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 "exec/address-spaces.h"
  42 #include "trace.h"
  43 #include "qapi-event.h"
  44 #include "hw/s390x/s390-pci-inst.h"
  45 #include "hw/s390x/s390-pci-bus.h"
  46 #include "hw/s390x/ipl.h"
  47
  48 /* #define DEBUG_KVM */
  49
  50 #ifdef DEBUG_KVM
  51 #define DPRINTF(fmt, ...) \
  52     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  53 #else
  54 #define DPRINTF(fmt, ...) \
  55     do { } while (0)
  56 #endif
  57
  58 #define IPA0_DIAG                       0x8300
  59 #define IPA0_SIGP                       0xae00
  60 #define IPA0_B2                         0xb200
  61 #define IPA0_B9                         0xb900
  62 #define IPA0_EB                         0xeb00
  63 #define IPA0_E3                         0xe300
  64
  65 #define PRIV_B2_SCLP_CALL               0x20
  66 #define PRIV_B2_CSCH                    0x30
  67 #define PRIV_B2_HSCH                    0x31
  68 #define PRIV_B2_MSCH                    0x32
  69 #define PRIV_B2_SSCH                    0x33
  70 #define PRIV_B2_STSCH                   0x34
  71 #define PRIV_B2_TSCH                    0x35
  72 #define PRIV_B2_TPI                     0x36
  73 #define PRIV_B2_SAL                     0x37
  74 #define PRIV_B2_RSCH                    0x38
  75 #define PRIV_B2_STCRW                   0x39
  76 #define PRIV_B2_STCPS                   0x3a
  77 #define PRIV_B2_RCHP                    0x3b
  78 #define PRIV_B2_SCHM                    0x3c
  79 #define PRIV_B2_CHSC                    0x5f
  80 #define PRIV_B2_SIGA                    0x74
  81 #define PRIV_B2_XSCH                    0x76
  82
  83 #define PRIV_EB_SQBS                    0x8a
  84 #define PRIV_EB_PCISTB                  0xd0
  85 #define PRIV_EB_SIC                     0xd1
  86
  87 #define PRIV_B9_EQBS                    0x9c
  88 #define PRIV_B9_CLP                     0xa0
  89 #define PRIV_B9_PCISTG                  0xd0
  90 #define PRIV_B9_PCILG                   0xd2
  91 #define PRIV_B9_RPCIT                   0xd3
  92
  93 #define PRIV_E3_MPCIFC                  0xd0
  94 #define PRIV_E3_STPCIFC                 0xd4
  95
  96 #define DIAG_IPL                        0x308
  97 #define DIAG_KVM_HYPERCALL              0x500
  98 #define DIAG_KVM_BREAKPOINT             0x501
  99
 100 #define ICPT_INSTRUCTION                0x04
 101 #define ICPT_PROGRAM                    0x08
 102 #define ICPT_EXT_INT                    0x14
 103 #define ICPT_WAITPSW                    0x1c
 104 #define ICPT_SOFT_INTERCEPT             0x24
 105 #define ICPT_CPU_STOP                   0x28
 106 #define ICPT_IO                         0x40
 107
 108 static CPUWatchpoint hw_watchpoint;
 109 /*
 110  * We don't use a list because this structure is also used to transmit the
 111  * hardware breakpoints to the kernel.
 112  */
 113 static struct kvm_hw_breakpoint *hw_breakpoints;
 114 static int nb_hw_breakpoints;
 115
 116 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
 117     KVM_CAP_LAST_INFO
 118 };
 119
 120 static int cap_sync_regs;
 121 static int cap_async_pf;
 122
 123 static void *legacy_s390_alloc(size_t size, uint64_t *align);
 124
 125 static int kvm_s390_check_clear_cmma(KVMState *s)
 126 {
 127     struct kvm_device_attr attr = {
 128         .group = KVM_S390_VM_MEM_CTRL,
 129         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 130     };
 131
 132     return kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attr);
 133 }
 134
 135 static int kvm_s390_check_enable_cmma(KVMState *s)
 136 {
 137     struct kvm_device_attr attr = {
 138         .group = KVM_S390_VM_MEM_CTRL,
 139         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 140     };
 141
 142     return kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attr);
 143 }
 144
 145 void kvm_s390_clear_cmma_callback(void *opaque)
 146 {
 147     int rc;
 148     KVMState *s = opaque;
 149     struct kvm_device_attr attr = {
 150         .group = KVM_S390_VM_MEM_CTRL,
 151         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 152     };
 153
 154     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 155     trace_kvm_clear_cmma(rc);
 156 }
 157
 158 static void kvm_s390_enable_cmma(KVMState *s)
 159 {
 160     int rc;
 161     struct kvm_device_attr attr = {
 162         .group = KVM_S390_VM_MEM_CTRL,
 163         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 164     };
 165
 166     if (kvm_s390_check_enable_cmma(s) || kvm_s390_check_clear_cmma(s)) {
 167         return;
 168     }
 169
 170     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 171     if (!rc) {
 172         qemu_register_reset(kvm_s390_clear_cmma_callback, s);
 173     }
 174     trace_kvm_enable_cmma(rc);
 175 }
 176
 177 int kvm_arch_init(KVMState *s)
 178 {
 179     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
 180     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
 181
 182     if (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES)) {
 183         kvm_s390_enable_cmma(s);
 184     }
 185
 186     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
 187         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
 188         phys_mem_set_alloc(legacy_s390_alloc);
 189     }
 190     return 0;
 191 }
 192
 193 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
 194 {
 195     return cpu->cpu_index;
 196 }
 197
 198 int kvm_arch_init_vcpu(CPUState *cs)
 199 {
 200     S390CPU *cpu = S390_CPU(cs);
 201     kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
 202     return 0;
 203 }
 204
 205 void kvm_s390_reset_vcpu(S390CPU *cpu)
 206 {
 207     CPUState *cs = CPU(cpu);
 208
 209     /* The initial reset call is needed here to reset in-kernel
 210      * vcpu data that we can't access directly from QEMU
 211      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
 212      * Before this ioctl cpu_synchronize_state() is called in common kvm
 213      * code (kvm-all) */
 214     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
 215         error_report("Initial CPU reset failed on CPU %i\n", cs->cpu_index);
 216     }
 217 }
 218
 219 static int can_sync_regs(CPUState *cs, int regs)
 220 {
 221     return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
 222 }
 223
 224 int kvm_arch_put_registers(CPUState *cs, int level)
 225 {
 226     S390CPU *cpu = S390_CPU(cs);
 227     CPUS390XState *env = &cpu->env;
 228     struct kvm_sregs sregs;
 229     struct kvm_regs regs;
 230     struct kvm_fpu fpu = {};
 231     int r;
 232     int i;
 233
 234     /* always save the PSW  and the GPRS*/
 235     cs->kvm_run->psw_addr = env->psw.addr;
 236     cs->kvm_run->psw_mask = env->psw.mask;
 237
 238     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 239         for (i = 0; i < 16; i++) {
 240             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
 241             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
 242         }
 243     } else {
 244         for (i = 0; i < 16; i++) {
 245             regs.gprs[i] = env->regs[i];
 246         }
 247         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
 248         if (r < 0) {
 249             return r;
 250         }
 251     }
 252
 253     /* Floating point */
 254     for (i = 0; i < 16; i++) {
 255         fpu.fprs[i] = env->fregs[i].ll;
 256     }
 257     fpu.fpc = env->fpc;
 258
 259     r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
 260     if (r < 0) {
 261         return r;
 262     }
 263
 264     /* Do we need to save more than that? */
 265     if (level == KVM_PUT_RUNTIME_STATE) {
 266         return 0;
 267     }
 268
 269     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 270         cs->kvm_run->s.regs.cputm = env->cputm;
 271         cs->kvm_run->s.regs.ckc = env->ckc;
 272         cs->kvm_run->s.regs.todpr = env->todpr;
 273         cs->kvm_run->s.regs.gbea = env->gbea;
 274         cs->kvm_run->s.regs.pp = env->pp;
 275         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
 276     } else {
 277         /*
 278          * These ONE_REGS are not protected by a capability. As they are only
 279          * necessary for migration we just trace a possible error, but don't
 280          * return with an error return code.
