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
  27 #include <linux/kvm.h>
  28 #include <asm/ptrace.h>
  29
  30 #include "qemu-common.h"
  31 #include "cpu.h"
  32 #include "qemu/error-report.h"
  33 #include "qemu/timer.h"
  34 #include "sysemu/sysemu.h"
  35 #include "sysemu/kvm.h"
  36 #include "hw/hw.h"
  37 #include "sysemu/device_tree.h"
  38 #include "qapi/qmp/qjson.h"
  39 #include "exec/gdbstub.h"
  40 #include "exec/address-spaces.h"
  41 #include "trace.h"
  42 #include "qapi-event.h"
  43 #include "hw/s390x/s390-pci-inst.h"
  44 #include "hw/s390x/s390-pci-bus.h"
  45 #include "hw/s390x/ipl.h"
  46 #include "hw/s390x/ebcdic.h"
  47 #include "exec/memattrs.h"
  48 #include "hw/s390x/s390-virtio-ccw.h"
  49
  50 /* #define DEBUG_KVM */
  51
  52 #ifdef DEBUG_KVM
  53 #define DPRINTF(fmt, ...) \
  54     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  55 #else
  56 #define DPRINTF(fmt, ...) \
  57     do { } while (0)
  58 #endif
  59
  60 #define kvm_vm_check_mem_attr(s, attr) \
  61     kvm_vm_check_attr(s, KVM_S390_VM_MEM_CTRL, attr)
  62
  63 #define IPA0_DIAG                       0x8300
  64 #define IPA0_SIGP                       0xae00
  65 #define IPA0_B2                         0xb200
  66 #define IPA0_B9                         0xb900
  67 #define IPA0_EB                         0xeb00
  68 #define IPA0_E3                         0xe300
  69
  70 #define PRIV_B2_SCLP_CALL               0x20
  71 #define PRIV_B2_CSCH                    0x30
  72 #define PRIV_B2_HSCH                    0x31
  73 #define PRIV_B2_MSCH                    0x32
  74 #define PRIV_B2_SSCH                    0x33
  75 #define PRIV_B2_STSCH                   0x34
  76 #define PRIV_B2_TSCH                    0x35
  77 #define PRIV_B2_TPI                     0x36
  78 #define PRIV_B2_SAL                     0x37
  79 #define PRIV_B2_RSCH                    0x38
  80 #define PRIV_B2_STCRW                   0x39
  81 #define PRIV_B2_STCPS                   0x3a
  82 #define PRIV_B2_RCHP                    0x3b
  83 #define PRIV_B2_SCHM                    0x3c
  84 #define PRIV_B2_CHSC                    0x5f
  85 #define PRIV_B2_SIGA                    0x74
  86 #define PRIV_B2_XSCH                    0x76
  87
  88 #define PRIV_EB_SQBS                    0x8a
  89 #define PRIV_EB_PCISTB                  0xd0
  90 #define PRIV_EB_SIC                     0xd1
  91
  92 #define PRIV_B9_EQBS                    0x9c
  93 #define PRIV_B9_CLP                     0xa0
  94 #define PRIV_B9_PCISTG                  0xd0
  95 #define PRIV_B9_PCILG                   0xd2
  96 #define PRIV_B9_RPCIT                   0xd3
  97
  98 #define PRIV_E3_MPCIFC                  0xd0
  99 #define PRIV_E3_STPCIFC                 0xd4
 100
 101 #define DIAG_TIMEREVENT                 0x288
 102 #define DIAG_IPL                        0x308
 103 #define DIAG_KVM_HYPERCALL              0x500
 104 #define DIAG_KVM_BREAKPOINT             0x501
 105
 106 #define ICPT_INSTRUCTION                0x04
 107 #define ICPT_PROGRAM                    0x08
 108 #define ICPT_EXT_INT                    0x14
 109 #define ICPT_WAITPSW                    0x1c
 110 #define ICPT_SOFT_INTERCEPT             0x24
 111 #define ICPT_CPU_STOP                   0x28
 112 #define ICPT_OPEREXC                    0x2c
 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 static uint8_t const *sw_bp_inst;
 670 static uint8_t sw_bp_ilen;
 671
 672 static void determine_sw_breakpoint_instr(void)
 673 {
 674         /* DIAG 501 is used for sw breakpoints with old kernels */
 675         static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 676         /* Instruction 0x0000 is used for sw breakpoints with recent kernels */
 677         static const uint8_t instr_0x0000[] = {0x00, 0x00};
 678
 679         if (sw_bp_inst) {
 680             return;
 681         }
 682         if (kvm_vm_enable_cap(kvm_state, KVM_CAP_S390_USER_INSTR0, 0)) {
 683             sw_bp_inst = diag_501;
 684             sw_bp_ilen = sizeof(diag_501);
 685             DPRINTF("KVM: will use 4-byte sw breakpoints.\n");
 686         } else {
 687             sw_bp_inst = instr_0x0000;
 688             sw_bp_ilen = sizeof(instr_0x0000);
 689             DPRINTF("KVM: will use 2-byte sw breakpoints.\n");
 690         }
 691 }
 692
 693 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 694 {
 695     determine_sw_breakpoint_instr();
 696
 697     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 698                             sw_bp_ilen, 0) ||
 699         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)sw_bp_inst, sw_bp_ilen, 1)) {
 700         return -EINVAL;
 701     }
 702     return 0;
 703 }
 704
 705 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 706 {
 707     uint8_t t[MAX_ILEN];
 708
 709     if (cpu_memory_rw_debug(cs, bp->pc, t, sw_bp_ilen, 0)) {
 710         return -EINVAL;
 711     } else if (memcmp(t, sw_bp_inst, sw_bp_ilen)) {
 712         return -EINVAL;
 713     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 714                                    sw_bp_ilen, 1)) {
 715         return -EINVAL;
 716     }
 717
 718     return 0;
 719 }
 720
 721 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
 722                                                     int len, int type)
 723 {
 724     int n;
 725
 726     for (n = 0; n < nb_hw_breakpoints; n++) {
 727         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
 728             (hw_breakpoints[n].len == len || len == -1)) {
 729             return &hw_breakpoints[n];
 730         }
 731     }
 732
 733     return NULL;
 734 }
 735
 736 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
 737 {
 738     int size;
 739
 740     if (find_hw_breakpoint(addr, len, type)) {
 741         return -EEXIST;
 742     }
 743
 744     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
 745
 746     if (!hw_breakpoints) {
 747         nb_hw_breakpoints = 0;
 748         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
 749     } else {
 750         hw_breakpoints =
 751             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
 752     }
 753
 754     if (!hw_breakpoints) {
 755         nb_hw_breakpoints = 0;
 756         return -ENOMEM;
 757     }
 758
 759     hw_breakpoints[nb_hw_breakpoints].addr = addr;
 760     hw_breakpoints[nb_hw_breakpoints].len = len;
 761     hw_breakpoints[nb_hw_breakpoints].type = type;
 762
 763     nb_hw_breakpoints++;
 764
 765     return 0;
 766 }
 767
 768 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 769                                   target_ulong len, int type)
 770 {
 771     switch (type) {
 772     case GDB_BREAKPOINT_HW:
 773         type = KVM_HW_BP;
 774         break;
 775     case GDB_WATCHPOINT_WRITE:
 776         if (len < 1) {
 777             return -EINVAL;
 778         }
 779         type = KVM_HW_WP_WRITE;
 780         break;
 781     default:
 782         return -ENOSYS;
 783     }
 784     return insert_hw_breakpoint(addr, len, type);
 785 }
 786
 787 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 788                                   target_ulong len, int type)
 789 {
 790     int size;
 791     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
 792
 793     if (bp == NULL) {
 794         return -ENOENT;
 795     }
 796
 797     nb_hw_breakpoints--;
 798     if (nb_hw_breakpoints > 0) {
 799         /*
 800          * In order to trim the array, move the last element to the position to
 801          * be removed - if necessary.
