Pileus Git - ~andy/linux/blob - 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/export.h>
  24 #include <linux/err.h>
  25 #include <linux/slab.h>
  26
  27 #include <asm/reg.h>
  28 #include <asm/cputable.h>
  29 #include <asm/cacheflush.h>
  30 #include <asm/tlbflush.h>
  31 #include <asm/uaccess.h>
  32 #include <asm/io.h>
  33 #include <asm/kvm_ppc.h>
  34 #include <asm/kvm_book3s.h>
  35 #include <asm/mmu_context.h>
  36 #include <asm/switch_to.h>
  37 #include <asm/firmware.h>
  38 #include <asm/hvcall.h>
  39 #include <linux/gfp.h>
  40 #include <linux/sched.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/highmem.h>
  43
  44 #include "trace.h"
  45
  46 /* #define EXIT_DEBUG */
  47 /* #define DEBUG_EXT */
  48
  49 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
  50                              ulong msr);
  51
  52 /* Some compatibility defines */
  53 #ifdef CONFIG_PPC_BOOK3S_32
  54 #define MSR_USER32 MSR_USER
  55 #define MSR_USER64 MSR_USER
  56 #define HW_PAGE_SIZE PAGE_SIZE
  57 #endif
  58
  59 void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
  60 {
  61 #ifdef CONFIG_PPC_BOOK3S_64
  62         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  63         memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
  64         svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
  65         svcpu_put(svcpu);
  66 #endif
  67         vcpu->cpu = smp_processor_id();
  68 #ifdef CONFIG_PPC_BOOK3S_32
  69         current->thread.kvm_shadow_vcpu = to_book3s(vcpu)->shadow_vcpu;
  70 #endif
  71 }
  72
  73 void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
  74 {
  75 #ifdef CONFIG_PPC_BOOK3S_64
  76         struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
  77         memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
  78         to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
  79         svcpu_put(svcpu);
  80 #endif
  81
  82         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
  83         vcpu->cpu = -1;
  84 }
  85
  86 /* Copy data needed by real-mode code from vcpu to shadow vcpu */
  87 void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu *svcpu,
  88                           struct kvm_vcpu *vcpu)
  89 {
  90         svcpu->gpr[0] = vcpu->arch.gpr[0];
  91         svcpu->gpr[1] = vcpu->arch.gpr[1];
  92         svcpu->gpr[2] = vcpu->arch.gpr[2];
  93         svcpu->gpr[3] = vcpu->arch.gpr[3];
  94         svcpu->gpr[4] = vcpu->arch.gpr[4];
  95         svcpu->gpr[5] = vcpu->arch.gpr[5];
  96         svcpu->gpr[6] = vcpu->arch.gpr[6];
  97         svcpu->gpr[7] = vcpu->arch.gpr[7];
  98         svcpu->gpr[8] = vcpu->arch.gpr[8];
  99         svcpu->gpr[9] = vcpu->arch.gpr[9];
 100         svcpu->gpr[10] = vcpu->arch.gpr[10];
 101         svcpu->gpr[11] = vcpu->arch.gpr[11];
 102         svcpu->gpr[12] = vcpu->arch.gpr[12];
 103         svcpu->gpr[13] = vcpu->arch.gpr[13];
 104         svcpu->cr  = vcpu->arch.cr;
 105         svcpu->xer = vcpu->arch.xer;
 106         svcpu->ctr = vcpu->arch.ctr;
 107         svcpu->lr  = vcpu->arch.lr;
 108         svcpu->pc  = vcpu->arch.pc;
 109 }
 110
 111 /* Copy data touched by real-mode code from shadow vcpu back to vcpu */
 112 void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu,
 113                             struct kvmppc_book3s_shadow_vcpu *svcpu)
 114 {
 115         vcpu->arch.gpr[0] = svcpu->gpr[0];
 116         vcpu->arch.gpr[1] = svcpu->gpr[1];
 117         vcpu->arch.gpr[2] = svcpu->gpr[2];
 118         vcpu->arch.gpr[3] = svcpu->gpr[3];
 119         vcpu->arch.gpr[4] = svcpu->gpr[4];
 120         vcpu->arch.gpr[5] = svcpu->gpr[5];
 121         vcpu->arch.gpr[6] = svcpu->gpr[6];
 122         vcpu->arch.gpr[7] = svcpu->gpr[7];
 123         vcpu->arch.gpr[8] = svcpu->gpr[8];
 124         vcpu->arch.gpr[9] = svcpu->gpr[9];
 125         vcpu->arch.gpr[10] = svcpu->gpr[10];
 126         vcpu->arch.gpr[11] = svcpu->gpr[11];
 127         vcpu->arch.gpr[12] = svcpu->gpr[12];
 128         vcpu->arch.gpr[13] = svcpu->gpr[13];
 129         vcpu->arch.cr  = svcpu->cr;
 130         vcpu->arch.xer = svcpu->xer;
 131         vcpu->arch.ctr = svcpu->ctr;
 132         vcpu->arch.lr  = svcpu->lr;
 133         vcpu->arch.pc  = svcpu->pc;
 134         vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
 135         vcpu->arch.fault_dar   = svcpu->fault_dar;
 136         vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
 137         vcpu->arch.last_inst   = svcpu->last_inst;
 138 }
 139
 140 int kvmppc_core_check_requests(struct kvm_vcpu *vcpu)
 141 {
 142         int r = 1; /* Indicate we want to get back into the guest */
 143
 144         /* We misuse TLB_FLUSH to indicate that we want to clear
 145            all shadow cache entries */
 146         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
 147                 kvmppc_mmu_pte_flush(vcpu, 0, 0);
 148
 149         return r;
 150 }
 151
 152 /************* MMU Notifiers *************/
 153
 154 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
 155 {
 156         trace_kvm_unmap_hva(hva);
 157
 158         /*
 159          * Flush all shadow tlb entries everywhere. This is slow, but
 160          * we are 100% sure that we catch the to be unmapped page
 161          */
 162         kvm_flush_remote_tlbs(kvm);
 163
 164         return 0;
 165 }
 166
 167 int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
 168 {
 169         /* kvm_unmap_hva flushes everything anyways */
 170         kvm_unmap_hva(kvm, start);
 171
 172         return 0;
 173 }
 174
 175 int kvm_age_hva(struct kvm *kvm, unsigned long hva)
 176 {
 177         /* XXX could be more clever ;) */
 178         return 0;
 179 }
 180
 181 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
 182 {
 183         /* XXX could be more clever ;) */
 184         return 0;
 185 }
 186
 187 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
 188 {
 189         /* The page will get remapped properly on its next fault */
 190         kvm_unmap_hva(kvm, hva);
 191 }
 192
 193 /*****************************************/
 194
 195 static void kvmppc_recalc_shadow_msr(struct kvm_vcpu *vcpu)
 196 {
 197         ulong smsr = vcpu->arch.shared->msr;
 198
 199         /* Guest MSR values */
 200         smsr &= MSR_FE0 | MSR_FE1 | MSR_SF | MSR_SE | MSR_BE;