 281          */
 282         kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 283         kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 284         kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 285         kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 286         kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 287     }
 288
 289     /* pfault parameters */
 290     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 291         cs->kvm_run->s.regs.pft = env->pfault_token;
 292         cs->kvm_run->s.regs.pfs = env->pfault_select;
 293         cs->kvm_run->s.regs.pfc = env->pfault_compare;
 294         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
 295     } else if (cap_async_pf) {
 296         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 297         if (r < 0) {
 298             return r;
 299         }
 300         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 301         if (r < 0) {
 302             return r;
 303         }
 304         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 305         if (r < 0) {
 306             return r;
 307         }
 308     }
 309
 310     /* access registers and control registers*/
 311     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 312         for (i = 0; i < 16; i++) {
 313             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
 314             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
 315         }
 316         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
 317         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
 318     } else {
 319         for (i = 0; i < 16; i++) {
 320             sregs.acrs[i] = env->aregs[i];
 321             sregs.crs[i] = env->cregs[i];
 322         }
 323         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
 324         if (r < 0) {
 325             return r;
 326         }
 327     }
 328
 329     /* Finally the prefix */
 330     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 331         cs->kvm_run->s.regs.prefix = env->psa;
 332         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
 333     } else {
 334         /* prefix is only supported via sync regs */
 335     }
 336     return 0;
 337 }
 338
 339 int kvm_arch_get_registers(CPUState *cs)
 340 {
 341     S390CPU *cpu = S390_CPU(cs);
 342     CPUS390XState *env = &cpu->env;
 343     struct kvm_sregs sregs;
 344     struct kvm_regs regs;
 345     struct kvm_fpu fpu;
 346     int i, r;
 347
 348     /* get the PSW */
 349     env->psw.addr = cs->kvm_run->psw_addr;
 350     env->psw.mask = cs->kvm_run->psw_mask;
 351
 352     /* the GPRS */
 353     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 354         for (i = 0; i < 16; i++) {
 355             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
 356         }
 357     } else {
 358         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
 359         if (r < 0) {
 360             return r;
 361         }
 362          for (i = 0; i < 16; i++) {
 363             env->regs[i] = regs.gprs[i];
 364         }
 365     }
 366
 367     /* The ACRS and CRS */
 368     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 369         for (i = 0; i < 16; i++) {
 370             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
 371             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
 372         }
 373     } else {
 374         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
 375         if (r < 0) {
 376             return r;
 377         }
 378          for (i = 0; i < 16; i++) {
 379             env->aregs[i] = sregs.acrs[i];
 380             env->cregs[i] = sregs.crs[i];
 381         }
 382     }
 383
 384     /* Floating point */
 385     r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
 386     if (r < 0) {
 387         return r;
 388     }
 389     for (i = 0; i < 16; i++) {
 390         env->fregs[i].ll = fpu.fprs[i];
 391     }
 392     env->fpc = fpu.fpc;
 393
 394     /* The prefix */
 395     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 396         env->psa = cs->kvm_run->s.regs.prefix;
 397     }
 398
 399     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 400         env->cputm = cs->kvm_run->s.regs.cputm;
 401         env->ckc = cs->kvm_run->s.regs.ckc;
 402         env->todpr = cs->kvm_run->s.regs.todpr;
 403         env->gbea = cs->kvm_run->s.regs.gbea;
 404         env->pp = cs->kvm_run->s.regs.pp;
 405     } else {
 406         /*
 407          * These ONE_REGS are not protected by a capability. As they are only
 408          * necessary for migration we just trace a possible error, but don't
 409          * return with an error return code.
 410          */
 411         kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 412         kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 413         kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 414         kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 415         kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 416     }
 417
 418     /* pfault parameters */
 419     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 420         env->pfault_token = cs->kvm_run->s.regs.pft;
 421         env->pfault_select = cs->kvm_run->s.regs.pfs;
 422         env->pfault_compare = cs->kvm_run->s.regs.pfc;
 423     } else if (cap_async_pf) {
 424         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 425         if (r < 0) {
 426             return r;
 427         }
 428         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 429         if (r < 0) {
 430             return r;
 431         }
 432         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 433         if (r < 0) {
 434             return r;
 435         }
 436     }
 437
 438     return 0;
 439 }
 440
 441 /*
 442  * Legacy layout for s390:
 443  * Older S390 KVM requires the topmost vma of the RAM to be
 444  * smaller than an system defined value, which is at least 256GB.
 445  * Larger systems have larger values. We put the guest between
 446  * the end of data segment (system break) and this value. We
 447  * use 32GB as a base to have enough room for the system break
 448  * to grow. We also have to use MAP parameters that avoid
 449  * read-only mapping of guest pages.
