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 "qemu/osdep.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 "cpu.h"
  33 #include "qemu/error-report.h"
  34 #include "qemu/timer.h"
  35 #include "sysemu/sysemu.h"
  36 #include "sysemu/kvm.h"
  37 #include "hw/hw.h"
  38 #include "sysemu/device_tree.h"
  39 #include "qapi/qmp/qjson.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 #include "hw/s390x/ebcdic.h"
  48 #include "exec/memattrs.h"
  49 #include "hw/s390x/s390-virtio-ccw.h"
  50
  51 /* #define DEBUG_KVM */
  52
  53 #ifdef DEBUG_KVM
  54 #define DPRINTF(fmt, ...) \
  55     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  56 #else
  57 #define DPRINTF(fmt, ...) \
  58     do { } while (0)
  59 #endif
  60
  61 #define kvm_vm_check_mem_attr(s, attr) \
  62     kvm_vm_check_attr(s, KVM_S390_VM_MEM_CTRL, attr)
  63
  64 #define IPA0_DIAG                       0x8300
  65 #define IPA0_SIGP                       0xae00
  66 #define IPA0_B2                         0xb200
  67 #define IPA0_B9                         0xb900
  68 #define IPA0_EB                         0xeb00
  69 #define IPA0_E3                         0xe300
  70
  71 #define PRIV_B2_SCLP_CALL               0x20
  72 #define PRIV_B2_CSCH                    0x30
  73 #define PRIV_B2_HSCH                    0x31
  74 #define PRIV_B2_MSCH                    0x32
  75 #define PRIV_B2_SSCH                    0x33
  76 #define PRIV_B2_STSCH                   0x34
  77 #define PRIV_B2_TSCH                    0x35
  78 #define PRIV_B2_TPI                     0x36
  79 #define PRIV_B2_SAL                     0x37
  80 #define PRIV_B2_RSCH                    0x38
  81 #define PRIV_B2_STCRW                   0x39
  82 #define PRIV_B2_STCPS                   0x3a
  83 #define PRIV_B2_RCHP                    0x3b
  84 #define PRIV_B2_SCHM                    0x3c
  85 #define PRIV_B2_CHSC                    0x5f
  86 #define PRIV_B2_SIGA                    0x74
  87 #define PRIV_B2_XSCH                    0x76
  88
  89 #define PRIV_EB_SQBS                    0x8a
  90 #define PRIV_EB_PCISTB                  0xd0
  91 #define PRIV_EB_SIC                     0xd1
  92
  93 #define PRIV_B9_EQBS                    0x9c
  94 #define PRIV_B9_CLP                     0xa0
  95 #define PRIV_B9_PCISTG                  0xd0
  96 #define PRIV_B9_PCILG                   0xd2
  97 #define PRIV_B9_RPCIT                   0xd3
  98
  99 #define PRIV_E3_MPCIFC                  0xd0
 100 #define PRIV_E3_STPCIFC                 0xd4
 101
 102 #define DIAG_TIMEREVENT                 0x288
 103 #define DIAG_IPL                        0x308
 104 #define DIAG_KVM_HYPERCALL              0x500
 105 #define DIAG_KVM_BREAKPOINT             0x501
 106
 107 #define ICPT_INSTRUCTION                0x04
 108 #define ICPT_PROGRAM                    0x08
 109 #define ICPT_EXT_INT                    0x14
 110 #define ICPT_WAITPSW                    0x1c
 111 #define ICPT_SOFT_INTERCEPT             0x24
 112 #define ICPT_CPU_STOP                   0x28
 113 #define ICPT_IO                         0x40
 114
 115 #define NR_LOCAL_IRQS 32
 116 /*
 117  * Needs to be big enough to contain max_cpus emergency signals
 118  * and in addition NR_LOCAL_IRQS interrupts
 119  */
 120 #define VCPU_IRQ_BUF_SIZE (sizeof(struct kvm_s390_irq) * \
 121                            (max_cpus + NR_LOCAL_IRQS))
 122
 123 static CPUWatchpoint hw_watchpoint;
 124 /*
 125  * We don't use a list because this structure is also used to transmit the
 126  * hardware breakpoints to the kernel.
 127  */
 128 static struct kvm_hw_breakpoint *hw_breakpoints;
 129 static int nb_hw_breakpoints;
 130
 131 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
 132     KVM_CAP_LAST_INFO
 133 };
 134
 135 static int cap_sync_regs;
 136 static int cap_async_pf;
 137 static int cap_mem_op;
 138 static int cap_s390_irq;
 139 static int cap_ri;
 140
 141 static void *legacy_s390_alloc(size_t size, uint64_t *align);
 142
 143 static int kvm_s390_query_mem_limit(KVMState *s, uint64_t *memory_limit)
 144 {
 145     struct kvm_device_attr attr = {
 146         .group = KVM_S390_VM_MEM_CTRL,
 147         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
 148         .addr = (uint64_t) memory_limit,
 149     };
 150
 151     return kvm_vm_ioctl(s, KVM_GET_DEVICE_ATTR, &attr);
 152 }
 153
 154 int kvm_s390_set_mem_limit(KVMState *s, uint64_t new_limit, uint64_t *hw_limit)
 155 {
 156     int rc;
 157
 158     struct kvm_device_attr attr = {
 159         .group = KVM_S390_VM_MEM_CTRL,
 160         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
 161         .addr = (uint64_t) &new_limit,
 162     };
 163
 164     if (!kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_LIMIT_SIZE)) {
 165         return 0;
 166     }
 167
 168     rc = kvm_s390_query_mem_limit(s, hw_limit);
 169     if (rc) {
 170         return rc;
 171     } else if (*hw_limit < new_limit) {
 172         return -E2BIG;
 173     }
 174
 175     return kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 176 }
 177
 178 void kvm_s390_cmma_reset(void)
 179 {
 180     int rc;
 181     struct kvm_device_attr attr = {
 182         .group = KVM_S390_VM_MEM_CTRL,
 183         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 184     };
 185
 186     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 187     trace_kvm_clear_cmma(rc);
 188 }
 189
 190 static void kvm_s390_enable_cmma(KVMState *s)
 191 {
 192     int rc;
 193     struct kvm_device_attr attr = {
 194         .group = KVM_S390_VM_MEM_CTRL,
 195         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 196     };
 197
 198     if (!kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_ENABLE_CMMA) ||
 199         !kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_CLR_CMMA)) {
 200         return;
 201     }
 202
 203     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 204     trace_kvm_enable_cmma(rc);
 205 }
 206
 207 static void kvm_s390_set_attr(uint64_t attr)
 208 {
 209     struct kvm_device_attr attribute = {
 210         .group = KVM_S390_VM_CRYPTO,
 211         .attr  = attr,
 212     };
 213
 214     int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
 215
 216     if (ret) {
 217         error_report("Failed to set crypto device attribute %lu: %s",
 218                      attr, strerror(-ret));
 219     }
 220 }
 221
 222 static void kvm_s390_init_aes_kw(void)
 223 {
 224     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
 225
 226     if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
 227                                  NULL)) {
 228             attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
 229     }
 230
 231     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 232             kvm_s390_set_attr(attr);
 233     }
 234 }
 235
 236 static void kvm_s390_init_dea_kw(void)
 237 {
 238     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
 239
 240     if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
 241                                  NULL)) {
 242             attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
 243     }
 244
 245     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 246             kvm_s390_set_attr(attr);
 247     }
 248 }
 249
 250 void kvm_s390_crypto_reset(void)
 251 {
 252     kvm_s390_init_aes_kw();
 253     kvm_s390_init_dea_kw();
 254 }
 255
 256 int kvm_arch_init(MachineState *ms, KVMState *s)
 257 {
 258     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
 259     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
 260     cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
 261     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
 262
 263     if (!mem_path) {
 264         kvm_s390_enable_cmma(s);
 265     }
 266
 267     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
 268         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
 269         phys_mem_set_alloc(legacy_s390_alloc);
 270     }
 271
 272     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
 273     kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
 274     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_STSI, 0);
 275     if (ri_allowed()) {
 276         if (kvm_vm_enable_cap(s, KVM_CAP_S390_RI, 0) == 0) {
 277             cap_ri = 1;
 278         }
 279     }
 280
 281     return 0;
 282 }
 283
 284 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
 285 {
 286     return cpu->cpu_index;
 287 }
 288
 289 int kvm_arch_init_vcpu(CPUState *cs)
 290 {
 291     S390CPU *cpu = S390_CPU(cs);
 292     kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
 293     cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE);
 294     return 0;
 295 }
 296
 297 void kvm_s390_reset_vcpu(S390CPU *cpu)
 298 {
 299     CPUState *cs = CPU(cpu);
 300
 301     /* The initial reset call is needed here to reset in-kernel
