
 376                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 377                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 378
 379                 vcpu->arch.fpscr = t->fpscr.val;
 380                 break;
 381         case MSR_VEC:
 382 #ifdef CONFIG_ALTIVEC
 383                 giveup_altivec(current);
 384                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 385                 vcpu->arch.vscr = t->vscr;
 386 #endif
 387                 break;
 388         case MSR_VSX:
 389 #ifdef CONFIG_VSX
 390                 __giveup_vsx(current);
 391                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 392                         vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 393 #endif
 394                 break;
 395         default:
 396                 BUG();
 397         }
 398
 399         vcpu->arch.guest_owned_ext &= ~msr;
 400         current->thread.regs->msr &= ~msr;
 401         kvmppc_recalc_shadow_msr(vcpu);
 402 }
 403
 404 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 405 {
 406         ulong srr0 = kvmppc_get_pc(vcpu);
 407         u32 last_inst = kvmppc_get_last_inst(vcpu);
 408         int ret;
 409
 410         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 411         if (ret == -ENOENT) {
 412                 ulong msr = vcpu->arch.shared->msr;
 413
 414                 msr = kvmppc_set_field(msr, 33, 33, 1);
 415                 msr = kvmppc_set_field(msr, 34, 36, 0);
 416                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 417                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 418                 return EMULATE_AGAIN;
 419         }
 420
 421         return EMULATE_DONE;
 422 }
 423
 424 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 425 {
 426
 427         /* Need to do paired single emulation? */
 428         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 429                 return EMULATE_DONE;
 430
 431         /* Read out the instruction */
 432         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 433                 /* Need to emulate */
 434                 return EMULATE_FAIL;
 435
 436         return EMULATE_AGAIN;
 437 }
 438
 439 /* Handle external providers (FPU, Altivec, VSX) */
 440 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 441                              ulong msr)
 442 {
 443         struct thread_struct *t = &current->thread;
 444         u64 *vcpu_fpr = vcpu->arch.fpr;
 445 #ifdef CONFIG_VSX
 446         u64 *vcpu_vsx = vcpu->arch.vsr;
 447 #endif
 448         u64 *thread_fpr = (u64*)t->fpr;
 449         int i;
 450
 451         /* When we have paired singles, we emulate in software */
 452         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 453                 return RESUME_GUEST;
 454
 455         if (!(vcpu->arch.shared->msr & msr)) {
 456                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 457                 return RESUME_GUEST;
 458         }
 459
 460         /* We already own the ext */
 461         if (vcpu->arch.guest_owned_ext & msr) {
 462                 return RESUME_GUEST;
 463         }
 464
 465 #ifdef DEBUG_EXT
 466         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 467 #endif
 468
 469         current->thread.regs->msr |= msr;
 470
 471         switch (msr) {
 472         case MSR_FP:
 473                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 474                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 475
 476                 t->fpscr.val = vcpu->arch.fpscr;
 477                 t->fpexc_mode = 0;
 478                 kvmppc_load_up_fpu();
 479                 break;
 480         case MSR_VEC:
 481 #ifdef CONFIG_ALTIVEC
 482                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 483                 t->vscr = vcpu->arch.vscr;
 484                 t->vrsave = -1;
 485                 kvmppc_load_up_altivec();
 486 #endif
 487                 break;
 488         case MSR_VSX:
 489 #ifdef CONFIG_VSX
 490                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr); i++)
 491                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 492                 kvmppc_load_up_vsx();
 493 #endif
 494                 break;
 495         default:
 496                 BUG();
 497         }
 498
 499         vcpu->arch.guest_owned_ext |= msr;
 500
 501         kvmppc_recalc_shadow_msr(vcpu);
 502
 503         return RESUME_GUEST;
 504 }
 505
 506 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 507                        unsigned int exit_nr)
 508 {
 509         int r = RESUME_HOST;
 510
 511         vcpu->stat.sum_exits++;
 512
 513         run->exit_reason = KVM_EXIT_UNKNOWN;
 514         run->ready_for_interrupt_injection = 1;
 515
 516         trace_kvm_book3s_exit(exit_nr, vcpu);
 517         kvm_resched(vcpu);
 518         switch (exit_nr) {
 519         case BOOK3S_INTERRUPT_INST_STORAGE:
 520                 vcpu->stat.pf_instruc++;
 521
 522 #ifdef CONFIG_PPC_BOOK3S_32
 523                 /* We set segments as unused segments when invalidating them. So
 524                  * treat the respective fault as segment fault. */
 525                 if (to_svcpu(vcpu)->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT]
 526                     == SR_INVALID) {
 527                         kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 528                         r = RESUME_GUEST;
 529                         break;
 530                 }
 531 #endif
 532
 533                 /* only care about PTEG not found errors, but leave NX alone */
 534                 if (to_svcpu(vcpu)->shadow_srr1 & 0x40000000) {
 535                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 536                         vcpu->stat.sp_instruc++;
 537                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 538                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 539                         /*
 540                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 541                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 542                          *     that no guest that needs the dcbz hack does NX.