 450  */
 451 static void *legacy_s390_alloc(size_t size, uint64_t *align)
 452 {
 453     void *mem;
 454
 455     mem = mmap((void *) 0x800000000ULL, size,
 456                PROT_EXEC|PROT_READ|PROT_WRITE,
 457                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 458     return mem == MAP_FAILED ? NULL : mem;
 459 }
 460
 461 /* DIAG 501 is used for sw breakpoints */
 462 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 463
 464 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 465 {
 466
 467     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 468                             sizeof(diag_501), 0) ||
 469         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501,
 470                             sizeof(diag_501), 1)) {
 471         return -EINVAL;
 472     }
 473     return 0;
 474 }
 475
 476 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 477 {
 478     uint8_t t[sizeof(diag_501)];
 479
 480     if (cpu_memory_rw_debug(cs, bp->pc, t, sizeof(diag_501), 0)) {
 481         return -EINVAL;
 482     } else if (memcmp(t, diag_501, sizeof(diag_501))) {
 483         return -EINVAL;
 484     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 485                                    sizeof(diag_501), 1)) {
 486         return -EINVAL;
 487     }
 488
 489     return 0;
 490 }
 491
 492 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
 493                                                     int len, int type)
 494 {
 495     int n;
 496
 497     for (n = 0; n < nb_hw_breakpoints; n++) {
 498         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
 499             (hw_breakpoints[n].len == len || len == -1)) {
 500             return &hw_breakpoints[n];
 501         }
 502     }
 503
 504     return NULL;
 505 }
 506
 507 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
 508 {
 509     int size;
 510
 511     if (find_hw_breakpoint(addr, len, type)) {
 512         return -EEXIST;
 513     }
 514
 515     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
 516
 517     if (!hw_breakpoints) {
 518         nb_hw_breakpoints = 0;
 519         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
 520     } else {
 521         hw_breakpoints =
 522             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
 523     }
 524
 525     if (!hw_breakpoints) {
 526         nb_hw_breakpoints = 0;
 527         return -ENOMEM;
 528     }
 529
 530     hw_breakpoints[nb_hw_breakpoints].addr = addr;
 531     hw_breakpoints[nb_hw_breakpoints].len = len;
 532     hw_breakpoints[nb_hw_breakpoints].type = type;
 533
 534     nb_hw_breakpoints++;
 535
 536     return 0;
 537 }
 538
 539 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 540                                   target_ulong len, int type)
 541 {
 542     switch (type) {
 543     case GDB_BREAKPOINT_HW:
 544         type = KVM_HW_BP;
 545         break;
 546     case GDB_WATCHPOINT_WRITE:
 547         if (len < 1) {
 548             return -EINVAL;
 549         }
 550         type = KVM_HW_WP_WRITE;
 551         break;
 552     default:
 553         return -ENOSYS;
 554     }
 555     return insert_hw_breakpoint(addr, len, type);
 556 }
 557
 558 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 559                                   target_ulong len, int type)
 560 {
 561     int size;
 562     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
 563
 564     if (bp == NULL) {
 565         return -ENOENT;
 566     }
 567
 568     nb_hw_breakpoints--;
 569     if (nb_hw_breakpoints > 0) {
 570         /*
 571          * In order to trim the array, move the last element to the position to
 572          * be removed - if necessary.
 573          */
 574         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
 575             *bp = hw_breakpoints[nb_hw_breakpoints];
 576         }
 577         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
 578         hw_breakpoints =
 579              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
 580     } else {
 581         g_free(hw_breakpoints);
 582         hw_breakpoints = NULL;
 583     }
 584
 585     return 0;
 586 }
 587
 588 void kvm_arch_remove_all_hw_breakpoints(void)
 589 {
 590     nb_hw_breakpoints = 0;
 591     g_free(hw_breakpoints);
 592     hw_breakpoints = NULL;
 593 }
 594
 595 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 596 {
 597     int i;
 598
 599     if (nb_hw_breakpoints > 0) {
 600         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
 601         dbg->arch.hw_bp = hw_breakpoints;
 602
 603         for (i = 0; i < nb_hw_breakpoints; ++i) {
 604             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
 605                                                        hw_breakpoints[i].addr);
 606         }
 607         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 608     } else {
 609         dbg->arch.nr_hw_bp = 0;
 610         dbg->arch.hw_bp = NULL;
 611     }
 612 }
 613
 614 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 615 {
 616 }
 617
 618 void kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
 619 {
 620 }
 621
 622 int kvm_arch_process_async_events(CPUState *cs)
 623 {
 624     return cs->halted;
 625 }
 626
 627 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
 628                                      struct kvm_s390_interrupt *interrupt)
 629 {
 630     int r = 0;
 631
 632     interrupt->type = irq->type;
 633     switch (irq->type) {
 634     case KVM_S390_INT_VIRTIO:
 635         interrupt->parm = irq->u.ext.ext_params;
 636         /* fall through */
 637     case KVM_S390_INT_PFAULT_INIT:
 638     case KVM_S390_INT_PFAULT_DONE:
 639         interrupt->parm64 = irq->u.ext.ext_params2;
 640         break;
 641     case KVM_S390_PROGRAM_INT:
 642         interrupt->parm = irq->u.pgm.code;
 643         break;
 644     case KVM_S390_SIGP_SET_PREFIX:
 645         interrupt->parm = irq->u.prefix.address;
 646         break;
 647     case KVM_S390_INT_SERVICE:
 648         interrupt->parm = irq->u.ext.ext_params;
 649         break;
 650     case KVM_S390_MCHK:
 651         interrupt->parm = irq->u.mchk.cr14;
 652         interrupt->parm64 = irq->u.mchk.mcic;
 653         break;
 654     case KVM_S390_INT_EXTERNAL_CALL:
 655         interrupt->parm = irq->u.extcall.code;
 656         break;
 657     case KVM_S390_INT_EMERGENCY:
 658         interrupt->parm = irq->u.emerg.code;
 659         break;
 660     case KVM_S390_SIGP_STOP:
 661     case KVM_S390_RESTART:
 662         break; /* These types have no parameters */
 663     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
 664         interrupt->parm = irq->u.io.subchannel_id << 16;
 665         interrupt->parm |= irq->u.io.subchannel_nr;
 666         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
 667         interrupt->parm64 |= irq->u.io.io_int_word;
 668         break;
 669     default:
 670         r = -EINVAL;
 671         break;
 672     }
 673     return r;
 674 }
 675
 676 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
