target-ppc/kvm.c

   1 /*
   2  * PowerPC implementation of KVM hooks
   3  *
   4  * Copyright IBM Corp. 2007
   5  * Copyright (C) 2011 Freescale Semiconductor, Inc.
   6  *
   7  * Authors:
   8  *  Jerone Young <jyoung5@us.ibm.com>
   9  *  Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
  10  *  Hollis Blanchard <hollisb@us.ibm.com>
  11  *
  12  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  13  * See the COPYING file in the top-level directory.
  14  *
  15  */
  16
  17 #include <dirent.h>
  18 #include <sys/types.h>
  19 #include <sys/ioctl.h>
  20 #include <sys/mman.h>
  21
  22 #include <linux/kvm.h>
  23
  24 #include "qemu-common.h"
  25 #include "qemu-timer.h"
  26 #include "sysemu.h"
  27 #include "kvm.h"
  28 #include "kvm_ppc.h"
  29 #include "cpu.h"
  30 #include "cpus.h"
  31 #include "device_tree.h"
  32 #include "hw/sysbus.h"
  33 #include "hw/spapr.h"
  34
  35 #include "hw/sysbus.h"
  36 #include "hw/spapr.h"
  37 #include "hw/spapr_vio.h"
  38
  39 //#define DEBUG_KVM
  40
  41 #ifdef DEBUG_KVM
  42 #define dprintf(fmt, ...) \
  43     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  44 #else
  45 #define dprintf(fmt, ...) \
  46     do { } while (0)
  47 #endif
  48
  49 #define PROC_DEVTREE_CPU      "/proc/device-tree/cpus/"
  50
  51 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
  52     KVM_CAP_LAST_INFO
  53 };
  54
  55 static int cap_interrupt_unset = false;
  56 static int cap_interrupt_level = false;
  57 static int cap_segstate;
  58 static int cap_booke_sregs;
  59 static int cap_ppc_smt;
  60 static int cap_ppc_rma;
  61 static int cap_spapr_tce;
  62
  63 /* XXX We have a race condition where we actually have a level triggered
  64  *     interrupt, but the infrastructure can't expose that yet, so the guest
  65  *     takes but ignores it, goes to sleep and never gets notified that there's
  66  *     still an interrupt pending.
  67  *
  68  *     As a quick workaround, let's just wake up again 20 ms after we injected
  69  *     an interrupt. That way we can assure that we're always reinjecting
  70  *     interrupts in case the guest swallowed them.
  71  */
  72 static QEMUTimer *idle_timer;
  73
  74 static void kvm_kick_env(void *env)
  75 {
  76     qemu_cpu_kick(env);
  77 }
  78
  79 int kvm_arch_init(KVMState *s)
  80 {
  81     cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
  82     cap_interrupt_level = kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL);
  83     cap_segstate = kvm_check_extension(s, KVM_CAP_PPC_SEGSTATE);
  84     cap_booke_sregs = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_SREGS);
  85     cap_ppc_smt = kvm_check_extension(s, KVM_CAP_PPC_SMT);
  86     cap_ppc_rma = kvm_check_extension(s, KVM_CAP_PPC_RMA);
  87     cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
  88
  89     if (!cap_interrupt_level) {
  90         fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the "
  91                         "VM to stall at times!\n");
  92     }
  93
  94     return 0;
  95 }
  96
  97 static int kvm_arch_sync_sregs(CPUPPCState *cenv)
  98 {
  99     struct kvm_sregs sregs;
 100     int ret;
 101
 102     if (cenv->excp_model == POWERPC_EXCP_BOOKE) {
 103         /* What we're really trying to say is "if we're on BookE, we use
 104            the native PVR for now". This is the only sane way to check
 105            it though, so we potentially confuse users that they can run
 106            BookE guests on BookS. Let's hope nobody dares enough :) */
 107         return 0;
 108     } else {
 109         if (!cap_segstate) {
 110             fprintf(stderr, "kvm error: missing PVR setting capability\n");
 111             return -ENOSYS;