 543                          */
 544                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 545                         r = RESUME_GUEST;
 546                 } else {
 547                         vcpu->arch.shared->msr |=
 548                                 to_svcpu(vcpu)->shadow_srr1 & 0x58000000;
 549                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 550                         r = RESUME_GUEST;
 551                 }
 552                 break;
 553         case BOOK3S_INTERRUPT_DATA_STORAGE:
 554         {
 555                 ulong dar = kvmppc_get_fault_dar(vcpu);
 556                 vcpu->stat.pf_storage++;
 557
 558 #ifdef CONFIG_PPC_BOOK3S_32
 559                 /* We set segments as unused segments when invalidating them. So
 560                  * treat the respective fault as segment fault. */
 561                 if ((to_svcpu(vcpu)->sr[dar >> SID_SHIFT]) == SR_INVALID) {
 562                         kvmppc_mmu_map_segment(vcpu, dar);
 563                         r = RESUME_GUEST;
 564                         break;
 565                 }
 566 #endif
 567
 568                 /* The only case we need to handle is missing shadow PTEs */
 569                 if (to_svcpu(vcpu)->fault_dsisr & DSISR_NOHPTE) {
 570                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 571                 } else {
 572                         vcpu->arch.shared->dar = dar;
 573                         vcpu->arch.shared->dsisr = to_svcpu(vcpu)->fault_dsisr;
 574                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 575                         r = RESUME_GUEST;
 576                 }
 577                 break;
 578         }
 579         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 580                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 581                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 582                         kvmppc_book3s_queue_irqprio(vcpu,
 583                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 584                 }
 585                 r = RESUME_GUEST;
 586                 break;
 587         case BOOK3S_INTERRUPT_INST_SEGMENT:
 588                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 589                         kvmppc_book3s_queue_irqprio(vcpu,
 590                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 591                 }
 592                 r = RESUME_GUEST;
 593                 break;
 594         /* We're good on these - the host merely wanted to get our attention */
 595         case BOOK3S_INTERRUPT_DECREMENTER:
 596                 vcpu->stat.dec_exits++;
 597                 r = RESUME_GUEST;
 598                 break;
 599         case BOOK3S_INTERRUPT_EXTERNAL:
 600                 vcpu->stat.ext_intr_exits++;
 601                 r = RESUME_GUEST;
 602                 break;
 603         case BOOK3S_INTERRUPT_PERFMON:
 604                 r = RESUME_GUEST;
 605                 break;
 606         case BOOK3S_INTERRUPT_PROGRAM:
 607         {
 608                 enum emulation_result er;
 609                 ulong flags;
 610
 611 program_interrupt:
 612                 flags = to_svcpu(vcpu)->shadow_srr1 & 0x1f0000ull;
 613
 614                 if (vcpu->arch.shared->msr & MSR_PR) {
 615 #ifdef EXIT_DEBUG
 616                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 617 #endif
 618                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 619                             (INS_DCBZ & 0xfffffff7)) {
 620                                 kvmppc_core_queue_program(vcpu, flags);
 621                                 r = RESUME_GUEST;
 622                                 break;
 623                         }
 624                 }
 625
 626                 vcpu->stat.emulated_inst_exits++;
 627                 er = kvmppc_emulate_instruction(run, vcpu);
 628                 switch (er) {
 629                 case EMULATE_DONE:
 630                         r = RESUME_GUEST_NV;
 631                         break;
 632                 case EMULATE_AGAIN:
 633                         r = RESUME_GUEST;
 634                         break;
 635                 case EMULATE_FAIL:
 636                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 637                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 638                         kvmppc_core_queue_program(vcpu, flags);
 639                         r = RESUME_GUEST;
 640                         break;
 641                 case EMULATE_DO_MMIO:
 642                         run->exit_reason = KVM_EXIT_MMIO;
 643                         r = RESUME_HOST_NV;
 644                         break;
 645                 default:
 646                         BUG();
 647                 }
 648                 break;
 649         }
 650         case BOOK3S_INTERRUPT_SYSCALL:
 651                 if (vcpu->arch.osi_enabled &&
 652                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 653                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 654                         /* MOL hypercalls */
 655                         u64 *gprs = run->osi.gprs;
 656                         int i;
 657
 658                         run->exit_reason = KVM_EXIT_OSI;
 659                         for (i = 0; i < 32; i++)
 660                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 661                         vcpu->arch.osi_needed = 1;
 662                         r = RESUME_HOST_NV;
 663                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 664                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 665                         /* KVM PV hypercalls */
 666                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 667                         r = RESUME_GUEST;
 668                 } else {
 669                         /* Guest syscalls */
 670                         vcpu->stat.syscall_exits++;
 671                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 672                         r = RESUME_GUEST;
 673                 }
 674                 break;
 675         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 676         case BOOK3S_INTERRUPT_ALTIVEC:
 677         case BOOK3S_INTERRUPT_VSX:
 678         {
 679                 int ext_msr = 0;
 680
 681                 switch (exit_nr) {
 682                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 683                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 684                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 685                 }
 686
 687                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
 688                 case EMULATE_DONE:
 689                         /* everything ok - let's enable the ext */
 690                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
 691                         break;
 692                 case EMULATE_FAIL:
 693                         /* we need to emulate this instruction */
 694                         goto program_interrupt;
 695                         break;
 696                 default:
 697                         /* nothing to worry about - go again */
 698                         break;
 699                 }
 700                 break;
 701         }
 702         case BOOK3S_INTERRUPT_ALIGNMENT:
 703                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
 704                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
 705                                 kvmppc_get_last_inst(vcpu));
 706                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
 707                                 kvmppc_get_last_inst(vcpu));
 708                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 709                 }
 710                 r = RESUME_GUEST;
 711                 break;
 712         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 713         case BOOK3S_INTERRUPT_TRACE:
 714                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 715                 r = RESUME_GUEST;
 716                 break;
 717         default:
 718                 /* Ugh - bork here! What did we get? */
 719                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
 720                         exit_nr, kvmppc_get_pc(vcpu), to_svcpu(vcpu)->shadow_srr1);
 721                 r = RESUME_HOST;
 722                 BUG();
 723                 break;
 724         }
 725
 726
 727         if (!(r & RESUME_HOST)) {
 728                 /* To avoid clobbering exit_reason, only check for signals if
 729                  * we aren't already exiting to userspace for some other
 730                  * reason. */
 731                 if (signal_pending(current)) {
 732 #ifdef EXIT_DEBUG
 733                         printk(KERN_EMERG "KVM: Going back to host\n");
 734 #endif
 735                         vcpu->stat.signal_exits++;
 736                         run->exit_reason = KVM_EXIT_INTR;
 737                         r = -EINTR;
 738                 } else {
 739                         /* In case an interrupt came in that was triggered
 740                          * from userspace (like DEC), we need to check what
 741                          * to inject now! */
 742                         kvmppc_core_deliver_interrupts(vcpu);
 743                 }
 744         }
 745
 746         trace_kvm_book3s_reenter(r, vcpu);
 747
 748         return r;
 749 }
 750
 751 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 752                                   struct kvm_sregs *sregs)
 753 {
 754         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 755         int i;
 756