 302      * vcpu data that we can't access directly from QEMU
 303      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
 304      * Before this ioctl cpu_synchronize_state() is called in common kvm
 305      * code (kvm-all) */
 306     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
 307         error_report("Initial CPU reset failed on CPU %i", cs->cpu_index);
 308     }
 309 }
 310
 311 static int can_sync_regs(CPUState *cs, int regs)
 312 {
 313     return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
 314 }
 315
 316 int kvm_arch_put_registers(CPUState *cs, int level)
 317 {
 318     S390CPU *cpu = S390_CPU(cs);
 319     CPUS390XState *env = &cpu->env;
 320     struct kvm_sregs sregs;
 321     struct kvm_regs regs;
 322     struct kvm_fpu fpu = {};
 323     int r;
 324     int i;
 325
 326     /* always save the PSW  and the GPRS*/
 327     cs->kvm_run->psw_addr = env->psw.addr;
 328     cs->kvm_run->psw_mask = env->psw.mask;
 329
 330     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 331         for (i = 0; i < 16; i++) {
 332             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
 333             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
 334         }
 335     } else {
 336         for (i = 0; i < 16; i++) {
 337             regs.gprs[i] = env->regs[i];
 338         }
 339         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
 340         if (r < 0) {
 341             return r;
 342         }
 343     }
 344
 345     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 346         for (i = 0; i < 32; i++) {
 347             cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
 348             cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
 349         }
 350         cs->kvm_run->s.regs.fpc = env->fpc;
 351         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
 352     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
 353         for (i = 0; i < 16; i++) {
 354             cs->kvm_run->s.regs.fprs[i] = get_freg(env, i)->ll;
 355         }
 356         cs->kvm_run->s.regs.fpc = env->fpc;
 357         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_FPRS;
 358     } else {
 359         /* Floating point */
 360         for (i = 0; i < 16; i++) {
 361             fpu.fprs[i] = get_freg(env, i)->ll;
 362         }
 363         fpu.fpc = env->fpc;
 364
 365         r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
 366         if (r < 0) {
 367             return r;
 368         }
 369     }
 370
 371     /* Do we need to save more than that? */
 372     if (level == KVM_PUT_RUNTIME_STATE) {
 373         return 0;
 374     }
 375
 376     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 377         cs->kvm_run->s.regs.cputm = env->cputm;
 378         cs->kvm_run->s.regs.ckc = env->ckc;
 379         cs->kvm_run->s.regs.todpr = env->todpr;
 380         cs->kvm_run->s.regs.gbea = env->gbea;
 381         cs->kvm_run->s.regs.pp = env->pp;
 382         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
 383     } else {
 384         /*
 385          * These ONE_REGS are not protected by a capability. As they are only
 386          * necessary for migration we just trace a possible error, but don't
 387          * return with an error return code.
 388          */
 389         kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 390         kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 391         kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 392         kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 393         kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 394     }
 395
 396     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
 397         memcpy(cs->kvm_run->s.regs.riccb, env->riccb, 64);
 398         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_RICCB;
 399     }
 400
 401     /* pfault parameters */
 402     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 403         cs->kvm_run->s.regs.pft = env->pfault_token;
 404         cs->kvm_run->s.regs.pfs = env->pfault_select;
 405         cs->kvm_run->s.regs.pfc = env->pfault_compare;
 406         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
 407     } else if (cap_async_pf) {
 408         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 409         if (r < 0) {
 410             return r;
 411         }
 412         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 413         if (r < 0) {
 414             return r;
 415         }
 416         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 417         if (r < 0) {
 418             return r;
 419         }
 420     }
 421
 422     /* access registers and control registers*/
 423     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 424         for (i = 0; i < 16; i++) {
 425             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
 426             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
 427         }
 428         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
 429         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
 430     } else {
 431         for (i = 0; i < 16; i++) {
 432             sregs.acrs[i] = env->aregs[i];
 433             sregs.crs[i] = env->cregs[i];
 434         }
 435         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
 436         if (r < 0) {
 437             return r;
 438         }
 439     }
 440
 441     /* Finally the prefix */
 442     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 443         cs->kvm_run->s.regs.prefix = env->psa;
 444         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
 445     } else {
 446         /* prefix is only supported via sync regs */
 447     }
 448     return 0;
 449 }
 450
 451 int kvm_arch_get_registers(CPUState *cs)
 452 {
 453     S390CPU *cpu = S390_CPU(cs);
 454     CPUS390XState *env = &cpu->env;
 455     struct kvm_sregs sregs;
 456     struct kvm_regs regs;
 457     struct kvm_fpu fpu;
 458     int i, r;
 459
 460     /* get the PSW */
 461     env->psw.addr = cs->kvm_run->psw_addr;
 462     env->psw.mask = cs->kvm_run->psw_mask;
 463
 464     /* the GPRS */
 465     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 466         for (i = 0; i < 16; i++) {
 467             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
 468         }
 469     } else {
 470         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
 471         if (r < 0) {
 472             return r;
 473         }
 474          for (i = 0; i < 16; i++) {
 475             env->regs[i] = regs.gprs[i];
 476         }
 477     }
 478
 479     /* The ACRS and CRS */
 480     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 481         for (i = 0; i < 16; i++) {
 482             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
 483             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
 484         }
 485     } else {
 486         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
 487         if (r < 0) {
 488             return r;
 489         }
 490          for (i = 0; i < 16; i++) {
 491             env->aregs[i] = sregs.acrs[i];
 492             env->cregs[i] = sregs.crs[i];
 493         }
 494     }
 495
 496     /* Floating point and vector registers */
 497     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 498         for (i = 0; i < 32; i++) {
 499             env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
 500             env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
 501         }
 502         env->fpc = cs->kvm_run->s.regs.fpc;
 503     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
 504         for (i = 0; i < 16; i++) {
 505             get_freg(env, i)->ll = cs->kvm_run->s.regs.fprs[i];
 506         }
 507         env->fpc = cs->kvm_run->s.regs.fpc;
 508     } else {
 509         r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
 510         if (r < 0) {
 511             return r;
 512         }
 513         for (i = 0; i < 16; i++) {
 514             get_freg(env, i)->ll = fpu.fprs[i];
 515         }
 516         env->fpc = fpu.fpc;
 517     }
 518
 519     /* The prefix */
 520     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 521         env->psa = cs->kvm_run->s.regs.prefix;
 522     }
 523
 524     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 525         env->cputm = cs->kvm_run->s.regs.cputm;
 526         env->ckc = cs->kvm_run->s.regs.ckc;
 527         env->todpr = cs->kvm_run->s.regs.todpr;
 528         env->gbea = cs->kvm_run->s.regs.gbea;
 529         env->pp = cs->kvm_run->s.regs.pp;
 530     } else {
 531         /*
 532          * These ONE_REGS are not protected by a capability. As they are only
 533          * necessary for migration we just trace a possible error, but don't
 534          * return with an error return code.
