target-s390x/kvm.c

   1 /*
   2  * QEMU S390x KVM implementation
   3  *
   4  * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
   5  * Copyright IBM Corp. 2012
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * Contributions after 2012-10-29 are licensed under the terms of the
  18  * GNU GPL, version 2 or (at your option) any later version.
  19  *
  20  * You should have received a copy of the GNU (Lesser) General Public
  21  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  22  */
  23
  24 #include <sys/types.h>
  25 #include <sys/ioctl.h>
  26 #include <sys/mman.h>
  27
  28 #include <linux/kvm.h>
  29 #include <asm/ptrace.h>
  30
  31 #include "qemu-common.h"
  32 #include "qemu/timer.h"
  33 #include "sysemu/sysemu.h"
  34 #include "sysemu/kvm.h"
  35 #include "cpu.h"
  36 #include "sysemu/device_tree.h"
  37 #include "qapi/qmp/qjson.h"
  38 #include "monitor/monitor.h"
  39 #include "trace.h"
  40
  41 /* #define DEBUG_KVM */
  42
  43 #ifdef DEBUG_KVM
  44 #define DPRINTF(fmt, ...) \
  45     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  46 #else
  47 #define DPRINTF(fmt, ...) \
  48     do { } while (0)
  49 #endif
  50
  51 #define IPA0_DIAG                       0x8300
  52 #define IPA0_SIGP                       0xae00
  53 #define IPA0_B2                         0xb200
  54 #define IPA0_B9                         0xb900
  55 #define IPA0_EB                         0xeb00
  56
  57 #define PRIV_B2_SCLP_CALL               0x20
  58 #define PRIV_B2_CSCH                    0x30
  59 #define PRIV_B2_HSCH                    0x31
  60 #define PRIV_B2_MSCH                    0x32
  61 #define PRIV_B2_SSCH                    0x33
  62 #define PRIV_B2_STSCH                   0x34
  63 #define PRIV_B2_TSCH                    0x35
  64 #define PRIV_B2_TPI                     0x36
  65 #define PRIV_B2_SAL                     0x37
  66 #define PRIV_B2_RSCH                    0x38
  67 #define PRIV_B2_STCRW                   0x39
  68 #define PRIV_B2_STCPS                   0x3a
  69 #define PRIV_B2_RCHP                    0x3b
  70 #define PRIV_B2_SCHM                    0x3c
  71 #define PRIV_B2_CHSC                    0x5f
  72 #define PRIV_B2_SIGA                    0x74
  73 #define PRIV_B2_XSCH                    0x76
  74
  75 #define PRIV_EB_SQBS                    0x8a
  76
  77 #define PRIV_B9_EQBS                    0x9c
  78
  79 #define DIAG_IPL                        0x308
  80 #define DIAG_KVM_HYPERCALL              0x500
  81 #define DIAG_KVM_BREAKPOINT             0x501
  82
  83 #define ICPT_INSTRUCTION                0x04
  84 #define ICPT_WAITPSW                    0x1c
  85 #define ICPT_SOFT_INTERCEPT             0x24
  86 #define ICPT_CPU_STOP                   0x28
  87 #define ICPT_IO                         0x40
  88
  89 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
  90     KVM_CAP_LAST_INFO
  91 };
  92
  93 static int cap_sync_regs;
  94 static int cap_async_pf;
  95
  96 static void *legacy_s390_alloc(size_t size);
  97
  98 int kvm_arch_init(KVMState *s)
  99 {
 100     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
 101     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
 102     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
 103         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
 104         phys_mem_set_alloc(legacy_s390_alloc);
 105     }
 106     return 0;
 107 }
 108
 109 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
 110 {
 111     return cpu->cpu_index;
 112 }
 113
 114 int kvm_arch_init_vcpu(CPUState *cpu)
 115 {
 116     /* nothing todo yet */
 117     return 0;
 118 }
 119
 120 void kvm_s390_reset_vcpu(S390CPU *cpu)
 121 {
 122     CPUState *cs = CPU(cpu);
 123
 124     /* The initial reset call is needed here to reset in-kernel
 125      * vcpu data that we can't access directly from QEMU
 126      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
 127      * Before this ioctl cpu_synchronize_state() is called in common kvm
 128      * code (kvm-all) */
 129     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
 130         perror("Can't reset vcpu\n");
 131     }
 132 }
 133
 134 int kvm_arch_put_registers(CPUState *cs, int level)
 135 {
 136     S390CPU *cpu = S390_CPU(cs);
 137     CPUS390XState *env = &cpu->env;
 138     struct kvm_sregs sregs;
 139     struct kvm_regs regs;
 140     int r;
 141     int i;
 142
 143     /* always save the PSW  and the GPRS*/
 144     cs->kvm_run->psw_addr = env->psw.addr;
 145     cs->kvm_run->psw_mask = env->psw.mask;
 146
 147     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_GPRS) {