 201         /* Process MSR values */
 202         smsr |= MSR_ME | MSR_RI | MSR_IR | MSR_DR | MSR_PR | MSR_EE;
 203         /* External providers the guest reserved */
 204         smsr |= (vcpu->arch.shared->msr & vcpu->arch.guest_owned_ext);
 205         /* 64-bit Process MSR values */
 206 #ifdef CONFIG_PPC_BOOK3S_64
 207         smsr |= MSR_ISF | MSR_HV;
 208 #endif
 209         vcpu->arch.shadow_msr = smsr;
 210 }
 211
 212 void kvmppc_set_msr(struct kvm_vcpu *vcpu, u64 msr)
 213 {
 214         ulong old_msr = vcpu->arch.shared->msr;
 215
 216 #ifdef EXIT_DEBUG
 217         printk(KERN_INFO "KVM: Set MSR to 0x%llx\n", msr);
 218 #endif
 219
 220         msr &= to_book3s(vcpu)->msr_mask;
 221         vcpu->arch.shared->msr = msr;
 222         kvmppc_recalc_shadow_msr(vcpu);
 223
 224         if (msr & MSR_POW) {
 225                 if (!vcpu->arch.pending_exceptions) {
 226                         kvm_vcpu_block(vcpu);
 227                         clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
 228                         vcpu->stat.halt_wakeup++;
 229
 230                         /* Unset POW bit after we woke up */
 231                         msr &= ~MSR_POW;
 232                         vcpu->arch.shared->msr = msr;
 233                 }
 234         }
 235
 236         if ((vcpu->arch.shared->msr & (MSR_PR|MSR_IR|MSR_DR)) !=
 237                    (old_msr & (MSR_PR|MSR_IR|MSR_DR))) {
 238                 kvmppc_mmu_flush_segments(vcpu);
 239                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 240
 241                 /* Preload magic page segment when in kernel mode */
 242                 if (!(msr & MSR_PR) && vcpu->arch.magic_page_pa) {
 243                         struct kvm_vcpu_arch *a = &vcpu->arch;
 244
 245                         if (msr & MSR_DR)
 246                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_ea);
 247                         else
 248                                 kvmppc_mmu_map_segment(vcpu, a->magic_page_pa);
 249                 }
 250         }
 251
 252         /*
 253          * When switching from 32 to 64-bit, we may have a stale 32-bit
 254          * magic page around, we need to flush it. Typically 32-bit magic
 255          * page will be instanciated when calling into RTAS. Note: We
 256          * assume that such transition only happens while in kernel mode,
 257          * ie, we never transition from user 32-bit to kernel 64-bit with
 258          * a 32-bit magic page around.
 259          */
 260         if (vcpu->arch.magic_page_pa &&
 261             !(old_msr & MSR_PR) && !(old_msr & MSR_SF) && (msr & MSR_SF)) {
 262                 /* going from RTAS to normal kernel code */
 263                 kvmppc_mmu_pte_flush(vcpu, (uint32_t)vcpu->arch.magic_page_pa,
 264                                      ~0xFFFUL);
 265         }
 266
 267         /* Preload FPU if it's enabled */
 268         if (vcpu->arch.shared->msr & MSR_FP)
 269                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
 270 }
 271
 272 void kvmppc_set_pvr(struct kvm_vcpu *vcpu, u32 pvr)
 273 {
 274         u32 host_pvr;
 275
 276         vcpu->arch.hflags &= ~BOOK3S_HFLAG_SLB;
 277         vcpu->arch.pvr = pvr;
 278 #ifdef CONFIG_PPC_BOOK3S_64
 279         if ((pvr >= 0x330000) && (pvr < 0x70330000)) {
 280                 kvmppc_mmu_book3s_64_init(vcpu);
 281                 if (!to_book3s(vcpu)->hior_explicit)
 282                         to_book3s(vcpu)->hior = 0xfff00000;
 283                 to_book3s(vcpu)->msr_mask = 0xffffffffffffffffULL;
 284                 vcpu->arch.cpu_type = KVM_CPU_3S_64;
 285         } else
 286 #endif
 287         {
 288                 kvmppc_mmu_book3s_32_init(vcpu);
 289                 if (!to_book3s(vcpu)->hior_explicit)
 290                         to_book3s(vcpu)->hior = 0;
 291                 to_book3s(vcpu)->msr_mask = 0xffffffffULL;
 292                 vcpu->arch.cpu_type = KVM_CPU_3S_32;
 293         }
 294
 295         kvmppc_sanity_check(vcpu);
 296
 297         /* If we are in hypervisor level on 970, we can tell the CPU to
 298          * treat DCBZ as 32 bytes store */
 299         vcpu->arch.hflags &= ~BOOK3S_HFLAG_DCBZ32;
 300         if (vcpu->arch.mmu.is_dcbz32(vcpu) && (mfmsr() & MSR_HV) &&
 301             !strcmp(cur_cpu_spec->platform, "ppc970"))
 302                 vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 303
 304         /* Cell performs badly if MSR_FEx are set. So let's hope nobody
 305            really needs them in a VM on Cell and force disable them. */
 306         if (!strcmp(cur_cpu_spec->platform, "ppc-cell-be"))
 307                 to_book3s(vcpu)->msr_mask &= ~(MSR_FE0 | MSR_FE1);
 308
 309 #ifdef CONFIG_PPC_BOOK3S_32
 310         /* 32 bit Book3S always has 32 byte dcbz */
 311         vcpu->arch.hflags |= BOOK3S_HFLAG_DCBZ32;
 312 #endif
 313
 314         /* On some CPUs we can execute paired single operations natively */
 315         asm ( "mfpvr %0" : "=r"(host_pvr));
 316         switch (host_pvr) {
 317         case 0x00080200:        /* lonestar 2.0 */
 318         case 0x00088202:        /* lonestar 2.2 */
 319         case 0x70000100:        /* gekko 1.0 */
 320         case 0x00080100:        /* gekko 2.0 */
 321         case 0x00083203:        /* gekko 2.3a */
 322         case 0x00083213:        /* gekko 2.3b */
 323         case 0x00083204:        /* gekko 2.4 */
 324         case 0x00083214:        /* gekko 2.4e (8SE) - retail HW2 */
 325         case 0x00087200:        /* broadway */
 326                 vcpu->arch.hflags |= BOOK3S_HFLAG_NATIVE_PS;
 327                 /* Enable HID2.PSE - in case we need it later */
 328                 mtspr(SPRN_HID2_GEKKO, mfspr(SPRN_HID2_GEKKO) | (1 << 29));
 329         }
 330 }
 331
 332 /* Book3s_32 CPUs always have 32 bytes cache line size, which Linux assumes. To
 333  * make Book3s_32 Linux work on Book3s_64, we have to make sure we trap dcbz to
 334  * emulate 32 bytes dcbz length.
 335  *
 336  * The Book3s_64 inventors also realized this case and implemented a special bit
 337  * in the HID5 register, which is a hypervisor ressource. Thus we can't use it.
 338  *
 339  * My approach here is to patch the dcbz instruction on executing pages.