 802          */
 803         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
 804             *bp = hw_breakpoints[nb_hw_breakpoints];
 805         }
 806         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
 807         hw_breakpoints =
 808              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
 809     } else {
 810         g_free(hw_breakpoints);
 811         hw_breakpoints = NULL;
 812     }
 813
 814     return 0;
 815 }
 816
 817 void kvm_arch_remove_all_hw_breakpoints(void)
 818 {
 819     nb_hw_breakpoints = 0;
 820     g_free(hw_breakpoints);
 821     hw_breakpoints = NULL;
 822 }
 823
 824 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 825 {
 826     int i;
 827
 828     if (nb_hw_breakpoints > 0) {
 829         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
 830         dbg->arch.hw_bp = hw_breakpoints;
 831
 832         for (i = 0; i < nb_hw_breakpoints; ++i) {
 833             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
 834                                                        hw_breakpoints[i].addr);
 835         }
 836         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 837     } else {
 838         dbg->arch.nr_hw_bp = 0;
 839         dbg->arch.hw_bp = NULL;
 840     }
 841 }
 842
 843 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 844 {
 845 }
 846
 847 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
 848 {
 849     return MEMTXATTRS_UNSPECIFIED;
 850 }
 851
 852 int kvm_arch_process_async_events(CPUState *cs)
 853 {
 854     return cs->halted;
 855 }
 856
 857 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
 858                                      struct kvm_s390_interrupt *interrupt)
 859 {
 860     int r = 0;
 861
 862     interrupt->type = irq->type;
 863     switch (irq->type) {
 864     case KVM_S390_INT_VIRTIO:
 865         interrupt->parm = irq->u.ext.ext_params;
 866         /* fall through */
 867     case KVM_S390_INT_PFAULT_INIT:
 868     case KVM_S390_INT_PFAULT_DONE:
 869         interrupt->parm64 = irq->u.ext.ext_params2;
 870         break;
 871     case KVM_S390_PROGRAM_INT:
 872         interrupt->parm = irq->u.pgm.code;
 873         break;
 874     case KVM_S390_SIGP_SET_PREFIX:
 875         interrupt->parm = irq->u.prefix.address;
 876         break;
 877     case KVM_S390_INT_SERVICE:
 878         interrupt->parm = irq->u.ext.ext_params;
 879         break;
 880     case KVM_S390_MCHK:
 881         interrupt->parm = irq->u.mchk.cr14;
 882         interrupt->parm64 = irq->u.mchk.mcic;
 883         break;
 884     case KVM_S390_INT_EXTERNAL_CALL:
 885         interrupt->parm = irq->u.extcall.code;
 886         break;
 887     case KVM_S390_INT_EMERGENCY:
 888         interrupt->parm = irq->u.emerg.code;
 889         break;
 890     case KVM_S390_SIGP_STOP:
 891     case KVM_S390_RESTART:
 892         break; /* These types have no parameters */
 893     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
 894         interrupt->parm = irq->u.io.subchannel_id << 16;
 895         interrupt->parm |= irq->u.io.subchannel_nr;
 896         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
 897         interrupt->parm64 |= irq->u.io.io_int_word;
 898         break;
 899     default:
 900         r = -EINVAL;
 901         break;
 902     }
 903     return r;
 904 }
 905
 906 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
 907 {
 908     struct kvm_s390_interrupt kvmint = {};
 909     int r;
 910
 911     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 912     if (r < 0) {
 913         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 914         exit(1);
 915     }
 916
 917     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
 918     if (r < 0) {
 919         fprintf(stderr, "KVM failed to inject interrupt\n");
 920         exit(1);
 921     }
 922 }
 923
 924 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
 925 {
 926     CPUState *cs = CPU(cpu);
 927     int r;
 928
 929     if (cap_s390_irq) {
 930         r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
 931         if (!r) {
 932             return;
 933         }
 934         error_report("KVM failed to inject interrupt %llx", irq->type);
 935         exit(1);
 936     }
 937
 938     inject_vcpu_irq_legacy(cs, irq);
 939 }
 940
 941 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 942 {
 943     struct kvm_s390_interrupt kvmint = {};
 944     int r;
 945
 946     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 947     if (r < 0) {
 948         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 949         exit(1);
 950     }
 951
 952     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
 953     if (r < 0) {
 954         fprintf(stderr, "KVM failed to inject interrupt\n");
 955         exit(1);
 956     }
 957 }
 958
 959 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 960 {
 961     static bool use_flic = true;
 962     int r;
 963
 964     if (use_flic) {
 965         r = kvm_s390_inject_flic(irq);
 966         if (r == -ENOSYS) {
 967             use_flic = false;
 968         }
 969         if (!r) {
 970             return;
 971         }
 972     }
 973     __kvm_s390_floating_interrupt(irq);
 974 }
 975
 976 void kvm_s390_service_interrupt(uint32_t parm)
 977 {
 978     struct kvm_s390_irq irq = {
 979         .type = KVM_S390_INT_SERVICE,
 980         .u.ext.ext_params = parm,
 981     };
 982
 983     kvm_s390_floating_interrupt(&irq);
 984 }
 985
 986 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
 987 {
 988     struct kvm_s390_irq irq = {
 989         .type = KVM_S390_PROGRAM_INT,
 990         .u.pgm.code = code,
 991     };
 992
 993     kvm_s390_vcpu_interrupt(cpu, &irq);
 994 }
 995
 996 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
 997 {
 998     struct kvm_s390_irq irq = {
 999         .type = KVM_S390_PROGRAM_INT,
1000         .u.pgm.code = code,
1001         .u.pgm.trans_exc_code = te_code,
1002         .u.pgm.exc_access_id = te_code & 3,
1003     };
1004
1005     kvm_s390_vcpu_interrupt(cpu, &irq);
1006 }
1007
1008 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
1009                                  uint16_t ipbh0)
1010 {
1011     CPUS390XState *env = &cpu->env;
1012     uint64_t sccb;
1013     uint32_t code;
1014     int r = 0;
1015
1016     cpu_synchronize_state(CPU(cpu));
1017     sccb = env->regs[ipbh0 & 0xf];
1018     code = env->regs[(ipbh0 & 0xf0) >> 4];
1019
1020     r = sclp_service_call(env, sccb, code);
1021     if (r < 0) {
1022         enter_pgmcheck(cpu, -r);
1023     } else {
1024         setcc(cpu, r);
1025     }
1026
1027     return 0;
1028 }
1029
1030 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1031 {
1032     CPUS390XState *env = &cpu->env;
1033     int rc = 0;
1034     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1035
1036     cpu_synchronize_state(CPU(cpu));
1037
1038     switch (ipa1) {
1039     case PRIV_B2_XSCH:
1040         ioinst_handle_xsch(cpu, env->regs[1]);