 677 {
 678     struct kvm_s390_interrupt kvmint = {};
 679     CPUState *cs = CPU(cpu);
 680     int r;
 681
 682     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 683     if (r < 0) {
 684         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 685         exit(1);
 686     }
 687
 688     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
 689     if (r < 0) {
 690         fprintf(stderr, "KVM failed to inject interrupt\n");
 691         exit(1);
 692     }
 693 }
 694
 695 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 696 {
 697     struct kvm_s390_interrupt kvmint = {};
 698     int r;
 699
 700     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 701     if (r < 0) {
 702         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 703         exit(1);
 704     }
 705
 706     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
 707     if (r < 0) {
 708         fprintf(stderr, "KVM failed to inject interrupt\n");
 709         exit(1);
 710     }
 711 }
 712
 713 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 714 {
 715     static bool use_flic = true;
 716     int r;
 717
 718     if (use_flic) {
 719         r = kvm_s390_inject_flic(irq);
 720         if (r == -ENOSYS) {
 721             use_flic = false;
 722         }
 723         if (!r) {
 724             return;
 725         }
 726     }
 727     __kvm_s390_floating_interrupt(irq);
 728 }
 729
 730 void kvm_s390_virtio_irq(int config_change, uint64_t token)
 731 {
 732     struct kvm_s390_irq irq = {
 733         .type = KVM_S390_INT_VIRTIO,
 734         .u.ext.ext_params = config_change,
 735         .u.ext.ext_params2 = token,
 736     };
 737
 738     kvm_s390_floating_interrupt(&irq);
 739 }
 740
 741 void kvm_s390_service_interrupt(uint32_t parm)
 742 {
 743     struct kvm_s390_irq irq = {
 744         .type = KVM_S390_INT_SERVICE,
 745         .u.ext.ext_params = parm,
 746     };
 747
 748     kvm_s390_floating_interrupt(&irq);
 749 }
 750
 751 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
 752 {
 753     struct kvm_s390_irq irq = {
 754         .type = KVM_S390_PROGRAM_INT,
 755         .u.pgm.code = code,
 756     };
 757
 758     kvm_s390_vcpu_interrupt(cpu, &irq);
 759 }
 760
 761 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
 762 {
 763     struct kvm_s390_irq irq = {
 764         .type = KVM_S390_PROGRAM_INT,
 765         .u.pgm.code = code,
 766         .u.pgm.trans_exc_code = te_code,
 767         .u.pgm.exc_access_id = te_code & 3,
 768     };
 769
 770     kvm_s390_vcpu_interrupt(cpu, &irq);
 771 }
 772
 773 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
 774                                  uint16_t ipbh0)
 775 {
 776     CPUS390XState *env = &cpu->env;
 777     uint64_t sccb;
 778     uint32_t code;
 779     int r = 0;
 780
 781     cpu_synchronize_state(CPU(cpu));
 782     sccb = env->regs[ipbh0 & 0xf];
 783     code = env->regs[(ipbh0 & 0xf0) >> 4];
 784
 785     r = sclp_service_call(env, sccb, code);
 786     if (r < 0) {
 787         enter_pgmcheck(cpu, -r);
 788     } else {
 789         setcc(cpu, r);
 790     }
 791
 792     return 0;
 793 }
 794
 795 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 796 {
 797     CPUS390XState *env = &cpu->env;
 798     int rc = 0;
 799     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
 800
 801     cpu_synchronize_state(CPU(cpu));
 802
 803     switch (ipa1) {
 804     case PRIV_B2_XSCH:
 805         ioinst_handle_xsch(cpu, env->regs[1]);
 806         break;
 807     case PRIV_B2_CSCH:
 808         ioinst_handle_csch(cpu, env->regs[1]);
 809         break;
 810     case PRIV_B2_HSCH:
 811         ioinst_handle_hsch(cpu, env->regs[1]);
 812         break;
 813     case PRIV_B2_MSCH:
 814         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
 815         break;
 816     case PRIV_B2_SSCH:
 817         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
 818         break;
 819     case PRIV_B2_STCRW:
 820         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
 821         break;
 822     case PRIV_B2_STSCH:
 823         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
 824         break;
 825     case PRIV_B2_TSCH:
 826         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
 827         fprintf(stderr, "Spurious tsch intercept\n");
 828         break;
 829     case PRIV_B2_CHSC:
 830         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
 831         break;
 832     case PRIV_B2_TPI:
 833         /* This should have been handled by kvm already. */
 834         fprintf(stderr, "Spurious tpi intercept\n");
 835         break;
 836     case PRIV_B2_SCHM:
 837         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
 838                            run->s390_sieic.ipb);
 839         break;
 840     case PRIV_B2_RSCH:
 841         ioinst_handle_rsch(cpu, env->regs[1]);
 842         break;
 843     case PRIV_B2_RCHP:
 844         ioinst_handle_rchp(cpu, env->regs[1]);
 845         break;
 846     case PRIV_B2_STCPS:
 847         /* We do not provide this instruction, it is suppressed. */
 848         break;
 849     case PRIV_B2_SAL:
 850         ioinst_handle_sal(cpu, env->regs[1]);
 851         break;
 852     case PRIV_B2_SIGA:
 853         /* Not provided, set CC = 3 for subchannel not operational */
 854         setcc(cpu, 3);
 855         break;
 856     case PRIV_B2_SCLP_CALL:
 857         rc = kvm_sclp_service_call(cpu, run, ipbh0);
 858         break;
 859     default:
 860         rc = -1;
 861         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
 862         break;
 863     }
 864
 865     return rc;
 866 }
 867
 868 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run)
 869 {
 870     CPUS390XState *env = &cpu->env;
 871     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
 872     uint32_t base2 = run->s390_sieic.ipb >> 28;
 873     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
 874                      ((run->s390_sieic.ipb & 0xff00) << 4);
 875
 876     if (disp2 & 0x80000) {
 877         disp2 += 0xfff00000;
 878     }
 879
 880     return (base2 ? env->regs[base2] : 0) +
 881            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
 882 }
 883
 884 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run)
 885 {
 886     CPUS390XState *env = &cpu->env;
 887     uint32_t base2 = run->s390_sieic.ipb >> 28;
 888     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
 889                      ((run->s390_sieic.ipb & 0xff00) << 4);
 890
 891     if (disp2 & 0x80000) {
 892         disp2 += 0xfff00000;
 893     }
 894
 895     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
 896 }
 897
 898 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
 899 {
 900     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
 901
 902     return clp_service_call(cpu, r2);
 903 }
 904
 905 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
 906 {
 907     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
 908     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