 340  */
 341 static void kvmppc_patch_dcbz(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
 342 {
 343         struct page *hpage;
 344         u64 hpage_offset;
 345         u32 *page;
 346         int i;
 347
 348         hpage = gfn_to_page(vcpu->kvm, pte->raddr >> PAGE_SHIFT);
 349         if (is_error_page(hpage))
 350                 return;
 351
 352         hpage_offset = pte->raddr & ~PAGE_MASK;
 353         hpage_offset &= ~0xFFFULL;
 354         hpage_offset /= 4;
 355
 356         get_page(hpage);
 357         page = kmap_atomic(hpage);
 358
 359         /* patch dcbz into reserved instruction, so we trap */
 360         for (i=hpage_offset; i < hpage_offset + (HW_PAGE_SIZE / 4); i++)
 361                 if ((page[i] & 0xff0007ff) == INS_DCBZ)
 362                         page[i] &= 0xfffffff7;
 363
 364         kunmap_atomic(page);
 365         put_page(hpage);
 366 }
 367
 368 static int kvmppc_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
 369 {
 370         ulong mp_pa = vcpu->arch.magic_page_pa;
 371
 372         if (!(vcpu->arch.shared->msr & MSR_SF))
 373                 mp_pa = (uint32_t)mp_pa;
 374
 375         if (unlikely(mp_pa) &&
 376             unlikely((mp_pa & KVM_PAM) >> PAGE_SHIFT == gfn)) {
 377                 return 1;
 378         }
 379
 380         return kvm_is_visible_gfn(vcpu->kvm, gfn);
 381 }
 382
 383 int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 384                             ulong eaddr, int vec)
 385 {
 386         bool data = (vec == BOOK3S_INTERRUPT_DATA_STORAGE);
 387         int r = RESUME_GUEST;
 388         int relocated;
 389         int page_found = 0;
 390         struct kvmppc_pte pte;
 391         bool is_mmio = false;
 392         bool dr = (vcpu->arch.shared->msr & MSR_DR) ? true : false;
 393         bool ir = (vcpu->arch.shared->msr & MSR_IR) ? true : false;
 394         u64 vsid;
 395
 396         relocated = data ? dr : ir;
 397
 398         /* Resolve real address if translation turned on */
 399         if (relocated) {
 400                 page_found = vcpu->arch.mmu.xlate(vcpu, eaddr, &pte, data);
 401         } else {
 402                 pte.may_execute = true;
 403                 pte.may_read = true;
 404                 pte.may_write = true;
 405                 pte.raddr = eaddr & KVM_PAM;
 406                 pte.eaddr = eaddr;
 407                 pte.vpage = eaddr >> 12;
 408         }
 409
 410         switch (vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) {
 411         case 0:
 412                 pte.vpage |= ((u64)VSID_REAL << (SID_SHIFT - 12));
 413                 break;
 414         case MSR_DR:
 415         case MSR_IR:
 416                 vcpu->arch.mmu.esid_to_vsid(vcpu, eaddr >> SID_SHIFT, &vsid);
 417
 418                 if ((vcpu->arch.shared->msr & (MSR_DR|MSR_IR)) == MSR_DR)
 419                         pte.vpage |= ((u64)VSID_REAL_DR << (SID_SHIFT - 12));
 420                 else
 421                         pte.vpage |= ((u64)VSID_REAL_IR << (SID_SHIFT - 12));
 422                 pte.vpage |= vsid;
 423
 424                 if (vsid == -1)
 425                         page_found = -EINVAL;
 426                 break;
 427         }
 428
 429         if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 430            (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 431                 /*
 432                  * If we do the dcbz hack, we have to NX on every execution,
 433                  * so we can patch the executing code. This renders our guest
 434                  * NX-less.
 435                  */
 436                 pte.may_execute = !data;
 437         }
 438
 439         if (page_found == -ENOENT) {
 440                 /* Page not found in guest PTE entries */
 441                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 442                 vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr;
 443                 vcpu->arch.shared->msr |=
 444                         vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
 445                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 446         } else if (page_found == -EPERM) {
 447                 /* Storage protection */
 448                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 449                 vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
 450                 vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 451                 vcpu->arch.shared->msr |=
 452                         vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
 453                 kvmppc_book3s_queue_irqprio(vcpu, vec);
 454         } else if (page_found == -EINVAL) {
 455                 /* Page not found in guest SLB */
 456                 vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 457                 kvmppc_book3s_queue_irqprio(vcpu, vec + 0x80);
 458         } else if (!is_mmio &&
 459                    kvmppc_visible_gfn(vcpu, pte.raddr >> PAGE_SHIFT)) {
 460                 /* The guest's PTE is not mapped yet. Map on the host */
 461                 kvmppc_mmu_map_page(vcpu, &pte);
 462                 if (data)
 463                         vcpu->stat.sp_storage++;
 464                 else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 465                         (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32)))
 466                         kvmppc_patch_dcbz(vcpu, &pte);
 467         } else {
 468                 /* MMIO */
 469                 vcpu->stat.mmio_exits++;
 470                 vcpu->arch.paddr_accessed = pte.raddr;
 471                 vcpu->arch.vaddr_accessed = pte.eaddr;
 472                 r = kvmppc_emulate_mmio(run, vcpu);
 473                 if ( r == RESUME_HOST_NV )
 474                         r = RESUME_HOST;
 475         }
 476
 477         return r;
 478 }
 479
 480 static inline int get_fpr_index(int i)
 481 {
 482         return i * TS_FPRWIDTH;
 483 }
 484
 485 /* Give up external provider (FPU, Altivec, VSX) */
 486 void kvmppc_giveup_ext(struct kvm_vcpu *vcpu, ulong msr)
 487 {
 488         struct thread_struct *t = &current->thread;
 489         u64 *vcpu_fpr = vcpu->arch.fpr;
 490 #ifdef CONFIG_VSX
 491         u64 *vcpu_vsx = vcpu->arch.vsr;
 492 #endif
 493         u64 *thread_fpr = (u64*)t->fpr;
 494         int i;
 495
 496         /*
 497          * VSX instructions can access FP and vector registers, so if
 498          * we are giving up VSX, make sure we give up FP and VMX as well.
 499          */
 500         if (msr & MSR_VSX)
 501                 msr |= MSR_FP | MSR_VEC;
 502
 503         msr &= vcpu->arch.guest_owned_ext;
 504         if (!msr)
 505                 return;
 506
 507 #ifdef DEBUG_EXT
 508         printk(KERN_INFO "Giving up ext 0x%lx\n", msr);
 509 #endif
 510
 511         if (msr & MSR_FP) {
 512                 /*
 513                  * Note that on CPUs with VSX, giveup_fpu stores
 514                  * both the traditional FP registers and the added VSX
 515                  * registers into thread.fpr[].
