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 "hw/hw.h"
  36 #include "cpu.h"
  37 #include "sysemu/device_tree.h"
  38 #include "qapi/qmp/qjson.h"
  39 #include "monitor/monitor.h"
  40 #include "exec/gdbstub.h"
  41 #include "trace.h"
  42
  43 /* #define DEBUG_KVM */
  44
  45 #ifdef DEBUG_KVM
  46 #define DPRINTF(fmt, ...) \
  47     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  48 #else
  49 #define DPRINTF(fmt, ...) \
  50     do { } while (0)
  51 #endif
  52
  53 #define IPA0_DIAG                       0x8300
  54 #define IPA0_SIGP                       0xae00
  55 #define IPA0_B2                         0xb200
  56 #define IPA0_B9                         0xb900
  57 #define IPA0_EB                         0xeb00
  58
  59 #define PRIV_B2_SCLP_CALL               0x20
  60 #define PRIV_B2_CSCH                    0x30
  61 #define PRIV_B2_HSCH                    0x31
  62 #define PRIV_B2_MSCH                    0x32
  63 #define PRIV_B2_SSCH                    0x33
  64 #define PRIV_B2_STSCH                   0x34
  65 #define PRIV_B2_TSCH                    0x35
  66 #define PRIV_B2_TPI                     0x36
  67 #define PRIV_B2_SAL                     0x37
  68 #define PRIV_B2_RSCH                    0x38
  69 #define PRIV_B2_STCRW                   0x39
  70 #define PRIV_B2_STCPS                   0x3a
  71 #define PRIV_B2_RCHP                    0x3b
  72 #define PRIV_B2_SCHM                    0x3c
  73 #define PRIV_B2_CHSC                    0x5f
  74 #define PRIV_B2_SIGA                    0x74
  75 #define PRIV_B2_XSCH                    0x76
  76
  77 #define PRIV_EB_SQBS                    0x8a
  78
  79 #define PRIV_B9_EQBS                    0x9c
  80
  81 #define DIAG_IPL                        0x308
  82 #define DIAG_KVM_HYPERCALL              0x500
  83 #define DIAG_KVM_BREAKPOINT             0x501
  84
  85 #define ICPT_INSTRUCTION                0x04
  86 #define ICPT_PROGRAM                    0x08
  87 #define ICPT_EXT_INT                    0x14
  88 #define ICPT_WAITPSW                    0x1c
  89 #define ICPT_SOFT_INTERCEPT             0x24
  90 #define ICPT_CPU_STOP                   0x28
  91 #define ICPT_IO                         0x40
  92
  93 static CPUWatchpoint hw_watchpoint;
  94 /*
  95  * We don't use a list because this structure is also used to transmit the
  96  * hardware breakpoints to the kernel.
  97  */
  98 static struct kvm_hw_breakpoint *hw_breakpoints;
  99 static int nb_hw_breakpoints;
 100
 101 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
 102     KVM_CAP_LAST_INFO
 103 };
 104
 105 static int cap_sync_regs;
 106 static int cap_async_pf;
 107
 108 static void *legacy_s390_alloc(size_t size);
 109
 110 static int kvm_s390_check_clear_cmma(KVMState *s)
 111 {
 112     struct kvm_device_attr attr = {
 113         .group = KVM_S390_VM_MEM_CTRL,
 114         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 115     };
 116
 117     return kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attr);
 118 }
 119
 120 static int kvm_s390_check_enable_cmma(KVMState *s)
 121 {
 122     struct kvm_device_attr attr = {
 123         .group = KVM_S390_VM_MEM_CTRL,
 124         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 125     };
 126
 127     return kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attr);
 128 }
 129
 130 void kvm_s390_clear_cmma_callback(void *opaque)
 131 {
 132     int rc;
 133     KVMState *s = opaque;
 134     struct kvm_device_attr attr = {
 135         .group = KVM_S390_VM_MEM_CTRL,
 136         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 137     };
 138
 139     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 140     trace_kvm_clear_cmma(rc);
 141 }
 142
 143 static void kvm_s390_enable_cmma(KVMState *s)
 144 {
 145     int rc;
 146     struct kvm_device_attr attr = {
 147         .group = KVM_S390_VM_MEM_CTRL,
 148         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 149     };
 150
 151     if (kvm_s390_check_enable_cmma(s) || kvm_s390_check_clear_cmma(s)) {
 152         return;
 153     }
 154
 155     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 156     if (!rc) {
 157         qemu_register_reset(kvm_s390_clear_cmma_callback, s);
 158     }
 159     trace_kvm_enable_cmma(rc);
 160 }
 161
 162 int kvm_arch_init(KVMState *s)
 163 {
 164     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
 165     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
 166
 167     if (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES)) {
 168         kvm_s390_enable_cmma(s);
 169     }
 170
 171     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
 172         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
 173         phys_mem_set_alloc(legacy_s390_alloc);
 174     }
 175     return 0;
 176 }
 177
 178 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
 179 {
 180     return cpu->cpu_index;
 181 }
 182
 183 int kvm_arch_init_vcpu(CPUState *cpu)
 184 {
 185     /* nothing todo yet */
 186     return 0;
 187 }
 188
 189 void kvm_s390_reset_vcpu(S390CPU *cpu)
 190 {
 191     CPUState *cs = CPU(cpu);
 192
 193     /* The initial reset call is needed here to reset in-kernel
 194      * vcpu data that we can't access directly from QEMU
 195      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