 148         for (i = 0; i < 16; i++) {
 149             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
 150             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
 151         }
 152     } else {
 153         for (i = 0; i < 16; i++) {
 154             regs.gprs[i] = env->regs[i];
 155         }
 156         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
 157         if (r < 0) {
 158             return r;
 159         }
 160     }
 161
 162     /* Do we need to save more than that? */
 163     if (level == KVM_PUT_RUNTIME_STATE) {
 164         return 0;
 165     }
 166
 167     /*
 168      * These ONE_REGS are not protected by a capability. As they are only
 169      * necessary for migration we just trace a possible error, but don't
 170      * return with an error return code.
 171      */
 172     kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 173     kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 174     kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 175     kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 176     kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 177
 178     if (cap_async_pf) {
 179         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 180         if (r < 0) {
 181             return r;
 182         }
 183         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 184         if (r < 0) {
 185             return r;
 186         }
 187         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 188         if (r < 0) {
 189             return r;
 190         }
 191     }
 192
 193     if (cap_sync_regs &&
 194         cs->kvm_run->kvm_valid_regs & KVM_SYNC_ACRS &&
 195         cs->kvm_run->kvm_valid_regs & KVM_SYNC_CRS) {
 196         for (i = 0; i < 16; i++) {
 197             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
 198             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
 199         }
 200         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
 201         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
 202     } else {
 203         for (i = 0; i < 16; i++) {
 204             sregs.acrs[i] = env->aregs[i];
 205             sregs.crs[i] = env->cregs[i];
 206         }
 207         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
 208         if (r < 0) {
 209             return r;
 210         }
 211     }
 212
 213     /* Finally the prefix */
 214     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_PREFIX) {
 215         cs->kvm_run->s.regs.prefix = env->psa;
 216         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
 217     } else {
 218         /* prefix is only supported via sync regs */
 219     }
 220     return 0;
 221 }
 222
 223 int kvm_arch_get_registers(CPUState *cs)
 224 {
 225     S390CPU *cpu = S390_CPU(cs);
 226     CPUS390XState *env = &cpu->env;
 227     struct kvm_sregs sregs;
 228     struct kvm_regs regs;
 229     int i, r;
 230
 231     /* get the PSW */
 232     env->psw.addr = cs->kvm_run->psw_addr;
 233     env->psw.mask = cs->kvm_run->psw_mask;
 234
 235     /* the GPRS */
 236     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_GPRS) {
 237         for (i = 0; i < 16; i++) {
 238             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
 239         }
 240     } else {
 241         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
 242         if (r < 0) {
 243             return r;
 244         }
 245          for (i = 0; i < 16; i++) {
 246             env->regs[i] = regs.gprs[i];
 247         }
 248     }
 249
 250     /* The ACRS and CRS */
 251     if (cap_sync_regs &&
 252         cs->kvm_run->kvm_valid_regs & KVM_SYNC_ACRS &&
 253         cs->kvm_run->kvm_valid_regs & KVM_SYNC_CRS) {
 254         for (i = 0; i < 16; i++) {
 255             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
 256             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
 257         }
 258     } else {
 259         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
 260         if (r < 0) {
 261             return r;
 262         }
 263          for (i = 0; i < 16; i++) {
 264             env->aregs[i] = sregs.acrs[i];
 265             env->cregs[i] = sregs.crs[i];
 266         }
 267     }
 268
 269     /* The prefix */
 270     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_PREFIX) {
 271         env->psa = cs->kvm_run->s.regs.prefix;
 272     }
 273
 274     /*
 275      * These ONE_REGS are not protected by a capability. As they are only
 276      * necessary for migration we just trace a possible error, but don't
 277      * return with an error return code.
