target-s390x/kvm.c

   1 /*
   2  * QEMU S390x KVM implementation
   3  *
   4  * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
   5  * Copyright IBM Corp. 2012
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * Contributions after 2012-10-29 are licensed under the terms of the
  18  * GNU GPL, version 2 or (at your option) any later version.
  19  *
  20  * You should have received a copy of the GNU (Lesser) General Public
  21  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  22  */
  23
  24 #include <sys/types.h>
  25 #include <sys/ioctl.h>
  26 #include <sys/mman.h>
  27
  28 #include <linux/kvm.h>
  29 #include <asm/ptrace.h>
  30
  31 #include "qemu-common.h"
  32 #include "qemu/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 "cpu.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
  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_IO                         0x40
 113
 114 #define NR_LOCAL_IRQS 32
 115 /*
 116  * Needs to be big enough to contain max_cpus emergency signals
 117  * and in addition NR_LOCAL_IRQS interrupts
 118  */
 119 #define VCPU_IRQ_BUF_SIZE (sizeof(struct kvm_s390_irq) * \
 120                            (max_cpus + NR_LOCAL_IRQS))
 121
 122 static CPUWatchpoint hw_watchpoint;
 123 /*
 124  * We don't use a list because this structure is also used to transmit the
 125  * hardware breakpoints to the kernel.
 126  */
 127 static struct kvm_hw_breakpoint *hw_breakpoints;
 128 static int nb_hw_breakpoints;
 129
 130 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
 131     KVM_CAP_LAST_INFO
 132 };
 133
 134 static int cap_sync_regs;
 135 static int cap_async_pf;
 136 static int cap_mem_op;
 137 static int cap_s390_irq;
 138
 139 static void *legacy_s390_alloc(size_t size, uint64_t *align);
 140
 141 static int kvm_s390_query_mem_limit(KVMState *s, uint64_t *memory_limit)
 142 {
 143     struct kvm_device_attr attr = {
 144         .group = KVM_S390_VM_MEM_CTRL,
 145         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
 146         .addr = (uint64_t) memory_limit,
 147     };
 148
 149     return kvm_vm_ioctl(s, KVM_GET_DEVICE_ATTR, &attr);
 150 }
 151
 152 int kvm_s390_set_mem_limit(KVMState *s, uint64_t new_limit, uint64_t *hw_limit)
 153 {
 154     int rc;
 155
 156     struct kvm_device_attr attr = {
 157         .group = KVM_S390_VM_MEM_CTRL,
 158         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
 159         .addr = (uint64_t) &new_limit,
 160     };
 161
 162     if (!kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_LIMIT_SIZE)) {
 163         return 0;
 164     }
 165
 166     rc = kvm_s390_query_mem_limit(s, hw_limit);
 167     if (rc) {
 168         return rc;
 169     } else if (*hw_limit < new_limit) {
 170         return -E2BIG;
 171     }
 172
 173     return kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 174 }
 175
 176 void kvm_s390_clear_cmma_callback(void *opaque)
 177 {
 178     int rc;
 179     KVMState *s = opaque;
 180     struct kvm_device_attr attr = {
 181         .group = KVM_S390_VM_MEM_CTRL,
 182         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 183     };
 184
 185     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 186     trace_kvm_clear_cmma(rc);
 187 }
 188
 189 static void kvm_s390_enable_cmma(KVMState *s)
 190 {
 191     int rc;
 192     struct kvm_device_attr attr = {
 193         .group = KVM_S390_VM_MEM_CTRL,
 194         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 195     };
 196
 197     if (!kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_ENABLE_CMMA) ||
 198         !kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_CLR_CMMA)) {
 199         return;
 200     }
 201
 202     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 203     if (!rc) {
 204         qemu_register_reset(kvm_s390_clear_cmma_callback, s);
 205     }
 206     trace_kvm_enable_cmma(rc);
 207 }
 208
 209 static void kvm_s390_set_attr(uint64_t attr)
 210 {
 211     struct kvm_device_attr attribute = {
 212         .group = KVM_S390_VM_CRYPTO,
 213         .attr  = attr,
 214     };
 215
 216     int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
 217
 218     if (ret) {
 219         error_report("Failed to set crypto device attribute %lu: %s",
 220                      attr, strerror(-ret));
 221     }
 222 }
 223
 224 static void kvm_s390_init_aes_kw(void)
 225 {
 226     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
 227
 228     if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
 229                                  NULL)) {
 230             attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
 231     }
 232
 233     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 234             kvm_s390_set_attr(attr);
 235     }
 236 }
 237
 238 static void kvm_s390_init_dea_kw(void)
 239 {
 240     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
 241
 242     if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
 243                                  NULL)) {
 244             attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
 245     }
 246
 247     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 248             kvm_s390_set_attr(attr);
 249     }
 250 }
 251
 252 static void kvm_s390_init_crypto(void)
 253 {
 254     kvm_s390_init_aes_kw();
 255     kvm_s390_init_dea_kw();
 256 }
 257
 258 int kvm_arch_init(MachineState *ms, KVMState *s)
 259 {
 260     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
 261     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
 262     cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
 263     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
 264
 265     kvm_s390_enable_cmma(s);
 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
 276     return 0;
 277 }
 278
 279 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
 280 {
 281     return cpu->cpu_index;
 282 }
 283
 284 int kvm_arch_init_vcpu(CPUState *cs)
 285 {
 286     S390CPU *cpu = S390_CPU(cs);
 287     kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
 288     cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE);
 289     return 0;
 290 }
 291
 292 void kvm_s390_reset_vcpu(S390CPU *cpu)
 293 {
 294     CPUState *cs = CPU(cpu);
 295
 296     /* The initial reset call is needed here to reset in-kernel
 297      * vcpu data that we can't access directly from QEMU
 298      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
 299      * Before this ioctl cpu_synchronize_state() is called in common kvm
 300      * code (kvm-all) */
 301     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
 302         error_report("Initial CPU reset failed on CPU %i", cs->cpu_index);
 303     }
 304
 305     kvm_s390_init_crypto();
 306 }
 307
 308 static int can_sync_regs(CPUState *cs, int regs)
 309 {
 310     return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
 311 }
 312
 313 int kvm_arch_put_registers(CPUState *cs, int level)
 314 {
 315     S390CPU *cpu = S390_CPU(cs);
 316     CPUS390XState *env = &cpu->env;
 317     struct kvm_sregs sregs;
 318     struct kvm_regs regs;
 319     struct kvm_fpu fpu = {};
 320     int r;
 321     int i;
 322
 323     /* always save the PSW  and the GPRS*/