 196      * Before this ioctl cpu_synchronize_state() is called in common kvm
 197      * code (kvm-all) */
 198     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
 199         perror("Can't reset vcpu\n");
 200     }
 201 }
 202
 203 int kvm_arch_put_registers(CPUState *cs, int level)
 204 {
 205     S390CPU *cpu = S390_CPU(cs);
 206     CPUS390XState *env = &cpu->env;
 207     struct kvm_sregs sregs;
 208     struct kvm_regs regs;
 209     int r;
 210     int i;
 211
 212     /* always save the PSW  and the GPRS*/
 213     cs->kvm_run->psw_addr = env->psw.addr;
 214     cs->kvm_run->psw_mask = env->psw.mask;
 215
 216     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_GPRS) {
 217         for (i = 0; i < 16; i++) {
 218             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
 219             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
 220         }
 221     } else {
 222         for (i = 0; i < 16; i++) {
 223             regs.gprs[i] = env->regs[i];
 224         }
 225         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
 226         if (r < 0) {
 227             return r;
 228         }
 229     }
 230
 231     /* Do we need to save more than that? */
 232     if (level == KVM_PUT_RUNTIME_STATE) {
 233         return 0;
 234     }
 235
 236     /*
 237      * These ONE_REGS are not protected by a capability. As they are only
 238      * necessary for migration we just trace a possible error, but don't
 239      * return with an error return code.
 240      */
 241     kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 242     kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 243     kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 244     kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 245     kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 246
 247     if (cap_async_pf) {
 248         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 249         if (r < 0) {
 250             return r;
 251         }
 252         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 253         if (r < 0) {
 254             return r;
 255         }
 256         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 257         if (r < 0) {
 258             return r;
 259         }
 260     }
 261
 262     if (cap_sync_regs &&
 263         cs->kvm_run->kvm_valid_regs & KVM_SYNC_ACRS &&
 264         cs->kvm_run->kvm_valid_regs & KVM_SYNC_CRS) {
 265         for (i = 0; i < 16; i++) {
 266             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
 267             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
 268         }
 269         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
 270         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
 271     } else {
 272         for (i = 0; i < 16; i++) {
 273             sregs.acrs[i] = env->aregs[i];
 274             sregs.crs[i] = env->cregs[i];
 275         }
 276         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
 277         if (r < 0) {
 278             return r;
 279         }
 280     }
 281
 282     /* Finally the prefix */
 283     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_PREFIX) {
 284         cs->kvm_run->s.regs.prefix = env->psa;
 285         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
 286     } else {
 287         /* prefix is only supported via sync regs */
 288     }
 289     return 0;
 290 }
 291
 292 int kvm_arch_get_registers(CPUState *cs)
 293 {
 294     S390CPU *cpu = S390_CPU(cs);
 295     CPUS390XState *env = &cpu->env;
 296     struct kvm_sregs sregs;
 297     struct kvm_regs regs;
 298     int i, r;
 299
 300     /* get the PSW */
 301     env->psw.addr = cs->kvm_run->psw_addr;
 302     env->psw.mask = cs->kvm_run->psw_mask;
 303
 304     /* the GPRS */
 305     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_GPRS) {
 306         for (i = 0; i < 16; i++) {
 307             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
 308         }
 309     } else {
 310         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
 311         if (r < 0) {
 312             return r;
 313         }
 314          for (i = 0; i < 16; i++) {
 315             env->regs[i] = regs.gprs[i];
 316         }
 317     }
 318
 319     /* The ACRS and CRS */
 320     if (cap_sync_regs &&
 321         cs->kvm_run->kvm_valid_regs & KVM_SYNC_ACRS &&
 322         cs->kvm_run->kvm_valid_regs & KVM_SYNC_CRS) {
 323         for (i = 0; i < 16; i++) {
 324             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
 325             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
 326         }
 327     } else {
 328         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
 329         if (r < 0) {
 330             return r;
 331         }
 332          for (i = 0; i < 16; i++) {
 333             env->aregs[i] = sregs.acrs[i];
 334             env->cregs[i] = sregs.crs[i];
 335         }
 336     }
 337
 338     /* The prefix */
 339     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_PREFIX) {
 340         env->psa = cs->kvm_run->s.regs.prefix;
 341     }
 342
 343     /*
 344      * These ONE_REGS are not protected by a capability. As they are only
 345      * necessary for migration we just trace a possible error, but don't
 346      * return with an error return code.