 278      */
 279     kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 280     kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 281     kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 282     kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 283     kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 284
 285     if (cap_async_pf) {
 286         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 287         if (r < 0) {
 288             return r;
 289         }
 290         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 291         if (r < 0) {
 292             return r;
 293         }
 294         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 295         if (r < 0) {
 296             return r;
 297         }
 298     }
 299
 300     return 0;
 301 }
 302
 303 /*
 304  * Legacy layout for s390:
 305  * Older S390 KVM requires the topmost vma of the RAM to be
 306  * smaller than an system defined value, which is at least 256GB.
 307  * Larger systems have larger values. We put the guest between
 308  * the end of data segment (system break) and this value. We
 309  * use 32GB as a base to have enough room for the system break
 310  * to grow. We also have to use MAP parameters that avoid
 311  * read-only mapping of guest pages.
 312  */
 313 static void *legacy_s390_alloc(size_t size)
 314 {
 315     void *mem;
 316
 317     mem = mmap((void *) 0x800000000ULL, size,
 318                PROT_EXEC|PROT_READ|PROT_WRITE,
 319                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 320     return mem == MAP_FAILED ? NULL : mem;
 321 }
 322
 323 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 324 {
 325     static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 326
 327     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 4, 0) ||
 328         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501, 4, 1)) {
 329         return -EINVAL;
 330     }
 331     return 0;
 332 }
 333
 334 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 335 {
 336     uint8_t t[4];
 337     static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 338
 339     if (cpu_memory_rw_debug(cs, bp->pc, t, 4, 0)) {
 340         return -EINVAL;
 341     } else if (memcmp(t, diag_501, 4)) {
 342         return -EINVAL;
 343     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1)) {
 344         return -EINVAL;
 345     }
 346
 347     return 0;
 348 }
 349
 350 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 351                                   target_ulong len, int type)
 352 {
 353     return -ENOSYS;
 354 }
 355
 356 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 357                                   target_ulong len, int type)
 358 {
 359     return -ENOSYS;
 360 }
 361
 362 void kvm_arch_remove_all_hw_breakpoints(void)
 363 {
 364 }
 365
 366 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 367 {
 368 }
 369
 370 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 371 {
 372 }
 373
 374 void kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
 375 {
 376 }
 377
 378 int kvm_arch_process_async_events(CPUState *cs)
 379 {
 380     return cs->halted;
 381 }
 382
 383 void kvm_s390_interrupt_internal(S390CPU *cpu, int type, uint32_t parm,
 384                                  uint64_t parm64, int vm)
 385 {
 386     CPUState *cs = CPU(cpu);
 387     struct kvm_s390_interrupt kvmint;
 388     int r;
 389
 390     if (!cs->kvm_state) {
 391         return;
 392     }
 393
 394     kvmint.type = type;
 395     kvmint.parm = parm;
 396     kvmint.parm64 = parm64;
 397
 398     if (vm) {
 399         r = kvm_vm_ioctl(cs->kvm_state, KVM_S390_INTERRUPT, &kvmint);
 400     } else {
 401         r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
 402     }
 403
 404     if (r < 0) {
 405         fprintf(stderr, "KVM failed to inject interrupt\n");
 406         exit(1);
 407     }
 408 }
 409
 410 void kvm_s390_virtio_irq(S390CPU *cpu, int config_change, uint64_t token)
 411 {
 412     kvm_s390_interrupt_internal(cpu, KVM_S390_INT_VIRTIO, config_change,
 413                                 token, 1);
 414 }
 415
 416 void kvm_s390_interrupt(S390CPU *cpu, int type, uint32_t code)
 417 {
 418     kvm_s390_interrupt_internal(cpu, type, code, 0, 0);
 419 }
 420
 421 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
 422 {
 423     kvm_s390_interrupt(cpu, KVM_S390_PROGRAM_INT, code);
 424 }
 425
 426 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
 427                                  uint16_t ipbh0)
 428 {
 429     CPUS390XState *env = &cpu->env;
 430     uint64_t sccb;
 431     uint32_t code;
 432     int r = 0;