 324     cs->kvm_run->psw_addr = env->psw.addr;
 325     cs->kvm_run->psw_mask = env->psw.mask;
 326
 327     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 328         for (i = 0; i < 16; i++) {
 329             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
 330             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
 331         }
 332     } else {
 333         for (i = 0; i < 16; i++) {
 334             regs.gprs[i] = env->regs[i];
 335         }
 336         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
 337         if (r < 0) {
 338             return r;
 339         }
 340     }
 341
 342     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 343         for (i = 0; i < 32; i++) {
 344             cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
 345             cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
 346         }
 347         cs->kvm_run->s.regs.fpc = env->fpc;
 348         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
 349     } else {
 350         /* Floating point */
 351         for (i = 0; i < 16; i++) {
 352             fpu.fprs[i] = get_freg(env, i)->ll;
 353         }
 354         fpu.fpc = env->fpc;
 355
 356         r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
 357         if (r < 0) {
 358             return r;
 359         }
 360     }
 361
 362     /* Do we need to save more than that? */
 363     if (level == KVM_PUT_RUNTIME_STATE) {
 364         return 0;
 365     }
 366
 367     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 368         cs->kvm_run->s.regs.cputm = env->cputm;
 369         cs->kvm_run->s.regs.ckc = env->ckc;
 370         cs->kvm_run->s.regs.todpr = env->todpr;
 371         cs->kvm_run->s.regs.gbea = env->gbea;
 372         cs->kvm_run->s.regs.pp = env->pp;
 373         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
 374     } else {
 375         /*
 376          * These ONE_REGS are not protected by a capability. As they are only
 377          * necessary for migration we just trace a possible error, but don't
 378          * return with an error return code.
 379          */
 380         kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 381         kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 382         kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 383         kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 384         kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 385     }
 386
 387     /* pfault parameters */
 388     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 389         cs->kvm_run->s.regs.pft = env->pfault_token;
 390         cs->kvm_run->s.regs.pfs = env->pfault_select;
 391         cs->kvm_run->s.regs.pfc = env->pfault_compare;
 392         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
 393     } else if (cap_async_pf) {
 394         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 395         if (r < 0) {
 396             return r;
 397         }
 398         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 399         if (r < 0) {
 400             return r;
 401         }
 402         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 403         if (r < 0) {
 404             return r;
 405         }
 406     }
 407
 408     /* access registers and control registers*/
 409     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 410         for (i = 0; i < 16; i++) {
 411             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
 412             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
 413         }
 414         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
 415         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
 416     } else {
 417         for (i = 0; i < 16; i++) {
 418             sregs.acrs[i] = env->aregs[i];
 419             sregs.crs[i] = env->cregs[i];
 420         }
 421         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
 422         if (r < 0) {
 423             return r;
 424         }
 425     }
 426
 427     /* Finally the prefix */
 428     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 429         cs->kvm_run->s.regs.prefix = env->psa;
 430         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
 431     } else {
 432         /* prefix is only supported via sync regs */
 433     }
 434     return 0;
 435 }
 436
 437 int kvm_arch_get_registers(CPUState *cs)
 438 {
 439     S390CPU *cpu = S390_CPU(cs);
 440     CPUS390XState *env = &cpu->env;
 441     struct kvm_sregs sregs;
 442     struct kvm_regs regs;
 443     struct kvm_fpu fpu;
 444     int i, r;
 445
 446     /* get the PSW */
 447     env->psw.addr = cs->kvm_run->psw_addr;
 448     env->psw.mask = cs->kvm_run->psw_mask;
 449
 450     /* the GPRS */
 451     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 452         for (i = 0; i < 16; i++) {
 453             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
 454         }
 455     } else {
 456         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
 457         if (r < 0) {
 458             return r;
 459         }
 460          for (i = 0; i < 16; i++) {
 461             env->regs[i] = regs.gprs[i];
 462         }
 463     }
 464
 465     /* The ACRS and CRS */
 466     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 467         for (i = 0; i < 16; i++) {
 468             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
 469             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
 470         }
 471     } else {
 472         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
 473         if (r < 0) {
 474             return r;
 475         }
 476          for (i = 0; i < 16; i++) {
 477             env->aregs[i] = sregs.acrs[i];
 478             env->cregs[i] = sregs.crs[i];
 479         }
 480     }
 481
 482     /* Floating point and vector registers */
 483     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 484         for (i = 0; i < 32; i++) {
 485             env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
 486             env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
 487         }
 488         env->fpc = cs->kvm_run->s.regs.fpc;
 489     } else {
 490         r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
 491         if (r < 0) {
 492             return r;
 493         }
 494         for (i = 0; i < 16; i++) {
 495             get_freg(env, i)->ll = fpu.fprs[i];
 496         }
 497         env->fpc = fpu.fpc;
 498     }
 499
 500     /* The prefix */
 501     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 502         env->psa = cs->kvm_run->s.regs.prefix;
 503     }
 504
 505     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 506         env->cputm = cs->kvm_run->s.regs.cputm;
 507         env->ckc = cs->kvm_run->s.regs.ckc;
 508         env->todpr = cs->kvm_run->s.regs.todpr;
 509         env->gbea = cs->kvm_run->s.regs.gbea;
 510         env->pp = cs->kvm_run->s.regs.pp;
 511     } else {
 512         /*
 513          * These ONE_REGS are not protected by a capability. As they are only
 514          * necessary for migration we just trace a possible error, but don't
 515          * return with an error return code.