 516                  */
 517                 if (current->thread.regs->msr & MSR_FP)
 518                         giveup_fpu(current);
 519                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 520                         vcpu_fpr[i] = thread_fpr[get_fpr_index(i)];
 521
 522                 vcpu->arch.fpscr = t->fpscr.val;
 523
 524 #ifdef CONFIG_VSX
 525                 if (cpu_has_feature(CPU_FTR_VSX))
 526                         for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
 527                                 vcpu_vsx[i] = thread_fpr[get_fpr_index(i) + 1];
 528 #endif
 529         }
 530
 531 #ifdef CONFIG_ALTIVEC
 532         if (msr & MSR_VEC) {
 533                 if (current->thread.regs->msr & MSR_VEC)
 534                         giveup_altivec(current);
 535                 memcpy(vcpu->arch.vr, t->vr, sizeof(vcpu->arch.vr));
 536                 vcpu->arch.vscr = t->vscr;
 537         }
 538 #endif
 539
 540         vcpu->arch.guest_owned_ext &= ~(msr | MSR_VSX);
 541         kvmppc_recalc_shadow_msr(vcpu);
 542 }
 543
 544 static int kvmppc_read_inst(struct kvm_vcpu *vcpu)
 545 {
 546         ulong srr0 = kvmppc_get_pc(vcpu);
 547         u32 last_inst = kvmppc_get_last_inst(vcpu);
 548         int ret;
 549
 550         ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
 551         if (ret == -ENOENT) {
 552                 ulong msr = vcpu->arch.shared->msr;
 553
 554                 msr = kvmppc_set_field(msr, 33, 33, 1);
 555                 msr = kvmppc_set_field(msr, 34, 36, 0);
 556                 vcpu->arch.shared->msr = kvmppc_set_field(msr, 42, 47, 0);
 557                 kvmppc_book3s_queue_irqprio(vcpu, BOOK3S_INTERRUPT_INST_STORAGE);
 558                 return EMULATE_AGAIN;
 559         }
 560
 561         return EMULATE_DONE;
 562 }
 563
 564 static int kvmppc_check_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr)
 565 {
 566
 567         /* Need to do paired single emulation? */
 568         if (!(vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE))
 569                 return EMULATE_DONE;
 570
 571         /* Read out the instruction */
 572         if (kvmppc_read_inst(vcpu) == EMULATE_DONE)
 573                 /* Need to emulate */
 574                 return EMULATE_FAIL;
 575
 576         return EMULATE_AGAIN;
 577 }
 578
 579 /* Handle external providers (FPU, Altivec, VSX) */
 580 static int kvmppc_handle_ext(struct kvm_vcpu *vcpu, unsigned int exit_nr,
 581                              ulong msr)
 582 {
 583         struct thread_struct *t = &current->thread;
 584         u64 *vcpu_fpr = vcpu->arch.fpr;
 585 #ifdef CONFIG_VSX
 586         u64 *vcpu_vsx = vcpu->arch.vsr;
 587 #endif
 588         u64 *thread_fpr = (u64*)t->fpr;
 589         int i;
 590
 591         /* When we have paired singles, we emulate in software */
 592         if (vcpu->arch.hflags & BOOK3S_HFLAG_PAIRED_SINGLE)
 593                 return RESUME_GUEST;
 594
 595         if (!(vcpu->arch.shared->msr & msr)) {
 596                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 597                 return RESUME_GUEST;
 598         }
 599
 600         if (msr == MSR_VSX) {
 601                 /* No VSX?  Give an illegal instruction interrupt */
 602 #ifdef CONFIG_VSX
 603                 if (!cpu_has_feature(CPU_FTR_VSX))
 604 #endif
 605                 {
 606                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 607                         return RESUME_GUEST;
 608                 }
 609
 610                 /*
 611                  * We have to load up all the FP and VMX registers before
 612                  * we can let the guest use VSX instructions.
 613                  */
 614                 msr = MSR_FP | MSR_VEC | MSR_VSX;
 615         }
 616
 617         /* See if we already own all the ext(s) needed */
 618         msr &= ~vcpu->arch.guest_owned_ext;
 619         if (!msr)
 620                 return RESUME_GUEST;
 621
 622 #ifdef DEBUG_EXT
 623         printk(KERN_INFO "Loading up ext 0x%lx\n", msr);
 624 #endif
 625
 626         if (msr & MSR_FP) {
 627                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.fpr); i++)
 628                         thread_fpr[get_fpr_index(i)] = vcpu_fpr[i];
 629 #ifdef CONFIG_VSX
 630                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.vsr) / 2; i++)
 631                         thread_fpr[get_fpr_index(i) + 1] = vcpu_vsx[i];
 632 #endif
 633                 t->fpscr.val = vcpu->arch.fpscr;
 634                 t->fpexc_mode = 0;
 635                 kvmppc_load_up_fpu();
 636         }
 637
 638         if (msr & MSR_VEC) {
 639 #ifdef CONFIG_ALTIVEC
 640                 memcpy(t->vr, vcpu->arch.vr, sizeof(vcpu->arch.vr));
 641                 t->vscr = vcpu->arch.vscr;
 642                 t->vrsave = -1;
 643                 kvmppc_load_up_altivec();
 644 #endif
 645         }
 646
 647         current->thread.regs->msr |= msr;
 648         vcpu->arch.guest_owned_ext |= msr;
 649         kvmppc_recalc_shadow_msr(vcpu);
 650
 651         return RESUME_GUEST;
 652 }
 653
 654 /*
 655  * Kernel code using FP or VMX could have flushed guest state to
 656  * the thread_struct; if so, get it back now.
 657  */
 658 static void kvmppc_handle_lost_ext(struct kvm_vcpu *vcpu)
 659 {
 660         unsigned long lost_ext;
 661
 662         lost_ext = vcpu->arch.guest_owned_ext & ~current->thread.regs->msr;
 663         if (!lost_ext)
 664                 return;
 665
 666         if (lost_ext & MSR_FP)
 667                 kvmppc_load_up_fpu();
 668 #ifdef CONFIG_ALTIVEC
 669         if (lost_ext & MSR_VEC)
 670                 kvmppc_load_up_altivec();
 671 #endif
 672         current->thread.regs->msr |= lost_ext;
 673 }
 674
 675 int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 676                        unsigned int exit_nr)
 677 {
 678         int r = RESUME_HOST;
 679         int s;
 680
 681         vcpu->stat.sum_exits++;
 682
 683         run->exit_reason = KVM_EXIT_UNKNOWN;
 684         run->ready_for_interrupt_injection = 1;
 685
 686         /* We get here with MSR.EE=1 */
 687
 688         trace_kvm_exit(exit_nr, vcpu);
 689         kvm_guest_exit();
 690
 691         switch (exit_nr) {
 692         case BOOK3S_INTERRUPT_INST_STORAGE:
 693         {
 694                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 695                 vcpu->stat.pf_instruc++;
 696
 697 #ifdef CONFIG_PPC_BOOK3S_32
 698                 /* We set segments as unused segments when invalidating them. So
 699                  * treat the respective fault as segment fault. */
 700                 {
 701                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 702                         u32 sr;
 703
 704                         svcpu = svcpu_get(vcpu);
 705                         sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
 706                         svcpu_put(svcpu);
 707                         if (sr == SR_INVALID) {
 708                                 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
 709                                 r = RESUME_GUEST;
 710                                 break;
 711                         }
 712                 }
 713 #endif
 714
 715                 /* only care about PTEG not found errors, but leave NX alone */
 716                 if (shadow_srr1 & 0x40000000) {
 717                         r = kvmppc_handle_pagefault(run, vcpu, kvmppc_get_pc(vcpu), exit_nr);
 718                         vcpu->stat.sp_instruc++;
 719                 } else if (vcpu->arch.mmu.is_dcbz32(vcpu) &&
 720                           (!(vcpu->arch.hflags & BOOK3S_HFLAG_DCBZ32))) {
 721                         /*
 722                          * XXX If we do the dcbz hack we use the NX bit to flush&patch the page,
 723                          *     so we can't use the NX bit inside the guest. Let's cross our fingers,
 724                          *     that no guest that needs the dcbz hack does NX.