 535          */
 536         kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 537         kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 538         kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 539         kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 540         kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 541     }
 542
 543     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
 544         memcpy(env->riccb, cs->kvm_run->s.regs.riccb, 64);
 545     }
 546
 547     /* pfault parameters */
 548     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 549         env->pfault_token = cs->kvm_run->s.regs.pft;
 550         env->pfault_select = cs->kvm_run->s.regs.pfs;
 551         env->pfault_compare = cs->kvm_run->s.regs.pfc;
 552     } else if (cap_async_pf) {
 553         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 554         if (r < 0) {
 555             return r;
 556         }
 557         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 558         if (r < 0) {
 559             return r;
 560         }
 561         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 562         if (r < 0) {
 563             return r;
 564         }
 565     }
 566
 567     return 0;
 568 }
 569
 570 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
 571 {
 572     int r;
 573     struct kvm_device_attr attr = {
 574         .group = KVM_S390_VM_TOD,
 575         .attr = KVM_S390_VM_TOD_LOW,
 576         .addr = (uint64_t)tod_low,
 577     };
 578
 579     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 580     if (r) {
 581         return r;
 582     }
 583
 584     attr.attr = KVM_S390_VM_TOD_HIGH;
 585     attr.addr = (uint64_t)tod_high;
 586     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 587 }
 588
 589 int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_low)
 590 {
 591     int r;
 592
 593     struct kvm_device_attr attr = {
 594         .group = KVM_S390_VM_TOD,
 595         .attr = KVM_S390_VM_TOD_LOW,
 596         .addr = (uint64_t)tod_low,
 597     };
 598
 599     r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 600     if (r) {
 601         return r;
 602     }
 603
 604     attr.attr = KVM_S390_VM_TOD_HIGH;
 605     attr.addr = (uint64_t)tod_high;
 606     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 607 }
 608
 609 /**
 610  * kvm_s390_mem_op:
 611  * @addr:      the logical start address in guest memory
 612  * @ar:        the access register number
 613  * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
 614  * @len:       length that should be transferred
 615  * @is_write:  true = write, false = read
 616  * Returns:    0 on success, non-zero if an exception or error occurred
 617  *
 618  * Use KVM ioctl to read/write from/to guest memory. An access exception
 619  * is injected into the vCPU in case of translation errors.
 620  */
 621 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
 622                     int len, bool is_write)
 623 {
 624     struct kvm_s390_mem_op mem_op = {
 625         .gaddr = addr,
 626         .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
 627         .size = len,
 628         .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
 629                        : KVM_S390_MEMOP_LOGICAL_READ,
 630         .buf = (uint64_t)hostbuf,
 631         .ar = ar,
 632     };
 633     int ret;
 634
 635     if (!cap_mem_op) {
 636         return -ENOSYS;
 637     }
 638     if (!hostbuf) {
 639         mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
 640     }
 641
 642     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
 643     if (ret < 0) {
 644         error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
 645     }
 646     return ret;
 647 }
 648
 649 /*
 650  * Legacy layout for s390:
 651  * Older S390 KVM requires the topmost vma of the RAM to be
 652  * smaller than an system defined value, which is at least 256GB.
 653  * Larger systems have larger values. We put the guest between
 654  * the end of data segment (system break) and this value. We
 655  * use 32GB as a base to have enough room for the system break
 656  * to grow. We also have to use MAP parameters that avoid
 657  * read-only mapping of guest pages.
 658  */
 659 static void *legacy_s390_alloc(size_t size, uint64_t *align)
 660 {
 661     void *mem;
 662
 663     mem = mmap((void *) 0x800000000ULL, size,
 664                PROT_EXEC|PROT_READ|PROT_WRITE,
 665                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 666     return mem == MAP_FAILED ? NULL : mem;
 667 }
 668
 669 /* DIAG 501 is used for sw breakpoints */
 670 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 671
 672 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 673 {
 674
 675     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 676                             sizeof(diag_501), 0) ||
 677         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501,
 678                             sizeof(diag_501), 1)) {
 679         return -EINVAL;
 680     }
 681     return 0;
 682 }
 683
 684 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 685 {
 686     uint8_t t[sizeof(diag_501)];
 687
 688     if (cpu_memory_rw_debug(cs, bp->pc, t, sizeof(diag_501), 0)) {
 689         return -EINVAL;
 690     } else if (memcmp(t, diag_501, sizeof(diag_501))) {
 691         return -EINVAL;
 692     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 693                                    sizeof(diag_501), 1)) {
 694         return -EINVAL;
 695     }
 696
 697     return 0;
 698 }
 699
 700 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
 701                                                     int len, int type)
 702 {
 703     int n;
 704
 705     for (n = 0; n < nb_hw_breakpoints; n++) {
 706         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
 707             (hw_breakpoints[n].len == len || len == -1)) {
 708             return &hw_breakpoints[n];
 709         }
 710     }
 711
 712     return NULL;
 713 }
 714
 715 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
 716 {
 717     int size;
 718
 719     if (find_hw_breakpoint(addr, len, type)) {
 720         return -EEXIST;
 721     }
 722
 723     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
 724
 725     if (!hw_breakpoints) {
 726         nb_hw_breakpoints = 0;
 727         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
 728     } else {
 729         hw_breakpoints =
 730             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
 731     }
 732
 733     if (!hw_breakpoints) {
 734         nb_hw_breakpoints = 0;
 735         return -ENOMEM;
 736     }
 737
 738     hw_breakpoints[nb_hw_breakpoints].addr = addr;
 739     hw_breakpoints[nb_hw_breakpoints].len = len;
 740     hw_breakpoints[nb_hw_breakpoints].type = type;
 741
 742     nb_hw_breakpoints++;
 743
 744     return 0;
 745 }
 746
 747 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 748                                   target_ulong len, int type)
 749 {
 750     switch (type) {
 751     case GDB_BREAKPOINT_HW:
 752         type = KVM_HW_BP;
 753         break;
 754     case GDB_WATCHPOINT_WRITE:
 755         if (len < 1) {
 756             return -EINVAL;
 757         }
 758         type = KVM_HW_WP_WRITE;
 759         break;
 760     default:
 761         return -ENOSYS;
 762     }
 763     return insert_hw_breakpoint(addr, len, type);
 764 }
 765
 766 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 767                                   target_ulong len, int type)
 768 {
 769     int size;
 770     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
 771
 772     if (bp == NULL) {
 773         return -ENOENT;
 774     }
 775
 776     nb_hw_breakpoints--;
 777     if (nb_hw_breakpoints > 0) {
 778         /*
 779          * In order to trim the array, move the last element to the position to
 780          * be removed - if necessary.
 781          */
 782         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
 783             *bp = hw_breakpoints[nb_hw_breakpoints];
 784         }
 785         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
 786         hw_breakpoints =
 787              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
 788     } else {
 789         g_free(hw_breakpoints);
 790         hw_breakpoints = NULL;
 791     }
 792
 793     return 0;
 794 }
 795
 796 void kvm_arch_remove_all_hw_breakpoints(void)
 797 {
 798     nb_hw_breakpoints = 0;
 799     g_free(hw_breakpoints);
 800     hw_breakpoints = NULL;
 801 }
 802
 803 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 804 {
 805     int i;
 806
 807     if (nb_hw_breakpoints > 0) {
 808         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
 809         dbg->arch.hw_bp = hw_breakpoints;
 810
 811         for (i = 0; i < nb_hw_breakpoints; ++i) {
 812             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