 433
 434     cpu_synchronize_state(CPU(cpu));
 435     sccb = env->regs[ipbh0 & 0xf];
 436     code = env->regs[(ipbh0 & 0xf0) >> 4];
 437
 438     r = sclp_service_call(env, sccb, code);
 439     if (r < 0) {
 440         enter_pgmcheck(cpu, -r);
 441     } else {
 442         setcc(cpu, r);
 443     }
 444
 445     return 0;
 446 }
 447
 448 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 449 {
 450     CPUS390XState *env = &cpu->env;
 451     int rc = 0;
 452     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
 453
 454     cpu_synchronize_state(CPU(cpu));
 455
 456     switch (ipa1) {
 457     case PRIV_B2_XSCH:
 458         ioinst_handle_xsch(cpu, env->regs[1]);
 459         break;
 460     case PRIV_B2_CSCH:
 461         ioinst_handle_csch(cpu, env->regs[1]);
 462         break;
 463     case PRIV_B2_HSCH:
 464         ioinst_handle_hsch(cpu, env->regs[1]);
 465         break;
 466     case PRIV_B2_MSCH:
 467         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
 468         break;
 469     case PRIV_B2_SSCH:
 470         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
 471         break;
 472     case PRIV_B2_STCRW:
 473         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
 474         break;
 475     case PRIV_B2_STSCH:
 476         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
 477         break;
 478     case PRIV_B2_TSCH:
 479         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
 480         fprintf(stderr, "Spurious tsch intercept\n");
 481         break;
 482     case PRIV_B2_CHSC:
 483         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
 484         break;
 485     case PRIV_B2_TPI:
 486         /* This should have been handled by kvm already. */
 487         fprintf(stderr, "Spurious tpi intercept\n");
 488         break;
 489     case PRIV_B2_SCHM:
 490         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
 491                            run->s390_sieic.ipb);
 492         break;
 493     case PRIV_B2_RSCH:
 494         ioinst_handle_rsch(cpu, env->regs[1]);
 495         break;
 496     case PRIV_B2_RCHP:
 497         ioinst_handle_rchp(cpu, env->regs[1]);
 498         break;
 499     case PRIV_B2_STCPS:
 500         /* We do not provide this instruction, it is suppressed. */
 501         break;
 502     case PRIV_B2_SAL:
 503         ioinst_handle_sal(cpu, env->regs[1]);
 504         break;
 505     case PRIV_B2_SIGA:
 506         /* Not provided, set CC = 3 for subchannel not operational */
 507         setcc(cpu, 3);
 508         break;
 509     case PRIV_B2_SCLP_CALL:
 510         rc = kvm_sclp_service_call(cpu, run, ipbh0);
 511         break;
 512     default:
 513         rc = -1;
 514         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
 515         break;
 516     }
 517
 518     return rc;
 519 }
 520
 521 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 522 {
 523     int r = 0;
 524
 525     switch (ipa1) {
 526     case PRIV_B9_EQBS:
 527         /* just inject exception */
 528         r = -1;
 529         break;
 530     default:
 531         r = -1;
 532         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
 533         break;
 534     }
 535
 536     return r;
 537 }
 538
 539 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 540 {
 541     int r = 0;
 542
 543     switch (ipa1) {
 544     case PRIV_EB_SQBS:
 545         /* just inject exception */
 546         r = -1;
 547         break;
 548     default:
 549         r = -1;
 550         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipa1);
 551         break;
 552     }
 553
 554     return r;
 555 }
 556
 557 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
 558 {
 559     CPUS390XState *env = &cpu->env;
 560     int ret;
 561
 562     cpu_synchronize_state(CPU(cpu));
 563     ret = s390_virtio_hypercall(env);
 564     if (ret == -EINVAL) {
 565         enter_pgmcheck(cpu, PGM_SPECIFICATION);
 566         return 0;
 567     }
 568
 569     return ret;
 570 }
 571
 572 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
 573 {
 574     uint64_t r1, r3;
 575
 576     cpu_synchronize_state(CPU(cpu));
 577     r1 = (run->s390_sieic.ipa & 0x00f0) >> 8;
 578     r3 = run->s390_sieic.ipa & 0x000f;
 579     handle_diag_308(&cpu->env, r1, r3);
 580 }
 581
 582 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
 583
 584 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
 585 {
 586     int r = 0;
 587     uint16_t func_code;
 588
 589     /*
 590      * For any diagnose call we support, bits 48-63 of the resulting
 591      * address specify the function code; the remainder is ignored.