 347      */
 348     kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 349     kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 350     kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 351     kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 352     kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 353
 354     if (cap_async_pf) {
 355         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 356         if (r < 0) {
 357             return r;
 358         }
 359         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 360         if (r < 0) {
 361             return r;
 362         }
 363         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 364         if (r < 0) {
 365             return r;
 366         }
 367     }
 368
 369     return 0;
 370 }
 371
 372 /*
 373  * Legacy layout for s390:
 374  * Older S390 KVM requires the topmost vma of the RAM to be
 375  * smaller than an system defined value, which is at least 256GB.
 376  * Larger systems have larger values. We put the guest between
 377  * the end of data segment (system break) and this value. We
 378  * use 32GB as a base to have enough room for the system break
 379  * to grow. We also have to use MAP parameters that avoid
 380  * read-only mapping of guest pages.
 381  */
 382 static void *legacy_s390_alloc(size_t size)
 383 {
 384     void *mem;
 385
 386     mem = mmap((void *) 0x800000000ULL, size,
 387                PROT_EXEC|PROT_READ|PROT_WRITE,
 388                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 389     return mem == MAP_FAILED ? NULL : mem;
 390 }
 391
 392 /* DIAG 501 is used for sw breakpoints */
 393 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 394
 395 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 396 {
 397
 398     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 399                             sizeof(diag_501), 0) ||
 400         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501,
 401                             sizeof(diag_501), 1)) {
 402         return -EINVAL;
 403     }
 404     return 0;
 405 }
 406
 407 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 408 {
 409     uint8_t t[sizeof(diag_501)];
 410
 411     if (cpu_memory_rw_debug(cs, bp->pc, t, sizeof(diag_501), 0)) {
 412         return -EINVAL;
 413     } else if (memcmp(t, diag_501, sizeof(diag_501))) {
 414         return -EINVAL;
 415     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 416                                    sizeof(diag_501), 1)) {
 417         return -EINVAL;
 418     }
 419
 420     return 0;
 421 }
 422
 423 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
 424                                                     int len, int type)
 425 {
 426     int n;
 427
 428     for (n = 0; n < nb_hw_breakpoints; n++) {
 429         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
 430             (hw_breakpoints[n].len == len || len == -1)) {
 431             return &hw_breakpoints[n];
 432         }
 433     }
 434
 435     return NULL;
 436 }
 437
 438 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
 439 {
 440     int size;
 441
 442     if (find_hw_breakpoint(addr, len, type)) {
 443         return -EEXIST;
 444     }
 445
 446     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
 447
 448     if (!hw_breakpoints) {
 449         nb_hw_breakpoints = 0;
 450         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
 451     } else {
 452         hw_breakpoints =
 453             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
 454     }
 455
 456     if (!hw_breakpoints) {
 457         nb_hw_breakpoints = 0;
 458         return -ENOMEM;
 459     }
 460
 461     hw_breakpoints[nb_hw_breakpoints].addr = addr;
 462     hw_breakpoints[nb_hw_breakpoints].len = len;
 463     hw_breakpoints[nb_hw_breakpoints].type = type;
 464
 465     nb_hw_breakpoints++;
 466
 467     return 0;
 468 }
 469
 470 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 471                                   target_ulong len, int type)
 472 {
 473     switch (type) {
 474     case GDB_BREAKPOINT_HW:
 475         type = KVM_HW_BP;
 476         break;
 477     case GDB_WATCHPOINT_WRITE:
 478         if (len < 1) {
 479             return -EINVAL;
 480         }
 481         type = KVM_HW_WP_WRITE;
 482         break;
 483     default:
 484         return -ENOSYS;
 485     }
 486     return insert_hw_breakpoint(addr, len, type);
 487 }
 488
 489 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 490                                   target_ulong len, int type)
 491 {
 492     int size;
 493     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
 494
 495     if (bp == NULL) {
 496         return -ENOENT;
 497     }
 498
 499     nb_hw_breakpoints--;
 500     if (nb_hw_breakpoints > 0) {
 501         /*
 502          * In order to trim the array, move the last element to the position to
 503          * be removed - if necessary.