 813                                                        hw_breakpoints[i].addr);
 814         }
 815         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 816     } else {
 817         dbg->arch.nr_hw_bp = 0;
 818         dbg->arch.hw_bp = NULL;
 819     }
 820 }
 821
 822 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 823 {
 824 }
 825
 826 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
 827 {
 828     return MEMTXATTRS_UNSPECIFIED;
 829 }
 830
 831 int kvm_arch_process_async_events(CPUState *cs)
 832 {
 833     return cs->halted;
 834 }
 835
 836 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
 837                                      struct kvm_s390_interrupt *interrupt)
 838 {
 839     int r = 0;
 840
 841     interrupt->type = irq->type;
 842     switch (irq->type) {
 843     case KVM_S390_INT_VIRTIO:
 844         interrupt->parm = irq->u.ext.ext_params;
 845         /* fall through */
 846     case KVM_S390_INT_PFAULT_INIT:
 847     case KVM_S390_INT_PFAULT_DONE:
 848         interrupt->parm64 = irq->u.ext.ext_params2;
 849         break;
 850     case KVM_S390_PROGRAM_INT:
 851         interrupt->parm = irq->u.pgm.code;
 852         break;
 853     case KVM_S390_SIGP_SET_PREFIX:
 854         interrupt->parm = irq->u.prefix.address;
 855         break;
 856     case KVM_S390_INT_SERVICE:
 857         interrupt->parm = irq->u.ext.ext_params;
 858         break;
 859     case KVM_S390_MCHK:
 860         interrupt->parm = irq->u.mchk.cr14;
 861         interrupt->parm64 = irq->u.mchk.mcic;
 862         break;
 863     case KVM_S390_INT_EXTERNAL_CALL:
 864         interrupt->parm = irq->u.extcall.code;
 865         break;
 866     case KVM_S390_INT_EMERGENCY:
 867         interrupt->parm = irq->u.emerg.code;
 868         break;
 869     case KVM_S390_SIGP_STOP:
 870     case KVM_S390_RESTART:
 871         break; /* These types have no parameters */
 872     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
 873         interrupt->parm = irq->u.io.subchannel_id << 16;
 874         interrupt->parm |= irq->u.io.subchannel_nr;
 875         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
 876         interrupt->parm64 |= irq->u.io.io_int_word;
 877         break;
 878     default:
 879         r = -EINVAL;
 880         break;
 881     }
 882     return r;
 883 }
 884
 885 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
 886 {
 887     struct kvm_s390_interrupt kvmint = {};
 888     int r;
 889
 890     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 891     if (r < 0) {
 892         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 893         exit(1);
 894     }
 895
 896     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
 897     if (r < 0) {
 898         fprintf(stderr, "KVM failed to inject interrupt\n");
 899         exit(1);
 900     }
 901 }
 902
 903 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
 904 {
 905     CPUState *cs = CPU(cpu);
 906     int r;
 907
 908     if (cap_s390_irq) {
 909         r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
 910         if (!r) {
 911             return;
 912         }
 913         error_report("KVM failed to inject interrupt %llx", irq->type);
 914         exit(1);
 915     }
 916
 917     inject_vcpu_irq_legacy(cs, irq);
 918 }
 919
 920 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 921 {
 922     struct kvm_s390_interrupt kvmint = {};
 923     int r;
 924
 925     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 926     if (r < 0) {
 927         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 928         exit(1);
 929     }
 930
 931     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
 932     if (r < 0) {
 933         fprintf(stderr, "KVM failed to inject interrupt\n");
 934         exit(1);
 935     }
 936 }
 937
 938 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 939 {
 940     static bool use_flic = true;
 941     int r;
 942
 943     if (use_flic) {
 944         r = kvm_s390_inject_flic(irq);
 945         if (r == -ENOSYS) {
 946             use_flic = false;
 947         }
 948         if (!r) {
 949             return;
 950         }
 951     }
 952     __kvm_s390_floating_interrupt(irq);
 953 }
 954
 955 void kvm_s390_service_interrupt(uint32_t parm)
 956 {
 957     struct kvm_s390_irq irq = {
 958         .type = KVM_S390_INT_SERVICE,
 959         .u.ext.ext_params = parm,
 960     };
 961
 962     kvm_s390_floating_interrupt(&irq);
 963 }
 964
 965 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
 966 {
 967     struct kvm_s390_irq irq = {
 968         .type = KVM_S390_PROGRAM_INT,
 969         .u.pgm.code = code,
 970     };
 971
 972     kvm_s390_vcpu_interrupt(cpu, &irq);
 973 }
 974
 975 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
 976 {
 977     struct kvm_s390_irq irq = {
 978         .type = KVM_S390_PROGRAM_INT,
 979         .u.pgm.code = code,
 980         .u.pgm.trans_exc_code = te_code,
 981         .u.pgm.exc_access_id = te_code & 3,
 982     };
 983
 984     kvm_s390_vcpu_interrupt(cpu, &irq);
 985 }
 986
 987 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
 988                                  uint16_t ipbh0)
 989 {
 990     CPUS390XState *env = &cpu->env;
 991     uint64_t sccb;
 992     uint32_t code;
 993     int r = 0;
 994
 995     cpu_synchronize_state(CPU(cpu));
 996     sccb = env->regs[ipbh0 & 0xf];
 997     code = env->regs[(ipbh0 & 0xf0) >> 4];
 998
 999     r = sclp_service_call(env, sccb, code);
1000     if (r < 0) {
1001         enter_pgmcheck(cpu, -r);
1002     } else {
1003         setcc(cpu, r);
1004     }
1005
1006     return 0;
1007 }
1008
1009 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1010 {
1011     CPUS390XState *env = &cpu->env;
1012     int rc = 0;
1013     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1014
1015     cpu_synchronize_state(CPU(cpu));
1016
1017     switch (ipa1) {
1018     case PRIV_B2_XSCH:
1019         ioinst_handle_xsch(cpu, env->regs[1]);
1020         break;
1021     case PRIV_B2_CSCH:
1022         ioinst_handle_csch(cpu, env->regs[1]);
1023         break;
1024     case PRIV_B2_HSCH:
1025         ioinst_handle_hsch(cpu, env->regs[1]);
1026         break;
1027     case PRIV_B2_MSCH:
1028         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
1029         break;
1030     case PRIV_B2_SSCH:
1031         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
1032         break;
1033     case PRIV_B2_STCRW:
1034         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
1035         break;
1036     case PRIV_B2_STSCH:
1037         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
1038         break;
1039     case PRIV_B2_TSCH:
1040         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1041         fprintf(stderr, "Spurious tsch intercept\n");
1042         break;
1043     case PRIV_B2_CHSC:
1044         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
1045         break;
1046     case PRIV_B2_TPI:
1047         /* This should have been handled by kvm already. */
1048         fprintf(stderr, "Spurious tpi intercept\n");
1049         break;
1050     case PRIV_B2_SCHM:
1051         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1052                            run->s390_sieic.ipb);
1053         break;
1054     case PRIV_B2_RSCH:
1055         ioinst_handle_rsch(cpu, env->regs[1]);
1056         break;
1057     case PRIV_B2_RCHP:
1058         ioinst_handle_rchp(cpu, env->regs[1]);
1059         break;
1060     case PRIV_B2_STCPS:
1061         /* We do not provide this instruction, it is suppressed. */
1062         break;
1063     case PRIV_B2_SAL:
1064         ioinst_handle_sal(cpu, env->regs[1]);
1065         break;
1066     case PRIV_B2_SIGA:
1067         /* Not provided, set CC = 3 for subchannel not operational */
1068         setcc(cpu, 3);
1069         break;
1070     case PRIV_B2_SCLP_CALL:
1071         rc = kvm_sclp_service_call(cpu, run, ipbh0);
1072         break;
1073     default:
1074         rc = -1;
1075         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1076         break;
1077     }
1078
1079     return rc;
1080 }
1081
1082 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1083                                   uint8_t *ar)
1084 {
1085     CPUS390XState *env = &cpu->env;
1086     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1087     uint32_t base2 = run->s390_sieic.ipb >> 28;
1088     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1089                      ((run->s390_sieic.ipb & 0xff00) << 4);
1090
1091     if (disp2 & 0x80000) {
1092         disp2 += 0xfff00000;
1093     }
1094     if (ar) {
1095         *ar = base2;
1096     }
1097
1098     return (base2 ? env->regs[base2] : 0) +
1099            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1100 }
1101
1102 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1103                                   uint8_t *ar)