 112         }
 113     }
 114
 115     ret = kvm_vcpu_ioctl(cenv, KVM_GET_SREGS, &sregs);
 116     if (ret) {
 117         return ret;
 118     }
 119
 120     sregs.pvr = cenv->spr[SPR_PVR];
 121     return kvm_vcpu_ioctl(cenv, KVM_SET_SREGS, &sregs);
 122 }
 123
 124 /* Set up a shared TLB array with KVM */
 125 static int kvm_booke206_tlb_init(CPUPPCState *env)
 126 {
 127     struct kvm_book3e_206_tlb_params params = {};
 128     struct kvm_config_tlb cfg = {};
 129     struct kvm_enable_cap encap = {};
 130     unsigned int entries = 0;
 131     int ret, i;
 132
 133     if (!kvm_enabled() ||
 134         !kvm_check_extension(env->kvm_state, KVM_CAP_SW_TLB)) {
 135         return 0;
 136     }
 137
 138     assert(ARRAY_SIZE(params.tlb_sizes) == BOOKE206_MAX_TLBN);
 139
 140     for (i = 0; i < BOOKE206_MAX_TLBN; i++) {
 141         params.tlb_sizes[i] = booke206_tlb_size(env, i);
 142         params.tlb_ways[i] = booke206_tlb_ways(env, i);
 143         entries += params.tlb_sizes[i];
 144     }
 145
 146     assert(entries == env->nb_tlb);
 147     assert(sizeof(struct kvm_book3e_206_tlb_entry) == sizeof(ppcmas_tlb_t));
 148
 149     env->tlb_dirty = true;
 150
 151     cfg.array = (uintptr_t)env->tlb.tlbm;
 152     cfg.array_len = sizeof(ppcmas_tlb_t) * entries;
 153     cfg.params = (uintptr_t)&params;
 154     cfg.mmu_type = KVM_MMU_FSL_BOOKE_NOHV;
 155
 156     encap.cap = KVM_CAP_SW_TLB;
 157     encap.args[0] = (uintptr_t)&cfg;
 158
 159     ret = kvm_vcpu_ioctl(env, KVM_ENABLE_CAP, &encap);
 160     if (ret < 0) {
 161         fprintf(stderr, "%s: couldn't enable KVM_CAP_SW_TLB: %s\n",
 162                 __func__, strerror(-ret));
 163         return ret;
 164     }
 165
 166     env->kvm_sw_tlb = true;
 167     return 0;
 168 }
 169
 170 int kvm_arch_init_vcpu(CPUPPCState *cenv)
 171 {
 172     int ret;
 173
 174     ret = kvm_arch_sync_sregs(cenv);
 175     if (ret) {
 176         return ret;
 177     }
 178
 179     idle_timer = qemu_new_timer_ns(vm_clock, kvm_kick_env, cenv);
 180
 181     /* Some targets support access to KVM's guest TLB. */
 182     switch (cenv->mmu_model) {
 183     case POWERPC_MMU_BOOKE206:
 184         ret = kvm_booke206_tlb_init(cenv);
 185         break;
 186     default:
 187         break;
 188     }
 189
 190     return ret;
 191 }
 192
 193 void kvm_arch_reset_vcpu(CPUPPCState *env)
 194 {
 195 }
 196
 197 static void kvm_sw_tlb_put(CPUPPCState *env)
 198 {
 199     struct kvm_dirty_tlb dirty_tlb;
 200     unsigned char *bitmap;
 201     int ret;
 202
 203     if (!env->kvm_sw_tlb) {
 204         return;
 205     }
 206
 207     bitmap = g_malloc((env->nb_tlb + 7) / 8);
 208     memset(bitmap, 0xFF, (env->nb_tlb + 7) / 8);
 209
 210     dirty_tlb.bitmap = (uintptr_t)bitmap;
 211     dirty_tlb.num_dirty = env->nb_tlb;
 212
 213     ret = kvm_vcpu_ioctl(env, KVM_DIRTY_TLB, &dirty_tlb);
 214     if (ret) {
 215         fprintf(stderr, "%s: KVM_DIRTY_TLB: %s\n",
 216                 __func__, strerror(-ret));
 217     }
 218
 219     g_free(bitmap);
 220 }
 221
 222 int kvm_arch_put_registers(CPUPPCState *env, int level)
 223 {
 224     struct kvm_regs regs;
 225     int ret;
 226     int i;
 227
 228     ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
 229     if (ret < 0)
 230         return ret;
 231
 232     regs.ctr = env->ctr;
 233     regs.lr  = env->lr;
 234     regs.xer = env->xer;
 235     regs.msr = env->msr;
 236     regs.pc = env->nip;
 237
 238     regs.srr0 = env->spr[SPR_SRR0];
 239     regs.srr1 = env->spr[SPR_SRR1];
 240
 241     regs.sprg0 = env->spr[SPR_SPRG0];
 242     regs.sprg1 = env->spr[SPR_SPRG1];
 243     regs.sprg2 = env->spr[SPR_SPRG2];
 244     regs.sprg3 = env->spr[SPR_SPRG3];