 516          */
 517         kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 518         kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 519         kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 520         kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 521         kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 522     }
 523
 524     /* pfault parameters */
 525     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 526         env->pfault_token = cs->kvm_run->s.regs.pft;
 527         env->pfault_select = cs->kvm_run->s.regs.pfs;
 528         env->pfault_compare = cs->kvm_run->s.regs.pfc;
 529     } else if (cap_async_pf) {
 530         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 531         if (r < 0) {
 532             return r;
 533         }
 534         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 535         if (r < 0) {
 536             return r;
 537         }
 538         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 539         if (r < 0) {
 540             return r;
 541         }
 542     }
 543
 544     return 0;
 545 }
 546
 547 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
 548 {
 549     int r;
 550     struct kvm_device_attr attr = {
 551         .group = KVM_S390_VM_TOD,
 552         .attr = KVM_S390_VM_TOD_LOW,
 553         .addr = (uint64_t)tod_low,
 554     };
 555
 556     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 557     if (r) {
 558         return r;
 559     }
 560
 561     attr.attr = KVM_S390_VM_TOD_HIGH;
 562     attr.addr = (uint64_t)tod_high;
 563     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 564 }
 565
 566 int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_low)
 567 {
 568     int r;
 569
 570     struct kvm_device_attr attr = {
 571         .group = KVM_S390_VM_TOD,
 572         .attr = KVM_S390_VM_TOD_LOW,
 573         .addr = (uint64_t)tod_low,
 574     };
 575
 576     r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 577     if (r) {
 578         return r;
 579     }
 580
 581     attr.attr = KVM_S390_VM_TOD_HIGH;
 582     attr.addr = (uint64_t)tod_high;
 583     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 584 }
 585
 586 /**
 587  * kvm_s390_mem_op:
 588  * @addr:      the logical start address in guest memory
 589  * @ar:        the access register number
 590  * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
 591  * @len:       length that should be transfered
 592  * @is_write:  true = write, false = read
 593  * Returns:    0 on success, non-zero if an exception or error occured
 594  *
 595  * Use KVM ioctl to read/write from/to guest memory. An access exception
 596  * is injected into the vCPU in case of translation errors.
 597  */
 598 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
 599                     int len, bool is_write)
 600 {
 601     struct kvm_s390_mem_op mem_op = {
 602         .gaddr = addr,
 603         .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
 604         .size = len,
 605         .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
 606                        : KVM_S390_MEMOP_LOGICAL_READ,
 607         .buf = (uint64_t)hostbuf,
 608         .ar = ar,
 609     };
 610     int ret;
 611
 612     if (!cap_mem_op) {
 613         return -ENOSYS;
 614     }
 615     if (!hostbuf) {
 616         mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
 617     }
 618
 619     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
 620     if (ret < 0) {
 621         error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
 622     }
 623     return ret;
 624 }
 625
 626 /*
 627  * Legacy layout for s390:
 628  * Older S390 KVM requires the topmost vma of the RAM to be
 629  * smaller than an system defined value, which is at least 256GB.
 630  * Larger systems have larger values. We put the guest between
 631  * the end of data segment (system break) and this value. We
 632  * use 32GB as a base to have enough room for the system break
 633  * to grow. We also have to use MAP parameters that avoid
 634  * read-only mapping of guest pages.