 725                          */
 726                         kvmppc_mmu_pte_flush(vcpu, kvmppc_get_pc(vcpu), ~0xFFFUL);
 727                         r = RESUME_GUEST;
 728                 } else {
 729                         vcpu->arch.shared->msr |= shadow_srr1 & 0x58000000;
 730                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 731                         r = RESUME_GUEST;
 732                 }
 733                 break;
 734         }
 735         case BOOK3S_INTERRUPT_DATA_STORAGE:
 736         {
 737                 ulong dar = kvmppc_get_fault_dar(vcpu);
 738                 u32 fault_dsisr = vcpu->arch.fault_dsisr;
 739                 vcpu->stat.pf_storage++;
 740
 741 #ifdef CONFIG_PPC_BOOK3S_32
 742                 /* We set segments as unused segments when invalidating them. So
 743                  * treat the respective fault as segment fault. */
 744                 {
 745                         struct kvmppc_book3s_shadow_vcpu *svcpu;
 746                         u32 sr;
 747
 748                         svcpu = svcpu_get(vcpu);
 749                         sr = svcpu->sr[dar >> SID_SHIFT];
 750                         svcpu_put(svcpu);
 751                         if (sr == SR_INVALID) {
 752                                 kvmppc_mmu_map_segment(vcpu, dar);
 753                                 r = RESUME_GUEST;
 754                                 break;
 755                         }
 756                 }
 757 #endif
 758
 759                 /* The only case we need to handle is missing shadow PTEs */
 760                 if (fault_dsisr & DSISR_NOHPTE) {
 761                         r = kvmppc_handle_pagefault(run, vcpu, dar, exit_nr);
 762                 } else {
 763                         vcpu->arch.shared->dar = dar;
 764                         vcpu->arch.shared->dsisr = fault_dsisr;
 765                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 766                         r = RESUME_GUEST;
 767                 }
 768                 break;
 769         }
 770         case BOOK3S_INTERRUPT_DATA_SEGMENT:
 771                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_fault_dar(vcpu)) < 0) {
 772                         vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
 773                         kvmppc_book3s_queue_irqprio(vcpu,
 774                                 BOOK3S_INTERRUPT_DATA_SEGMENT);
 775                 }
 776                 r = RESUME_GUEST;
 777                 break;
 778         case BOOK3S_INTERRUPT_INST_SEGMENT:
 779                 if (kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu)) < 0) {
 780                         kvmppc_book3s_queue_irqprio(vcpu,
 781                                 BOOK3S_INTERRUPT_INST_SEGMENT);
 782                 }
 783                 r = RESUME_GUEST;
 784                 break;
 785         /* We're good on these - the host merely wanted to get our attention */
 786         case BOOK3S_INTERRUPT_DECREMENTER:
 787         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 788                 vcpu->stat.dec_exits++;
 789                 r = RESUME_GUEST;
 790                 break;
 791         case BOOK3S_INTERRUPT_EXTERNAL:
 792         case BOOK3S_INTERRUPT_EXTERNAL_LEVEL:
 793         case BOOK3S_INTERRUPT_EXTERNAL_HV:
 794                 vcpu->stat.ext_intr_exits++;
 795                 r = RESUME_GUEST;
 796                 break;
 797         case BOOK3S_INTERRUPT_PERFMON:
 798                 r = RESUME_GUEST;
 799                 break;
 800         case BOOK3S_INTERRUPT_PROGRAM:
 801         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 802         {
 803                 enum emulation_result er;
 804                 ulong flags;
 805
 806 program_interrupt:
 807                 flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
 808
 809                 if (vcpu->arch.shared->msr & MSR_PR) {
 810 #ifdef EXIT_DEBUG
 811                         printk(KERN_INFO "Userspace triggered 0x700 exception at 0x%lx (0x%x)\n", kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 812 #endif
 813                         if ((kvmppc_get_last_inst(vcpu) & 0xff0007ff) !=
 814                             (INS_DCBZ & 0xfffffff7)) {
 815                                 kvmppc_core_queue_program(vcpu, flags);
 816                                 r = RESUME_GUEST;
 817                                 break;
 818                         }
 819                 }
 820
 821                 vcpu->stat.emulated_inst_exits++;
 822                 er = kvmppc_emulate_instruction(run, vcpu);
 823                 switch (er) {
 824                 case EMULATE_DONE:
 825                         r = RESUME_GUEST_NV;
 826                         break;
 827                 case EMULATE_AGAIN:
 828                         r = RESUME_GUEST;
 829                         break;
 830                 case EMULATE_FAIL:
 831                         printk(KERN_CRIT "%s: emulation at %lx failed (%08x)\n",
 832                                __func__, kvmppc_get_pc(vcpu), kvmppc_get_last_inst(vcpu));
 833                         kvmppc_core_queue_program(vcpu, flags);
 834                         r = RESUME_GUEST;
 835                         break;
 836                 case EMULATE_DO_MMIO:
 837                         run->exit_reason = KVM_EXIT_MMIO;
 838                         r = RESUME_HOST_NV;
 839                         break;
 840                 case EMULATE_EXIT_USER:
 841                         r = RESUME_HOST_NV;
 842                         break;
 843                 default:
 844                         BUG();
 845                 }
 846                 break;
 847         }
 848         case BOOK3S_INTERRUPT_SYSCALL:
 849                 if (vcpu->arch.papr_enabled &&
 850                     (kvmppc_get_last_sc(vcpu) == 0x44000022) &&
 851                     !(vcpu->arch.shared->msr & MSR_PR)) {
 852                         /* SC 1 papr hypercalls */
 853                         ulong cmd = kvmppc_get_gpr(vcpu, 3);
 854                         int i;
 855
 856 #ifdef CONFIG_KVM_BOOK3S_64_PR
 857                         if (kvmppc_h_pr(vcpu, cmd) == EMULATE_DONE) {
 858                                 r = RESUME_GUEST;
 859                                 break;
 860                         }
 861 #endif
 862
 863                         run->papr_hcall.nr = cmd;
 864                         for (i = 0; i < 9; ++i) {
 865                                 ulong gpr = kvmppc_get_gpr(vcpu, 4 + i);
 866                                 run->papr_hcall.args[i] = gpr;
 867                         }
 868                         run->exit_reason = KVM_EXIT_PAPR_HCALL;
 869                         vcpu->arch.hcall_needed = 1;
 870                         r = RESUME_HOST;
 871                 } else if (vcpu->arch.osi_enabled &&
 872                     (((u32)kvmppc_get_gpr(vcpu, 3)) == OSI_SC_MAGIC_R3) &&
 873                     (((u32)kvmppc_get_gpr(vcpu, 4)) == OSI_SC_MAGIC_R4)) {
 874                         /* MOL hypercalls */