 245     regs.sprg4 = env->spr[SPR_SPRG4];
 246     regs.sprg5 = env->spr[SPR_SPRG5];
 247     regs.sprg6 = env->spr[SPR_SPRG6];
 248     regs.sprg7 = env->spr[SPR_SPRG7];
 249
 250     regs.pid = env->spr[SPR_BOOKE_PID];
 251
 252     for (i = 0;i < 32; i++)
 253         regs.gpr[i] = env->gpr[i];
 254
 255     ret = kvm_vcpu_ioctl(env, KVM_SET_REGS, &regs);
 256     if (ret < 0)
 257         return ret;
 258
 259     if (env->tlb_dirty) {
 260         kvm_sw_tlb_put(env);
 261         env->tlb_dirty = false;
 262     }
 263
 264     return ret;
 265 }
 266
 267 int kvm_arch_get_registers(CPUPPCState *env)
 268 {
 269     struct kvm_regs regs;
 270     struct kvm_sregs sregs;
 271     uint32_t cr;
 272     int i, ret;
 273
 274     ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
 275     if (ret < 0)
 276         return ret;
 277
 278     cr = regs.cr;
 279     for (i = 7; i >= 0; i--) {
 280         env->crf[i] = cr & 15;
 281         cr >>= 4;
 282     }
 283
 284     env->ctr = regs.ctr;
 285     env->lr = regs.lr;
 286     env->xer = regs.xer;
 287     env->msr = regs.msr;
 288     env->nip = regs.pc;
 289
 290     env->spr[SPR_SRR0] = regs.srr0;
 291     env->spr[SPR_SRR1] = regs.srr1;
 292
 293     env->spr[SPR_SPRG0] = regs.sprg0;
 294     env->spr[SPR_SPRG1] = regs.sprg1;
 295     env->spr[SPR_SPRG2] = regs.sprg2;
 296     env->spr[SPR_SPRG3] = regs.sprg3;
 297     env->spr[SPR_SPRG4] = regs.sprg4;
 298     env->spr[SPR_SPRG5] = regs.sprg5;
 299     env->spr[SPR_SPRG6] = regs.sprg6;
 300     env->spr[SPR_SPRG7] = regs.sprg7;
 301
 302     env->spr[SPR_BOOKE_PID] = regs.pid;
 303
 304     for (i = 0;i < 32; i++)
 305         env->gpr[i] = regs.gpr[i];
 306
 307     if (cap_booke_sregs) {
 308         ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
 309         if (ret < 0) {
 310             return ret;
 311         }
 312
 313         if (sregs.u.e.features & KVM_SREGS_E_BASE) {
 314             env->spr[SPR_BOOKE_CSRR0] = sregs.u.e.csrr0;
 315             env->spr[SPR_BOOKE_CSRR1] = sregs.u.e.csrr1;
 316             env->spr[SPR_BOOKE_ESR] = sregs.u.e.esr;
 317             env->spr[SPR_BOOKE_DEAR] = sregs.u.e.dear;
 318             env->spr[SPR_BOOKE_MCSR] = sregs.u.e.mcsr;
 319             env->spr[SPR_BOOKE_TSR] = sregs.u.e.tsr;
 320             env->spr[SPR_BOOKE_TCR] = sregs.u.e.tcr;
 321             env->spr[SPR_DECR] = sregs.u.e.dec;
 322             env->spr[SPR_TBL] = sregs.u.e.tb & 0xffffffff;
 323             env->spr[SPR_TBU] = sregs.u.e.tb >> 32;
 324             env->spr[SPR_VRSAVE] = sregs.u.e.vrsave;
 325         }
 326
 327         if (sregs.u.e.features & KVM_SREGS_E_ARCH206) {
 328             env->spr[SPR_BOOKE_PIR] = sregs.u.e.pir;
 329             env->spr[SPR_BOOKE_MCSRR0] = sregs.u.e.mcsrr0;
 330             env->spr[SPR_BOOKE_MCSRR1] = sregs.u.e.mcsrr1;
 331             env->spr[SPR_BOOKE_DECAR] = sregs.u.e.decar;
 332             env->spr[SPR_BOOKE_IVPR] = sregs.u.e.ivpr;
 333         }
 334
 335         if (sregs.u.e.features & KVM_SREGS_E_64) {
 336             env->spr[SPR_BOOKE_EPCR] = sregs.u.e.epcr;
 337         }
 338
 339         if (sregs.u.e.features & KVM_SREGS_E_SPRG8) {
 340             env->spr[SPR_BOOKE_SPRG8] = sregs.u.e.sprg8;
 341         }
 342
 343         if (sregs.u.e.features & KVM_SREGS_E_IVOR) {
 344             env->spr[SPR_BOOKE_IVOR0] = sregs.u.e.ivor_low[0];
 345             env->spr[SPR_BOOKE_IVOR1] = sregs.u.e.ivor_low[1];
 346             env->spr[SPR_BOOKE_IVOR2] = sregs.u.e.ivor_low[2];
 347             env->spr[SPR_BOOKE_IVOR3] = sregs.u.e.ivor_low[3];
 348             env->spr[SPR_BOOKE_IVOR4] = sregs.u.e.ivor_low[4];
 349             env->spr[SPR_BOOKE_IVOR5] = sregs.u.e.ivor_low[5];