 757         sregs->pvr = vcpu->arch.pvr;
 758
 759         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 760         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 761                 for (i = 0; i < 64; i++) {
 762                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
 763                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 764                 }
 765         } else {
 766                 for (i = 0; i < 16; i++)
 767                         sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
 768
 769                 for (i = 0; i < 8; i++) {
 770                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 771                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 772                 }
 773         }
 774
 775         if (sregs->u.s.flags & KVM_SREGS_S_HIOR)
 776                 sregs->u.s.hior = to_book3s(vcpu)->hior;
 777
 778         return 0;
 779 }
 780
 781 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 782                                   struct kvm_sregs *sregs)
 783 {
 784         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 785         int i;
 786
 787         kvmppc_set_pvr(vcpu, sregs->pvr);
 788
 789         vcpu3s->sdr1 = sregs->u.s.sdr1;
 790         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 791                 for (i = 0; i < 64; i++) {
 792                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
 793                                                     sregs->u.s.ppc64.slb[i].slbe);
 794                 }
 795         } else {
 796                 for (i = 0; i < 16; i++) {
 797                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
 798                 }
 799                 for (i = 0; i < 8; i++) {
 800                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
 801                                        (u32)sregs->u.s.ppc32.ibat[i]);
 802                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
 803                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
 804                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
 805                                        (u32)sregs->u.s.ppc32.dbat[i]);
 806                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
 807                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
 808                 }
 809         }
 810
 811         /* Flush the MMU after messing with the segments */
 812         kvmppc_mmu_pte_flush(vcpu, 0, 0);
 813
 814         if (sregs->u.s.flags & KVM_SREGS_S_HIOR) {
 815                 to_book3s(vcpu)->hior_sregs = true;
 816                 to_book3s(vcpu)->hior = sregs->u.s.hior;
 817         }
 818
 819         return 0;
 820 }
 821
 822 int kvmppc_core_check_processor_compat(void)
 823 {
 824         return 0;
 825 }
 826
 827 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 828 {
 829         struct kvmppc_vcpu_book3s *vcpu_book3s;
 830         struct kvm_vcpu *vcpu;
 831         int err = -ENOMEM;
 832         unsigned long p;
 833
 834         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
 835         if (!vcpu_book3s)
 836                 goto out;
 837
 838         vcpu_book3s->shadow_vcpu = (struct kvmppc_book3s_shadow_vcpu *)
 839                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
 840         if (!vcpu_book3s->shadow_vcpu)
 841                 goto free_vcpu;
 842
 843         vcpu = &vcpu_book3s->vcpu;
 844         err = kvm_vcpu_init(vcpu, kvm, id);
 845         if (err)
 846                 goto free_shadow_vcpu;
 847
 848         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
 849         /* the real shared page fills the last 4k of our page */
 850         vcpu->arch.shared = (void*)(p + PAGE_SIZE - 4096);
 851         if (!p)
 852                 goto uninit_vcpu;
 853
 854         vcpu->arch.host_retip = kvm_return_point;
 855         vcpu->arch.host_msr = mfmsr();
 856 #ifdef CONFIG_PPC_BOOK3S_64
 857         /* default to book3s_64 (970fx) */
 858         vcpu->arch.pvr = 0x3C0301;
 859 #else
 860         /* default to book3s_32 (750) */
 861         vcpu->arch.pvr = 0x84202;
 862 #endif
 863         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
 864         vcpu->arch.slb_nr = 64;
 865
 866         /* remember where some real-mode handlers are */
 867         vcpu->arch.trampoline_lowmem = __pa(kvmppc_handler_lowmem_trampoline);
 868         vcpu->arch.trampoline_enter = __pa(kvmppc_handler_trampoline_enter);
 869         vcpu->arch.highmem_handler = (ulong)kvmppc_handler_highmem;
 870 #ifdef CONFIG_PPC_BOOK3S_64
 871         vcpu->arch.rmcall = *(ulong*)kvmppc_rmcall;
 872 #else
 873         vcpu->arch.rmcall = (ulong)kvmppc_rmcall;
 874 #endif
 875
 876         vcpu->arch.shadow_msr = MSR_USER64;
 877
 878         err = kvmppc_mmu_init(vcpu);
 879         if (err < 0)
