target-ppc/kvm.c

   1 /*
   2  * PowerPC implementation of KVM hooks
   3  *
   4  * Copyright IBM Corp. 2007
   5  *
   6  * Authors:
   7  *  Jerone Young <jyoung5@us.ibm.com>
   8  *  Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
   9  *  Hollis Blanchard <hollisb@us.ibm.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  12  * See the COPYING file in the top-level directory.
  13  *
  14  */
  15
  16 #include <sys/types.h>
  17 #include <sys/ioctl.h>
  18 #include <sys/mman.h>
  19
  20 #include <linux/kvm.h>
  21
  22 #include "qemu-common.h"
  23 #include "qemu-timer.h"
  24 #include "sysemu.h"
  25 #include "kvm.h"
  26 #include "kvm_ppc.h"
  27 #include "cpu.h"
  28 #include "device_tree.h"
  29
  30 //#define DEBUG_KVM
  31
  32 #ifdef DEBUG_KVM
  33 #define dprintf(fmt, ...) \
  34     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  35 #else
  36 #define dprintf(fmt, ...) \
  37     do { } while (0)
  38 #endif
  39
  40 static int cap_interrupt_unset = false;
  41 static int cap_interrupt_level = false;
  42
  43 /* XXX We have a race condition where we actually have a level triggered
  44  *     interrupt, but the infrastructure can't expose that yet, so the guest
  45  *     takes but ignores it, goes to sleep and never gets notified that there's
  46  *     still an interrupt pending.
  47  *
  48  *     As a quick workaround, let's just wake up again 20 ms after we injected
  49  *     an interrupt. That way we can assure that we're always reinjecting
  50  *     interrupts in case the guest swallowed them.
  51  */
  52 static QEMUTimer *idle_timer;
  53
  54 static void kvm_kick_env(void *env)
  55 {
  56     qemu_cpu_kick(env);
  57 }
  58
  59 int kvm_arch_init(KVMState *s, int smp_cpus)
  60 {
  61 #ifdef KVM_CAP_PPC_UNSET_IRQ
  62     cap_interrupt_unset = kvm_check_extension(s, KVM_CAP_PPC_UNSET_IRQ);
  63 #endif
  64 #ifdef KVM_CAP_PPC_IRQ_LEVEL
  65     cap_interrupt_level = kvm_check_extension(s, KVM_CAP_PPC_IRQ_LEVEL);
  66 #endif
  67
  68     if (!cap_interrupt_level) {
  69         fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the "
  70                         "VM to stall at times!\n");
  71     }
  72
  73     return 0;
  74 }
  75
  76 int kvm_arch_init_vcpu(CPUState *cenv)
  77 {
  78     int ret = 0;
  79     struct kvm_sregs sregs;
  80
  81     sregs.pvr = cenv->spr[SPR_PVR];
  82     ret = kvm_vcpu_ioctl(cenv, KVM_SET_SREGS, &sregs);
  83
  84     idle_timer = qemu_new_timer(vm_clock, kvm_kick_env, cenv);
  85
  86     return ret;
  87 }
  88
  89 void kvm_arch_reset_vcpu(CPUState *env)
  90 {
  91 }
  92
  93 int kvm_arch_put_registers(CPUState *env, int level)
  94 {
  95     struct kvm_regs regs;
  96     int ret;
  97     int i;
  98
  99     ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
 100     if (ret < 0)
 101         return ret;
 102
 103     regs.ctr = env->ctr;
 104     regs.lr  = env->lr;
 105     regs.xer = env->xer;
 106     regs.msr = env->msr;
 107     regs.pc = env->nip;
 108
 109     regs.srr0 = env->spr[SPR_SRR0];
 110     regs.srr1 = env->spr[SPR_SRR1];
 111
 112     regs.sprg0 = env->spr[SPR_SPRG0];
 113     regs.sprg1 = env->spr[SPR_SPRG1];
 114     regs.sprg2 = env->spr[SPR_SPRG2];
 115     regs.sprg3 = env->spr[SPR_SPRG3];
 116     regs.sprg4 = env->spr[SPR_SPRG4];
 117     regs.sprg5 = env->spr[SPR_SPRG5];
 118     regs.sprg6 = env->spr[SPR_SPRG6];
 119     regs.sprg7 = env->spr[SPR_SPRG7];
 120
 121     for (i = 0;i < 32; i++)
 122         regs.gpr[i] = env->gpr[i];