1104 {
1105     CPUS390XState *env = &cpu->env;
1106     uint32_t base2 = run->s390_sieic.ipb >> 28;
1107     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1108                      ((run->s390_sieic.ipb & 0xff00) << 4);
1109
1110     if (disp2 & 0x80000) {
1111         disp2 += 0xfff00000;
1112     }
1113     if (ar) {
1114         *ar = base2;
1115     }
1116
1117     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1118 }
1119
1120 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1121 {
1122     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1123
1124     return clp_service_call(cpu, r2);
1125 }
1126
1127 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1128 {
1129     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1130     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1131
1132     return pcilg_service_call(cpu, r1, r2);
1133 }
1134
1135 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1136 {
1137     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1138     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1139
1140     return pcistg_service_call(cpu, r1, r2);
1141 }
1142
1143 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1144 {
1145     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1146     uint64_t fiba;
1147     uint8_t ar;
1148
1149     cpu_synchronize_state(CPU(cpu));
1150     fiba = get_base_disp_rxy(cpu, run, &ar);
1151
1152     return stpcifc_service_call(cpu, r1, fiba, ar);
1153 }
1154
1155 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1156 {
1157     /* NOOP */
1158     return 0;
1159 }
1160
1161 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1162 {
1163     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1164     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1165
1166     return rpcit_service_call(cpu, r1, r2);
1167 }
1168
1169 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1170 {
1171     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1172     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1173     uint64_t gaddr;
1174     uint8_t ar;
1175
1176     cpu_synchronize_state(CPU(cpu));
1177     gaddr = get_base_disp_rsy(cpu, run, &ar);
1178
1179     return pcistb_service_call(cpu, r1, r3, gaddr, ar);
1180 }
1181
1182 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1183 {
1184     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1185     uint64_t fiba;
1186     uint8_t ar;
1187
1188     cpu_synchronize_state(CPU(cpu));
1189     fiba = get_base_disp_rxy(cpu, run, &ar);
1190
1191     return mpcifc_service_call(cpu, r1, fiba, ar);
1192 }
1193
1194 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1195 {
1196     int r = 0;
1197
1198     switch (ipa1) {
1199     case PRIV_B9_CLP:
1200         r = kvm_clp_service_call(cpu, run);
1201         break;
1202     case PRIV_B9_PCISTG:
1203         r = kvm_pcistg_service_call(cpu, run);
1204         break;
1205     case PRIV_B9_PCILG:
1206         r = kvm_pcilg_service_call(cpu, run);
1207         break;
1208     case PRIV_B9_RPCIT:
1209         r = kvm_rpcit_service_call(cpu, run);
1210         break;
1211     case PRIV_B9_EQBS:
1212         /* just inject exception */
1213         r = -1;
1214         break;
1215     default:
1216         r = -1;
1217         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1218         break;
1219     }
1220
1221     return r;
1222 }
1223
1224 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1225 {
1226     int r = 0;
1227
1228     switch (ipbl) {
1229     case PRIV_EB_PCISTB:
1230         r = kvm_pcistb_service_call(cpu, run);
1231         break;
1232     case PRIV_EB_SIC:
1233         r = kvm_sic_service_call(cpu, run);
1234         break;
1235     case PRIV_EB_SQBS:
1236         /* just inject exception */
1237         r = -1;
1238         break;
1239     default:
1240         r = -1;
1241         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1242         break;
1243     }
1244
1245     return r;
1246 }
1247
1248 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1249 {
1250     int r = 0;
1251
1252     switch (ipbl) {
1253     case PRIV_E3_MPCIFC:
1254         r = kvm_mpcifc_service_call(cpu, run);
1255         break;
1256     case PRIV_E3_STPCIFC:
1257         r = kvm_stpcifc_service_call(cpu, run);
1258         break;
1259     default:
1260         r = -1;
1261         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1262         break;
1263     }
1264
1265     return r;
1266 }
1267
1268 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1269 {
1270     CPUS390XState *env = &cpu->env;
1271     int ret;
1272
1273     cpu_synchronize_state(CPU(cpu));
1274     ret = s390_virtio_hypercall(env);
1275     if (ret == -EINVAL) {
1276         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1277         return 0;
1278     }
1279
1280     return ret;
1281 }
1282
1283 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1284 {
1285     uint64_t r1, r3;
1286     int rc;
1287
1288     cpu_synchronize_state(CPU(cpu));
1289     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1290     r3 = run->s390_sieic.ipa & 0x000f;
1291     rc = handle_diag_288(&cpu->env, r1, r3);
1292     if (rc) {
1293         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1294     }
1295 }
1296
1297 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1298 {
1299     uint64_t r1, r3;
1300
1301     cpu_synchronize_state(CPU(cpu));
1302     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1303     r3 = run->s390_sieic.ipa & 0x000f;
1304     handle_diag_308(&cpu->env, r1, r3);
1305 }
1306
1307 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1308 {
1309     CPUS390XState *env = &cpu->env;
1310     unsigned long pc;
1311
1312     cpu_synchronize_state(CPU(cpu));
1313
1314     pc = env->psw.addr - 4;
1315     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1316         env->psw.addr = pc;
1317         return EXCP_DEBUG;
1318     }
1319
1320     return -ENOENT;
1321 }
1322
1323 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1324
1325 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1326 {
1327     int r = 0;
1328     uint16_t func_code;
1329
1330     /*
1331      * For any diagnose call we support, bits 48-63 of the resulting
1332      * address specify the function code; the remainder is ignored.
1333      */
1334     func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1335     switch (func_code) {
1336     case DIAG_TIMEREVENT:
1337         kvm_handle_diag_288(cpu, run);
1338         break;
1339     case DIAG_IPL:
1340         kvm_handle_diag_308(cpu, run);
1341         break;
1342     case DIAG_KVM_HYPERCALL:
1343         r = handle_hypercall(cpu, run);
1344         break;
1345     case DIAG_KVM_BREAKPOINT:
1346         r = handle_sw_breakpoint(cpu, run);
1347         break;
1348     default:
1349         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1350         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1351         break;
1352     }
1353
1354     return r;
1355 }
1356
1357 typedef struct SigpInfo {
1358     S390CPU *cpu;
1359     uint64_t param;
1360     int cc;
1361     uint64_t *status_reg;
1362 } SigpInfo;
1363
1364 static void set_sigp_status(SigpInfo *si, uint64_t status)
1365 {
1366     *si->status_reg &= 0xffffffff00000000ULL;
1367     *si->status_reg |= status;
1368     si->cc = SIGP_CC_STATUS_STORED;
1369 }
1370
1371 static void sigp_start(void *arg)
1372 {
1373     SigpInfo *si = arg;
1374
1375     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1376         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1377         return;
1378     }
1379
1380     s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1381     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1382 }
1383
1384 static void sigp_stop(void *arg)
1385 {
1386     SigpInfo *si = arg;
1387     struct kvm_s390_irq irq = {
1388         .type = KVM_S390_SIGP_STOP,
1389     };
1390
1391     if (s390_cpu_get_state(si->cpu) != CPU_STATE_OPERATING) {
1392         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1393         return;
1394     }
1395
1396     /* disabled wait - sleeping in user space */
1397     if (CPU(si->cpu)->halted) {
1398         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1399     } else {
1400         /* execute the stop function */
1401         si->cpu->env.sigp_order = SIGP_STOP;
1402         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1403     }
1404     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1405 }
1406
1407 #define ADTL_SAVE_AREA_SIZE 1024
1408 static int kvm_s390_store_adtl_status(S390CPU *cpu, hwaddr addr)
1409 {
1410     void *mem;
1411     hwaddr len = ADTL_SAVE_AREA_SIZE;
1412
1413     mem = cpu_physical_memory_map(addr, &len, 1);
1414     if (!mem) {
1415         return -EFAULT;
1416     }
1417     if (len != ADTL_SAVE_AREA_SIZE) {
1418         cpu_physical_memory_unmap(mem, len, 1, 0);
1419         return -EFAULT;