 875                         u64 *gprs = run->osi.gprs;
 876                         int i;
 877
 878                         run->exit_reason = KVM_EXIT_OSI;
 879                         for (i = 0; i < 32; i++)
 880                                 gprs[i] = kvmppc_get_gpr(vcpu, i);
 881                         vcpu->arch.osi_needed = 1;
 882                         r = RESUME_HOST_NV;
 883                 } else if (!(vcpu->arch.shared->msr & MSR_PR) &&
 884                     (((u32)kvmppc_get_gpr(vcpu, 0)) == KVM_SC_MAGIC_R0)) {
 885                         /* KVM PV hypercalls */
 886                         kvmppc_set_gpr(vcpu, 3, kvmppc_kvm_pv(vcpu));
 887                         r = RESUME_GUEST;
 888                 } else {
 889                         /* Guest syscalls */
 890                         vcpu->stat.syscall_exits++;
 891                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 892                         r = RESUME_GUEST;
 893                 }
 894                 break;
 895         case BOOK3S_INTERRUPT_FP_UNAVAIL:
 896         case BOOK3S_INTERRUPT_ALTIVEC:
 897         case BOOK3S_INTERRUPT_VSX:
 898         {
 899                 int ext_msr = 0;
 900
 901                 switch (exit_nr) {
 902                 case BOOK3S_INTERRUPT_FP_UNAVAIL: ext_msr = MSR_FP;  break;
 903                 case BOOK3S_INTERRUPT_ALTIVEC:    ext_msr = MSR_VEC; break;
 904                 case BOOK3S_INTERRUPT_VSX:        ext_msr = MSR_VSX; break;
 905                 }
 906
 907                 switch (kvmppc_check_ext(vcpu, exit_nr)) {
 908                 case EMULATE_DONE:
 909                         /* everything ok - let's enable the ext */
 910                         r = kvmppc_handle_ext(vcpu, exit_nr, ext_msr);
 911                         break;
 912                 case EMULATE_FAIL:
 913                         /* we need to emulate this instruction */
 914                         goto program_interrupt;
 915                         break;
 916                 default:
 917                         /* nothing to worry about - go again */
 918                         break;
 919                 }
 920                 break;
 921         }
 922         case BOOK3S_INTERRUPT_ALIGNMENT:
 923                 if (kvmppc_read_inst(vcpu) == EMULATE_DONE) {
 924                         vcpu->arch.shared->dsisr = kvmppc_alignment_dsisr(vcpu,
 925                                 kvmppc_get_last_inst(vcpu));
 926                         vcpu->arch.shared->dar = kvmppc_alignment_dar(vcpu,
 927                                 kvmppc_get_last_inst(vcpu));
 928                         kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 929                 }
 930                 r = RESUME_GUEST;
 931                 break;
 932         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 933         case BOOK3S_INTERRUPT_TRACE:
 934                 kvmppc_book3s_queue_irqprio(vcpu, exit_nr);
 935                 r = RESUME_GUEST;
 936                 break;
 937         default:
 938         {
 939                 ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 940                 /* Ugh - bork here! What did we get? */
 941                 printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
 942                         exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
 943                 r = RESUME_HOST;
 944                 BUG();
 945                 break;
 946         }
 947         }
 948
 949         if (!(r & RESUME_HOST)) {
 950                 /* To avoid clobbering exit_reason, only check for signals if
 951                  * we aren't already exiting to userspace for some other
 952                  * reason. */
 953
 954                 /*
 955                  * Interrupts could be timers for the guest which we have to
 956                  * inject again, so let's postpone them until we're in the guest
 957                  * and if we really did time things so badly, then we just exit
 958                  * again due to a host external interrupt.
 959                  */
 960                 local_irq_disable();
 961                 s = kvmppc_prepare_to_enter(vcpu);
 962                 if (s <= 0) {
 963                         local_irq_enable();
 964                         r = s;
 965                 } else {
 966                         kvmppc_fix_ee_before_entry();
 967                 }
 968                 kvmppc_handle_lost_ext(vcpu);
 969         }
 970
 971         trace_kvm_book3s_reenter(r, vcpu);
 972
 973         return r;
 974 }
 975
 976 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
 977                                   struct kvm_sregs *sregs)
 978 {
 979         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
 980         int i;
 981
 982         sregs->pvr = vcpu->arch.pvr;
 983
 984         sregs->u.s.sdr1 = to_book3s(vcpu)->sdr1;
 985         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
 986                 for (i = 0; i < 64; i++) {
 987                         sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige | i;
 988                         sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
 989                 }
 990         } else {
 991                 for (i = 0; i < 16; i++)
 992                         sregs->u.s.ppc32.sr[i] = vcpu->arch.shared->sr[i];
 993
 994                 for (i = 0; i < 8; i++) {
 995                         sregs->u.s.ppc32.ibat[i] = vcpu3s->ibat[i].raw;
 996                         sregs->u.s.ppc32.dbat[i] = vcpu3s->dbat[i].raw;
 997                 }
 998         }
 999
1000         return 0;
1001 }
1002
1003 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1004                                   struct kvm_sregs *sregs)
1005 {
1006         struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
1007         int i;
1008
1009         kvmppc_set_pvr(vcpu, sregs->pvr);
1010
1011         vcpu3s->sdr1 = sregs->u.s.sdr1;
1012         if (vcpu->arch.hflags & BOOK3S_HFLAG_SLB) {
1013                 for (i = 0; i < 64; i++) {
1014                         vcpu->arch.mmu.slbmte(vcpu, sregs->u.s.ppc64.slb[i].slbv,
1015                                                     sregs->u.s.ppc64.slb[i].slbe);
1016                 }
1017         } else {
1018                 for (i = 0; i < 16; i++) {
1019                         vcpu->arch.mmu.mtsrin(vcpu, i, sregs->u.s.ppc32.sr[i]);
1020                 }
1021                 for (i = 0; i < 8; i++) {
1022                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), false,
1023                                        (u32)sregs->u.s.ppc32.ibat[i]);
1024                         kvmppc_set_bat(vcpu, &(vcpu3s->ibat[i]), true,
1025                                        (u32)(sregs->u.s.ppc32.ibat[i] >> 32));
1026                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), false,
1027                                        (u32)sregs->u.s.ppc32.dbat[i]);
1028                         kvmppc_set_bat(vcpu, &(vcpu3s->dbat[i]), true,
1029                                        (u32)(sregs->u.s.ppc32.dbat[i] >> 32));