 592      */
 593     func_code = decode_basedisp_rs(&cpu->env, ipb) & DIAG_KVM_CODE_MASK;
 594     switch (func_code) {
 595     case DIAG_IPL:
 596         kvm_handle_diag_308(cpu, run);
 597         break;
 598     case DIAG_KVM_HYPERCALL:
 599         r = handle_hypercall(cpu, run);
 600         break;
 601     case DIAG_KVM_BREAKPOINT:
 602         sleep(10);
 603         break;
 604     default:
 605         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
 606         r = -1;
 607         break;
 608     }
 609
 610     return r;
 611 }
 612
 613 static int kvm_s390_cpu_start(S390CPU *cpu)
 614 {
 615     s390_add_running_cpu(cpu);
 616     qemu_cpu_kick(CPU(cpu));
 617     DPRINTF("DONE: KVM cpu start: %p\n", &cpu->env);
 618     return 0;
 619 }
 620
 621 int kvm_s390_cpu_restart(S390CPU *cpu)
 622 {
 623     kvm_s390_interrupt(cpu, KVM_S390_RESTART, 0);
 624     s390_add_running_cpu(cpu);
 625     qemu_cpu_kick(CPU(cpu));
 626     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
 627     return 0;
 628 }
 629
 630 static void sigp_initial_cpu_reset(void *arg)
 631 {
 632     CPUState *cpu = arg;
 633     S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
 634
 635     cpu_synchronize_state(cpu);
 636     scc->initial_cpu_reset(cpu);
 637 }
 638
 639 static void sigp_cpu_reset(void *arg)
 640 {
 641     CPUState *cpu = arg;
 642     S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
 643
 644     cpu_synchronize_state(cpu);
 645     scc->cpu_reset(cpu);
 646 }
 647
 648 #define SIGP_ORDER_MASK 0x000000ff
 649
 650 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 651 {
 652     CPUS390XState *env = &cpu->env;
 653     uint8_t order_code;
 654     uint16_t cpu_addr;
 655     S390CPU *target_cpu;
 656     uint64_t *statusreg = &env->regs[ipa1 >> 4];
 657     int cc;
 658
 659     cpu_synchronize_state(CPU(cpu));
 660
 661     /* get order code */
 662     order_code = decode_basedisp_rs(env, run->s390_sieic.ipb) & SIGP_ORDER_MASK;
 663
 664     cpu_addr = env->regs[ipa1 & 0x0f];
 665     target_cpu = s390_cpu_addr2state(cpu_addr);
 666     if (target_cpu == NULL) {
 667         cc = 3;    /* not operational */
 668         goto out;
 669     }
 670
 671     switch (order_code) {
 672     case SIGP_START:
 673         cc = kvm_s390_cpu_start(target_cpu);
 674         break;
 675     case SIGP_RESTART:
 676         cc = kvm_s390_cpu_restart(target_cpu);
 677         break;
 678     case SIGP_SET_ARCH:
 679         *statusreg &= 0xffffffff00000000UL;
 680         *statusreg |= SIGP_STAT_INVALID_PARAMETER;
 681         cc = 1;   /* status stored */
 682         break;
 683     case SIGP_INITIAL_CPU_RESET:
 684         run_on_cpu(CPU(target_cpu), sigp_initial_cpu_reset, CPU(target_cpu));
 685         cc = 0;
 686         break;
 687     case SIGP_CPU_RESET:
 688         run_on_cpu(CPU(target_cpu), sigp_cpu_reset, CPU(target_cpu));
 689         cc = 0;
 690         break;
 691     default:
 692         DPRINTF("KVM: unknown SIGP: 0x%x\n", order_code);
 693         *statusreg &= 0xffffffff00000000UL;
 694         *statusreg |= SIGP_STAT_INVALID_ORDER;
 695         cc = 1;   /* status stored */
 696         break;
 697     }
 698
 699 out:
 700     setcc(cpu, cc);
 701     return 0;
 702 }
 703
 704 static void handle_instruction(S390CPU *cpu, struct kvm_run *run)
 705 {
 706     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