 350             env->spr[SPR_BOOKE_IVOR6] = sregs.u.e.ivor_low[6];
 351             env->spr[SPR_BOOKE_IVOR7] = sregs.u.e.ivor_low[7];
 352             env->spr[SPR_BOOKE_IVOR8] = sregs.u.e.ivor_low[8];
 353             env->spr[SPR_BOOKE_IVOR9] = sregs.u.e.ivor_low[9];
 354             env->spr[SPR_BOOKE_IVOR10] = sregs.u.e.ivor_low[10];
 355             env->spr[SPR_BOOKE_IVOR11] = sregs.u.e.ivor_low[11];
 356             env->spr[SPR_BOOKE_IVOR12] = sregs.u.e.ivor_low[12];
 357             env->spr[SPR_BOOKE_IVOR13] = sregs.u.e.ivor_low[13];
 358             env->spr[SPR_BOOKE_IVOR14] = sregs.u.e.ivor_low[14];
 359             env->spr[SPR_BOOKE_IVOR15] = sregs.u.e.ivor_low[15];
 360
 361             if (sregs.u.e.features & KVM_SREGS_E_SPE) {
 362                 env->spr[SPR_BOOKE_IVOR32] = sregs.u.e.ivor_high[0];
 363                 env->spr[SPR_BOOKE_IVOR33] = sregs.u.e.ivor_high[1];
 364                 env->spr[SPR_BOOKE_IVOR34] = sregs.u.e.ivor_high[2];
 365             }
 366
 367             if (sregs.u.e.features & KVM_SREGS_E_PM) {
 368                 env->spr[SPR_BOOKE_IVOR35] = sregs.u.e.ivor_high[3];
 369             }
 370
 371             if (sregs.u.e.features & KVM_SREGS_E_PC) {
 372                 env->spr[SPR_BOOKE_IVOR36] = sregs.u.e.ivor_high[4];
 373                 env->spr[SPR_BOOKE_IVOR37] = sregs.u.e.ivor_high[5];
 374             }
 375         }
 376
 377         if (sregs.u.e.features & KVM_SREGS_E_ARCH206_MMU) {
 378             env->spr[SPR_BOOKE_MAS0] = sregs.u.e.mas0;
 379             env->spr[SPR_BOOKE_MAS1] = sregs.u.e.mas1;
 380             env->spr[SPR_BOOKE_MAS2] = sregs.u.e.mas2;
 381             env->spr[SPR_BOOKE_MAS3] = sregs.u.e.mas7_3 & 0xffffffff;
 382             env->spr[SPR_BOOKE_MAS4] = sregs.u.e.mas4;
 383             env->spr[SPR_BOOKE_MAS6] = sregs.u.e.mas6;
 384             env->spr[SPR_BOOKE_MAS7] = sregs.u.e.mas7_3 >> 32;
 385             env->spr[SPR_MMUCFG] = sregs.u.e.mmucfg;
 386             env->spr[SPR_BOOKE_TLB0CFG] = sregs.u.e.tlbcfg[0];
 387             env->spr[SPR_BOOKE_TLB1CFG] = sregs.u.e.tlbcfg[1];
 388         }
 389
 390         if (sregs.u.e.features & KVM_SREGS_EXP) {
 391             env->spr[SPR_BOOKE_EPR] = sregs.u.e.epr;
 392         }
 393
 394         if (sregs.u.e.features & KVM_SREGS_E_PD) {
 395             env->spr[SPR_BOOKE_EPLC] = sregs.u.e.eplc;
 396             env->spr[SPR_BOOKE_EPSC] = sregs.u.e.epsc;
 397         }
 398
 399         if (sregs.u.e.impl_id == KVM_SREGS_E_IMPL_FSL) {
 400             env->spr[SPR_E500_SVR] = sregs.u.e.impl.fsl.svr;
 401             env->spr[SPR_Exxx_MCAR] = sregs.u.e.impl.fsl.mcar;
 402             env->spr[SPR_HID0] = sregs.u.e.impl.fsl.hid0;
 403
 404             if (sregs.u.e.impl.fsl.features & KVM_SREGS_E_FSL_PIDn) {
 405                 env->spr[SPR_BOOKE_PID1] = sregs.u.e.impl.fsl.pid1;
 406                 env->spr[SPR_BOOKE_PID2] = sregs.u.e.impl.fsl.pid2;
 407             }
 408         }
 409     }
 410
 411     if (cap_segstate) {
 412         ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
 413         if (ret < 0) {
 414             return ret;
 415         }
 416
 417         ppc_store_sdr1(env, sregs.u.s.sdr1);
 418
 419         /* Sync SLB */
 420 #ifdef TARGET_PPC64
 421         for (i = 0; i < 64; i++) {
 422             ppc_store_slb(env, sregs.u.s.ppc64.slb[i].slbe,
 423                                sregs.u.s.ppc64.slb[i].slbv);
 424         }
 425 #endif
 426
 427         /* Sync SRs */
 428         for (i = 0; i < 16; i++) {
 429             env->sr[i] = sregs.u.s.ppc32.sr[i];
 430         }
 431
 432         /* Sync BATs */
 433         for (i = 0; i < 8; i++) {