 635  */
 636 static void *legacy_s390_alloc(size_t size, uint64_t *align)
 637 {
 638     void *mem;
 639
 640     mem = mmap((void *) 0x800000000ULL, size,
 641                PROT_EXEC|PROT_READ|PROT_WRITE,
 642                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 643     return mem == MAP_FAILED ? NULL : mem;
 644 }
 645
 646 /* DIAG 501 is used for sw breakpoints */
 647 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 648
 649 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 650 {
 651
 652     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 653                             sizeof(diag_501), 0) ||
 654         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501,
 655                             sizeof(diag_501), 1)) {
 656         return -EINVAL;
 657     }
 658     return 0;
 659 }
 660
 661 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 662 {
 663     uint8_t t[sizeof(diag_501)];
 664
 665     if (cpu_memory_rw_debug(cs, bp->pc, t, sizeof(diag_501), 0)) {
 666         return -EINVAL;
 667     } else if (memcmp(t, diag_501, sizeof(diag_501))) {
 668         return -EINVAL;
 669     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 670                                    sizeof(diag_501), 1)) {
 671         return -EINVAL;
 672     }
 673
 674     return 0;
 675 }
 676
 677 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
 678                                                     int len, int type)
 679 {
 680     int n;
 681
 682     for (n = 0; n < nb_hw_breakpoints; n++) {
 683         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
 684             (hw_breakpoints[n].len == len || len == -1)) {
 685             return &hw_breakpoints[n];
 686         }
 687     }
 688
 689     return NULL;
 690 }
 691
 692 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
 693 {
 694     int size;
 695
 696     if (find_hw_breakpoint(addr, len, type)) {
 697         return -EEXIST;
 698     }
 699
 700     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
 701
 702     if (!hw_breakpoints) {
 703         nb_hw_breakpoints = 0;
 704         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
 705     } else {
 706         hw_breakpoints =
 707             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
 708     }
 709
 710     if (!hw_breakpoints) {
 711         nb_hw_breakpoints = 0;
 712         return -ENOMEM;
 713     }
 714
 715     hw_breakpoints[nb_hw_breakpoints].addr = addr;
 716     hw_breakpoints[nb_hw_breakpoints].len = len;
 717     hw_breakpoints[nb_hw_breakpoints].type = type;
 718
 719     nb_hw_breakpoints++;
 720
 721     return 0;
 722 }
 723
 724 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 725                                   target_ulong len, int type)
 726 {
 727     switch (type) {
 728     case GDB_BREAKPOINT_HW:
 729         type = KVM_HW_BP;
 730         break;
 731     case GDB_WATCHPOINT_WRITE:
 732         if (len < 1) {
 733             return -EINVAL;
 734         }
 735         type = KVM_HW_WP_WRITE;
 736         break;
 737     default:
 738         return -ENOSYS;
 739     }
 740     return insert_hw_breakpoint(addr, len, type);
 741 }
 742
 743 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 744                                   target_ulong len, int type)
 745 {
 746     int size;
 747     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
 748
 749     if (bp == NULL) {
 750         return -ENOENT;
 751     }
 752
 753     nb_hw_breakpoints--;
 754     if (nb_hw_breakpoints > 0) {
 755         /*
 756          * In order to trim the array, move the last element to the position to
 757          * be removed - if necessary.
 758          */
 759         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
 760             *bp = hw_breakpoints[nb_hw_breakpoints];
 761         }
 762         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
 763         hw_breakpoints =
 764              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
 765     } else {
 766         g_free(hw_breakpoints);
 767         hw_breakpoints = NULL;
 768     }
 769
 770     return 0;
 771 }
 772
 773 void kvm_arch_remove_all_hw_breakpoints(void)
 774 {
 775     nb_hw_breakpoints = 0;
 776     g_free(hw_breakpoints);
 777     hw_breakpoints = NULL;
 778 }
 779
 780 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 781 {
 782     int i;
 783
 784     if (nb_hw_breakpoints > 0) {
 785         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
 786         dbg->arch.hw_bp = hw_breakpoints;
 787
 788         for (i = 0; i < nb_hw_breakpoints; ++i) {
 789             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
 790                                                        hw_breakpoints[i].addr);
 791         }
 792         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 793     } else {
 794         dbg->arch.nr_hw_bp = 0;
 795         dbg->arch.hw_bp = NULL;
 796     }
 797 }
 798
 799 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 800 {
 801 }
 802
 803 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
 804 {
 805     return MEMTXATTRS_UNSPECIFIED;
 806 }
 807
 808 int kvm_arch_process_async_events(CPUState *cs)
 809 {
 810     return cs->halted;
 811 }
 812
 813 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
 814                                      struct kvm_s390_interrupt *interrupt)
 815 {
 816     int r = 0;
 817
 818     interrupt->type = irq->type;
 819     switch (irq->type) {
 820     case KVM_S390_INT_VIRTIO:
 821         interrupt->parm = irq->u.ext.ext_params;
 822         /* fall through */
 823     case KVM_S390_INT_PFAULT_INIT:
 824     case KVM_S390_INT_PFAULT_DONE:
 825         interrupt->parm64 = irq->u.ext.ext_params2;
 826         break;
 827     case KVM_S390_PROGRAM_INT:
 828         interrupt->parm = irq->u.pgm.code;
 829         break;
 830     case KVM_S390_SIGP_SET_PREFIX:
 831         interrupt->parm = irq->u.prefix.address;
 832         break;
 833     case KVM_S390_INT_SERVICE:
 834         interrupt->parm = irq->u.ext.ext_params;
 835         break;
 836     case KVM_S390_MCHK:
 837         interrupt->parm = irq->u.mchk.cr14;
 838         interrupt->parm64 = irq->u.mchk.mcic;
 839         break;
 840     case KVM_S390_INT_EXTERNAL_CALL:
 841         interrupt->parm = irq->u.extcall.code;
 842         break;
 843     case KVM_S390_INT_EMERGENCY:
 844         interrupt->parm = irq->u.emerg.code;
 845         break;
 846     case KVM_S390_SIGP_STOP:
 847     case KVM_S390_RESTART:
 848         break; /* These types have no parameters */
 849     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
 850         interrupt->parm = irq->u.io.subchannel_id << 16;