1420     }
1421
1422     memcpy(mem, &cpu->env.vregs, 512);
1423
1424     cpu_physical_memory_unmap(mem, len, 1, len);
1425
1426     return 0;
1427 }
1428
1429 #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, floating_pt_save_area)
1430 #define SAVE_AREA_SIZE 512
1431 static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
1432 {
1433     static const uint8_t ar_id = 1;
1434     uint64_t ckc = cpu->env.ckc >> 8;
1435     void *mem;
1436     int i;
1437     hwaddr len = SAVE_AREA_SIZE;
1438
1439     mem = cpu_physical_memory_map(addr, &len, 1);
1440     if (!mem) {
1441         return -EFAULT;
1442     }
1443     if (len != SAVE_AREA_SIZE) {
1444         cpu_physical_memory_unmap(mem, len, 1, 0);
1445         return -EFAULT;
1446     }
1447
1448     if (store_arch) {
1449         cpu_physical_memory_write(offsetof(LowCore, ar_access_id), &ar_id, 1);
1450     }
1451     for (i = 0; i < 16; ++i) {
1452         *((uint64_t *)mem + i) = get_freg(&cpu->env, i)->ll;
1453     }
1454     memcpy(mem + 128, &cpu->env.regs, 128);
1455     memcpy(mem + 256, &cpu->env.psw, 16);
1456     memcpy(mem + 280, &cpu->env.psa, 4);
1457     memcpy(mem + 284, &cpu->env.fpc, 4);
1458     memcpy(mem + 292, &cpu->env.todpr, 4);
1459     memcpy(mem + 296, &cpu->env.cputm, 8);
1460     memcpy(mem + 304, &ckc, 8);
1461     memcpy(mem + 320, &cpu->env.aregs, 64);
1462     memcpy(mem + 384, &cpu->env.cregs, 128);
1463
1464     cpu_physical_memory_unmap(mem, len, 1, len);
1465
1466     return 0;
1467 }
1468
1469 static void sigp_stop_and_store_status(void *arg)
1470 {
1471     SigpInfo *si = arg;
1472     struct kvm_s390_irq irq = {
1473         .type = KVM_S390_SIGP_STOP,
1474     };
1475
1476     /* disabled wait - sleeping in user space */
1477     if (s390_cpu_get_state(si->cpu) == CPU_STATE_OPERATING &&
1478         CPU(si->cpu)->halted) {
1479         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1480     }
1481
1482     switch (s390_cpu_get_state(si->cpu)) {
1483     case CPU_STATE_OPERATING:
1484         si->cpu->env.sigp_order = SIGP_STOP_STORE_STATUS;
1485         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1486         /* store will be performed when handling the stop intercept */
1487         break;
1488     case CPU_STATE_STOPPED:
1489         /* already stopped, just store the status */
1490         cpu_synchronize_state(CPU(si->cpu));
1491         kvm_s390_store_status(si->cpu, KVM_S390_STORE_STATUS_DEF_ADDR, true);
1492         break;
1493     }
1494     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1495 }
1496
1497 static void sigp_store_status_at_address(void *arg)
1498 {
1499     SigpInfo *si = arg;
1500     uint32_t address = si->param & 0x7ffffe00u;
1501
1502     /* cpu has to be stopped */
1503     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1504         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1505         return;
1506     }
1507
1508     cpu_synchronize_state(CPU(si->cpu));
1509
1510     if (kvm_s390_store_status(si->cpu, address, false)) {
1511         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1512         return;
1513     }
1514     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1515 }
1516
1517 static void sigp_store_adtl_status(void *arg)
1518 {
1519     SigpInfo *si = arg;
1520
1521     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_VECTOR_REGISTERS)) {
1522         set_sigp_status(si, SIGP_STAT_INVALID_ORDER);
1523         return;
1524     }
1525
1526     /* cpu has to be stopped */
1527     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1528         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1529         return;
1530     }
1531
1532     /* parameter must be aligned to 1024-byte boundary */
1533     if (si->param & 0x3ff) {
1534         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1535         return;
1536     }
1537
1538     cpu_synchronize_state(CPU(si->cpu));
1539
1540     if (kvm_s390_store_adtl_status(si->cpu, si->param)) {
1541         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1542         return;
1543     }
1544     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1545 }
1546
1547 static void sigp_restart(void *arg)
1548 {
1549     SigpInfo *si = arg;
1550     struct kvm_s390_irq irq = {
1551         .type = KVM_S390_RESTART,
1552     };
1553
1554     switch (s390_cpu_get_state(si->cpu)) {
1555     case CPU_STATE_STOPPED:
1556         /* the restart irq has to be delivered prior to any other pending irq */
1557         cpu_synchronize_state(CPU(si->cpu));
1558         do_restart_interrupt(&si->cpu->env);
1559         s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1560         break;
1561     case CPU_STATE_OPERATING:
1562         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1563         break;
1564     }
1565     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1566 }
1567
1568 int kvm_s390_cpu_restart(S390CPU *cpu)
1569 {
1570     SigpInfo si = {
1571         .cpu = cpu,
1572     };
1573
1574     run_on_cpu(CPU(cpu), sigp_restart, &si);
1575     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1576     return 0;
1577 }
1578
1579 static void sigp_initial_cpu_reset(void *arg)
1580 {
1581     SigpInfo *si = arg;
1582     CPUState *cs = CPU(si->cpu);
1583     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1584
1585     cpu_synchronize_state(cs);
1586     scc->initial_cpu_reset(cs);
1587     cpu_synchronize_post_reset(cs);
1588     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1589 }
1590
1591 static void sigp_cpu_reset(void *arg)
1592 {
1593     SigpInfo *si = arg;
1594     CPUState *cs = CPU(si->cpu);
1595     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1596
1597     cpu_synchronize_state(cs);
1598     scc->cpu_reset(cs);
1599     cpu_synchronize_post_reset(cs);
1600     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1601 }
1602
1603 static void sigp_set_prefix(void *arg)
1604 {
1605     SigpInfo *si = arg;
1606     uint32_t addr = si->param & 0x7fffe000u;
1607
1608     cpu_synchronize_state(CPU(si->cpu));
1609
1610     if (!address_space_access_valid(&address_space_memory, addr,
1611                                     sizeof(struct LowCore), false)) {
1612         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1613         return;
1614     }
1615
1616     /* cpu has to be stopped */
1617     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1618         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1619         return;
1620     }
1621
1622     si->cpu->env.psa = addr;
1623     cpu_synchronize_post_init(CPU(si->cpu));
1624     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1625 }
1626
1627 static int handle_sigp_single_dst(S390CPU *dst_cpu, uint8_t order,
1628                                   uint64_t param, uint64_t *status_reg)
1629 {
1630     SigpInfo si = {
1631         .cpu = dst_cpu,
1632         .param = param,
1633         .status_reg = status_reg,
1634     };
1635
1636     /* cpu available? */
1637     if (dst_cpu == NULL) {
1638         return SIGP_CC_NOT_OPERATIONAL;
1639     }
1640
1641     /* only resets can break pending orders */
1642     if (dst_cpu->env.sigp_order != 0 &&
1643         order != SIGP_CPU_RESET &&
1644         order != SIGP_INITIAL_CPU_RESET) {
1645         return SIGP_CC_BUSY;
1646     }
1647
1648     switch (order) {
1649     case SIGP_START:
1650         run_on_cpu(CPU(dst_cpu), sigp_start, &si);
1651         break;
1652     case SIGP_STOP:
1653         run_on_cpu(CPU(dst_cpu), sigp_stop, &si);
1654         break;
1655     case SIGP_RESTART:
1656         run_on_cpu(CPU(dst_cpu), sigp_restart, &si);
1657         break;
1658     case SIGP_STOP_STORE_STATUS:
1659         run_on_cpu(CPU(dst_cpu), sigp_stop_and_store_status, &si);
1660         break;
1661     case SIGP_STORE_STATUS_ADDR:
1662         run_on_cpu(CPU(dst_cpu), sigp_store_status_at_address, &si);
1663         break;
1664     case SIGP_STORE_ADTL_STATUS:
1665         run_on_cpu(CPU(dst_cpu), sigp_store_adtl_status, &si);
1666         break;
1667     case SIGP_SET_PREFIX:
1668         run_on_cpu(CPU(dst_cpu), sigp_set_prefix, &si);
1669         break;
1670     case SIGP_INITIAL_CPU_RESET:
1671         run_on_cpu(CPU(dst_cpu), sigp_initial_cpu_reset, &si);
1672         break;
1673     case SIGP_CPU_RESET:
1674         run_on_cpu(CPU(dst_cpu), sigp_cpu_reset, &si);
1675         break;
1676     default:
1677         DPRINTF("KVM: unknown SIGP: 0x%x\n", order);
1678         set_sigp_status(&si, SIGP_STAT_INVALID_ORDER);
1679     }
1680
1681     return si.cc;
1682 }
1683
1684 static int sigp_set_architecture(S390CPU *cpu, uint32_t param,
1685                                  uint64_t *status_reg)
1686 {
1687     CPUState *cur_cs;
1688     S390CPU *cur_cpu;
1689
1690     /* due to the BQL, we are the only active cpu */
1691     CPU_FOREACH(cur_cs) {
1692         cur_cpu = S390_CPU(cur_cs);
1693         if (cur_cpu->env.sigp_order != 0) {