1041         break;
1042     case PRIV_B2_CSCH:
1043         ioinst_handle_csch(cpu, env->regs[1]);
1044         break;
1045     case PRIV_B2_HSCH:
1046         ioinst_handle_hsch(cpu, env->regs[1]);
1047         break;
1048     case PRIV_B2_MSCH:
1049         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
1050         break;
1051     case PRIV_B2_SSCH:
1052         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
1053         break;
1054     case PRIV_B2_STCRW:
1055         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
1056         break;
1057     case PRIV_B2_STSCH:
1058         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
1059         break;
1060     case PRIV_B2_TSCH:
1061         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1062         fprintf(stderr, "Spurious tsch intercept\n");
1063         break;
1064     case PRIV_B2_CHSC:
1065         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
1066         break;
1067     case PRIV_B2_TPI:
1068         /* This should have been handled by kvm already. */
1069         fprintf(stderr, "Spurious tpi intercept\n");
1070         break;
1071     case PRIV_B2_SCHM:
1072         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1073                            run->s390_sieic.ipb);
1074         break;
1075     case PRIV_B2_RSCH:
1076         ioinst_handle_rsch(cpu, env->regs[1]);
1077         break;
1078     case PRIV_B2_RCHP:
1079         ioinst_handle_rchp(cpu, env->regs[1]);
1080         break;
1081     case PRIV_B2_STCPS:
1082         /* We do not provide this instruction, it is suppressed. */
1083         break;
1084     case PRIV_B2_SAL:
1085         ioinst_handle_sal(cpu, env->regs[1]);
1086         break;
1087     case PRIV_B2_SIGA:
1088         /* Not provided, set CC = 3 for subchannel not operational */
1089         setcc(cpu, 3);
1090         break;
1091     case PRIV_B2_SCLP_CALL:
1092         rc = kvm_sclp_service_call(cpu, run, ipbh0);
1093         break;
1094     default:
1095         rc = -1;
1096         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1097         break;
1098     }
1099
1100     return rc;
1101 }
1102
1103 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1104                                   uint8_t *ar)
1105 {
1106     CPUS390XState *env = &cpu->env;
1107     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1108     uint32_t base2 = run->s390_sieic.ipb >> 28;
1109     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1110                      ((run->s390_sieic.ipb & 0xff00) << 4);
1111
1112     if (disp2 & 0x80000) {
1113         disp2 += 0xfff00000;
1114     }
1115     if (ar) {
1116         *ar = base2;
1117     }
1118
1119     return (base2 ? env->regs[base2] : 0) +
1120            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1121 }
1122
1123 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1124                                   uint8_t *ar)
1125 {
1126     CPUS390XState *env = &cpu->env;
1127     uint32_t base2 = run->s390_sieic.ipb >> 28;
1128     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1129                      ((run->s390_sieic.ipb & 0xff00) << 4);
1130
1131     if (disp2 & 0x80000) {
1132         disp2 += 0xfff00000;
1133     }
1134     if (ar) {
1135         *ar = base2;
1136     }
1137
1138     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1139 }
1140
1141 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1142 {
1143     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1144
1145     return clp_service_call(cpu, r2);
1146 }
1147
1148 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1149 {
1150     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1151     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1152
1153     return pcilg_service_call(cpu, r1, r2);
1154 }
1155
1156 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1157 {
1158     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1159     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1160
1161     return pcistg_service_call(cpu, r1, r2);
1162 }
1163
1164 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1165 {
1166     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1167     uint64_t fiba;
1168     uint8_t ar;
1169
1170     cpu_synchronize_state(CPU(cpu));
1171     fiba = get_base_disp_rxy(cpu, run, &ar);
1172
1173     return stpcifc_service_call(cpu, r1, fiba, ar);
1174 }
1175
1176 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1177 {
1178     /* NOOP */
1179     return 0;
1180 }
1181
1182 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1183 {
1184     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1185     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1186
1187     return rpcit_service_call(cpu, r1, r2);
1188 }
1189
1190 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1191 {
1192     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1193     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1194     uint64_t gaddr;
1195     uint8_t ar;
1196
1197     cpu_synchronize_state(CPU(cpu));
1198     gaddr = get_base_disp_rsy(cpu, run, &ar);
1199
1200     return pcistb_service_call(cpu, r1, r3, gaddr, ar);
1201 }
1202
1203 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1204 {
1205     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1206     uint64_t fiba;
1207     uint8_t ar;
1208
1209     cpu_synchronize_state(CPU(cpu));
1210     fiba = get_base_disp_rxy(cpu, run, &ar);
1211
1212     return mpcifc_service_call(cpu, r1, fiba, ar);
1213 }
1214
1215 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1216 {
1217     int r = 0;
1218
1219     switch (ipa1) {
1220     case PRIV_B9_CLP:
1221         r = kvm_clp_service_call(cpu, run);
1222         break;
1223     case PRIV_B9_PCISTG:
1224         r = kvm_pcistg_service_call(cpu, run);
1225         break;
1226     case PRIV_B9_PCILG:
1227         r = kvm_pcilg_service_call(cpu, run);
1228         break;
1229     case PRIV_B9_RPCIT:
1230         r = kvm_rpcit_service_call(cpu, run);
1231         break;
1232     case PRIV_B9_EQBS:
1233         /* just inject exception */
1234         r = -1;
1235         break;
1236     default:
1237         r = -1;
1238         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1239         break;
1240     }
1241
1242     return r;
1243 }
1244
1245 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1246 {
1247     int r = 0;
1248
1249     switch (ipbl) {
1250     case PRIV_EB_PCISTB:
1251         r = kvm_pcistb_service_call(cpu, run);
1252         break;
1253     case PRIV_EB_SIC:
1254         r = kvm_sic_service_call(cpu, run);
1255         break;
1256     case PRIV_EB_SQBS:
1257         /* just inject exception */
1258         r = -1;
1259         break;
1260     default:
1261         r = -1;
1262         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1263         break;
1264     }
1265
1266     return r;
1267 }
1268
1269 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1270 {