 434             env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff;
 435             env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32;
 436             env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff;
 437             env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32;
 438         }
 439     }
 440
 441     return 0;
 442 }
 443
 444 int kvmppc_set_interrupt(CPUPPCState *env, int irq, int level)
 445 {
 446     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 447
 448     if (irq != PPC_INTERRUPT_EXT) {
 449         return 0;
 450     }
 451
 452     if (!kvm_enabled() || !cap_interrupt_unset || !cap_interrupt_level) {
 453         return 0;
 454     }
 455
 456     kvm_vcpu_ioctl(env, KVM_INTERRUPT, &virq);
 457
 458     return 0;
 459 }
 460
 461 #if defined(TARGET_PPCEMB)
 462 #define PPC_INPUT_INT PPC40x_INPUT_INT
 463 #elif defined(TARGET_PPC64)
 464 #define PPC_INPUT_INT PPC970_INPUT_INT
 465 #else
 466 #define PPC_INPUT_INT PPC6xx_INPUT_INT
 467 #endif
 468
 469 void kvm_arch_pre_run(CPUPPCState *env, struct kvm_run *run)
 470 {
 471     int r;
 472     unsigned irq;
 473
 474     /* PowerPC QEMU tracks the various core input pins (interrupt, critical
 475      * interrupt, reset, etc) in PPC-specific env->irq_input_state. */
 476     if (!cap_interrupt_level &&
 477         run->ready_for_interrupt_injection &&
 478         (env->interrupt_request & CPU_INTERRUPT_HARD) &&
 479         (env->irq_input_state & (1<<PPC_INPUT_INT)))
 480     {
 481         /* For now KVM disregards the 'irq' argument. However, in the
 482          * future KVM could cache it in-kernel to avoid a heavyweight exit
 483          * when reading the UIC.
 484          */
 485         irq = KVM_INTERRUPT_SET;
 486
 487         dprintf("injected interrupt %d\n", irq);
 488         r = kvm_vcpu_ioctl(env, KVM_INTERRUPT, &irq);
 489         if (r < 0)
 490             printf("cpu %d fail inject %x\n", env->cpu_index, irq);
 491
 492         /* Always wake up soon in case the interrupt was level based */
 493         qemu_mod_timer(idle_timer, qemu_get_clock_ns(vm_clock) +
 494                        (get_ticks_per_sec() / 50));
 495     }
 496
 497     /* We don't know if there are more interrupts pending after this. However,
 498      * the guest will return to userspace in the course of handling this one
 499      * anyways, so we will get a chance to deliver the rest. */
 500 }
 501
 502 void kvm_arch_post_run(CPUPPCState *env, struct kvm_run *run)
 503 {
 504 }
 505
 506 int kvm_arch_process_async_events(CPUPPCState *env)
 507 {
 508     return env->halted;
 509 }
 510
 511 static int kvmppc_handle_halt(CPUPPCState *env)
 512 {
 513     if (!(env->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
 514         env->halted = 1;
 515         env->exception_index = EXCP_HLT;
 516     }
 517
 518     return 0;
 519 }
 520
 521 /* map dcr access to existing qemu dcr emulation */
 522 static int kvmppc_handle_dcr_read(CPUPPCState *env, uint32_t dcrn, uint32_t *data)
 523 {
 524     if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0)
 525         fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
 526
 527     return 0;
 528 }
 529
 530 static int kvmppc_handle_dcr_write(CPUPPCState *env, uint32_t dcrn, uint32_t data)
 531 {
 532     if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0)
 533         fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
 534
 535     return 0;
 536 }
 537
 538 int kvm_arch_handle_exit(CPUPPCState *env, struct kvm_run *run)
 539 {
 540     int ret;
 541
 542     switch (run->exit_reason) {
 543     case KVM_EXIT_DCR:
 544         if (run->dcr.is_write) {
 545             dprintf("handle dcr write\n");