1030                 }
1031         }
1032
1033         /* Flush the MMU after messing with the segments */
1034         kvmppc_mmu_pte_flush(vcpu, 0, 0);
1035
1036         return 0;
1037 }
1038
1039 int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
1040 {
1041         int r = 0;
1042
1043         switch (id) {
1044         case KVM_REG_PPC_HIOR:
1045                 *val = get_reg_val(id, to_book3s(vcpu)->hior);
1046                 break;
1047 #ifdef CONFIG_VSX
1048         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1049                 long int i = id - KVM_REG_PPC_VSR0;
1050
1051                 if (!cpu_has_feature(CPU_FTR_VSX)) {
1052                         r = -ENXIO;
1053                         break;
1054                 }
1055                 val->vsxval[0] = vcpu->arch.fpr[i];
1056                 val->vsxval[1] = vcpu->arch.vsr[i];
1057                 break;
1058         }
1059 #endif /* CONFIG_VSX */
1060         default:
1061                 r = -EINVAL;
1062                 break;
1063         }
1064
1065         return r;
1066 }
1067
1068 int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id, union kvmppc_one_reg *val)
1069 {
1070         int r = 0;
1071
1072         switch (id) {
1073         case KVM_REG_PPC_HIOR:
1074                 to_book3s(vcpu)->hior = set_reg_val(id, *val);
1075                 to_book3s(vcpu)->hior_explicit = true;
1076                 break;
1077 #ifdef CONFIG_VSX
1078         case KVM_REG_PPC_VSR0 ... KVM_REG_PPC_VSR31: {
1079                 long int i = id - KVM_REG_PPC_VSR0;
1080
1081                 if (!cpu_has_feature(CPU_FTR_VSX)) {
1082                         r = -ENXIO;
1083                         break;
1084                 }
1085                 vcpu->arch.fpr[i] = val->vsxval[0];
1086                 vcpu->arch.vsr[i] = val->vsxval[1];
1087                 break;
1088         }
1089 #endif /* CONFIG_VSX */
1090         default:
1091                 r = -EINVAL;
1092                 break;
1093         }
1094
1095         return r;
1096 }
1097
1098 int kvmppc_core_check_processor_compat(void)
1099 {
1100         return 0;
1101 }
1102
1103 struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
1104 {
1105         struct kvmppc_vcpu_book3s *vcpu_book3s;
1106         struct kvm_vcpu *vcpu;
1107         int err = -ENOMEM;
1108         unsigned long p;
1109
1110         vcpu_book3s = vzalloc(sizeof(struct kvmppc_vcpu_book3s));
1111         if (!vcpu_book3s)
1112                 goto out;
1113
1114 #ifdef CONFIG_KVM_BOOK3S_32
1115         vcpu_book3s->shadow_vcpu =
1116                 kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
1117         if (!vcpu_book3s->shadow_vcpu)
1118                 goto free_vcpu;
1119 #endif
1120         vcpu = &vcpu_book3s->vcpu;
1121         err = kvm_vcpu_init(vcpu, kvm, id);
1122         if (err)
1123                 goto free_shadow_vcpu;
1124
1125         err = -ENOMEM;
1126         p = __get_free_page(GFP_KERNEL|__GFP_ZERO);
1127         if (!p)
1128                 goto uninit_vcpu;
1129         /* the real shared page fills the last 4k of our page */
1130         vcpu->arch.shared = (void *)(p + PAGE_SIZE - 4096);
1131
1132 #ifdef CONFIG_PPC_BOOK3S_64
1133         /* default to book3s_64 (970fx) */
1134         vcpu->arch.pvr = 0x3C0301;
1135 #else
1136         /* default to book3s_32 (750) */
1137         vcpu->arch.pvr = 0x84202;
1138 #endif
1139         kvmppc_set_pvr(vcpu, vcpu->arch.pvr);
1140         vcpu->arch.slb_nr = 64;
1141
1142         vcpu->arch.shadow_msr = MSR_USER64;
1143
1144         err = kvmppc_mmu_init(vcpu);
1145         if (err < 0)
1146                 goto uninit_vcpu;
1147
1148         return vcpu;
1149
1150 uninit_vcpu:
1151         kvm_vcpu_uninit(vcpu);
1152 free_shadow_vcpu:
1153 #ifdef CONFIG_KVM_BOOK3S_32
1154         kfree(vcpu_book3s->shadow_vcpu);
1155 free_vcpu:
1156 #endif
1157         vfree(vcpu_book3s);
1158 out:
1159         return ERR_PTR(err);
1160 }
1161
1162 void kvmppc_core_vcpu_free(struct kvm_vcpu *vcpu)
1163 {
1164         struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
1165
1166         free_page((unsigned long)vcpu->arch.shared & PAGE_MASK);
1167         kvm_vcpu_uninit(vcpu);
1168         kfree(vcpu_book3s->shadow_vcpu);
1169         vfree(vcpu_book3s);
1170 }
1171
1172 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1173 {
1174         int ret;
1175         double fpr[32][TS_FPRWIDTH];
1176         unsigned int fpscr;
1177         int fpexc_mode;
1178 #ifdef CONFIG_ALTIVEC
1179         vector128 vr[32];
1180         vector128 vscr;
1181         unsigned long uninitialized_var(vrsave);
1182         int used_vr;
1183 #endif
1184 #ifdef CONFIG_VSX
1185         int used_vsr;
1186 #endif
1187         ulong ext_msr;
1188
1189         /* Check if we can run the vcpu at all */
1190         if (!vcpu->arch.sane) {
1191                 kvm_run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1192                 ret = -EINVAL;
1193                 goto out;
1194         }
1195
1196         /*
1197          * Interrupts could be timers for the guest which we have to inject
1198          * again, so let's postpone them until we're in the guest and if we
1199          * really did time things so badly, then we just exit again due to
1200          * a host external interrupt.
1201          */
1202         local_irq_disable();
1203         ret = kvmppc_prepare_to_enter(vcpu);
1204         if (ret <= 0) {
1205                 local_irq_enable();
1206                 goto out;
1207         }
1208
1209         /* Save FPU state in stack */
1210         if (current->thread.regs->msr & MSR_FP)
1211                 giveup_fpu(current);
1212         memcpy(fpr, current->thread.fpr, sizeof(current->thread.fpr));
1213         fpscr = current->thread.fpscr.val;
1214         fpexc_mode = current->thread.fpexc_mode;
1215
1216 #ifdef CONFIG_ALTIVEC
1217         /* Save Altivec state in stack */
1218         used_vr = current->thread.used_vr;
1219         if (used_vr) {
1220                 if (current->thread.regs->msr & MSR_VEC)
1221                         giveup_altivec(current);
1222                 memcpy(vr, current->thread.vr, sizeof(current->thread.vr));
1223                 vscr = current->thread.vscr;
1224                 vrsave = current->thread.vrsave;
1225         }
1226 #endif
1227
1228 #ifdef CONFIG_VSX
1229         /* Save VSX state in stack */
1230         used_vsr = current->thread.used_vsr;
1231         if (used_vsr && (current->thread.regs->msr & MSR_VSX))
1232                 __giveup_vsx(current);
1233 #endif
1234
1235         /* Remember the MSR with disabled extensions */
1236         ext_msr = current->thread.regs->msr;
1237
1238         /* Preload FPU if it's enabled */
1239         if (vcpu->arch.shared->msr & MSR_FP)
1240                 kvmppc_handle_ext(vcpu, BOOK3S_INTERRUPT_FP_UNAVAIL, MSR_FP);
1241
1242         kvmppc_fix_ee_before_entry();
1243
1244         ret = __kvmppc_vcpu_run(kvm_run, vcpu);
1245
1246         /* No need for kvm_guest_exit. It's done in handle_exit.