 504          */
 505         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
 506             *bp = hw_breakpoints[nb_hw_breakpoints];
 507         }
 508         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
 509         hw_breakpoints =
 510              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
 511     } else {
 512         g_free(hw_breakpoints);
 513         hw_breakpoints = NULL;
 514     }
 515
 516     return 0;
 517 }
 518
 519 void kvm_arch_remove_all_hw_breakpoints(void)
 520 {
 521     nb_hw_breakpoints = 0;
 522     g_free(hw_breakpoints);
 523     hw_breakpoints = NULL;
 524 }
 525
 526 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 527 {
 528     int i;
 529
 530     if (nb_hw_breakpoints > 0) {
 531         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
 532         dbg->arch.hw_bp = hw_breakpoints;
 533
 534         for (i = 0; i < nb_hw_breakpoints; ++i) {
 535             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
 536                                                        hw_breakpoints[i].addr);
 537         }
 538         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 539     } else {
 540         dbg->arch.nr_hw_bp = 0;
 541         dbg->arch.hw_bp = NULL;
 542     }
 543 }
 544
 545 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 546 {
 547 }
 548
 549 void kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
 550 {
 551 }
 552
 553 int kvm_arch_process_async_events(CPUState *cs)
 554 {
 555     return cs->halted;
 556 }
 557
 558 void kvm_s390_interrupt_internal(S390CPU *cpu, int type, uint32_t parm,
 559                                  uint64_t parm64, int vm)
 560 {
 561     CPUState *cs = CPU(cpu);
 562     struct kvm_s390_interrupt kvmint;
 563     int r;
 564
 565     if (!cs->kvm_state) {
 566         return;
 567     }
 568
 569     kvmint.type = type;
 570     kvmint.parm = parm;
 571     kvmint.parm64 = parm64;
 572
 573     if (vm) {
 574         r = kvm_vm_ioctl(cs->kvm_state, KVM_S390_INTERRUPT, &kvmint);
 575     } else {
 576         r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
 577     }
 578
 579     if (r < 0) {
 580         fprintf(stderr, "KVM failed to inject interrupt\n");
 581         exit(1);
 582     }
 583 }
 584
 585 void kvm_s390_virtio_irq(S390CPU *cpu, int config_change, uint64_t token)
 586 {
 587     kvm_s390_interrupt_internal(cpu, KVM_S390_INT_VIRTIO, config_change,
 588                                 token, 1);
 589 }
 590
 591 void kvm_s390_interrupt(S390CPU *cpu, int type, uint32_t code)
 592 {
 593     kvm_s390_interrupt_internal(cpu, type, code, 0, 0);
 594 }
 595
 596 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
 597 {
 598     kvm_s390_interrupt(cpu, KVM_S390_PROGRAM_INT, code);
 599 }
 600
 601 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
 602                                  uint16_t ipbh0)
 603 {
 604     CPUS390XState *env = &cpu->env;
 605     uint64_t sccb;
 606     uint32_t code;
 607     int r = 0;
 608
 609     cpu_synchronize_state(CPU(cpu));
 610     sccb = env->regs[ipbh0 & 0xf];
 611     code = env->regs[(ipbh0 & 0xf0) >> 4];
 612
 613     r = sclp_service_call(env, sccb, code);
 614     if (r < 0) {
 615         enter_pgmcheck(cpu, -r);
 616     } else {
 617         setcc(cpu, r);
 618     }
 619
 620     return 0;
 621 }
 622
 623 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 624 {
 625     CPUS390XState *env = &cpu->env;
 626     int rc = 0;
 627     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
 628
 629     cpu_synchronize_state(CPU(cpu));
 630
 631     switch (ipa1) {
 632     case PRIV_B2_XSCH:
 633         ioinst_handle_xsch(cpu, env->regs[1]);
 634         break;
 635     case PRIV_B2_CSCH:
 636         ioinst_handle_csch(cpu, env->regs[1]);
 637         break;
 638     case PRIV_B2_HSCH:
 639         ioinst_handle_hsch(cpu, env->regs[1]);
 640         break;
 641     case PRIV_B2_MSCH:
 642         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
 643         break;