 546             ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data);
 547         } else {
 548             dprintf("handle dcr read\n");
 549             ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
 550         }
 551         break;
 552     case KVM_EXIT_HLT:
 553         dprintf("handle halt\n");
 554         ret = kvmppc_handle_halt(env);
 555         break;
 556 #ifdef CONFIG_PSERIES
 557     case KVM_EXIT_PAPR_HCALL:
 558         dprintf("handle PAPR hypercall\n");
 559         run->papr_hcall.ret = spapr_hypercall(env, run->papr_hcall.nr,
 560                                               run->papr_hcall.args);
 561         ret = 1;
 562         break;
 563 #endif
 564     default:
 565         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
 566         ret = -1;
 567         break;
 568     }
 569
 570     return ret;
 571 }
 572
 573 static int read_cpuinfo(const char *field, char *value, int len)
 574 {
 575     FILE *f;
 576     int ret = -1;
 577     int field_len = strlen(field);
 578     char line[512];
 579
 580     f = fopen("/proc/cpuinfo", "r");
 581     if (!f) {
 582         return -1;
 583     }
 584
 585     do {
 586         if(!fgets(line, sizeof(line), f)) {
 587             break;
 588         }
 589         if (!strncmp(line, field, field_len)) {
 590             strncpy(value, line, len);
 591             ret = 0;
 592             break;
 593         }
 594     } while(*line);
 595
 596     fclose(f);
 597
 598     return ret;
 599 }
 600
 601 uint32_t kvmppc_get_tbfreq(void)
 602 {
 603     char line[512];
 604     char *ns;
 605     uint32_t retval = get_ticks_per_sec();
 606
 607     if (read_cpuinfo("timebase", line, sizeof(line))) {
 608         return retval;
 609     }
 610
 611     if (!(ns = strchr(line, ':'))) {
 612         return retval;
 613     }
 614
 615     ns++;
 616
 617     retval = atoi(ns);
 618     return retval;
 619 }
 620
 621 /* Try to find a device tree node for a CPU with clock-frequency property */
 622 static int kvmppc_find_cpu_dt(char *buf, int buf_len)
 623 {
 624     struct dirent *dirp;
 625     DIR *dp;
 626
 627     if ((dp = opendir(PROC_DEVTREE_CPU)) == NULL) {
 628         printf("Can't open directory " PROC_DEVTREE_CPU "\n");
 629         return -1;
 630     }
 631
 632     buf[0] = '\0';
 633     while ((dirp = readdir(dp)) != NULL) {
 634         FILE *f;
 635         snprintf(buf, buf_len, "%s%s/clock-frequency", PROC_DEVTREE_CPU,
 636                  dirp->d_name);
 637         f = fopen(buf, "r");
 638         if (f) {
 639             snprintf(buf, buf_len, "%s%s", PROC_DEVTREE_CPU, dirp->d_name);
 640             fclose(f);
 641             break;
 642         }
 643         buf[0] = '\0';
 644     }
 645     closedir(dp);
 646     if (buf[0] == '\0') {
 647         printf("Unknown host!\n");
 648         return -1;
 649     }
 650
 651     return 0;
 652 }
 653
 654 /* Read a CPU node property from the host device tree that's a single
 655  * integer (32-bit or 64-bit).  Returns 0 if anything goes wrong
 656  * (can't find or open the property, or doesn't understand the
 657  * format) */
 658 static uint64_t kvmppc_read_int_cpu_dt(const char *propname)
 659 {
 660     char buf[PATH_MAX];
 661     union {
 662         uint32_t v32;
 663         uint64_t v64;
 664     } u;
 665     FILE *f;
 666     int len;
 667
 668     if (kvmppc_find_cpu_dt(buf, sizeof(buf))) {
 669         return -1;
 670     }
 671
 672     strncat(buf, "/", sizeof(buf) - strlen(buf));
 673     strncat(buf, propname, sizeof(buf) - strlen(buf));
 674
 675     f = fopen(buf, "rb");
 676     if (!f) {
 677         return -1;
 678     }
 679
 680     len = fread(&u, 1, sizeof(u), f);