 707     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
 708     int r = -1;
 709
 710     DPRINTF("handle_instruction 0x%x 0x%x\n",
 711             run->s390_sieic.ipa, run->s390_sieic.ipb);
 712     switch (ipa0) {
 713     case IPA0_B2:
 714         r = handle_b2(cpu, run, ipa1);
 715         break;
 716     case IPA0_B9:
 717         r = handle_b9(cpu, run, ipa1);
 718         break;
 719     case IPA0_EB:
 720         r = handle_eb(cpu, run, ipa1);
 721         break;
 722     case IPA0_DIAG:
 723         r = handle_diag(cpu, run, run->s390_sieic.ipb);
 724         break;
 725     case IPA0_SIGP:
 726         r = handle_sigp(cpu, run, ipa1);
 727         break;
 728     }
 729
 730     if (r < 0) {
 731         enter_pgmcheck(cpu, 0x0001);
 732     }
 733 }
 734
 735 static bool is_special_wait_psw(CPUState *cs)
 736 {
 737     /* signal quiesce */
 738     return cs->kvm_run->psw_addr == 0xfffUL;
 739 }
 740
 741 static int handle_intercept(S390CPU *cpu)
 742 {
 743     CPUState *cs = CPU(cpu);
 744     struct kvm_run *run = cs->kvm_run;
 745     int icpt_code = run->s390_sieic.icptcode;
 746     int r = 0;
 747
 748     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
 749             (long)cs->kvm_run->psw_addr);
 750     switch (icpt_code) {
 751         case ICPT_INSTRUCTION:
 752             handle_instruction(cpu, run);
 753             break;
 754         case ICPT_WAITPSW:
 755             /* disabled wait, since enabled wait is handled in kernel */
 756             if (s390_del_running_cpu(cpu) == 0) {
 757                 if (is_special_wait_psw(cs)) {
 758                     qemu_system_shutdown_request();
 759                 } else {
 760                     QObject *data;
 761
 762                     data = qobject_from_jsonf("{ 'action': %s }", "pause");
 763                     monitor_protocol_event(QEVENT_GUEST_PANICKED, data);
 764                     qobject_decref(data);
 765                     vm_stop(RUN_STATE_GUEST_PANICKED);
 766                 }
 767             }
 768             r = EXCP_HALTED;
 769             break;
 770         case ICPT_CPU_STOP:
 771             if (s390_del_running_cpu(cpu) == 0) {
 772                 qemu_system_shutdown_request();
 773             }
 774             r = EXCP_HALTED;
 775             break;
 776         case ICPT_SOFT_INTERCEPT:
 777             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
 778             exit(1);
 779             break;
 780         case ICPT_IO:
 781             fprintf(stderr, "KVM unimplemented icpt IO\n");
 782             exit(1);
 783             break;
 784         default:
 785             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
 786             exit(1);
 787             break;
 788     }
 789
 790     return r;
 791 }
 792
 793 static int handle_tsch(S390CPU *cpu)
 794 {
 795     CPUS390XState *env = &cpu->env;
 796     CPUState *cs = CPU(cpu);
 797     struct kvm_run *run = cs->kvm_run;
 798     int ret;
 799
 800     cpu_synchronize_state(cs);
 801
 802     ret = ioinst_handle_tsch(env, env->regs[1], run->s390_tsch.ipb);
 803     if (ret >= 0) {
 804         /* Success; set condition code. */
 805         setcc(cpu, ret);
 806         ret = 0;
 807     } else if (ret < -1) {
 808         /*
 809          * Failure.
 810          * If an I/O interrupt had been dequeued, we have to reinject it.