1694             return SIGP_CC_BUSY;
1695         }
1696         cpu_synchronize_state(cur_cs);
1697         /* all but the current one have to be stopped */
1698         if (cur_cpu != cpu &&
1699             s390_cpu_get_state(cur_cpu) != CPU_STATE_STOPPED) {
1700             *status_reg &= 0xffffffff00000000ULL;
1701             *status_reg |= SIGP_STAT_INCORRECT_STATE;
1702             return SIGP_CC_STATUS_STORED;
1703         }
1704     }
1705
1706     switch (param & 0xff) {
1707     case SIGP_MODE_ESA_S390:
1708         /* not supported */
1709         return SIGP_CC_NOT_OPERATIONAL;
1710     case SIGP_MODE_Z_ARCH_TRANS_ALL_PSW:
1711     case SIGP_MODE_Z_ARCH_TRANS_CUR_PSW:
1712         CPU_FOREACH(cur_cs) {
1713             cur_cpu = S390_CPU(cur_cs);
1714             cur_cpu->env.pfault_token = -1UL;
1715         }
1716         break;
1717     default:
1718         *status_reg &= 0xffffffff00000000ULL;
1719         *status_reg |= SIGP_STAT_INVALID_PARAMETER;
1720         return SIGP_CC_STATUS_STORED;
1721     }
1722
1723     return SIGP_CC_ORDER_CODE_ACCEPTED;
1724 }
1725
1726 #define SIGP_ORDER_MASK 0x000000ff
1727
1728 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1729 {
1730     CPUS390XState *env = &cpu->env;
1731     const uint8_t r1 = ipa1 >> 4;
1732     const uint8_t r3 = ipa1 & 0x0f;
1733     int ret;
1734     uint8_t order;
1735     uint64_t *status_reg;
1736     uint64_t param;
1737     S390CPU *dst_cpu = NULL;
1738
1739     cpu_synchronize_state(CPU(cpu));
1740
1741     /* get order code */
1742     order = decode_basedisp_rs(env, run->s390_sieic.ipb, NULL)
1743         & SIGP_ORDER_MASK;
1744     status_reg = &env->regs[r1];
1745     param = (r1 % 2) ? env->regs[r1] : env->regs[r1 + 1];
1746
1747     switch (order) {
1748     case SIGP_SET_ARCH:
1749         ret = sigp_set_architecture(cpu, param, status_reg);
1750         break;
1751     default:
1752         /* all other sigp orders target a single vcpu */
1753         dst_cpu = s390_cpu_addr2state(env->regs[r3]);
1754         ret = handle_sigp_single_dst(dst_cpu, order, param, status_reg);
1755     }
1756
1757     trace_kvm_sigp_finished(order, CPU(cpu)->cpu_index,
1758                             dst_cpu ? CPU(dst_cpu)->cpu_index : -1, ret);
1759
1760     if (ret >= 0) {
1761         setcc(cpu, ret);
1762         return 0;
1763     }
1764
1765     return ret;
1766 }
1767
1768 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1769 {
1770     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1771     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1772     int r = -1;
1773
1774     DPRINTF("handle_instruction 0x%x 0x%x\n",
1775             run->s390_sieic.ipa, run->s390_sieic.ipb);
1776     switch (ipa0) {
1777     case IPA0_B2:
1778         r = handle_b2(cpu, run, ipa1);
1779         break;
1780     case IPA0_B9:
1781         r = handle_b9(cpu, run, ipa1);
1782         break;
1783     case IPA0_EB:
1784         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1785         break;
1786     case IPA0_E3:
1787         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1788         break;
1789     case IPA0_DIAG:
1790         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1791         break;
1792     case IPA0_SIGP:
1793         r = handle_sigp(cpu, run, ipa1);
1794         break;
1795     }
1796
1797     if (r < 0) {
1798         r = 0;
1799         enter_pgmcheck(cpu, 0x0001);
1800     }
1801
1802     return r;
1803 }
1804
1805 static bool is_special_wait_psw(CPUState *cs)
1806 {
1807     /* signal quiesce */
1808     return cs->kvm_run->psw_addr == 0xfffUL;
1809 }
1810
1811 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1812 {
1813     CPUState *cs = CPU(cpu);
1814
1815     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1816                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1817                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1818     s390_cpu_halt(cpu);
1819     qemu_system_guest_panicked();
1820 }
1821
1822 static int handle_intercept(S390CPU *cpu)
1823 {
1824     CPUState *cs = CPU(cpu);
1825     struct kvm_run *run = cs->kvm_run;
1826     int icpt_code = run->s390_sieic.icptcode;
1827     int r = 0;
1828
1829     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1830             (long)cs->kvm_run->psw_addr);
1831     switch (icpt_code) {
1832         case ICPT_INSTRUCTION:
1833             r = handle_instruction(cpu, run);
1834             break;
1835         case ICPT_PROGRAM:
1836             unmanageable_intercept(cpu, "program interrupt",
1837                                    offsetof(LowCore, program_new_psw));
1838             r = EXCP_HALTED;
1839             break;
1840         case ICPT_EXT_INT:
1841             unmanageable_intercept(cpu, "external interrupt",
1842                                    offsetof(LowCore, external_new_psw));
1843             r = EXCP_HALTED;
1844             break;
1845         case ICPT_WAITPSW:
1846             /* disabled wait, since enabled wait is handled in kernel */
1847             cpu_synchronize_state(cs);
1848             if (s390_cpu_halt(cpu) == 0) {
1849                 if (is_special_wait_psw(cs)) {
1850                     qemu_system_shutdown_request();
1851                 } else {
1852                     qemu_system_guest_panicked();
1853                 }
1854             }
1855             r = EXCP_HALTED;
1856             break;
1857         case ICPT_CPU_STOP:
1858             if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
1859                 qemu_system_shutdown_request();
1860             }
1861             if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
1862                 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR,
1863                                       true);
1864             }
1865             cpu->env.sigp_order = 0;
1866             r = EXCP_HALTED;
1867             break;
1868         case ICPT_SOFT_INTERCEPT:
1869             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1870             exit(1);
1871             break;
1872         case ICPT_IO:
1873             fprintf(stderr, "KVM unimplemented icpt IO\n");
1874             exit(1);
1875             break;
1876         default:
1877             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1878             exit(1);
1879             break;
1880     }
1881
1882     return r;
1883 }
1884
1885 static int handle_tsch(S390CPU *cpu)
1886 {
1887     CPUState *cs = CPU(cpu);
1888     struct kvm_run *run = cs->kvm_run;
1889     int ret;
1890
1891     cpu_synchronize_state(cs);
1892
1893     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb);
1894     if (ret < 0) {
1895         /*
1896          * Failure.
1897          * If an I/O interrupt had been dequeued, we have to reinject it.
1898          */
1899         if (run->s390_tsch.dequeued) {
1900             kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1901                                   run->s390_tsch.subchannel_nr,
1902                                   run->s390_tsch.io_int_parm,
1903                                   run->s390_tsch.io_int_word);
1904         }
1905         ret = 0;
1906     }
1907     return ret;
1908 }
1909
1910 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1911 {
1912     struct sysib_322 sysib;
1913     int del;
1914
1915     if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1916         return;
1917     }
1918     /* Shift the stack of Extended Names to prepare for our own data */
1919     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1920             sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1921     /* First virt level, that doesn't provide Ext Names delimits stack. It is
1922      * assumed it's not capable of managing Extended Names for lower levels.
1923      */
1924     for (del = 1; del < sysib.count; del++) {
1925         if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1926             break;
1927         }
1928     }
1929     if (del < sysib.count) {
1930         memset(sysib.ext_names[del], 0,
1931                sizeof(sysib.ext_names[0]) * (sysib.count - del));
1932     }
1933     /* Insert short machine name in EBCDIC, padded with blanks */
1934     if (qemu_name) {
1935         memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1936         ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1937                                                     strlen(qemu_name)));
1938     }
1939     sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1940     memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1941     /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1942      * considered by s390 as not capable of providing any Extended Name.
1943      * Therefore if no name was specified on qemu invocation, we go with the
1944      * same "KVMguest" default, which KVM has filled into short name field.