 681     fclose(f);
 682     switch (len) {
 683     case 4:
 684         /* property is a 32-bit quantity */
 685         return be32_to_cpu(u.v32);
 686     case 8:
 687         return be64_to_cpu(u.v64);
 688     }
 689
 690     return 0;
 691 }
 692
 693 uint64_t kvmppc_get_clockfreq(void)
 694 {
 695     return kvmppc_read_int_cpu_dt("clock-frequency");
 696 }
 697
 698 uint32_t kvmppc_get_vmx(void)
 699 {
 700     return kvmppc_read_int_cpu_dt("ibm,vmx");
 701 }
 702
 703 uint32_t kvmppc_get_dfp(void)
 704 {
 705     return kvmppc_read_int_cpu_dt("ibm,dfp");
 706 }
 707
 708 int kvmppc_get_hypercall(CPUPPCState *env, uint8_t *buf, int buf_len)
 709 {
 710     uint32_t *hc = (uint32_t*)buf;
 711
 712     struct kvm_ppc_pvinfo pvinfo;
 713
 714     if (kvm_check_extension(env->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
 715         !kvm_vm_ioctl(env->kvm_state, KVM_PPC_GET_PVINFO, &pvinfo)) {
 716         memcpy(buf, pvinfo.hcall, buf_len);
 717
 718         return 0;
 719     }
 720
 721     /*
 722      * Fallback to always fail hypercalls:
 723      *
 724      *     li r3, -1
 725      *     nop
 726      *     nop
 727      *     nop
 728      */
 729
 730     hc[0] = 0x3860ffff;
 731     hc[1] = 0x60000000;
 732     hc[2] = 0x60000000;
 733     hc[3] = 0x60000000;
 734
 735     return 0;
 736 }
 737
 738 void kvmppc_set_papr(CPUPPCState *env)
 739 {
 740     struct kvm_enable_cap cap = {};
 741     struct kvm_one_reg reg = {};
 742     struct kvm_sregs sregs = {};
 743     int ret;
 744     uint64_t hior = env->spr[SPR_HIOR];
 745
 746     cap.cap = KVM_CAP_PPC_PAPR;
 747     ret = kvm_vcpu_ioctl(env, KVM_ENABLE_CAP, &cap);
 748
 749     if (ret) {
 750         goto fail;
 751     }
 752
 753     /*
 754      * XXX We set HIOR here. It really should be a qdev property of
 755      *     the CPU node, but we don't have CPUs converted to qdev yet.
 756      *
 757      *     Once we have qdev CPUs, move HIOR to a qdev property and
 758      *     remove this chunk.
 759      */
 760     reg.id = KVM_REG_PPC_HIOR;
 761     reg.addr = (uintptr_t)&hior;
 762     ret = kvm_vcpu_ioctl(env, KVM_SET_ONE_REG, &reg);
 763     if (ret) {
 764         fprintf(stderr, "Couldn't set HIOR. Maybe you're running an old \n"
 765                         "kernel with support for HV KVM but no PAPR PR \n"
 766                         "KVM in which case things will work. If they don't \n"
 767                         "please update your host kernel!\n");
 768     }
 769
 770     /* Set SDR1 so kernel space finds the HTAB */
 771     ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
 772     if (ret) {
 773         goto fail;
 774     }
 775
 776     sregs.u.s.sdr1 = env->spr[SPR_SDR1];
 777
 778     ret = kvm_vcpu_ioctl(env, KVM_SET_SREGS, &sregs);
 779     if (ret) {
 780         goto fail;
 781     }
 782
 783     return;
 784
 785 fail:
 786     cpu_abort(env, "This KVM version does not support PAPR\n");
 787 }
 788
 789 int kvmppc_smt_threads(void)
 790 {
 791     return cap_ppc_smt ? cap_ppc_smt : 1;
 792 }
 793
 794 off_t kvmppc_alloc_rma(const char *name, MemoryRegion *sysmem)
 795 {
 796     void *rma;
 797     off_t size;
 798     int fd;
 799     struct kvm_allocate_rma ret;
 800     MemoryRegion *rma_region;
 801
 802     /* If cap_ppc_rma == 0, contiguous RMA allocation is not supported
 803      * if cap_ppc_rma == 1, contiguous RMA allocation is supported, but
 804      *                      not necessary on this hardware
 805      * if cap_ppc_rma == 2, contiguous RMA allocation is needed on this hardware
 806      *
 807      * FIXME: We should allow the user to force contiguous RMA
 808      * allocation in the cap_ppc_rma==1 case.