 644     case PRIV_B2_SSCH:
 645         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
 646         break;
 647     case PRIV_B2_STCRW:
 648         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
 649         break;
 650     case PRIV_B2_STSCH:
 651         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
 652         break;
 653     case PRIV_B2_TSCH:
 654         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
 655         fprintf(stderr, "Spurious tsch intercept\n");
 656         break;
 657     case PRIV_B2_CHSC:
 658         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
 659         break;
 660     case PRIV_B2_TPI:
 661         /* This should have been handled by kvm already. */
 662         fprintf(stderr, "Spurious tpi intercept\n");
 663         break;
 664     case PRIV_B2_SCHM:
 665         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
 666                            run->s390_sieic.ipb);
 667         break;
 668     case PRIV_B2_RSCH:
 669         ioinst_handle_rsch(cpu, env->regs[1]);
 670         break;
 671     case PRIV_B2_RCHP:
 672         ioinst_handle_rchp(cpu, env->regs[1]);
 673         break;
 674     case PRIV_B2_STCPS:
 675         /* We do not provide this instruction, it is suppressed. */
 676         break;
 677     case PRIV_B2_SAL:
 678         ioinst_handle_sal(cpu, env->regs[1]);
 679         break;
 680     case PRIV_B2_SIGA:
 681         /* Not provided, set CC = 3 for subchannel not operational */
 682         setcc(cpu, 3);
 683         break;
 684     case PRIV_B2_SCLP_CALL:
 685         rc = kvm_sclp_service_call(cpu, run, ipbh0);
 686         break;
 687     default:
 688         rc = -1;
 689         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
 690         break;
 691     }
 692
 693     return rc;
 694 }
 695
 696 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 697 {
 698     int r = 0;
 699
 700     switch (ipa1) {
 701     case PRIV_B9_EQBS:
 702         /* just inject exception */
 703         r = -1;
 704         break;
 705     default:
 706         r = -1;
 707         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
 708         break;
 709     }
 710
 711     return r;
 712 }
 713
 714 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 715 {
 716     int r = 0;
 717
 718     switch (ipa1) {
 719     case PRIV_EB_SQBS:
 720         /* just inject exception */
 721         r = -1;
 722         break;
 723     default:
 724         r = -1;
 725         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipa1);
 726         break;
 727     }
 728
 729     return r;
 730 }
 731
 732 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
 733 {
 734     CPUS390XState *env = &cpu->env;
 735     int ret;
 736
 737     cpu_synchronize_state(CPU(cpu));
 738     ret = s390_virtio_hypercall(env);
 739     if (ret == -EINVAL) {
 740         enter_pgmcheck(cpu, PGM_SPECIFICATION);
 741         return 0;
 742     }
 743
 744     return ret;
 745 }
 746
 747 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
 748 {
 749     uint64_t r1, r3;
 750
 751     cpu_synchronize_state(CPU(cpu));
 752     r1 = (run->s390_sieic.ipa & 0x00f0) >> 8;
 753     r3 = run->s390_sieic.ipa & 0x000f;
 754     handle_diag_308(&cpu->env, r1, r3);
 755 }
 756
 757 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
 758 {
 759     CPUS390XState *env = &cpu->env;
 760     unsigned long pc;
 761
 762     cpu_synchronize_state(CPU(cpu));
 763
 764     pc = env->psw.addr - 4;
 765     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
 766         env->psw.addr = pc;
 767         return EXCP_DEBUG;
 768     }
 769
 770     return -ENOENT;
 771 }
 772
 773 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
 774
 775 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
 776 {
 777     int r = 0;
 778     uint16_t func_code;
 779
 780     /*
 781      * For any diagnose call we support, bits 48-63 of the resulting
 782      * address specify the function code; the remainder is ignored.