 909
 910     return pcilg_service_call(cpu, r1, r2);
 911 }
 912
 913 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
 914 {
 915     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
 916     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
 917
 918     return pcistg_service_call(cpu, r1, r2);
 919 }
 920
 921 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
 922 {
 923     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
 924     uint64_t fiba;
 925
 926     cpu_synchronize_state(CPU(cpu));
 927     fiba = get_base_disp_rxy(cpu, run);
 928
 929     return stpcifc_service_call(cpu, r1, fiba);
 930 }
 931
 932 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
 933 {
 934     /* NOOP */
 935     return 0;
 936 }
 937
 938 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
 939 {
 940     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
 941     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
 942
 943     return rpcit_service_call(cpu, r1, r2);
 944 }
 945
 946 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
 947 {
 948     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
 949     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
 950     uint64_t gaddr;
 951
 952     cpu_synchronize_state(CPU(cpu));
 953     gaddr = get_base_disp_rsy(cpu, run);
 954
 955     return pcistb_service_call(cpu, r1, r3, gaddr);
 956 }
 957
 958 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
 959 {
 960     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
 961     uint64_t fiba;
 962
 963     cpu_synchronize_state(CPU(cpu));
 964     fiba = get_base_disp_rxy(cpu, run);
 965
 966     return mpcifc_service_call(cpu, r1, fiba);
 967 }
 968
 969 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 970 {
 971     int r = 0;
 972
 973     switch (ipa1) {
 974     case PRIV_B9_CLP:
 975         r = kvm_clp_service_call(cpu, run);
 976         break;
 977     case PRIV_B9_PCISTG:
 978         r = kvm_pcistg_service_call(cpu, run);
 979         break;
 980     case PRIV_B9_PCILG:
 981         r = kvm_pcilg_service_call(cpu, run);
 982         break;
 983     case PRIV_B9_RPCIT:
 984         r = kvm_rpcit_service_call(cpu, run);
 985         break;
 986     case PRIV_B9_EQBS:
 987         /* just inject exception */
 988         r = -1;
 989         break;
 990     default:
 991         r = -1;
 992         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
 993         break;
 994     }
 995
 996     return r;
 997 }
 998
 999 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1000 {
1001     int r = 0;
1002
1003     switch (ipbl) {
1004     case PRIV_EB_PCISTB:
1005         r = kvm_pcistb_service_call(cpu, run);
1006         break;
1007     case PRIV_EB_SIC:
1008         r = kvm_sic_service_call(cpu, run);
1009         break;
1010     case PRIV_EB_SQBS:
1011         /* just inject exception */
1012         r = -1;
1013         break;
1014     default:
1015         r = -1;
1016         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1017         break;
1018     }
1019
1020     return r;
1021 }
1022
1023 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1024 {
1025     int r = 0;
1026
1027     switch (ipbl) {
1028     case PRIV_E3_MPCIFC:
1029         r = kvm_mpcifc_service_call(cpu, run);
1030         break;
1031     case PRIV_E3_STPCIFC:
1032         r = kvm_stpcifc_service_call(cpu, run);
1033         break;
1034     default:
1035         r = -1;
1036         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1037         break;
1038     }
1039
1040     return r;
1041 }
1042
1043 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1044 {
1045     CPUS390XState *env = &cpu->env;
1046     int ret;
1047
1048     cpu_synchronize_state(CPU(cpu));
1049     ret = s390_virtio_hypercall(env);
1050     if (ret == -EINVAL) {
1051         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1052         return 0;
1053     }
1054
1055     return ret;
1056 }
1057
1058 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1059 {
1060     uint64_t r1, r3;
1061
1062     cpu_synchronize_state(CPU(cpu));
1063     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1064     r3 = run->s390_sieic.ipa & 0x000f;
1065     handle_diag_308(&cpu->env, r1, r3);
1066 }
1067
1068 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1069 {
1070     CPUS390XState *env = &cpu->env;
1071     unsigned long pc;
1072
1073     cpu_synchronize_state(CPU(cpu));
1074
1075     pc = env->psw.addr - 4;
1076     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1077         env->psw.addr = pc;
1078         return EXCP_DEBUG;
1079     }
1080
1081     return -ENOENT;
1082 }
1083
1084 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1085
1086 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1087 {
1088     int r = 0;
1089     uint16_t func_code;
1090
1091     /*
1092      * For any diagnose call we support, bits 48-63 of the resulting
1093      * address specify the function code; the remainder is ignored.
1094      */
1095     func_code = decode_basedisp_rs(&cpu->env, ipb) & DIAG_KVM_CODE_MASK;
1096     switch (func_code) {
1097     case DIAG_IPL:
1098         kvm_handle_diag_308(cpu, run);
1099         break;
1100     case DIAG_KVM_HYPERCALL:
1101         r = handle_hypercall(cpu, run);
1102         break;
1103     case DIAG_KVM_BREAKPOINT:
1104         r = handle_sw_breakpoint(cpu, run);
1105         break;
1106     default:
1107         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1108         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1109         break;
1110     }
1111
1112     return r;
1113 }
1114
1115 typedef struct SigpInfo {
1116     S390CPU *cpu;
1117     uint64_t param;
1118     int cc;
1119     uint64_t *status_reg;
1120 } SigpInfo;
1121
1122 static void set_sigp_status(SigpInfo *si, uint64_t status)
1123 {
1124     *si->status_reg &= 0xffffffff00000000ULL;
1125     *si->status_reg |= status;
1126     si->cc = SIGP_CC_STATUS_STORED;
1127 }
1128
1129 static void sigp_start(void *arg)
1130 {
1131     SigpInfo *si = arg;
1132
1133     s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1134     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1135 }
1136
1137 static void sigp_stop(void *arg)
1138 {
1139     SigpInfo *si = arg;
1140     struct kvm_s390_irq irq = {
1141         .type = KVM_S390_SIGP_STOP,
1142     };
1143
1144     if (s390_cpu_get_state(si->cpu) != CPU_STATE_OPERATING) {
1145         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1146         return;
1147     }
1148
1149     /* disabled wait - sleeping in user space */
1150     if (CPU(si->cpu)->halted) {
1151         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1152     } else {
1153         /* execute the stop function */
1154         si->cpu->env.sigp_order = SIGP_STOP;
1155         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1156     }
1157     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1158 }
1159
1160 #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, floating_pt_save_area)