 851         interrupt->parm |= irq->u.io.subchannel_nr;
 852         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
 853         interrupt->parm64 |= irq->u.io.io_int_word;
 854         break;
 855     default:
 856         r = -EINVAL;
 857         break;
 858     }
 859     return r;
 860 }
 861
 862 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
 863 {
 864     struct kvm_s390_interrupt kvmint = {};
 865     int r;
 866
 867     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 868     if (r < 0) {
 869         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 870         exit(1);
 871     }
 872
 873     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
 874     if (r < 0) {
 875         fprintf(stderr, "KVM failed to inject interrupt\n");
 876         exit(1);
 877     }
 878 }
 879
 880 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
 881 {
 882     CPUState *cs = CPU(cpu);
 883     int r;
 884
 885     if (cap_s390_irq) {
 886         r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
 887         if (!r) {
 888             return;
 889         }
 890         error_report("KVM failed to inject interrupt %llx", irq->type);
 891         exit(1);
 892     }
 893
 894     inject_vcpu_irq_legacy(cs, irq);
 895 }
 896
 897 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 898 {
 899     struct kvm_s390_interrupt kvmint = {};
 900     int r;
 901
 902     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 903     if (r < 0) {
 904         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 905         exit(1);
 906     }
 907
 908     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
 909     if (r < 0) {
 910         fprintf(stderr, "KVM failed to inject interrupt\n");
 911         exit(1);
 912     }
 913 }
 914
 915 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 916 {
 917     static bool use_flic = true;
 918     int r;
 919
 920     if (use_flic) {
 921         r = kvm_s390_inject_flic(irq);
 922         if (r == -ENOSYS) {
 923             use_flic = false;
 924         }
 925         if (!r) {
 926             return;
 927         }
 928     }
 929     __kvm_s390_floating_interrupt(irq);
 930 }
 931
 932 void kvm_s390_virtio_irq(int config_change, uint64_t token)
 933 {
 934     struct kvm_s390_irq irq = {
 935         .type = KVM_S390_INT_VIRTIO,
 936         .u.ext.ext_params = config_change,
 937         .u.ext.ext_params2 = token,
 938     };
 939
 940     kvm_s390_floating_interrupt(&irq);
 941 }
 942
 943 void kvm_s390_service_interrupt(uint32_t parm)
 944 {
 945     struct kvm_s390_irq irq = {
 946         .type = KVM_S390_INT_SERVICE,
 947         .u.ext.ext_params = parm,
 948     };
 949
 950     kvm_s390_floating_interrupt(&irq);
 951 }
 952
 953 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
 954 {
 955     struct kvm_s390_irq irq = {
 956         .type = KVM_S390_PROGRAM_INT,
 957         .u.pgm.code = code,
 958     };
 959
 960     kvm_s390_vcpu_interrupt(cpu, &irq);
 961 }
 962
 963 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
 964 {
 965     struct kvm_s390_irq irq = {
 966         .type = KVM_S390_PROGRAM_INT,
 967         .u.pgm.code = code,
 968         .u.pgm.trans_exc_code = te_code,
 969         .u.pgm.exc_access_id = te_code & 3,
 970     };
 971
 972     kvm_s390_vcpu_interrupt(cpu, &irq);
 973 }
 974
 975 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
 976                                  uint16_t ipbh0)
 977 {
 978     CPUS390XState *env = &cpu->env;
 979     uint64_t sccb;
 980     uint32_t code;
 981     int r = 0;
 982
 983     cpu_synchronize_state(CPU(cpu));
 984     sccb = env->regs[ipbh0 & 0xf];
 985     code = env->regs[(ipbh0 & 0xf0) >> 4];
 986
 987     r = sclp_service_call(env, sccb, code);
 988     if (r < 0) {
 989         enter_pgmcheck(cpu, -r);
 990     } else {
 991         setcc(cpu, r);
 992     }
 993
 994     return 0;
 995 }
 996
 997 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 998 {
 999     CPUS390XState *env = &cpu->env;
1000     int rc = 0;
1001     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1002
1003     cpu_synchronize_state(CPU(cpu));
1004
1005     switch (ipa1) {
1006     case PRIV_B2_XSCH:
1007         ioinst_handle_xsch(cpu, env->regs[1]);
1008         break;
1009     case PRIV_B2_CSCH:
1010         ioinst_handle_csch(cpu, env->regs[1]);
1011         break;
1012     case PRIV_B2_HSCH:
1013         ioinst_handle_hsch(cpu, env->regs[1]);
1014         break;
1015     case PRIV_B2_MSCH:
1016         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
1017         break;
1018     case PRIV_B2_SSCH:
1019         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
1020         break;
1021     case PRIV_B2_STCRW:
1022         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
1023         break;
1024     case PRIV_B2_STSCH:
1025         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
1026         break;
1027     case PRIV_B2_TSCH:
1028         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1029         fprintf(stderr, "Spurious tsch intercept\n");
1030         break;
1031     case PRIV_B2_CHSC:
1032         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
1033         break;
1034     case PRIV_B2_TPI:
1035         /* This should have been handled by kvm already. */
1036         fprintf(stderr, "Spurious tpi intercept\n");
1037         break;
1038     case PRIV_B2_SCHM:
1039         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1040                            run->s390_sieic.ipb);
1041         break;
1042     case PRIV_B2_RSCH:
1043         ioinst_handle_rsch(cpu, env->regs[1]);
1044         break;
1045     case PRIV_B2_RCHP:
1046         ioinst_handle_rchp(cpu, env->regs[1]);
1047         break;
1048     case PRIV_B2_STCPS:
1049         /* We do not provide this instruction, it is suppressed. */
1050         break;
1051     case PRIV_B2_SAL:
1052         ioinst_handle_sal(cpu, env->regs[1]);
1053         break;
1054     case PRIV_B2_SIGA:
1055         /* Not provided, set CC = 3 for subchannel not operational */
1056         setcc(cpu, 3);
1057         break;
1058     case PRIV_B2_SCLP_CALL:
1059         rc = kvm_sclp_service_call(cpu, run, ipbh0);
1060         break;
1061     default:
1062         rc = -1;
1063         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1064         break;
1065     }
1066
1067     return rc;
1068 }
1069
1070 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1071                                   uint8_t *ar)
1072 {
1073     CPUS390XState *env = &cpu->env;
1074     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1075     uint32_t base2 = run->s390_sieic.ipb >> 28;
1076     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1077                      ((run->s390_sieic.ipb & 0xff00) << 4);