 783      */
 784     func_code = decode_basedisp_rs(&cpu->env, ipb) & DIAG_KVM_CODE_MASK;
 785     switch (func_code) {
 786     case DIAG_IPL:
 787         kvm_handle_diag_308(cpu, run);
 788         break;
 789     case DIAG_KVM_HYPERCALL:
 790         r = handle_hypercall(cpu, run);
 791         break;
 792     case DIAG_KVM_BREAKPOINT:
 793         r = handle_sw_breakpoint(cpu, run);
 794         break;
 795     default:
 796         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
 797         r = -1;
 798         break;
 799     }
 800
 801     return r;
 802 }
 803
 804 static int kvm_s390_cpu_start(S390CPU *cpu)
 805 {
 806     s390_add_running_cpu(cpu);
 807     qemu_cpu_kick(CPU(cpu));
 808     DPRINTF("DONE: KVM cpu start: %p\n", &cpu->env);
 809     return 0;
 810 }
 811
 812 int kvm_s390_cpu_restart(S390CPU *cpu)
 813 {
 814     kvm_s390_interrupt(cpu, KVM_S390_RESTART, 0);
 815     s390_add_running_cpu(cpu);
 816     qemu_cpu_kick(CPU(cpu));
 817     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
 818     return 0;
 819 }
 820
 821 static void sigp_initial_cpu_reset(void *arg)
 822 {
 823     CPUState *cpu = arg;
 824     S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
 825
 826     cpu_synchronize_state(cpu);
 827     scc->initial_cpu_reset(cpu);
 828 }
 829
 830 static void sigp_cpu_reset(void *arg)
 831 {
 832     CPUState *cpu = arg;
 833     S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
 834
 835     cpu_synchronize_state(cpu);
 836     scc->cpu_reset(cpu);
 837 }
 838
 839 #define SIGP_ORDER_MASK 0x000000ff
 840
 841 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 842 {
 843     CPUS390XState *env = &cpu->env;
 844     uint8_t order_code;
 845     uint16_t cpu_addr;
 846     S390CPU *target_cpu;
 847     uint64_t *statusreg = &env->regs[ipa1 >> 4];
 848     int cc;
 849
 850     cpu_synchronize_state(CPU(cpu));
 851
 852     /* get order code */
 853     order_code = decode_basedisp_rs(env, run->s390_sieic.ipb) & SIGP_ORDER_MASK;
 854
 855     cpu_addr = env->regs[ipa1 & 0x0f];
 856     target_cpu = s390_cpu_addr2state(cpu_addr);
 857     if (target_cpu == NULL) {
 858         cc = 3;    /* not operational */
 859         goto out;
 860     }
 861
 862     switch (order_code) {
 863     case SIGP_START:
 864         cc = kvm_s390_cpu_start(target_cpu);
 865         break;
 866     case SIGP_RESTART:
 867         cc = kvm_s390_cpu_restart(target_cpu);
 868         break;
 869     case SIGP_SET_ARCH:
 870         *statusreg &= 0xffffffff00000000UL;
 871         *statusreg |= SIGP_STAT_INVALID_PARAMETER;
 872         cc = 1;   /* status stored */
 873         break;
 874     case SIGP_INITIAL_CPU_RESET:
 875         run_on_cpu(CPU(target_cpu), sigp_initial_cpu_reset, CPU(target_cpu));
 876         cc = 0;
 877         break;
 878     case SIGP_CPU_RESET:
 879         run_on_cpu(CPU(target_cpu), sigp_cpu_reset, CPU(target_cpu));
 880         cc = 0;
 881         break;
 882     default:
 883         DPRINTF("KVM: unknown SIGP: 0x%x\n", order_code);
 884         *statusreg &= 0xffffffff00000000UL;
 885         *statusreg |= SIGP_STAT_INVALID_ORDER;
 886         cc = 1;   /* status stored */
 887         break;
 888     }
 889
 890 out:
 891     setcc(cpu, cc);
 892     return 0;
 893 }
 894
 895 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
 896 {
 897     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
 898     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
 899     int r = -1;
 900
 901     DPRINTF("handle_instruction 0x%x 0x%x\n",
 902             run->s390_sieic.ipa, run->s390_sieic.ipb);
 903     switch (ipa0) {
 904     case IPA0_B2:
 905         r = handle_b2(cpu, run, ipa1);
 906         break;
 907     case IPA0_B9:
 908         r = handle_b9(cpu, run, ipa1);
 909         break;
 910     case IPA0_EB:
 911         r = handle_eb(cpu, run, ipa1);