1271     int r = 0;
1272
1273     switch (ipbl) {
1274     case PRIV_E3_MPCIFC:
1275         r = kvm_mpcifc_service_call(cpu, run);
1276         break;
1277     case PRIV_E3_STPCIFC:
1278         r = kvm_stpcifc_service_call(cpu, run);
1279         break;
1280     default:
1281         r = -1;
1282         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1283         break;
1284     }
1285
1286     return r;
1287 }
1288
1289 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1290 {
1291     CPUS390XState *env = &cpu->env;
1292     int ret;
1293
1294     cpu_synchronize_state(CPU(cpu));
1295     ret = s390_virtio_hypercall(env);
1296     if (ret == -EINVAL) {
1297         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1298         return 0;
1299     }
1300
1301     return ret;
1302 }
1303
1304 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1305 {
1306     uint64_t r1, r3;
1307     int rc;
1308
1309     cpu_synchronize_state(CPU(cpu));
1310     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1311     r3 = run->s390_sieic.ipa & 0x000f;
1312     rc = handle_diag_288(&cpu->env, r1, r3);
1313     if (rc) {
1314         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1315     }
1316 }
1317
1318 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1319 {
1320     uint64_t r1, r3;
1321
1322     cpu_synchronize_state(CPU(cpu));
1323     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1324     r3 = run->s390_sieic.ipa & 0x000f;
1325     handle_diag_308(&cpu->env, r1, r3);
1326 }
1327
1328 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1329 {
1330     CPUS390XState *env = &cpu->env;
1331     unsigned long pc;
1332
1333     cpu_synchronize_state(CPU(cpu));
1334
1335     pc = env->psw.addr - sw_bp_ilen;
1336     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1337         env->psw.addr = pc;
1338         return EXCP_DEBUG;
1339     }
1340
1341     return -ENOENT;
1342 }
1343
1344 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1345
1346 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1347 {
1348     int r = 0;
1349     uint16_t func_code;
1350
1351     /*
1352      * For any diagnose call we support, bits 48-63 of the resulting
1353      * address specify the function code; the remainder is ignored.
1354      */
1355     func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1356     switch (func_code) {
1357     case DIAG_TIMEREVENT:
1358         kvm_handle_diag_288(cpu, run);
1359         break;
1360     case DIAG_IPL:
1361         kvm_handle_diag_308(cpu, run);
1362         break;
1363     case DIAG_KVM_HYPERCALL:
1364         r = handle_hypercall(cpu, run);
1365         break;
1366     case DIAG_KVM_BREAKPOINT:
1367         r = handle_sw_breakpoint(cpu, run);
1368         break;
1369     default:
1370         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1371         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1372         break;
1373     }
1374
1375     return r;
1376 }
1377
1378 typedef struct SigpInfo {
1379     S390CPU *cpu;
1380     uint64_t param;
1381     int cc;
1382     uint64_t *status_reg;
1383 } SigpInfo;
1384
1385 static void set_sigp_status(SigpInfo *si, uint64_t status)
1386 {
1387     *si->status_reg &= 0xffffffff00000000ULL;
1388     *si->status_reg |= status;
1389     si->cc = SIGP_CC_STATUS_STORED;
1390 }
1391
1392 static void sigp_start(void *arg)
1393 {
1394     SigpInfo *si = arg;
1395
1396     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1397         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1398         return;
1399     }
1400
1401     s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1402     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1403 }
1404
1405 static void sigp_stop(void *arg)
1406 {
1407     SigpInfo *si = arg;
1408     struct kvm_s390_irq irq = {
1409         .type = KVM_S390_SIGP_STOP,
1410     };
1411
1412     if (s390_cpu_get_state(si->cpu) != CPU_STATE_OPERATING) {
1413         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1414         return;
1415     }
1416
1417     /* disabled wait - sleeping in user space */
1418     if (CPU(si->cpu)->halted) {
1419         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1420     } else {
1421         /* execute the stop function */
1422         si->cpu->env.sigp_order = SIGP_STOP;
1423         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1424     }
1425     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1426 }
1427
1428 #define ADTL_SAVE_AREA_SIZE 1024
1429 static int kvm_s390_store_adtl_status(S390CPU *cpu, hwaddr addr)
1430 {
1431     void *mem;
1432     hwaddr len = ADTL_SAVE_AREA_SIZE;
1433
1434     mem = cpu_physical_memory_map(addr, &len, 1);
1435     if (!mem) {
1436         return -EFAULT;
1437     }
1438     if (len != ADTL_SAVE_AREA_SIZE) {
1439         cpu_physical_memory_unmap(mem, len, 1, 0);
1440         return -EFAULT;
1441     }
1442
1443     memcpy(mem, &cpu->env.vregs, 512);
1444
1445     cpu_physical_memory_unmap(mem, len, 1, len);
1446
1447     return 0;
1448 }
1449
1450 #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, floating_pt_save_area)
1451 #define SAVE_AREA_SIZE 512
1452 static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
1453 {
1454     static const uint8_t ar_id = 1;
1455     uint64_t ckc = cpu->env.ckc >> 8;
1456     void *mem;
1457     int i;
1458     hwaddr len = SAVE_AREA_SIZE;
1459
1460     mem = cpu_physical_memory_map(addr, &len, 1);
1461     if (!mem) {
1462         return -EFAULT;
1463     }
1464     if (len != SAVE_AREA_SIZE) {
1465         cpu_physical_memory_unmap(mem, len, 1, 0);
1466         return -EFAULT;
1467     }
1468
1469     if (store_arch) {
1470         cpu_physical_memory_write(offsetof(LowCore, ar_access_id), &ar_id, 1);
1471     }
1472     for (i = 0; i < 16; ++i) {
1473         *((uint64_t *)mem + i) = get_freg(&cpu->env, i)->ll;
1474     }
1475     memcpy(mem + 128, &cpu->env.regs, 128);
1476     memcpy(mem + 256, &cpu->env.psw, 16);
1477     memcpy(mem + 280, &cpu->env.psa, 4);
1478     memcpy(mem + 284, &cpu->env.fpc, 4);
1479     memcpy(mem + 292, &cpu->env.todpr, 4);
1480     memcpy(mem + 296, &cpu->env.cputm, 8);
1481     memcpy(mem + 304, &ckc, 8);
1482     memcpy(mem + 320, &cpu->env.aregs, 64);
1483     memcpy(mem + 384, &cpu->env.cregs, 128);
1484
1485     cpu_physical_memory_unmap(mem, len, 1, len);
1486
1487     return 0;
1488 }
1489
1490 static void sigp_stop_and_store_status(void *arg)
1491 {
1492     SigpInfo *si = arg;
1493     struct kvm_s390_irq irq = {
1494         .type = KVM_S390_SIGP_STOP,
1495     };
1496
1497     /* disabled wait - sleeping in user space */
1498     if (s390_cpu_get_state(si->cpu) == CPU_STATE_OPERATING &&
1499         CPU(si->cpu)->halted) {
1500         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1501     }
1502
1503     switch (s390_cpu_get_state(si->cpu)) {
1504     case CPU_STATE_OPERATING:
1505         si->cpu->env.sigp_order = SIGP_STOP_STORE_STATUS;