1078
1079     if (disp2 & 0x80000) {
1080         disp2 += 0xfff00000;
1081     }
1082     if (ar) {
1083         *ar = base2;
1084     }
1085
1086     return (base2 ? env->regs[base2] : 0) +
1087            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1088 }
1089
1090 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1091                                   uint8_t *ar)
1092 {
1093     CPUS390XState *env = &cpu->env;
1094     uint32_t base2 = run->s390_sieic.ipb >> 28;
1095     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1096                      ((run->s390_sieic.ipb & 0xff00) << 4);
1097
1098     if (disp2 & 0x80000) {
1099         disp2 += 0xfff00000;
1100     }
1101     if (ar) {
1102         *ar = base2;
1103     }
1104
1105     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1106 }
1107
1108 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1109 {
1110     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1111
1112     return clp_service_call(cpu, r2);
1113 }
1114
1115 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1116 {
1117     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1118     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1119
1120     return pcilg_service_call(cpu, r1, r2);
1121 }
1122
1123 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1124 {
1125     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1126     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1127
1128     return pcistg_service_call(cpu, r1, r2);
1129 }
1130
1131 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1132 {
1133     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1134     uint64_t fiba;
1135     uint8_t ar;
1136
1137     cpu_synchronize_state(CPU(cpu));
1138     fiba = get_base_disp_rxy(cpu, run, &ar);
1139
1140     return stpcifc_service_call(cpu, r1, fiba, ar);
1141 }
1142
1143 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1144 {
1145     /* NOOP */
1146     return 0;
1147 }
1148
1149 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1150 {
1151     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1152     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1153
1154     return rpcit_service_call(cpu, r1, r2);
1155 }
1156
1157 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1158 {
1159     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1160     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1161     uint64_t gaddr;
1162     uint8_t ar;
1163
1164     cpu_synchronize_state(CPU(cpu));
1165     gaddr = get_base_disp_rsy(cpu, run, &ar);
1166
1167     return pcistb_service_call(cpu, r1, r3, gaddr, ar);
1168 }
1169
1170 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1171 {
1172     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1173     uint64_t fiba;
1174     uint8_t ar;
1175
1176     cpu_synchronize_state(CPU(cpu));
1177     fiba = get_base_disp_rxy(cpu, run, &ar);
1178
1179     return mpcifc_service_call(cpu, r1, fiba, ar);
1180 }
1181
1182 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1183 {
1184     int r = 0;
1185
1186     switch (ipa1) {
1187     case PRIV_B9_CLP:
1188         r = kvm_clp_service_call(cpu, run);
1189         break;
1190     case PRIV_B9_PCISTG:
1191         r = kvm_pcistg_service_call(cpu, run);
1192         break;
1193     case PRIV_B9_PCILG:
1194         r = kvm_pcilg_service_call(cpu, run);
1195         break;
1196     case PRIV_B9_RPCIT:
1197         r = kvm_rpcit_service_call(cpu, run);
1198         break;
1199     case PRIV_B9_EQBS:
1200         /* just inject exception */
1201         r = -1;
1202         break;
1203     default:
1204         r = -1;
1205         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1206         break;
1207     }
1208
1209     return r;
1210 }
1211
1212 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1213 {
1214     int r = 0;
1215
1216     switch (ipbl) {
1217     case PRIV_EB_PCISTB:
1218         r = kvm_pcistb_service_call(cpu, run);
1219         break;
1220     case PRIV_EB_SIC:
1221         r = kvm_sic_service_call(cpu, run);
1222         break;
1223     case PRIV_EB_SQBS:
1224         /* just inject exception */
1225         r = -1;
1226         break;
1227     default:
1228         r = -1;
1229         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1230         break;
1231     }
1232
1233     return r;
1234 }
1235
1236 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1237 {
1238     int r = 0;
1239
1240     switch (ipbl) {
1241     case PRIV_E3_MPCIFC:
1242         r = kvm_mpcifc_service_call(cpu, run);
1243         break;
1244     case PRIV_E3_STPCIFC:
1245         r = kvm_stpcifc_service_call(cpu, run);
1246         break;
1247     default:
1248         r = -1;
1249         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1250         break;
1251     }
1252
1253     return r;
1254 }
1255
1256 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1257 {
1258     CPUS390XState *env = &cpu->env;
1259     int ret;
1260
1261     cpu_synchronize_state(CPU(cpu));
1262     ret = s390_virtio_hypercall(env);
1263     if (ret == -EINVAL) {
1264         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1265         return 0;
1266     }
1267
1268     return ret;
1269 }
1270
1271 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1272 {
1273     uint64_t r1, r3;
1274     int rc;
1275
1276     cpu_synchronize_state(CPU(cpu));
1277     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1278     r3 = run->s390_sieic.ipa & 0x000f;
1279     rc = handle_diag_288(&cpu->env, r1, r3);
1280     if (rc) {
1281         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1282     }
1283 }
1284
1285 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1286 {
1287     uint64_t r1, r3;
1288
1289     cpu_synchronize_state(CPU(cpu));
1290     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1291     r3 = run->s390_sieic.ipa & 0x000f;
1292     handle_diag_308(&cpu->env, r1, r3);
1293 }
1294
1295 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1296 {
1297     CPUS390XState *env = &cpu->env;
1298     unsigned long pc;
1299
1300     cpu_synchronize_state(CPU(cpu));
1301
1302     pc = env->psw.addr - 4;
1303     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1304         env->psw.addr = pc;
1305         return EXCP_DEBUG;
1306     }
1307
1308     return -ENOENT;
1309 }
1310
1311 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1312
1313 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1314 {
1315     int r = 0;
1316     uint16_t func_code;
1317
1318     /*
1319      * For any diagnose call we support, bits 48-63 of the resulting
1320      * address specify the function code; the remainder is ignored.