1247            We also get here with interrupts enabled. */
1248
1249         /* Make sure we save the guest FPU/Altivec/VSX state */
1250         kvmppc_giveup_ext(vcpu, MSR_FP | MSR_VEC | MSR_VSX);
1251
1252         current->thread.regs->msr = ext_msr;
1253
1254         /* Restore FPU/VSX state from stack */
1255         memcpy(current->thread.fpr, fpr, sizeof(current->thread.fpr));
1256         current->thread.fpscr.val = fpscr;
1257         current->thread.fpexc_mode = fpexc_mode;
1258
1259 #ifdef CONFIG_ALTIVEC
1260         /* Restore Altivec state from stack */
1261         if (used_vr && current->thread.used_vr) {
1262                 memcpy(current->thread.vr, vr, sizeof(current->thread.vr));
1263                 current->thread.vscr = vscr;
1264                 current->thread.vrsave = vrsave;
1265         }
1266         current->thread.used_vr = used_vr;
1267 #endif
1268
1269 #ifdef CONFIG_VSX
1270         current->thread.used_vsr = used_vsr;
1271 #endif
1272
1273 out:
1274         vcpu->mode = OUTSIDE_GUEST_MODE;
1275         return ret;
1276 }
1277
1278 /*
1279  * Get (and clear) the dirty memory log for a memory slot.
1280  */
1281 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1282                                       struct kvm_dirty_log *log)
1283 {
1284         struct kvm_memory_slot *memslot;
1285         struct kvm_vcpu *vcpu;
1286         ulong ga, ga_end;
1287         int is_dirty = 0;
1288         int r;
1289         unsigned long n;
1290
1291         mutex_lock(&kvm->slots_lock);
1292
1293         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1294         if (r)
1295                 goto out;
1296
1297         /* If nothing is dirty, don't bother messing with page tables. */
1298         if (is_dirty) {
1299                 memslot = id_to_memslot(kvm->memslots, log->slot);
1300
1301                 ga = memslot->base_gfn << PAGE_SHIFT;
1302                 ga_end = ga + (memslot->npages << PAGE_SHIFT);
1303
1304                 kvm_for_each_vcpu(n, vcpu, kvm)
1305                         kvmppc_mmu_pte_pflush(vcpu, ga, ga_end);
1306
1307                 n = kvm_dirty_bitmap_bytes(memslot);
1308                 memset(memslot->dirty_bitmap, 0, n);
1309         }
1310
1311         r = 0;
1312 out:
1313         mutex_unlock(&kvm->slots_lock);
1314         return r;
1315 }
1316
1317 #ifdef CONFIG_PPC64
1318 int kvm_vm_ioctl_get_smmu_info(struct kvm *kvm, struct kvm_ppc_smmu_info *info)
1319 {
1320         info->flags = KVM_PPC_1T_SEGMENTS;
1321
1322         /* SLB is always 64 entries */
1323         info->slb_size = 64;
1324
1325         /* Standard 4k base page size segment */
1326         info->sps[0].page_shift = 12;
1327         info->sps[0].slb_enc = 0;
1328         info->sps[0].enc[0].page_shift = 12;
1329         info->sps[0].enc[0].pte_enc = 0;
1330
1331         /* Standard 16M large page size segment */
1332         info->sps[1].page_shift = 24;
1333         info->sps[1].slb_enc = SLB_VSID_L;
1334         info->sps[1].enc[0].page_shift = 24;
1335         info->sps[1].enc[0].pte_enc = 0;
1336
1337         return 0;
1338 }
1339 #endif /* CONFIG_PPC64 */
1340
1341 void kvmppc_core_free_memslot(struct kvm_memory_slot *free,
1342                               struct kvm_memory_slot *dont)
1343 {
1344 }
1345
1346 int kvmppc_core_create_memslot(struct kvm_memory_slot *slot,
1347                                unsigned long npages)
1348 {
1349         return 0;
1350 }
1351
1352 int kvmppc_core_prepare_memory_region(struct kvm *kvm,
1353                                       struct kvm_memory_slot *memslot,
1354                                       struct kvm_userspace_memory_region *mem)
1355 {
1356         return 0;
1357 }
1358
1359 void kvmppc_core_commit_memory_region(struct kvm *kvm,
1360                                 struct kvm_userspace_memory_region *mem,
1361                                 const struct kvm_memory_slot *old)
1362 {
1363 }
1364
1365 void kvmppc_core_flush_memslot(struct kvm *kvm, struct kvm_memory_slot *memslot)
1366 {
1367 }
1368
1369 static unsigned int kvm_global_user_count = 0;
1370 static DEFINE_SPINLOCK(kvm_global_user_count_lock);
1371
1372 int kvmppc_core_init_vm(struct kvm *kvm)
1373 {
1374 #ifdef CONFIG_PPC64
1375         INIT_LIST_HEAD(&kvm->arch.spapr_tce_tables);
1376         INIT_LIST_HEAD(&kvm->arch.rtas_tokens);
1377 #endif
1378
1379         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1380                 spin_lock(&kvm_global_user_count_lock);
1381                 if (++kvm_global_user_count == 1)
1382                         pSeries_disable_reloc_on_exc();
1383                 spin_unlock(&kvm_global_user_count_lock);
1384         }
1385         return 0;
1386 }
1387
1388 void kvmppc_core_destroy_vm(struct kvm *kvm)
1389 {
1390 #ifdef CONFIG_PPC64
1391         WARN_ON(!list_empty(&kvm->arch.spapr_tce_tables));
1392 #endif
1393
1394         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
1395                 spin_lock(&kvm_global_user_count_lock);
1396                 BUG_ON(kvm_global_user_count == 0);
1397                 if (--kvm_global_user_count == 0)
1398                         pSeries_enable_reloc_on_exc();
1399                 spin_unlock(&kvm_global_user_count_lock);
1400         }
1401 }
1402
1403 static int kvmppc_book3s_init(void)
1404 {
1405         int r;
1406
1407         r = kvm_init(NULL, sizeof(struct kvmppc_vcpu_book3s), 0,
1408                      THIS_MODULE);
1409
1410         if (r)
1411                 return r;
1412
1413         r = kvmppc_mmu_hpte_sysinit();
1414
1415         return r;
1416 }
1417
1418 static void kvmppc_book3s_exit(void)
1419 {
1420         kvmppc_mmu_hpte_sysexit();
1421         kvm_exit();
1422 }
1423
1424 module_init(kvmppc_book3s_init);
1425 module_exit(kvmppc_book3s_exit);