1506         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1507         /* store will be performed when handling the stop intercept */
1508         break;
1509     case CPU_STATE_STOPPED:
1510         /* already stopped, just store the status */
1511         cpu_synchronize_state(CPU(si->cpu));
1512         kvm_s390_store_status(si->cpu, KVM_S390_STORE_STATUS_DEF_ADDR, true);
1513         break;
1514     }
1515     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1516 }
1517
1518 static void sigp_store_status_at_address(void *arg)
1519 {
1520     SigpInfo *si = arg;
1521     uint32_t address = si->param & 0x7ffffe00u;
1522
1523     /* cpu has to be stopped */
1524     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1525         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1526         return;
1527     }
1528
1529     cpu_synchronize_state(CPU(si->cpu));
1530
1531     if (kvm_s390_store_status(si->cpu, address, false)) {
1532         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1533         return;
1534     }
1535     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1536 }
1537
1538 static void sigp_store_adtl_status(void *arg)
1539 {
1540     SigpInfo *si = arg;
1541
1542     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_VECTOR_REGISTERS)) {
1543         set_sigp_status(si, SIGP_STAT_INVALID_ORDER);
1544         return;
1545     }
1546
1547     /* cpu has to be stopped */
1548     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1549         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1550         return;
1551     }
1552
1553     /* parameter must be aligned to 1024-byte boundary */
1554     if (si->param & 0x3ff) {
1555         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1556         return;
1557     }
1558
1559     cpu_synchronize_state(CPU(si->cpu));
1560
1561     if (kvm_s390_store_adtl_status(si->cpu, si->param)) {
1562         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1563         return;
1564     }
1565     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1566 }
1567
1568 static void sigp_restart(void *arg)
1569 {
1570     SigpInfo *si = arg;
1571     struct kvm_s390_irq irq = {
1572         .type = KVM_S390_RESTART,
1573     };
1574
1575     switch (s390_cpu_get_state(si->cpu)) {
1576     case CPU_STATE_STOPPED:
1577         /* the restart irq has to be delivered prior to any other pending irq */
1578         cpu_synchronize_state(CPU(si->cpu));
1579         do_restart_interrupt(&si->cpu->env);
1580         s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1581         break;
1582     case CPU_STATE_OPERATING:
1583         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1584         break;
1585     }
1586     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1587 }
1588
1589 int kvm_s390_cpu_restart(S390CPU *cpu)
1590 {
1591     SigpInfo si = {
1592         .cpu = cpu,
1593     };
1594
1595     run_on_cpu(CPU(cpu), sigp_restart, &si);
1596     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1597     return 0;
1598 }
1599
1600 static void sigp_initial_cpu_reset(void *arg)
1601 {
1602     SigpInfo *si = arg;
1603     CPUState *cs = CPU(si->cpu);
1604     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1605
1606     cpu_synchronize_state(cs);
1607     scc->initial_cpu_reset(cs);
1608     cpu_synchronize_post_reset(cs);
1609     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1610 }
1611
1612 static void sigp_cpu_reset(void *arg)
1613 {
1614     SigpInfo *si = arg;
1615     CPUState *cs = CPU(si->cpu);
1616     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1617
1618     cpu_synchronize_state(cs);
1619     scc->cpu_reset(cs);
1620     cpu_synchronize_post_reset(cs);
1621     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1622 }
1623
1624 static void sigp_set_prefix(void *arg)
1625 {
1626     SigpInfo *si = arg;
1627     uint32_t addr = si->param & 0x7fffe000u;
1628
1629     cpu_synchronize_state(CPU(si->cpu));
1630
1631     if (!address_space_access_valid(&address_space_memory, addr,
1632                                     sizeof(struct LowCore), false)) {
1633         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1634         return;
1635     }
1636
1637     /* cpu has to be stopped */
1638     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1639         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1640         return;
1641     }
1642
1643     si->cpu->env.psa = addr;
1644     cpu_synchronize_post_init(CPU(si->cpu));
1645     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1646 }
1647
1648 static int handle_sigp_single_dst(S390CPU *dst_cpu, uint8_t order,
1649                                   uint64_t param, uint64_t *status_reg)
1650 {
1651     SigpInfo si = {
1652         .cpu = dst_cpu,
1653         .param = param,
1654         .status_reg = status_reg,
1655     };
1656
1657     /* cpu available? */
1658     if (dst_cpu == NULL) {
1659         return SIGP_CC_NOT_OPERATIONAL;
1660     }
1661
1662     /* only resets can break pending orders */
1663     if (dst_cpu->env.sigp_order != 0 &&
1664         order != SIGP_CPU_RESET &&
1665         order != SIGP_INITIAL_CPU_RESET) {
1666         return SIGP_CC_BUSY;
1667     }
1668
1669     switch (order) {
1670     case SIGP_START:
1671         run_on_cpu(CPU(dst_cpu), sigp_start, &si);
1672         break;
1673     case SIGP_STOP:
1674         run_on_cpu(CPU(dst_cpu), sigp_stop, &si);
1675         break;
1676     case SIGP_RESTART:
1677         run_on_cpu(CPU(dst_cpu), sigp_restart, &si);
1678         break;
1679     case SIGP_STOP_STORE_STATUS:
1680         run_on_cpu(CPU(dst_cpu), sigp_stop_and_store_status, &si);
1681         break;
1682     case SIGP_STORE_STATUS_ADDR:
1683         run_on_cpu(CPU(dst_cpu), sigp_store_status_at_address, &si);
1684         break;
1685     case SIGP_STORE_ADTL_STATUS:
1686         run_on_cpu(CPU(dst_cpu), sigp_store_adtl_status, &si);
1687         break;
1688     case SIGP_SET_PREFIX:
1689         run_on_cpu(CPU(dst_cpu), sigp_set_prefix, &si);
1690         break;
1691     case SIGP_INITIAL_CPU_RESET:
1692         run_on_cpu(CPU(dst_cpu), sigp_initial_cpu_reset, &si);
1693         break;
1694     case SIGP_CPU_RESET:
1695         run_on_cpu(CPU(dst_cpu), sigp_cpu_reset, &si);
1696         break;
1697     default:
1698         DPRINTF("KVM: unknown SIGP: 0x%x\n", order);
1699         set_sigp_status(&si, SIGP_STAT_INVALID_ORDER);
1700     }
1701
1702     return si.cc;
1703 }
1704
1705 static int sigp_set_architecture(S390CPU *cpu, uint32_t param,
1706                                  uint64_t *status_reg)
1707 {
1708     CPUState *cur_cs;
1709     S390CPU *cur_cpu;
1710
1711     /* due to the BQL, we are the only active cpu */
1712     CPU_FOREACH(cur_cs) {
1713         cur_cpu = S390_CPU(cur_cs);
1714         if (cur_cpu->env.sigp_order != 0) {
1715             return SIGP_CC_BUSY;
1716         }
1717         cpu_synchronize_state(cur_cs);