1321      */
1322     func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1323     switch (func_code) {
1324     case DIAG_TIMEREVENT:
1325         kvm_handle_diag_288(cpu, run);
1326         break;
1327     case DIAG_IPL:
1328         kvm_handle_diag_308(cpu, run);
1329         break;
1330     case DIAG_KVM_HYPERCALL:
1331         r = handle_hypercall(cpu, run);
1332         break;
1333     case DIAG_KVM_BREAKPOINT:
1334         r = handle_sw_breakpoint(cpu, run);
1335         break;
1336     default:
1337         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1338         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1339         break;
1340     }
1341
1342     return r;
1343 }
1344
1345 typedef struct SigpInfo {
1346     S390CPU *cpu;
1347     uint64_t param;
1348     int cc;
1349     uint64_t *status_reg;
1350 } SigpInfo;
1351
1352 static void set_sigp_status(SigpInfo *si, uint64_t status)
1353 {
1354     *si->status_reg &= 0xffffffff00000000ULL;
1355     *si->status_reg |= status;
1356     si->cc = SIGP_CC_STATUS_STORED;
1357 }
1358
1359 static void sigp_start(void *arg)
1360 {
1361     SigpInfo *si = arg;
1362
1363     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1364         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1365         return;
1366     }
1367
1368     s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1369     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1370 }
1371
1372 static void sigp_stop(void *arg)
1373 {
1374     SigpInfo *si = arg;
1375     struct kvm_s390_irq irq = {
1376         .type = KVM_S390_SIGP_STOP,
1377     };
1378
1379     if (s390_cpu_get_state(si->cpu) != CPU_STATE_OPERATING) {
1380         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1381         return;
1382     }
1383
1384     /* disabled wait - sleeping in user space */
1385     if (CPU(si->cpu)->halted) {
1386         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1387     } else {
1388         /* execute the stop function */
1389         si->cpu->env.sigp_order = SIGP_STOP;
1390         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1391     }
1392     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1393 }
1394
1395 #define ADTL_SAVE_AREA_SIZE 1024
1396 static int kvm_s390_store_adtl_status(S390CPU *cpu, hwaddr addr)
1397 {
1398     void *mem;
1399     hwaddr len = ADTL_SAVE_AREA_SIZE;
1400
1401     mem = cpu_physical_memory_map(addr, &len, 1);
1402     if (!mem) {
1403         return -EFAULT;
1404     }
1405     if (len != ADTL_SAVE_AREA_SIZE) {
1406         cpu_physical_memory_unmap(mem, len, 1, 0);
1407         return -EFAULT;
1408     }
1409
1410     memcpy(mem, &cpu->env.vregs, 512);
1411
1412     cpu_physical_memory_unmap(mem, len, 1, len);
1413
1414     return 0;
1415 }
1416
1417 #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, floating_pt_save_area)
1418 #define SAVE_AREA_SIZE 512
1419 static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
1420 {
1421     static const uint8_t ar_id = 1;
1422     uint64_t ckc = cpu->env.ckc >> 8;
1423     void *mem;
1424     int i;
1425     hwaddr len = SAVE_AREA_SIZE;
1426
1427     mem = cpu_physical_memory_map(addr, &len, 1);
1428     if (!mem) {
1429         return -EFAULT;
1430     }
1431     if (len != SAVE_AREA_SIZE) {
1432         cpu_physical_memory_unmap(mem, len, 1, 0);
1433         return -EFAULT;
1434     }
1435
1436     if (store_arch) {
1437         cpu_physical_memory_write(offsetof(LowCore, ar_access_id), &ar_id, 1);
1438     }
1439     for (i = 0; i < 16; ++i) {
1440         *((uint64 *)mem + i) = get_freg(&cpu->env, i)->ll;
1441     }
1442     memcpy(mem + 128, &cpu->env.regs, 128);
1443     memcpy(mem + 256, &cpu->env.psw, 16);
1444     memcpy(mem + 280, &cpu->env.psa, 4);
1445     memcpy(mem + 284, &cpu->env.fpc, 4);
1446     memcpy(mem + 292, &cpu->env.todpr, 4);
1447     memcpy(mem + 296, &cpu->env.cputm, 8);
1448     memcpy(mem + 304, &ckc, 8);
1449     memcpy(mem + 320, &cpu->env.aregs, 64);
1450     memcpy(mem + 384, &cpu->env.cregs, 128);
1451
1452     cpu_physical_memory_unmap(mem, len, 1, len);
1453
1454     return 0;
1455 }
1456
1457 static void sigp_stop_and_store_status(void *arg)
1458 {
1459     SigpInfo *si = arg;
1460     struct kvm_s390_irq irq = {
1461         .type = KVM_S390_SIGP_STOP,
1462     };
1463
1464     /* disabled wait - sleeping in user space */
1465     if (s390_cpu_get_state(si->cpu) == CPU_STATE_OPERATING &&
1466         CPU(si->cpu)->halted) {
1467         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1468     }
1469
1470     switch (s390_cpu_get_state(si->cpu)) {
1471     case CPU_STATE_OPERATING:
1472         si->cpu->env.sigp_order = SIGP_STOP_STORE_STATUS;
1473         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1474         /* store will be performed when handling the stop intercept */
1475         break;
1476     case CPU_STATE_STOPPED:
1477         /* already stopped, just store the status */
1478         cpu_synchronize_state(CPU(si->cpu));
1479         kvm_s390_store_status(si->cpu, KVM_S390_STORE_STATUS_DEF_ADDR, true);
1480         break;
1481     }
1482     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1483 }
1484
1485 static void sigp_store_status_at_address(void *arg)
1486 {
1487     SigpInfo *si = arg;
1488     uint32_t address = si->param & 0x7ffffe00u;
1489
1490     /* cpu has to be stopped */