1161 #define SAVE_AREA_SIZE 512
1162 static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
1163 {
1164     static const uint8_t ar_id = 1;
1165     uint64_t ckc = cpu->env.ckc >> 8;
1166     void *mem;
1167     hwaddr len = SAVE_AREA_SIZE;
1168
1169     mem = cpu_physical_memory_map(addr, &len, 1);
1170     if (!mem) {
1171         return -EFAULT;
1172     }
1173     if (len != SAVE_AREA_SIZE) {
1174         cpu_physical_memory_unmap(mem, len, 1, 0);
1175         return -EFAULT;
1176     }
1177
1178     if (store_arch) {
1179         cpu_physical_memory_write(offsetof(LowCore, ar_access_id), &ar_id, 1);
1180     }
1181     memcpy(mem, &cpu->env.fregs, 128);
1182     memcpy(mem + 128, &cpu->env.regs, 128);
1183     memcpy(mem + 256, &cpu->env.psw, 16);
1184     memcpy(mem + 280, &cpu->env.psa, 4);
1185     memcpy(mem + 284, &cpu->env.fpc, 4);
1186     memcpy(mem + 292, &cpu->env.todpr, 4);
1187     memcpy(mem + 296, &cpu->env.cputm, 8);
1188     memcpy(mem + 304, &ckc, 8);
1189     memcpy(mem + 320, &cpu->env.aregs, 64);
1190     memcpy(mem + 384, &cpu->env.cregs, 128);
1191
1192     cpu_physical_memory_unmap(mem, len, 1, len);
1193
1194     return 0;
1195 }
1196
1197 static void sigp_stop_and_store_status(void *arg)
1198 {
1199     SigpInfo *si = arg;
1200     struct kvm_s390_irq irq = {
1201         .type = KVM_S390_SIGP_STOP,
1202     };
1203
1204     /* disabled wait - sleeping in user space */
1205     if (s390_cpu_get_state(si->cpu) == CPU_STATE_OPERATING &&
1206         CPU(si->cpu)->halted) {
1207         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1208     }
1209
1210     switch (s390_cpu_get_state(si->cpu)) {
1211     case CPU_STATE_OPERATING:
1212         si->cpu->env.sigp_order = SIGP_STOP_STORE_STATUS;
1213         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1214         /* store will be performed when handling the stop intercept */
1215         break;
1216     case CPU_STATE_STOPPED:
1217         /* already stopped, just store the status */
1218         cpu_synchronize_state(CPU(si->cpu));
1219         kvm_s390_store_status(si->cpu, KVM_S390_STORE_STATUS_DEF_ADDR, true);
1220         break;
1221     }
1222     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1223 }
1224
1225 static void sigp_store_status_at_address(void *arg)
1226 {
1227     SigpInfo *si = arg;
1228     uint32_t address = si->param & 0x7ffffe00u;
1229
1230     /* cpu has to be stopped */
1231     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1232         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1233         return;
1234     }
1235
1236     cpu_synchronize_state(CPU(si->cpu));
1237
1238     if (kvm_s390_store_status(si->cpu, address, false)) {
1239         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1240         return;
1241     }
1242     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1243 }
1244
1245 static void sigp_restart(void *arg)
1246 {
1247     SigpInfo *si = arg;
1248     struct kvm_s390_irq irq = {
1249         .type = KVM_S390_RESTART,
1250     };
1251
1252     kvm_s390_vcpu_interrupt(si->cpu, &irq);
1253     s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1254     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1255 }
1256
1257 int kvm_s390_cpu_restart(S390CPU *cpu)
1258 {
1259     SigpInfo si = {
1260         .cpu = cpu,
1261     };
1262
1263     run_on_cpu(CPU(cpu), sigp_restart, &si);
1264     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1265     return 0;
1266 }
1267
1268 static void sigp_initial_cpu_reset(void *arg)
1269 {
1270     SigpInfo *si = arg;
1271     CPUState *cs = CPU(si->cpu);
1272     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1273
1274     cpu_synchronize_state(cs);
1275     scc->initial_cpu_reset(cs);
1276     cpu_synchronize_post_reset(cs);
1277     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1278 }
1279
1280 static void sigp_cpu_reset(void *arg)
1281 {
1282     SigpInfo *si = arg;
1283     CPUState *cs = CPU(si->cpu);
1284     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1285
1286     cpu_synchronize_state(cs);
1287     scc->cpu_reset(cs);
1288     cpu_synchronize_post_reset(cs);
1289     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1290 }
1291
1292 static void sigp_set_prefix(void *arg)
1293 {
1294     SigpInfo *si = arg;
1295     uint32_t addr = si->param & 0x7fffe000u;
1296
1297     cpu_synchronize_state(CPU(si->cpu));
1298
1299     if (!address_space_access_valid(&address_space_memory, addr,
1300                                     sizeof(struct LowCore), false)) {
1301         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1302         return;
1303     }
1304
1305     /* cpu has to be stopped */
1306     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1307         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1308         return;
1309     }
1310
1311     si->cpu->env.psa = addr;
1312     cpu_synchronize_post_init(CPU(si->cpu));
1313     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1314 }
1315
1316 static int handle_sigp_single_dst(S390CPU *dst_cpu, uint8_t order,
1317                                   uint64_t param, uint64_t *status_reg)
1318 {
1319     SigpInfo si = {
1320         .cpu = dst_cpu,
1321         .param = param,
1322         .status_reg = status_reg,
1323     };
1324
1325     /* cpu available? */
1326     if (dst_cpu == NULL) {
1327         return SIGP_CC_NOT_OPERATIONAL;
1328     }
1329
1330     /* only resets can break pending orders */
1331     if (dst_cpu->env.sigp_order != 0 &&
1332         order != SIGP_CPU_RESET &&
1333         order != SIGP_INITIAL_CPU_RESET) {
1334         return SIGP_CC_BUSY;
1335     }
1336
1337     switch (order) {
1338     case SIGP_START:
1339         run_on_cpu(CPU(dst_cpu), sigp_start, &si);
1340         break;
1341     case SIGP_STOP:
1342         run_on_cpu(CPU(dst_cpu), sigp_stop, &si);
1343         break;
1344     case SIGP_RESTART:
1345         run_on_cpu(CPU(dst_cpu), sigp_restart, &si);
1346         break;
1347     case SIGP_STOP_STORE_STATUS:
1348         run_on_cpu(CPU(dst_cpu), sigp_stop_and_store_status, &si);
1349         break;
1350     case SIGP_STORE_STATUS_ADDR:
1351         run_on_cpu(CPU(dst_cpu), sigp_store_status_at_address, &si);
1352         break;
1353     case SIGP_SET_PREFIX:
1354         run_on_cpu(CPU(dst_cpu), sigp_set_prefix, &si);
1355         break;
1356     case SIGP_INITIAL_CPU_RESET:
1357         run_on_cpu(CPU(dst_cpu), sigp_initial_cpu_reset, &si);
1358         break;
1359     case SIGP_CPU_RESET:
1360         run_on_cpu(CPU(dst_cpu), sigp_cpu_reset, &si);
1361         break;
1362     default:
1363         DPRINTF("KVM: unknown SIGP: 0x%x\n", order);
1364         set_sigp_status(&si, SIGP_STAT_INVALID_ORDER);
1365     }
1366
1367     return si.cc;
1368 }
1369
1370 static int sigp_set_architecture(S390CPU *cpu, uint32_t param,
1371                                  uint64_t *status_reg)
1372 {
1373     CPUState *cur_cs;
1374     S390CPU *cur_cpu;
1375
1376     /* due to the BQL, we are the only active cpu */