 880                 goto uninit_vcpu;
 881
 882         return vcpu;
 883
 884 uninit_vcpu:
 885         kvm_vcpu_uninit(vcpu);
 886 free_shadow_vcpu:
 887         kfree(vcpu_book3s->shadow_vcpu);
 888 free_vcpu:
 889         vfree(vcpu_book3s);
 890 out:
 891         return ERR_PTR(err);
 892 }
 893
 894 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
 895 {
 896         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
 897
 898         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
 899         kvm_vcpu_uninit(vcpu);
 900         kfree(vcpu_book3s->shadow_vcpu);
 901         vfree(vcpu_book3s);
 902 }
 903
 904 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
 905 {
 906         int ret;
 907         double fpr[32][TS_FPRWIDTH];
 908         unsigned int fpscr;
 909         int fpexc_mode;
 910 #ifdef CONFIG_ALTIVEC
 911         vector128 vr[32];
 912         vector128 vscr;
 913         unsigned long uninitialized_var(vrsave);
 914         int used_vr;
 915 #endif
 916 #ifdef CONFIG_VSX
 917         int used_vsr;
 918 #endif
 919         ulong ext_msr;
 920
 921         /* No need to go into the guest when all we do is going out */
 922         if (signal_pending(current)) {
 923                 kvm_run->exit_reason = KVM_EXIT_INTR;
 924                 return -EINTR;
 925         }
 926
 927         /* Save FPU state in stack */
 928         if (current->thread.regs->msr & MSR_FP)
 929                 giveup_fpu(current);
 930         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
 931         fpscr = current->thread.fpscr.val;
 932         fpexc_mode = current->thread.fpexc_mode;
 933
 934 #ifdef CONFIG_ALTIVEC
 935         /* Save Altivec state in stack */
 936         used_vr = current->thread.used_vr;
 937         if (used_vr) {
 938                 if (current->thread.regs->msr & MSR_VEC)
 939                         giveup_altivec(current);
 940                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
 941                 vscr = current->thread.vscr;
 942                 vrsave = current->thread.vrsave;
 943         }
 944 #endif
 945
 946 #ifdef CONFIG_VSX
 947         /* Save VSX state in stack */
 948         used_vsr = current->thread.used_vsr;
 949         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
 950                         __giveup_vsx(current);
 951 #endif
 952
 953         /* Remember the MSR with disabled extensions */
 954         ext_msr = current->thread.regs->msr;
 955
 956         /* Preload FPU if it's enabled */
 957         if (vcpu->arch.shared->msr & MSR_FP)
 958                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 959
 960         kvm_guest_enter();
 961
 962         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
 963
 964         kvm_guest_exit();
 965
 966         local_irq_disable();
 967
 968         current->thread.regs->msr = ext_msr;
 969
 970         /* Make sure we save the guest FPU/Altivec/VSX state */
 971         kvmppc_giveup_ext(vcpu, MSR_FP);
 972         kvmppc_giveup_ext(vcpu, MSR_VEC);
 973         kvmppc_giveup_ext(vcpu, MSR_VSX);
 974
 975         /* Restore FPU state from stack */
 976         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
 977         current->thread.fpscr.val = fpscr;
 978         current->thread.fpexc_mode = fpexc_mode;
 979
 980 #ifdef CONFIG_ALTIVEC
 981         /* Restore Altivec state from stack */
 982         if (used_vr && current->thread.used_vr) {
 983                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
 984                 current->thread.vscr = vscr;
 985                 current->thread.vrsave = vrsave;
 986         }
 987         current->thread.used_vr = used_vr;
 988 #endif
 989
 990 #ifdef CONFIG_VSX
 991         current->thread.used_vsr = used_vsr;
 992 #endif
 993
 994         return ret;
 995 }
 996
 997 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
 998                                       struct kvm_userspace_memory_region *mem)
 999 {
1000         return 0;
1001 }
1002
1003 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1004                                 struct kvm_userspace_memory_region *mem)
1005 {
1006 }
1007
1008 int kvmppc_core_init_vm(struct kvm *kvm)
1009 {
1010         return 0;
1011 }
1012
1013 void kvmppc_core_destroy_vm(struct kvm *kvm)
1014 {
1015 }
1016
1017 static int kvmppc_book3s_init(void)
1018 {
1019         int r;
1020
1021         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1022                      THIS_MODULE);
1023
1024         if (r)
1025                 return r;
1026
1027         r = kvmppc_mmu_hpte_sysinit();
1028
1029         return r;
1030 }
1031
1032 static void kvmppc_book3s_exit(void)
1033 {
1034         kvmppc_mmu_hpte_sysexit();
1035         kvm_exit();
1036 }
1037
1038 module_init(kvmppc_book3s_init);
1039 module_exit(kvmppc_book3s_exit);