 123
 124     ret = kvm_vcpu_ioctl(env, KVM_SET_REGS, &regs);
 125     if (ret < 0)
 126         return ret;
 127
 128     return ret;
 129 }
 130
 131 int kvm_arch_get_registers(CPUState *env)
 132 {
 133     struct kvm_regs regs;
 134     struct kvm_sregs sregs;
 135     int i, ret;
 136
 137     ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
 138     if (ret < 0)
 139         return ret;
 140
 141     ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
 142     if (ret < 0)
 143         return ret;
 144
 145     env->ctr = regs.ctr;
 146     env->lr = regs.lr;
 147     env->xer = regs.xer;
 148     env->msr = regs.msr;
 149     env->nip = regs.pc;
 150
 151     env->spr[SPR_SRR0] = regs.srr0;
 152     env->spr[SPR_SRR1] = regs.srr1;
 153
 154     env->spr[SPR_SPRG0] = regs.sprg0;
 155     env->spr[SPR_SPRG1] = regs.sprg1;
 156     env->spr[SPR_SPRG2] = regs.sprg2;
 157     env->spr[SPR_SPRG3] = regs.sprg3;
 158     env->spr[SPR_SPRG4] = regs.sprg4;
 159     env->spr[SPR_SPRG5] = regs.sprg5;
 160     env->spr[SPR_SPRG6] = regs.sprg6;
 161     env->spr[SPR_SPRG7] = regs.sprg7;
 162
 163     for (i = 0;i < 32; i++)
 164         env->gpr[i] = regs.gpr[i];
 165
 166 #ifdef KVM_CAP_PPC_SEGSTATE
 167     if (kvm_check_extension(env->kvm_state, KVM_CAP_PPC_SEGSTATE)) {
 168         env->sdr1 = sregs.u.s.sdr1;
 169
 170         /* Sync SLB */
 171 #ifdef TARGET_PPC64
 172         for (i = 0; i < 64; i++) {
 173             ppc_store_slb(env, sregs.u.s.ppc64.slb[i].slbe,
 174                                sregs.u.s.ppc64.slb[i].slbv);
 175         }
 176 #endif
 177
 178         /* Sync SRs */
 179         for (i = 0; i < 16; i++) {
 180             env->sr[i] = sregs.u.s.ppc32.sr[i];
 181         }
 182
 183         /* Sync BATs */
 184         for (i = 0; i < 8; i++) {
 185             env->DBAT[0][i] = sregs.u.s.ppc32.dbat[i] & 0xffffffff;
 186             env->DBAT[1][i] = sregs.u.s.ppc32.dbat[i] >> 32;
 187             env->IBAT[0][i] = sregs.u.s.ppc32.ibat[i] & 0xffffffff;
 188             env->IBAT[1][i] = sregs.u.s.ppc32.ibat[i] >> 32;
 189         }
 190     }
 191 #endif
 192
 193     return 0;
 194 }
 195
 196 int kvmppc_set_interrupt(CPUState *env, int irq, int level)
 197 {
 198     unsigned virq = level ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
 199
 200     if (irq != PPC_INTERRUPT_EXT) {
 201         return 0;
 202     }
 203
 204     if (!kvm_enabled() || !cap_interrupt_unset || !cap_interrupt_level) {
 205         return 0;
 206     }
 207
 208     kvm_vcpu_ioctl(env, KVM_INTERRUPT, &virq);
 209
 210     return 0;
 211 }
 212
 213 #if defined(TARGET_PPCEMB)
 214 #define PPC_INPUT_INT PPC40x_INPUT_INT
 215 #elif defined(TARGET_PPC64)
 216 #define PPC_INPUT_INT PPC970_INPUT_INT
 217 #else
 218 #define PPC_INPUT_INT PPC6xx_INPUT_INT
 219 #endif
 220
 221 int kvm_arch_pre_run(CPUState *env, struct kvm_run *run)
 222 {
 223     int r;
 224     unsigned irq;
 225
 226     /* PowerPC Qemu tracks the various core input pins (interrupt, critical
 227      * interrupt, reset, etc) in PPC-specific env->irq_input_state. */
 228     if (!cap_interrupt_level &&
 229         run->ready_for_interrupt_injection &&
 230         (env->interrupt_request & CPU_INTERRUPT_HARD) &&
 231         (env->irq_input_state & (1<<PPC_INPUT_INT)))
 232     {
 233         /* For now KVM disregards the 'irq' argument. However, in the
 234          * future KVM could cache it in-kernel to avoid a heavyweight exit
 235          * when reading the UIC.