1718         /* all but the current one have to be stopped */
1719         if (cur_cpu != cpu &&
1720             s390_cpu_get_state(cur_cpu) != CPU_STATE_STOPPED) {
1721             *status_reg &= 0xffffffff00000000ULL;
1722             *status_reg |= SIGP_STAT_INCORRECT_STATE;
1723             return SIGP_CC_STATUS_STORED;
1724         }
1725     }
1726
1727     switch (param & 0xff) {
1728     case SIGP_MODE_ESA_S390:
1729         /* not supported */
1730         return SIGP_CC_NOT_OPERATIONAL;
1731     case SIGP_MODE_Z_ARCH_TRANS_ALL_PSW:
1732     case SIGP_MODE_Z_ARCH_TRANS_CUR_PSW:
1733         CPU_FOREACH(cur_cs) {
1734             cur_cpu = S390_CPU(cur_cs);
1735             cur_cpu->env.pfault_token = -1UL;
1736         }
1737         break;
1738     default:
1739         *status_reg &= 0xffffffff00000000ULL;
1740         *status_reg |= SIGP_STAT_INVALID_PARAMETER;
1741         return SIGP_CC_STATUS_STORED;
1742     }
1743
1744     return SIGP_CC_ORDER_CODE_ACCEPTED;
1745 }
1746
1747 #define SIGP_ORDER_MASK 0x000000ff
1748
1749 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1750 {
1751     CPUS390XState *env = &cpu->env;
1752     const uint8_t r1 = ipa1 >> 4;
1753     const uint8_t r3 = ipa1 & 0x0f;
1754     int ret;
1755     uint8_t order;
1756     uint64_t *status_reg;
1757     uint64_t param;
1758     S390CPU *dst_cpu = NULL;
1759
1760     cpu_synchronize_state(CPU(cpu));
1761
1762     /* get order code */
1763     order = decode_basedisp_rs(env, run->s390_sieic.ipb, NULL)
1764         & SIGP_ORDER_MASK;
1765     status_reg = &env->regs[r1];
1766     param = (r1 % 2) ? env->regs[r1] : env->regs[r1 + 1];
1767
1768     switch (order) {
1769     case SIGP_SET_ARCH:
1770         ret = sigp_set_architecture(cpu, param, status_reg);
1771         break;
1772     default:
1773         /* all other sigp orders target a single vcpu */
1774         dst_cpu = s390_cpu_addr2state(env->regs[r3]);
1775         ret = handle_sigp_single_dst(dst_cpu, order, param, status_reg);
1776     }
1777
1778     trace_kvm_sigp_finished(order, CPU(cpu)->cpu_index,
1779                             dst_cpu ? CPU(dst_cpu)->cpu_index : -1, ret);
1780
1781     if (ret >= 0) {
1782         setcc(cpu, ret);
1783         return 0;
1784     }
1785
1786     return ret;
1787 }
1788
1789 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1790 {
1791     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1792     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1793     int r = -1;
1794
1795     DPRINTF("handle_instruction 0x%x 0x%x\n",
1796             run->s390_sieic.ipa, run->s390_sieic.ipb);
1797     switch (ipa0) {
1798     case IPA0_B2:
1799         r = handle_b2(cpu, run, ipa1);
1800         break;
1801     case IPA0_B9:
1802         r = handle_b9(cpu, run, ipa1);
1803         break;
1804     case IPA0_EB:
1805         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1806         break;
1807     case IPA0_E3:
1808         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1809         break;
1810     case IPA0_DIAG:
1811         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1812         break;
1813     case IPA0_SIGP:
1814         r = handle_sigp(cpu, run, ipa1);
1815         break;
1816     }
1817
1818     if (r < 0) {
1819         r = 0;
1820         enter_pgmcheck(cpu, 0x0001);
1821     }
1822
1823     return r;
1824 }
1825
1826 static bool is_special_wait_psw(CPUState *cs)
1827 {
1828     /* signal quiesce */
1829     return cs->kvm_run->psw_addr == 0xfffUL;
1830 }
1831
1832 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1833 {
1834     CPUState *cs = CPU(cpu);
1835
1836     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1837                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1838                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1839     s390_cpu_halt(cpu);
1840     qemu_system_guest_panicked();
1841 }
1842
1843 static int handle_intercept(S390CPU *cpu)
1844 {
1845     CPUState *cs = CPU(cpu);
1846     struct kvm_run *run = cs->kvm_run;
1847     int icpt_code = run->s390_sieic.icptcode;
1848     int r = 0;
1849
1850     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1851             (long)cs->kvm_run->psw_addr);
1852     switch (icpt_code) {
1853         case ICPT_INSTRUCTION:
1854             r = handle_instruction(cpu, run);
1855             break;
1856         case ICPT_PROGRAM:
1857             unmanageable_intercept(cpu, "program interrupt",
1858                                    offsetof(LowCore, program_new_psw));
1859             r = EXCP_HALTED;
1860             break;
1861         case ICPT_EXT_INT:
1862             unmanageable_intercept(cpu, "external interrupt",
1863                                    offsetof(LowCore, external_new_psw));
1864             r = EXCP_HALTED;
1865             break;
1866         case ICPT_WAITPSW:
1867             /* disabled wait, since enabled wait is handled in kernel */
1868             cpu_synchronize_state(cs);
1869             if (s390_cpu_halt(cpu) == 0) {
1870                 if (is_special_wait_psw(cs)) {
1871                     qemu_system_shutdown_request();
1872                 } else {
1873                     qemu_system_guest_panicked();
1874                 }
1875             }
1876             r = EXCP_HALTED;
1877             break;
1878         case ICPT_CPU_STOP:
1879             if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
1880                 qemu_system_shutdown_request();
1881             }
1882             if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
1883                 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR,
1884                                       true);
1885             }
1886             cpu->env.sigp_order = 0;
1887             r = EXCP_HALTED;
1888             break;
1889         case ICPT_OPEREXC:
1890             /* currently only instr 0x0000 after enabled via capability */
1891             r = handle_sw_breakpoint(cpu, run);
1892             if (r == -ENOENT) {
1893                 enter_pgmcheck(cpu, PGM_OPERATION);
1894                 r = 0;
1895             }
1896             break;
1897         case ICPT_SOFT_INTERCEPT:
1898             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1899             exit(1);
1900             break;
1901         case ICPT_IO:
1902             fprintf(stderr, "KVM unimplemented icpt IO\n");
1903             exit(1);
1904             break;
1905         default:
1906             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1907             exit(1);
1908             break;
1909     }
1910
1911     return r;
1912 }
1913
1914 static int handle_tsch(S390CPU *cpu)
1915 {
1916     CPUState *cs = CPU(cpu);
1917     struct kvm_run *run = cs->kvm_run;
1918     int ret;
1919
1920     cpu_synchronize_state(cs);
1921
1922     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb);
1923     if (ret < 0) {
1924         /*
1925          * Failure.