 912         break;
 913     case IPA0_DIAG:
 914         r = handle_diag(cpu, run, run->s390_sieic.ipb);
 915         break;
 916     case IPA0_SIGP:
 917         r = handle_sigp(cpu, run, ipa1);
 918         break;
 919     }
 920
 921     if (r < 0) {
 922         r = 0;
 923         enter_pgmcheck(cpu, 0x0001);
 924     }
 925
 926     return r;
 927 }
 928
 929 static bool is_special_wait_psw(CPUState *cs)
 930 {
 931     /* signal quiesce */
 932     return cs->kvm_run->psw_addr == 0xfffUL;
 933 }
 934
 935 static void guest_panicked(void)
 936 {
 937     QObject *data;
 938
 939     data = qobject_from_jsonf("{ 'action': %s }", "pause");
 940     monitor_protocol_event(QEVENT_GUEST_PANICKED, data);
 941     qobject_decref(data);
 942
 943     vm_stop(RUN_STATE_GUEST_PANICKED);
 944 }
 945
 946 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
 947 {
 948     CPUState *cs = CPU(cpu);
 949
 950     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
 951                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
 952                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
 953     s390_del_running_cpu(cpu);
 954     guest_panicked();
 955 }
 956
 957 static int handle_intercept(S390CPU *cpu)
 958 {
 959     CPUState *cs = CPU(cpu);
 960     struct kvm_run *run = cs->kvm_run;
 961     int icpt_code = run->s390_sieic.icptcode;
 962     int r = 0;
 963
 964     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
 965             (long)cs->kvm_run->psw_addr);
 966     switch (icpt_code) {
 967         case ICPT_INSTRUCTION:
 968             r = handle_instruction(cpu, run);
 969             break;
 970         case ICPT_PROGRAM:
 971             unmanageable_intercept(cpu, "program interrupt",
 972                                    offsetof(LowCore, program_new_psw));
 973             r = EXCP_HALTED;
 974             break;
 975         case ICPT_EXT_INT:
 976             unmanageable_intercept(cpu, "external interrupt",
 977                                    offsetof(LowCore, external_new_psw));
 978             r = EXCP_HALTED;
 979             break;
 980         case ICPT_WAITPSW:
 981             /* disabled wait, since enabled wait is handled in kernel */
 982             if (s390_del_running_cpu(cpu) == 0) {
 983                 if (is_special_wait_psw(cs)) {
 984                     qemu_system_shutdown_request();
 985                 } else {
 986                     guest_panicked();
 987                 }
 988             }
 989             r = EXCP_HALTED;
 990             break;
 991         case ICPT_CPU_STOP:
 992             if (s390_del_running_cpu(cpu) == 0) {
 993                 qemu_system_shutdown_request();
 994             }
 995             r = EXCP_HALTED;
 996             break;
 997         case ICPT_SOFT_INTERCEPT:
 998             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
 999             exit(1);
1000             break;
1001         case ICPT_IO:
1002             fprintf(stderr, "KVM unimplemented icpt IO\n");
1003             exit(1);
1004             break;
1005         default:
1006             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1007             exit(1);
1008             break;
1009     }
1010
1011     return r;
1012 }
1013
1014 static int handle_tsch(S390CPU *cpu)
1015 {
1016     CPUS390XState *env = &cpu->env;
1017     CPUState *cs = CPU(cpu);
1018     struct kvm_run *run = cs->kvm_run;
1019     int ret;
1020
1021     cpu_synchronize_state(cs);
1022
1023     ret = ioinst_handle_tsch(env, env->regs[1], run->s390_tsch.ipb);
1024     if (ret >= 0) {
1025         /* Success; set condition code. */
1026         setcc(cpu, ret);
1027         ret = 0;
1028     } else if (ret < -1) {
1029         /*
1030          * Failure.
1031          * If an I/O interrupt had been dequeued, we have to reinject it.