 236          */
 237         irq = KVM_INTERRUPT_SET;
 238
 239         dprintf("injected interrupt %d\n", irq);
 240         r = kvm_vcpu_ioctl(env, KVM_INTERRUPT, &irq);
 241         if (r < 0)
 242             printf("cpu %d fail inject %x\n", env->cpu_index, irq);
 243
 244         /* Always wake up soon in case the interrupt was level based */
 245         qemu_mod_timer(idle_timer, qemu_get_clock(vm_clock) +
 246                        (get_ticks_per_sec() / 50));
 247     }
 248
 249     /* We don't know if there are more interrupts pending after this. However,
 250      * the guest will return to userspace in the course of handling this one
 251      * anyways, so we will get a chance to deliver the rest. */
 252     return 0;
 253 }
 254
 255 int kvm_arch_post_run(CPUState *env, struct kvm_run *run)
 256 {
 257     return 0;
 258 }
 259
 260 int kvm_arch_process_irqchip_events(CPUState *env)
 261 {
 262     return 0;
 263 }
 264
 265 static int kvmppc_handle_halt(CPUState *env)
 266 {
 267     if (!(env->interrupt_request & CPU_INTERRUPT_HARD) && (msr_ee)) {
 268         env->halted = 1;
 269         env->exception_index = EXCP_HLT;
 270     }
 271
 272     return 1;
 273 }
 274
 275 /* map dcr access to existing qemu dcr emulation */
 276 static int kvmppc_handle_dcr_read(CPUState *env, uint32_t dcrn, uint32_t *data)
 277 {
 278     if (ppc_dcr_read(env->dcr_env, dcrn, data) < 0)
 279         fprintf(stderr, "Read to unhandled DCR (0x%x)\n", dcrn);
 280
 281     return 1;
 282 }
 283
 284 static int kvmppc_handle_dcr_write(CPUState *env, uint32_t dcrn, uint32_t data)
 285 {
 286     if (ppc_dcr_write(env->dcr_env, dcrn, data) < 0)
 287         fprintf(stderr, "Write to unhandled DCR (0x%x)\n", dcrn);
 288
 289     return 1;
 290 }
 291
 292 int kvm_arch_handle_exit(CPUState *env, struct kvm_run *run)
 293 {
 294     int ret = 0;
 295
 296     switch (run->exit_reason) {
 297     case KVM_EXIT_DCR:
 298         if (run->dcr.is_write) {
 299             dprintf("handle dcr write\n");
 300             ret = kvmppc_handle_dcr_write(env, run->dcr.dcrn, run->dcr.data);
 301         } else {
 302             dprintf("handle dcr read\n");
 303             ret = kvmppc_handle_dcr_read(env, run->dcr.dcrn, &run->dcr.data);
 304         }
 305         break;
 306     case KVM_EXIT_HLT:
 307         dprintf("handle halt\n");
 308         ret = kvmppc_handle_halt(env);
 309         break;
 310     }
 311
 312     return ret;
 313 }
 314
 315 static int read_cpuinfo(const char *field, char *value, int len)
 316 {
 317     FILE *f;
 318     int ret = -1;
 319     int field_len = strlen(field);
 320     char line[512];
 321
 322     f = fopen("/proc/cpuinfo", "r");
 323     if (!f) {
 324         return -1;
 325     }
 326
 327     do {
 328         if(!fgets(line, sizeof(line), f)) {
 329             break;
 330         }
 331         if (!strncmp(line, field, field_len)) {
 332             strncpy(value, line, len);
 333             ret = 0;
 334             break;
 335         }
 336     } while(*line);
 337
 338     fclose(f);
 339
 340     return ret;
 341 }
 342
 343 uint32_t kvmppc_get_tbfreq(void)
 344 {
 345     char line[512];
 346     char *ns;
 347     uint32_t retval = get_ticks_per_sec();
 348
 349     if (read_cpuinfo("timebase", line, sizeof(line))) {
 350         return retval;
 351     }
 352
 353     if (!(ns = strchr(line, ':'))) {
 354         return retval;
 355     }
 356
 357     ns++;
 358
 359     retval = atoi(ns);
 360     return retval;
 361 }
 362
 363 int kvmppc_get_hypercall(CPUState *env, uint8_t *buf, int buf_len)
 364 {
 365     uint32_t *hc = (uint32_t*)buf;
 366
 367 #ifdef KVM_CAP_PPC_GET_PVINFO
 368     struct kvm_ppc_pvinfo pvinfo;
 369
 370     if (kvm_check_extension(env->kvm_state, KVM_CAP_PPC_GET_PVINFO) &&
 371         !kvm_vm_ioctl(env->kvm_state, KVM_PPC_GET_PVINFO, &pvinfo)) {
 372         memcpy(buf, pvinfo.hcall, buf_len);
 373
 374         return 0;
 375     }
 376 #endif
 377
 378     /*
 379      * Fallback to always fail hypercalls:
 380      *
 381      *     li r3, -1
 382      *     nop
 383      *     nop
 384      *     nop
 385      */
 386
 387     hc[0] = 0x3860ffff;
 388     hc[1] = 0x60000000;
 389     hc[2] = 0x60000000;
 390     hc[3] = 0x60000000;
 391
 392     return 0;
 393 }
 394
 395 bool kvm_arch_stop_on_emulation_error(CPUState *env)
 396 {
 397     return true;
 398 }