1926          * If an I/O interrupt had been dequeued, we have to reinject it.
1927          */
1928         if (run->s390_tsch.dequeued) {
1929             kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1930                                   run->s390_tsch.subchannel_nr,
1931                                   run->s390_tsch.io_int_parm,
1932                                   run->s390_tsch.io_int_word);
1933         }
1934         ret = 0;
1935     }
1936     return ret;
1937 }
1938
1939 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1940 {
1941     struct sysib_322 sysib;
1942     int del;
1943
1944     if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1945         return;
1946     }
1947     /* Shift the stack of Extended Names to prepare for our own data */
1948     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1949             sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1950     /* First virt level, that doesn't provide Ext Names delimits stack. It is
1951      * assumed it's not capable of managing Extended Names for lower levels.
1952      */
1953     for (del = 1; del < sysib.count; del++) {
1954         if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1955             break;
1956         }
1957     }
1958     if (del < sysib.count) {
1959         memset(sysib.ext_names[del], 0,
1960                sizeof(sysib.ext_names[0]) * (sysib.count - del));
1961     }
1962     /* Insert short machine name in EBCDIC, padded with blanks */
1963     if (qemu_name) {
1964         memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1965         ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1966                                                     strlen(qemu_name)));
1967     }
1968     sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1969     memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1970     /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1971      * considered by s390 as not capable of providing any Extended Name.
1972      * Therefore if no name was specified on qemu invocation, we go with the
1973      * same "KVMguest" default, which KVM has filled into short name field.
1974      */
1975     if (qemu_name) {
1976         strncpy((char *)sysib.ext_names[0], qemu_name,
1977                 sizeof(sysib.ext_names[0]));
1978     } else {
1979         strcpy((char *)sysib.ext_names[0], "KVMguest");
1980     }
1981     /* Insert UUID */
1982     memcpy(sysib.vm[0].uuid, qemu_uuid, sizeof(sysib.vm[0].uuid));
1983
1984     s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1985 }
1986
1987 static int handle_stsi(S390CPU *cpu)
1988 {
1989     CPUState *cs = CPU(cpu);
1990     struct kvm_run *run = cs->kvm_run;
1991
1992     switch (run->s390_stsi.fc) {
1993     case 3:
1994         if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
1995             return 0;
1996         }
1997         /* Only sysib 3.2.2 needs post-handling for now. */
1998         insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
1999         return 0;
2000     default:
2001         return 0;
2002     }
2003 }
2004
2005 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
2006 {
2007     CPUState *cs = CPU(cpu);
2008     struct kvm_run *run = cs->kvm_run;
2009
2010     int ret = 0;
2011     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
2012
2013     switch (arch_info->type) {
2014     case KVM_HW_WP_WRITE:
2015         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
2016             cs->watchpoint_hit = &hw_watchpoint;
2017             hw_watchpoint.vaddr = arch_info->addr;
2018             hw_watchpoint.flags = BP_MEM_WRITE;
2019             ret = EXCP_DEBUG;
2020         }
2021         break;
2022     case KVM_HW_BP:
2023         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
2024             ret = EXCP_DEBUG;
2025         }
2026         break;
2027     case KVM_SINGLESTEP:
2028         if (cs->singlestep_enabled) {
2029             ret = EXCP_DEBUG;
2030         }
2031         break;
2032     default:
2033         ret = -ENOSYS;
2034     }
2035
2036     return ret;
2037 }
2038
2039 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
2040 {
2041     S390CPU *cpu = S390_CPU(cs);
2042     int ret = 0;
2043
2044     qemu_mutex_lock_iothread();
2045
2046     switch (run->exit_reason) {
2047         case KVM_EXIT_S390_SIEIC:
2048             ret = handle_intercept(cpu);
2049             break;
2050         case KVM_EXIT_S390_RESET:
2051             s390_reipl_request();
2052             break;
2053         case KVM_EXIT_S390_TSCH:
2054             ret = handle_tsch(cpu);
2055             break;
2056         case KVM_EXIT_S390_STSI:
2057             ret = handle_stsi(cpu);
2058             break;
2059         case KVM_EXIT_DEBUG:
2060             ret = kvm_arch_handle_debug_exit(cpu);
2061             break;
2062         default:
2063             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
2064             break;
2065     }
2066     qemu_mutex_unlock_iothread();
2067
2068     if (ret == 0) {
2069         ret = EXCP_INTERRUPT;
2070     }
2071     return ret;
2072 }
2073
2074 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2075 {
2076     return true;
2077 }
2078
2079 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2080 {
2081     return 1;
2082 }
2083
2084 int kvm_arch_on_sigbus(int code, void *addr)
2085 {
2086     return 1;
2087 }
2088
2089 void kvm_s390_io_interrupt(uint16_t subchannel_id,
2090                            uint16_t subchannel_nr, uint32_t io_int_parm,
2091                            uint32_t io_int_word)
2092 {
2093     struct kvm_s390_irq irq = {
2094         .u.io.subchannel_id = subchannel_id,
2095         .u.io.subchannel_nr = subchannel_nr,
2096         .u.io.io_int_parm = io_int_parm,
2097         .u.io.io_int_word = io_int_word,
2098     };
2099
2100     if (io_int_word & IO_INT_WORD_AI) {
2101         irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
2102     } else {
2103         irq.type = KVM_S390_INT_IO(0, (subchannel_id & 0xff00) >> 8,
2104                                       (subchannel_id & 0x0006),
2105                                       subchannel_nr);
2106     }
2107     kvm_s390_floating_interrupt(&irq);
2108 }
2109
2110 static uint64_t build_channel_report_mcic(void)