 811          */
 812         if (run->s390_tsch.dequeued) {
 813             uint16_t subchannel_id = run->s390_tsch.subchannel_id;
 814             uint16_t subchannel_nr = run->s390_tsch.subchannel_nr;
 815             uint32_t io_int_parm = run->s390_tsch.io_int_parm;
 816             uint32_t io_int_word = run->s390_tsch.io_int_word;
 817             uint32_t type = ((subchannel_id & 0xff00) << 24) |
 818                 ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
 819
 820             kvm_s390_interrupt_internal(cpu, type,
 821                                         ((uint32_t)subchannel_id << 16)
 822                                         | subchannel_nr,
 823                                         ((uint64_t)io_int_parm << 32)
 824                                         | io_int_word, 1);
 825         }
 826         ret = 0;
 827     }
 828     return ret;
 829 }
 830
 831 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
 832 {
 833     return -ENOSYS;
 834 }
 835
 836 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
 837 {
 838     S390CPU *cpu = S390_CPU(cs);
 839     int ret = 0;
 840
 841     switch (run->exit_reason) {
 842         case KVM_EXIT_S390_SIEIC:
 843             ret = handle_intercept(cpu);
 844             break;
 845         case KVM_EXIT_S390_RESET:
 846             qemu_system_reset_request();
 847             break;
 848         case KVM_EXIT_S390_TSCH:
 849             ret = handle_tsch(cpu);
 850             break;
 851         case KVM_EXIT_DEBUG:
 852             ret = kvm_arch_handle_debug_exit(cpu);
 853             break;
 854         default:
 855             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
 856             break;
 857     }
 858
 859     if (ret == 0) {
 860         ret = EXCP_INTERRUPT;
 861     }
 862     return ret;
 863 }
 864
 865 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
 866 {
 867     return true;
 868 }
 869
 870 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
 871 {
 872     return 1;
 873 }
 874
 875 int kvm_arch_on_sigbus(int code, void *addr)
 876 {
 877     return 1;
 878 }
 879
 880 void kvm_s390_io_interrupt(S390CPU *cpu, uint16_t subchannel_id,
 881                            uint16_t subchannel_nr, uint32_t io_int_parm,
 882                            uint32_t io_int_word)
 883 {
 884     uint32_t type;
 885
 886     if (io_int_word & IO_INT_WORD_AI) {
 887         type = KVM_S390_INT_IO(1, 0, 0, 0);
 888     } else {
 889         type = ((subchannel_id & 0xff00) << 24) |
 890             ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
 891     }
 892     kvm_s390_interrupt_internal(cpu, type,
 893                                 ((uint32_t)subchannel_id << 16) | subchannel_nr,
 894                                 ((uint64_t)io_int_parm << 32) | io_int_word, 1);
 895 }
 896
 897 void kvm_s390_crw_mchk(S390CPU *cpu)
 898 {
 899     kvm_s390_interrupt_internal(cpu, KVM_S390_MCHK, 1 << 28,
 900                                 0x00400f1d40330000, 1);
 901 }
 902
 903 void kvm_s390_enable_css_support(S390CPU *cpu)
 904 {
 905     int r;
 906
 907     /* Activate host kernel channel subsystem support. */
 908     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
 909     assert(r == 0);
 910 }
 911
 912 void kvm_arch_init_irq_routing(KVMState *s)
 913 {
 914     /*
 915      * Note that while irqchip capabilities generally imply that cpustates
 916      * are handled in-kernel, it is not true for s390 (yet); therefore, we
 917      * have to override the common code kvm_halt_in_kernel_allowed setting.
 918      */
 919     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
 920         kvm_irqfds_allowed = true;
 921         kvm_gsi_routing_allowed = true;
 922         kvm_halt_in_kernel_allowed = false;
 923     }
 924 }
 925
 926 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
 927                                     int vq, bool assign)
 928 {
 929     struct kvm_ioeventfd kick = {
 930         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
 931         KVM_IOEVENTFD_FLAG_DATAMATCH,
 932         .fd = event_notifier_get_fd(notifier),
 933         .datamatch = vq,
 934         .addr = sch,
 935         .len = 8,
 936     };
 937     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
 938         return -ENOSYS;
 939     }
 940     if (!assign) {
 941         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
 942     }
 943     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
 944 }