 809      */
 810     if (cap_ppc_rma < 2) {
 811         return 0;
 812     }
 813
 814     fd = kvm_vm_ioctl(kvm_state, KVM_ALLOCATE_RMA, &ret);
 815     if (fd < 0) {
 816         fprintf(stderr, "KVM: Error on KVM_ALLOCATE_RMA: %s\n",
 817                 strerror(errno));
 818         return -1;
 819     }
 820
 821     size = MIN(ret.rma_size, 256ul << 20);
 822
 823     rma = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 824     if (rma == MAP_FAILED) {
 825         fprintf(stderr, "KVM: Error mapping RMA: %s\n", strerror(errno));
 826         return -1;
 827     };
 828
 829     rma_region = g_new(MemoryRegion, 1);
 830     memory_region_init_ram_ptr(rma_region, name, size, rma);
 831     vmstate_register_ram_global(rma_region);
 832     memory_region_add_subregion(sysmem, 0, rma_region);
 833
 834     return size;
 835 }
 836
 837 void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t window_size, int *pfd)
 838 {
 839     struct kvm_create_spapr_tce args = {
 840         .liobn = liobn,
 841         .window_size = window_size,
 842     };
 843     long len;
 844     int fd;
 845     void *table;
 846
 847     /* Must set fd to -1 so we don't try to munmap when called for
 848      * destroying the table, which the upper layers -will- do
 849      */
 850     *pfd = -1;
 851     if (!cap_spapr_tce) {
 852         return NULL;
 853     }
 854
 855     fd = kvm_vm_ioctl(kvm_state, KVM_CREATE_SPAPR_TCE, &args);
 856     if (fd < 0) {
 857         fprintf(stderr, "KVM: Failed to create TCE table for liobn 0x%x\n",
 858                 liobn);
 859         return NULL;
 860     }
 861
 862     len = (window_size / SPAPR_VIO_TCE_PAGE_SIZE) * sizeof(VIOsPAPR_RTCE);
 863     /* FIXME: round this up to page size */
 864
 865     table = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 866     if (table == MAP_FAILED) {
 867         fprintf(stderr, "KVM: Failed to map TCE table for liobn 0x%x\n",
 868                 liobn);
 869         close(fd);
 870         return NULL;
 871     }
 872
 873     *pfd = fd;
 874     return table;
 875 }
 876
 877 int kvmppc_remove_spapr_tce(void *table, int fd, uint32_t window_size)
 878 {
 879     long len;
 880
 881     if (fd < 0) {
 882         return -1;
 883     }
 884
 885     len = (window_size / SPAPR_VIO_TCE_PAGE_SIZE)*sizeof(VIOsPAPR_RTCE);
 886     if ((munmap(table, len) < 0) ||
 887         (close(fd) < 0)) {
 888         fprintf(stderr, "KVM: Unexpected error removing TCE table: %s",
 889                 strerror(errno));
 890         /* Leak the table */
 891     }
 892
 893     return 0;
 894 }
 895
 896 static inline uint32_t mfpvr(void)
 897 {
 898     uint32_t pvr;
 899
 900     asm ("mfpvr %0"
 901          : "=r"(pvr));
 902     return pvr;
 903 }
 904
 905 static void alter_insns(uint64_t *word, uint64_t flags, bool on)
 906 {
 907     if (on) {
 908         *word |= flags;
 909     } else {
 910         *word &= ~flags;
 911     }
 912 }
 913
 914 const ppc_def_t *kvmppc_host_cpu_def(void)
 915 {
 916     uint32_t host_pvr = mfpvr();
 917     const ppc_def_t *base_spec;
 918     ppc_def_t *spec;
 919     uint32_t vmx = kvmppc_get_vmx();
 920     uint32_t dfp = kvmppc_get_dfp();
 921
 922     base_spec = ppc_find_by_pvr(host_pvr);
 923
 924     spec = g_malloc0(sizeof(*spec));
 925     memcpy(spec, base_spec, sizeof(*spec));
 926
 927     /* Now fix up the spec with information we can query from the host */
 928
 929     if (vmx != -1) {
 930         /* Only override when we know what the host supports */
 931         alter_insns(&spec->insns_flags, PPC_ALTIVEC, vmx > 0);
 932         alter_insns(&spec->insns_flags2, PPC2_VSX, vmx > 1);
 933     }
 934     if (dfp != -1) {
 935         /* Only override when we know what the host supports */
 936         alter_insns(&spec->insns_flags2, PPC2_DFP, dfp);
 937     }
 938
 939     return spec;
 940 }
 941
 942 int kvmppc_fixup_cpu(CPUPPCState *env)
 943 {
 944     int smt;
 945
 946     /* Adjust cpu index for SMT */
 947     smt = kvmppc_smt_threads();
 948     env->cpu_index = (env->cpu_index / smp_threads) * smt
 949         + (env->cpu_index % smp_threads);
 950
 951     return 0;
 952 }
 953
 954
 955 bool kvm_arch_stop_on_emulation_error(CPUPPCState *env)
 956 {
 957     return true;
 958 }
 959
 960 int kvm_arch_on_sigbus_vcpu(CPUPPCState *env, int code, void *addr)
 961 {
 962     return 1;
 963 }
 964
 965 int kvm_arch_on_sigbus(int code, void *addr)
 966 {
 967     return 1;
 968 }