2111 {
2112     uint64_t mcic;
2113
2114     /* subclass: indicate channel report pending */
2115     mcic = MCIC_SC_CP |
2116     /* subclass modifiers: none */
2117     /* storage errors: none */
2118     /* validity bits: no damage */
2119         MCIC_VB_WP | MCIC_VB_MS | MCIC_VB_PM | MCIC_VB_IA | MCIC_VB_FP |
2120         MCIC_VB_GR | MCIC_VB_CR | MCIC_VB_ST | MCIC_VB_AR | MCIC_VB_PR |
2121         MCIC_VB_FC | MCIC_VB_CT | MCIC_VB_CC;
2122     if (kvm_check_extension(kvm_state, KVM_CAP_S390_VECTOR_REGISTERS)) {
2123         mcic |= MCIC_VB_VR;
2124     }
2125     return mcic;
2126 }
2127
2128 void kvm_s390_crw_mchk(void)
2129 {
2130     struct kvm_s390_irq irq = {
2131         .type = KVM_S390_MCHK,
2132         .u.mchk.cr14 = 1 << 28,
2133         .u.mchk.mcic = build_channel_report_mcic(),
2134     };
2135     kvm_s390_floating_interrupt(&irq);
2136 }
2137
2138 void kvm_s390_enable_css_support(S390CPU *cpu)
2139 {
2140     int r;
2141
2142     /* Activate host kernel channel subsystem support. */
2143     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
2144     assert(r == 0);
2145 }
2146
2147 void kvm_arch_init_irq_routing(KVMState *s)
2148 {
2149     /*
2150      * Note that while irqchip capabilities generally imply that cpustates
2151      * are handled in-kernel, it is not true for s390 (yet); therefore, we
2152      * have to override the common code kvm_halt_in_kernel_allowed setting.
2153      */
2154     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2155         kvm_gsi_routing_allowed = true;
2156         kvm_halt_in_kernel_allowed = false;
2157     }
2158 }
2159
2160 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
2161                                     int vq, bool assign)
2162 {
2163     struct kvm_ioeventfd kick = {
2164         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
2165         KVM_IOEVENTFD_FLAG_DATAMATCH,
2166         .fd = event_notifier_get_fd(notifier),
2167         .datamatch = vq,
2168         .addr = sch,
2169         .len = 8,
2170     };
2171     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
2172         return -ENOSYS;
2173     }
2174     if (!assign) {
2175         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
2176     }
2177     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
2178 }
2179
2180 int kvm_s390_get_memslot_count(KVMState *s)
2181 {
2182     return kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
2183 }
2184
2185 int kvm_s390_get_ri(void)
2186 {
2187     return cap_ri;
2188 }
2189
2190 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
2191 {
2192     struct kvm_mp_state mp_state = {};
2193     int ret;
2194
2195     /* the kvm part might not have been initialized yet */
2196     if (CPU(cpu)->kvm_state == NULL) {
2197         return 0;
2198     }
2199
2200     switch (cpu_state) {
2201     case CPU_STATE_STOPPED:
2202         mp_state.mp_state = KVM_MP_STATE_STOPPED;
2203         break;
2204     case CPU_STATE_CHECK_STOP:
2205         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
2206         break;
2207     case CPU_STATE_OPERATING:
2208         mp_state.mp_state = KVM_MP_STATE_OPERATING;
2209         break;
2210     case CPU_STATE_LOAD:
2211         mp_state.mp_state = KVM_MP_STATE_LOAD;
2212         break;
2213     default:
2214         error_report("Requested CPU state is not a valid S390 CPU state: %u",
2215                      cpu_state);
2216         exit(1);
2217     }
2218
2219     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2220     if (ret) {
2221         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
2222                                        strerror(-ret));
2223     }
2224
2225     return ret;
2226 }
2227
2228 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
2229 {
2230     struct kvm_s390_irq_state irq_state;
2231     CPUState *cs = CPU(cpu);
2232     int32_t bytes;
2233
2234     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2235         return;
2236     }
2237
2238     irq_state.buf = (uint64_t) cpu->irqstate;
2239     irq_state.len = VCPU_IRQ_BUF_SIZE;
2240
2241     bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
2242     if (bytes < 0) {
2243         cpu->irqstate_saved_size = 0;
2244         error_report("Migration of interrupt state failed");
2245         return;
2246     }
2247
2248     cpu->irqstate_saved_size = bytes;
2249 }
2250
2251 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
2252 {
2253     CPUState *cs = CPU(cpu);
2254     struct kvm_s390_irq_state irq_state;
2255     int r;
2256
2257     if (cpu->irqstate_saved_size == 0) {
2258         return 0;
2259     }
2260
2261     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2262         return -ENOSYS;
2263     }
2264
2265     irq_state.buf = (uint64_t) cpu->irqstate;
2266     irq_state.len = cpu->irqstate_saved_size;
2267
2268     r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
2269     if (r) {
2270         error_report("Setting interrupt state failed %d", r);
2271     }
2272     return r;
2273 }
2274
2275 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2276                              uint64_t address, uint32_t data, PCIDevice *dev)
2277 {
2278     S390PCIBusDevice *pbdev;
2279     uint32_t idx = data >> ZPCI_MSI_VEC_BITS;
2280     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
2281
2282     pbdev = s390_pci_find_dev_by_idx(idx);
2283     if (!pbdev) {
2284         DPRINTF("add_msi_route no dev\n");
2285         return -ENODEV;
2286     }
2287
2288     pbdev->routes.adapter.ind_offset = vec;
2289
2290     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2291     route->flags = 0;
2292     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2293     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2294     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2295     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset;
2296     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2297     return 0;
2298 }
2299
2300 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2301                                 int vector, PCIDevice *dev)
2302 {
2303     return 0;
2304 }
2305
2306 int kvm_arch_release_virq_post(int virq)
2307 {
2308     return 0;
2309 }
2310
2311 int kvm_arch_msi_data_to_gsi(uint32_t data)
2312 {
2313     abort();
2314 }