1377     CPU_FOREACH(cur_cs) {
1378         cur_cpu = S390_CPU(cur_cs);
1379         if (cur_cpu->env.sigp_order != 0) {
1380             return SIGP_CC_BUSY;
1381         }
1382         cpu_synchronize_state(cur_cs);
1383         /* all but the current one have to be stopped */
1384         if (cur_cpu != cpu &&
1385             s390_cpu_get_state(cur_cpu) != CPU_STATE_STOPPED) {
1386             *status_reg &= 0xffffffff00000000ULL;
1387             *status_reg |= SIGP_STAT_INCORRECT_STATE;
1388             return SIGP_CC_STATUS_STORED;
1389         }
1390     }
1391
1392     switch (param & 0xff) {
1393     case SIGP_MODE_ESA_S390:
1394         /* not supported */
1395         return SIGP_CC_NOT_OPERATIONAL;
1396     case SIGP_MODE_Z_ARCH_TRANS_ALL_PSW:
1397     case SIGP_MODE_Z_ARCH_TRANS_CUR_PSW:
1398         CPU_FOREACH(cur_cs) {
1399             cur_cpu = S390_CPU(cur_cs);
1400             cur_cpu->env.pfault_token = -1UL;
1401         }
1402         break;
1403     default:
1404         *status_reg &= 0xffffffff00000000ULL;
1405         *status_reg |= SIGP_STAT_INVALID_PARAMETER;
1406         return SIGP_CC_STATUS_STORED;
1407     }
1408
1409     return SIGP_CC_ORDER_CODE_ACCEPTED;
1410 }
1411
1412 #define SIGP_ORDER_MASK 0x000000ff
1413
1414 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1415 {
1416     CPUS390XState *env = &cpu->env;
1417     const uint8_t r1 = ipa1 >> 4;
1418     const uint8_t r3 = ipa1 & 0x0f;
1419     int ret;
1420     uint8_t order;
1421     uint64_t *status_reg;
1422     uint64_t param;
1423     S390CPU *dst_cpu = NULL;
1424
1425     cpu_synchronize_state(CPU(cpu));
1426
1427     /* get order code */
1428     order = decode_basedisp_rs(env, run->s390_sieic.ipb) & SIGP_ORDER_MASK;
1429     status_reg = &env->regs[r1];
1430     param = (r1 % 2) ? env->regs[r1] : env->regs[r1 + 1];
1431
1432     switch (order) {
1433     case SIGP_SET_ARCH:
1434         ret = sigp_set_architecture(cpu, param, status_reg);
1435         break;
1436     default:
1437         /* all other sigp orders target a single vcpu */
1438         dst_cpu = s390_cpu_addr2state(env->regs[r3]);
1439         ret = handle_sigp_single_dst(dst_cpu, order, param, status_reg);
1440     }
1441
1442     trace_kvm_sigp_finished(order, CPU(cpu)->cpu_index,
1443                             dst_cpu ? CPU(dst_cpu)->cpu_index : -1, ret);
1444
1445     if (ret >= 0) {
1446         setcc(cpu, ret);
1447         return 0;
1448     }
1449
1450     return ret;
1451 }
1452
1453 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1454 {
1455     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1456     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1457     int r = -1;
1458
1459     DPRINTF("handle_instruction 0x%x 0x%x\n",
1460             run->s390_sieic.ipa, run->s390_sieic.ipb);
1461     switch (ipa0) {
1462     case IPA0_B2:
1463         r = handle_b2(cpu, run, ipa1);
1464         break;
1465     case IPA0_B9:
1466         r = handle_b9(cpu, run, ipa1);
1467         break;
1468     case IPA0_EB:
1469         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1470         break;
1471     case IPA0_E3:
1472         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1473         break;
1474     case IPA0_DIAG:
1475         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1476         break;
1477     case IPA0_SIGP:
1478         r = handle_sigp(cpu, run, ipa1);
1479         break;
1480     }
1481
1482     if (r < 0) {
1483         r = 0;
1484         enter_pgmcheck(cpu, 0x0001);
1485     }
1486
1487     return r;
1488 }
1489
1490 static bool is_special_wait_psw(CPUState *cs)
1491 {
1492     /* signal quiesce */
1493     return cs->kvm_run->psw_addr == 0xfffUL;
1494 }
1495
1496 static void guest_panicked(void)
1497 {
1498     qapi_event_send_guest_panicked(GUEST_PANIC_ACTION_PAUSE,
1499                                    &error_abort);
1500     vm_stop(RUN_STATE_GUEST_PANICKED);
1501 }
1502
1503 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1504 {
1505     CPUState *cs = CPU(cpu);
1506
1507     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1508                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1509                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1510     s390_cpu_halt(cpu);
1511     guest_panicked();
1512 }
1513
1514 static int handle_intercept(S390CPU *cpu)
1515 {
1516     CPUState *cs = CPU(cpu);
1517     struct kvm_run *run = cs->kvm_run;
1518     int icpt_code = run->s390_sieic.icptcode;
1519     int r = 0;
1520
1521     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1522             (long)cs->kvm_run->psw_addr);
1523     switch (icpt_code) {
1524         case ICPT_INSTRUCTION:
1525             r = handle_instruction(cpu, run);
1526             break;
1527         case ICPT_PROGRAM:
1528             unmanageable_intercept(cpu, "program interrupt",
1529                                    offsetof(LowCore, program_new_psw));
1530             r = EXCP_HALTED;
1531             break;
1532         case ICPT_EXT_INT:
1533             unmanageable_intercept(cpu, "external interrupt",
1534                                    offsetof(LowCore, external_new_psw));
1535             r = EXCP_HALTED;
1536             break;
1537         case ICPT_WAITPSW:
1538             /* disabled wait, since enabled wait is handled in kernel */
1539             cpu_synchronize_state(cs);
1540             if (s390_cpu_halt(cpu) == 0) {
1541                 if (is_special_wait_psw(cs)) {
1542                     qemu_system_shutdown_request();
1543                 } else {
1544                     guest_panicked();
1545                 }
1546             }
1547             r = EXCP_HALTED;
1548             break;
1549         case ICPT_CPU_STOP:
1550             if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
1551                 qemu_system_shutdown_request();
1552             }
1553             if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
1554                 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR,
1555                                       true);
1556             }
1557             cpu->env.sigp_order = 0;
1558             r = EXCP_HALTED;
1559             break;
1560         case ICPT_SOFT_INTERCEPT:
1561             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1562             exit(1);
1563             break;
1564         case ICPT_IO:
1565             fprintf(stderr, "KVM unimplemented icpt IO\n");
1566             exit(1);
1567             break;
1568         default:
1569             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1570             exit(1);
1571             break;
1572     }
1573
1574     return r;
1575 }
1576
1577 static int handle_tsch(S390CPU *cpu)
1578 {
1579     CPUState *cs = CPU(cpu);
1580     struct kvm_run *run = cs->kvm_run;
1581     int ret;
1582
1583     cpu_synchronize_state(cs);
1584
1585     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb);
1586     if (ret < 0) {
1587         /*
1588          * Failure.