1032          */
1033         if (run->s390_tsch.dequeued) {
1034             uint16_t subchannel_id = run->s390_tsch.subchannel_id;
1035             uint16_t subchannel_nr = run->s390_tsch.subchannel_nr;
1036             uint32_t io_int_parm = run->s390_tsch.io_int_parm;
1037             uint32_t io_int_word = run->s390_tsch.io_int_word;
1038             uint32_t type = ((subchannel_id & 0xff00) << 24) |
1039                 ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
1040
1041             kvm_s390_interrupt_internal(cpu, type,
1042                                         ((uint32_t)subchannel_id << 16)
1043                                         | subchannel_nr,
1044                                         ((uint64_t)io_int_parm << 32)
1045                                         | io_int_word, 1);
1046         }
1047         ret = 0;
1048     }
1049     return ret;
1050 }
1051
1052 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1053 {
1054     CPUState *cs = CPU(cpu);
1055     struct kvm_run *run = cs->kvm_run;
1056
1057     int ret = 0;
1058     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1059
1060     switch (arch_info->type) {
1061     case KVM_HW_WP_WRITE:
1062         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1063             cs->watchpoint_hit = &hw_watchpoint;
1064             hw_watchpoint.vaddr = arch_info->addr;
1065             hw_watchpoint.flags = BP_MEM_WRITE;
1066             ret = EXCP_DEBUG;
1067         }
1068         break;
1069     case KVM_HW_BP:
1070         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1071             ret = EXCP_DEBUG;
1072         }
1073         break;
1074     case KVM_SINGLESTEP:
1075         if (cs->singlestep_enabled) {
1076             ret = EXCP_DEBUG;
1077         }
1078         break;
1079     default:
1080         ret = -ENOSYS;
1081     }
1082
1083     return ret;
1084 }
1085
1086 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1087 {
1088     S390CPU *cpu = S390_CPU(cs);
1089     int ret = 0;
1090
1091     switch (run->exit_reason) {
1092         case KVM_EXIT_S390_SIEIC:
1093             ret = handle_intercept(cpu);
1094             break;
1095         case KVM_EXIT_S390_RESET:
1096             qemu_system_reset_request();
1097             break;
1098         case KVM_EXIT_S390_TSCH:
1099             ret = handle_tsch(cpu);
1100             break;
1101         case KVM_EXIT_DEBUG:
1102             ret = kvm_arch_handle_debug_exit(cpu);
1103             break;
1104         default:
1105             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
1106             break;
1107     }
1108
1109     if (ret == 0) {
1110         ret = EXCP_INTERRUPT;
1111     }
1112     return ret;
1113 }
1114
1115 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
1116 {
1117     return true;
1118 }
1119
1120 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
1121 {
1122     return 1;
1123 }
1124
1125 int kvm_arch_on_sigbus(int code, void *addr)
1126 {
1127     return 1;
1128 }
1129
1130 void kvm_s390_io_interrupt(S390CPU *cpu, uint16_t subchannel_id,
1131                            uint16_t subchannel_nr, uint32_t io_int_parm,
1132                            uint32_t io_int_word)
1133 {
1134     uint32_t type;
1135
1136     if (io_int_word & IO_INT_WORD_AI) {
1137         type = KVM_S390_INT_IO(1, 0, 0, 0);
1138     } else {
1139         type = ((subchannel_id & 0xff00) << 24) |
1140             ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
1141     }
1142     kvm_s390_interrupt_internal(cpu, type,
1143                                 ((uint32_t)subchannel_id << 16) | subchannel_nr,
1144                                 ((uint64_t)io_int_parm << 32) | io_int_word, 1);
1145 }
1146
1147 void kvm_s390_crw_mchk(S390CPU *cpu)
1148 {
1149     kvm_s390_interrupt_internal(cpu, KVM_S390_MCHK, 1 << 28,
1150                                 0x00400f1d40330000, 1);
1151 }
1152
1153 void kvm_s390_enable_css_support(S390CPU *cpu)
1154 {
1155     int r;
1156
1157     /* Activate host kernel channel subsystem support. */
1158     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
1159     assert(r == 0);
1160 }
1161
1162 void kvm_arch_init_irq_routing(KVMState *s)
1163 {
1164     /*
1165      * Note that while irqchip capabilities generally imply that cpustates
1166      * are handled in-kernel, it is not true for s390 (yet); therefore, we
1167      * have to override the common code kvm_halt_in_kernel_allowed setting.
1168      */
1169     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
1170         kvm_irqfds_allowed = true;
1171         kvm_gsi_routing_allowed = true;
1172         kvm_halt_in_kernel_allowed = false;
1173     }
1174 }
1175
1176 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
1177                                     int vq, bool assign)
1178 {
1179     struct kvm_ioeventfd kick = {
1180         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
1181         KVM_IOEVENTFD_FLAG_DATAMATCH,
1182         .fd = event_notifier_get_fd(notifier),
1183         .datamatch = vq,
1184         .addr = sch,
1185         .len = 8,
1186     };
1187     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
1188         return -ENOSYS;
1189     }
1190     if (!assign) {
1191         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
1192     }
1193     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
1194 }