1945      */
1946     if (qemu_name) {
1947         strncpy((char *)sysib.ext_names[0], qemu_name,
1948                 sizeof(sysib.ext_names[0]));
1949     } else {
1950         strcpy((char *)sysib.ext_names[0], "KVMguest");
1951     }
1952     /* Insert UUID */
1953     memcpy(sysib.vm[0].uuid, qemu_uuid, sizeof(sysib.vm[0].uuid));
1954
1955     s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1956 }
1957
1958 static int handle_stsi(S390CPU *cpu)
1959 {
1960     CPUState *cs = CPU(cpu);
1961     struct kvm_run *run = cs->kvm_run;
1962
1963     switch (run->s390_stsi.fc) {
1964     case 3:
1965         if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
1966             return 0;
1967         }
1968         /* Only sysib 3.2.2 needs post-handling for now. */
1969         insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
1970         return 0;
1971     default:
1972         return 0;
1973     }
1974 }
1975
1976 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1977 {
1978     CPUState *cs = CPU(cpu);
1979     struct kvm_run *run = cs->kvm_run;
1980
1981     int ret = 0;
1982     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1983
1984     switch (arch_info->type) {
1985     case KVM_HW_WP_WRITE:
1986         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1987             cs->watchpoint_hit = &hw_watchpoint;
1988             hw_watchpoint.vaddr = arch_info->addr;
1989             hw_watchpoint.flags = BP_MEM_WRITE;
1990             ret = EXCP_DEBUG;
1991         }
1992         break;
1993     case KVM_HW_BP:
1994         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1995             ret = EXCP_DEBUG;
1996         }
1997         break;
1998     case KVM_SINGLESTEP:
1999         if (cs->singlestep_enabled) {
2000             ret = EXCP_DEBUG;
2001         }
2002         break;
2003     default:
2004         ret = -ENOSYS;
2005     }
2006
2007     return ret;
2008 }
2009
2010 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
2011 {
2012     S390CPU *cpu = S390_CPU(cs);
2013     int ret = 0;
2014
2015     qemu_mutex_lock_iothread();
2016
2017     switch (run->exit_reason) {
2018         case KVM_EXIT_S390_SIEIC:
2019             ret = handle_intercept(cpu);
2020             break;
2021         case KVM_EXIT_S390_RESET:
2022             s390_reipl_request();
2023             break;
2024         case KVM_EXIT_S390_TSCH:
2025             ret = handle_tsch(cpu);
2026             break;
2027         case KVM_EXIT_S390_STSI:
2028             ret = handle_stsi(cpu);
2029             break;
2030         case KVM_EXIT_DEBUG:
2031             ret = kvm_arch_handle_debug_exit(cpu);
2032             break;
2033         default:
2034             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
2035             break;
2036     }
2037     qemu_mutex_unlock_iothread();
2038
2039     if (ret == 0) {
2040         ret = EXCP_INTERRUPT;
2041     }
2042     return ret;
2043 }
2044
2045 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2046 {
2047     return true;
2048 }
2049
2050 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2051 {
2052     return 1;
2053 }
2054
2055 int kvm_arch_on_sigbus(int code, void *addr)
2056 {
2057     return 1;
2058 }
2059
2060 void kvm_s390_io_interrupt(uint16_t subchannel_id,
2061                            uint16_t subchannel_nr, uint32_t io_int_parm,
2062                            uint32_t io_int_word)
2063 {
2064     struct kvm_s390_irq irq = {
2065         .u.io.subchannel_id = subchannel_id,
2066         .u.io.subchannel_nr = subchannel_nr,
2067         .u.io.io_int_parm = io_int_parm,
2068         .u.io.io_int_word = io_int_word,
2069     };
2070
2071     if (io_int_word & IO_INT_WORD_AI) {
2072         irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
2073     } else {
2074         irq.type = KVM_S390_INT_IO(0, (subchannel_id & 0xff00) >> 8,
2075                                       (subchannel_id & 0x0006),
2076                                       subchannel_nr);
2077     }
2078     kvm_s390_floating_interrupt(&irq);
2079 }
2080
2081 static uint64_t build_channel_report_mcic(void)
2082 {
2083     uint64_t mcic;
2084
2085     /* subclass: indicate channel report pending */
2086     mcic = MCIC_SC_CP |
2087     /* subclass modifiers: none */
2088     /* storage errors: none */
2089     /* validity bits: no damage */
2090         MCIC_VB_WP | MCIC_VB_MS | MCIC_VB_PM | MCIC_VB_IA | MCIC_VB_FP |
2091         MCIC_VB_GR | MCIC_VB_CR | MCIC_VB_ST | MCIC_VB_AR | MCIC_VB_PR |
2092         MCIC_VB_FC | MCIC_VB_CT | MCIC_VB_CC;
2093     if (kvm_check_extension(kvm_state, KVM_CAP_S390_VECTOR_REGISTERS)) {
2094         mcic |= MCIC_VB_VR;
2095     }
2096     return mcic;
2097 }
2098
2099 void kvm_s390_crw_mchk(void)
2100 {
2101     struct kvm_s390_irq irq = {
2102         .type = KVM_S390_MCHK,
2103         .u.mchk.cr14 = 1 << 28,
2104         .u.mchk.mcic = build_channel_report_mcic(),
2105     };
2106     kvm_s390_floating_interrupt(&irq);
2107 }
2108
2109 void kvm_s390_enable_css_support(S390CPU *cpu)
2110 {
2111     int r;
2112
2113     /* Activate host kernel channel subsystem support. */
2114     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
2115     assert(r == 0);
2116 }
2117
2118 void kvm_arch_init_irq_routing(KVMState *s)
2119 {
2120     /*
2121      * Note that while irqchip capabilities generally imply that cpustates
2122      * are handled in-kernel, it is not true for s390 (yet); therefore, we
2123      * have to override the common code kvm_halt_in_kernel_allowed setting.
2124      */
2125     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2126         kvm_gsi_routing_allowed = true;
2127         kvm_halt_in_kernel_allowed = false;
2128     }
2129 }
2130
2131 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
2132                                     int vq, bool assign)
2133 {
2134     struct kvm_ioeventfd kick = {
2135         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
2136         KVM_IOEVENTFD_FLAG_DATAMATCH,
2137         .fd = event_notifier_get_fd(notifier),
2138         .datamatch = vq,
2139         .addr = sch,
2140         .len = 8,
2141     };
2142     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
2143         return -ENOSYS;
2144     }
2145     if (!assign) {
2146         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
2147     }
2148     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
2149 }
2150
2151 int kvm_s390_get_memslot_count(KVMState *s)
2152 {
2153     return kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
2154 }
2155
2156 int kvm_s390_get_ri(void)
2157 {
2158     return cap_ri;
2159 }
2160
2161 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
2162 {
2163     struct kvm_mp_state mp_state = {};
2164     int ret;
2165
2166     /* the kvm part might not have been initialized yet */
2167     if (CPU(cpu)->kvm_state == NULL) {
2168         return 0;
2169     }
2170
2171     switch (cpu_state) {
2172     case CPU_STATE_STOPPED:
2173         mp_state.mp_state = KVM_MP_STATE_STOPPED;
2174         break;
2175     case CPU_STATE_CHECK_STOP:
2176         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
2177         break;
2178     case CPU_STATE_OPERATING:
2179         mp_state.mp_state = KVM_MP_STATE_OPERATING;
2180         break;
2181     case CPU_STATE_LOAD:
2182         mp_state.mp_state = KVM_MP_STATE_LOAD;
2183         break;
2184     default:
2185         error_report("Requested CPU state is not a valid S390 CPU state: %u",
2186                      cpu_state);
2187         exit(1);
2188     }
2189
2190     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2191     if (ret) {
2192         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
2193                                        strerror(-ret));
2194     }
2195
2196     return ret;
2197 }
2198
2199 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
2200 {
2201     struct kvm_s390_irq_state irq_state;
2202     CPUState *cs = CPU(cpu);
2203     int32_t bytes;
2204
2205     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2206         return;
2207     }
2208
2209     irq_state.buf = (uint64_t) cpu->irqstate;
2210     irq_state.len = VCPU_IRQ_BUF_SIZE;
2211
2212     bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
2213     if (bytes < 0) {
2214         cpu->irqstate_saved_size = 0;
2215         error_report("Migration of interrupt state failed");
2216         return;
2217     }
2218
2219     cpu->irqstate_saved_size = bytes;
2220 }
2221
2222 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
2223 {
2224     CPUState *cs = CPU(cpu);
2225     struct kvm_s390_irq_state irq_state;
2226     int r;
2227
2228     if (cpu->irqstate_saved_size == 0) {
2229         return 0;
2230     }
2231
2232     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2233         return -ENOSYS;
2234     }
2235
2236     irq_state.buf = (uint64_t) cpu->irqstate;
2237     irq_state.len = cpu->irqstate_saved_size;
2238
2239     r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
2240     if (r) {
2241         error_report("Setting interrupt state failed %d", r);
2242     }
2243     return r;
2244 }
2245
2246 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2247                              uint64_t address, uint32_t data, PCIDevice *dev)
2248 {
2249     S390PCIBusDevice *pbdev;
2250     uint32_t fid = data >> ZPCI_MSI_VEC_BITS;
2251     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
2252
2253     pbdev = s390_pci_find_dev_by_fid(fid);
2254     if (!pbdev) {
2255         DPRINTF("add_msi_route no dev\n");
2256         return -ENODEV;
2257     }
2258
2259     pbdev->routes.adapter.ind_offset = vec;
2260
2261     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2262     route->flags = 0;
2263     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2264     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2265     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2266     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset;
2267     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2268     return 0;
2269 }
2270
2271 int kvm_arch_msi_data_to_gsi(uint32_t data)
2272 {
2273     abort();
2274 }