1589          * If an I/O interrupt had been dequeued, we have to reinject it.
1590          */
1591         if (run->s390_tsch.dequeued) {
1592             kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1593                                   run->s390_tsch.subchannel_nr,
1594                                   run->s390_tsch.io_int_parm,
1595                                   run->s390_tsch.io_int_word);
1596         }
1597         ret = 0;
1598     }
1599     return ret;
1600 }
1601
1602 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1603 {
1604     CPUState *cs = CPU(cpu);
1605     struct kvm_run *run = cs->kvm_run;
1606
1607     int ret = 0;
1608     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1609
1610     switch (arch_info->type) {
1611     case KVM_HW_WP_WRITE:
1612         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1613             cs->watchpoint_hit = &hw_watchpoint;
1614             hw_watchpoint.vaddr = arch_info->addr;
1615             hw_watchpoint.flags = BP_MEM_WRITE;
1616             ret = EXCP_DEBUG;
1617         }
1618         break;
1619     case KVM_HW_BP:
1620         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1621             ret = EXCP_DEBUG;
1622         }
1623         break;
1624     case KVM_SINGLESTEP:
1625         if (cs->singlestep_enabled) {
1626             ret = EXCP_DEBUG;
1627         }
1628         break;
1629     default:
1630         ret = -ENOSYS;
1631     }
1632
1633     return ret;
1634 }
1635
1636 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1637 {
1638     S390CPU *cpu = S390_CPU(cs);
1639     int ret = 0;
1640
1641     switch (run->exit_reason) {
1642         case KVM_EXIT_S390_SIEIC:
1643             ret = handle_intercept(cpu);
1644             break;
1645         case KVM_EXIT_S390_RESET:
1646             s390_reipl_request();
1647             break;
1648         case KVM_EXIT_S390_TSCH:
1649             ret = handle_tsch(cpu);
1650             break;
1651         case KVM_EXIT_DEBUG:
1652             ret = kvm_arch_handle_debug_exit(cpu);
1653             break;
1654         default:
1655             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
1656             break;
1657     }
1658
1659     if (ret == 0) {
1660         ret = EXCP_INTERRUPT;
1661     }
1662     return ret;
1663 }
1664
1665 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
1666 {
1667     return true;
1668 }
1669
1670 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
1671 {
1672     return 1;
1673 }
1674
1675 int kvm_arch_on_sigbus(int code, void *addr)
1676 {
1677     return 1;
1678 }
1679
1680 void kvm_s390_io_interrupt(uint16_t subchannel_id,
1681                            uint16_t subchannel_nr, uint32_t io_int_parm,
1682                            uint32_t io_int_word)
1683 {
1684     struct kvm_s390_irq irq = {
1685         .u.io.subchannel_id = subchannel_id,
1686         .u.io.subchannel_nr = subchannel_nr,
1687         .u.io.io_int_parm = io_int_parm,
1688         .u.io.io_int_word = io_int_word,
1689     };
1690
1691     if (io_int_word & IO_INT_WORD_AI) {
1692         irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
1693     } else {
1694         irq.type = ((subchannel_id & 0xff00) << 24) |
1695             ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
1696     }
1697     kvm_s390_floating_interrupt(&irq);
1698 }
1699
1700 void kvm_s390_crw_mchk(void)
1701 {
1702     struct kvm_s390_irq irq = {
1703         .type = KVM_S390_MCHK,
1704         .u.mchk.cr14 = 1 << 28,
1705         .u.mchk.mcic = 0x00400f1d40330000ULL,
1706     };
1707     kvm_s390_floating_interrupt(&irq);
1708 }
1709
1710 void kvm_s390_enable_css_support(S390CPU *cpu)
1711 {
1712     int r;
1713
1714     /* Activate host kernel channel subsystem support. */
1715     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
1716     assert(r == 0);
1717 }
1718
1719 void kvm_arch_init_irq_routing(KVMState *s)
1720 {
1721     /*
1722      * Note that while irqchip capabilities generally imply that cpustates
1723      * are handled in-kernel, it is not true for s390 (yet); therefore, we
1724      * have to override the common code kvm_halt_in_kernel_allowed setting.
1725      */
1726     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
1727         kvm_gsi_routing_allowed = true;
1728         kvm_halt_in_kernel_allowed = false;
1729     }
1730 }
1731
1732 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
1733                                     int vq, bool assign)
1734 {
1735     struct kvm_ioeventfd kick = {
1736         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
1737         KVM_IOEVENTFD_FLAG_DATAMATCH,
1738         .fd = event_notifier_get_fd(notifier),
1739         .datamatch = vq,
1740         .addr = sch,
1741         .len = 8,
1742     };
1743     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
1744         return -ENOSYS;
1745     }
1746     if (!assign) {
1747         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
1748     }
1749     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
1750 }
1751
1752 int kvm_s390_get_memslot_count(KVMState *s)
1753 {
1754     return kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1755 }
1756
1757 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
1758 {
1759     struct kvm_mp_state mp_state = {};
1760     int ret;
1761
1762     /* the kvm part might not have been initialized yet */
1763     if (CPU(cpu)->kvm_state == NULL) {
1764         return 0;
1765     }
1766
1767     switch (cpu_state) {
1768     case CPU_STATE_STOPPED:
1769         mp_state.mp_state = KVM_MP_STATE_STOPPED;
1770         break;
1771     case CPU_STATE_CHECK_STOP:
1772         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
1773         break;
1774     case CPU_STATE_OPERATING:
1775         mp_state.mp_state = KVM_MP_STATE_OPERATING;
1776         break;
1777     case CPU_STATE_LOAD:
1778         mp_state.mp_state = KVM_MP_STATE_LOAD;
1779         break;
1780     default:
1781         error_report("Requested CPU state is not a valid S390 CPU state: %u",
1782                      cpu_state);
1783         exit(1);
1784     }
1785
1786     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1787     if (ret) {
1788         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
1789                                        strerror(-ret));
1790     }
1791
1792     return ret;
1793 }
1794
1795 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1796                               uint64_t address, uint32_t data)
1797 {
1798     S390PCIBusDevice *pbdev;
1799     uint32_t fid = data >> ZPCI_MSI_VEC_BITS;
1800     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
1801
1802     pbdev = s390_pci_find_dev_by_fid(fid);
1803     if (!pbdev) {
1804         DPRINTF("add_msi_route no dev\n");
1805         return -ENODEV;
1806     }
1807
1808     pbdev->routes.adapter.ind_offset = vec;
1809
1810     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
1811     route->flags = 0;
1812     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
1813     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
1814     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
1815     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset;
1816     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
1817     return 0;
1818 }