1491     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1492         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1493         return;
1494     }
1495
1496     cpu_synchronize_state(CPU(si->cpu));
1497
1498     if (kvm_s390_store_status(si->cpu, address, false)) {
1499         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1500         return;
1501     }
1502     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1503 }
1504
1505 static void sigp_store_adtl_status(void *arg)
1506 {
1507     SigpInfo *si = arg;
1508
1509     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_VECTOR_REGISTERS)) {
1510         set_sigp_status(si, SIGP_STAT_INVALID_ORDER);
1511         return;
1512     }
1513
1514     /* cpu has to be stopped */
1515     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1516         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1517         return;
1518     }
1519
1520     /* parameter must be aligned to 1024-byte boundary */
1521     if (si->param & 0x3ff) {
1522         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1523         return;
1524     }
1525
1526     cpu_synchronize_state(CPU(si->cpu));
1527
1528     if (kvm_s390_store_adtl_status(si->cpu, si->param)) {
1529         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1530         return;
1531     }
1532     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1533 }
1534
1535 static void sigp_restart(void *arg)
1536 {
1537     SigpInfo *si = arg;
1538     struct kvm_s390_irq irq = {
1539         .type = KVM_S390_RESTART,
1540     };
1541
1542     switch (s390_cpu_get_state(si->cpu)) {
1543     case CPU_STATE_STOPPED:
1544         /* the restart irq has to be delivered prior to any other pending irq */
1545         cpu_synchronize_state(CPU(si->cpu));
1546         do_restart_interrupt(&si->cpu->env);
1547         s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1548         break;
1549     case CPU_STATE_OPERATING:
1550         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1551         break;
1552     }
1553     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1554 }
1555
1556 int kvm_s390_cpu_restart(S390CPU *cpu)
1557 {
1558     SigpInfo si = {
1559         .cpu = cpu,
1560     };
1561
1562     run_on_cpu(CPU(cpu), sigp_restart, &si);
1563     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1564     return 0;
1565 }
1566
1567 static void sigp_initial_cpu_reset(void *arg)
1568 {
1569     SigpInfo *si = arg;
1570     CPUState *cs = CPU(si->cpu);
1571     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1572
1573     cpu_synchronize_state(cs);
1574     scc->initial_cpu_reset(cs);
1575     cpu_synchronize_post_reset(cs);
1576     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1577 }
1578
1579 static void sigp_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->cpu_reset(cs);
1587     cpu_synchronize_post_reset(cs);
1588     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1589 }
1590
1591 static void sigp_set_prefix(void *arg)
1592 {
1593     SigpInfo *si = arg;
1594     uint32_t addr = si->param & 0x7fffe000u;
1595
1596     cpu_synchronize_state(CPU(si->cpu));
1597
1598     if (!address_space_access_valid(&address_space_memory, addr,
1599                                     sizeof(struct LowCore), false)) {
1600         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1601         return;
1602     }
1603
1604     /* cpu has to be stopped */
1605     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1606         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1607         return;
1608     }
1609
1610     si->cpu->env.psa = addr;
1611     cpu_synchronize_post_init(CPU(si->cpu));
1612     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1613 }
1614
1615 static int handle_sigp_single_dst(S390CPU *dst_cpu, uint8_t order,
1616                                   uint64_t param, uint64_t *status_reg)
1617 {
1618     SigpInfo si = {
1619         .cpu = dst_cpu,
1620         .param = param,
1621         .status_reg = status_reg,
1622     };
1623
1624     /* cpu available? */
1625     if (dst_cpu == NULL) {
1626         return SIGP_CC_NOT_OPERATIONAL;
1627     }
1628
1629     /* only resets can break pending orders */
1630     if (dst_cpu->env.sigp_order != 0 &&
1631         order != SIGP_CPU_RESET &&
1632         order != SIGP_INITIAL_CPU_RESET) {
1633         return SIGP_CC_BUSY;
1634     }
1635
1636     switch (order) {
1637     case SIGP_START:
1638         run_on_cpu(CPU(dst_cpu), sigp_start, &si);
1639         break;
1640     case SIGP_STOP:
1641         run_on_cpu(CPU(dst_cpu), sigp_stop, &si);
1642         break;
1643     case SIGP_RESTART:
1644         run_on_cpu(CPU(dst_cpu), sigp_restart, &si);
1645         break;
1646     case SIGP_STOP_STORE_STATUS:
1647         run_on_cpu(CPU(dst_cpu), sigp_stop_and_store_status, &si);
1648         break;
1649     case SIGP_STORE_STATUS_ADDR:
1650         run_on_cpu(CPU(dst_cpu), sigp_store_status_at_address, &si);
1651         break;
1652     case SIGP_STORE_ADTL_STATUS:
1653         run_on_cpu(CPU(dst_cpu), sigp_store_adtl_status, &si);
1654         break;
1655     case SIGP_SET_PREFIX:
1656         run_on_cpu(CPU(dst_cpu), sigp_set_prefix, &si);
1657         break;
1658     case SIGP_INITIAL_CPU_RESET:
1659         run_on_cpu(CPU(dst_cpu), sigp_initial_cpu_reset, &si);
1660         break;
1661     case SIGP_CPU_RESET:
1662         run_on_cpu(CPU(dst_cpu), sigp_cpu_reset, &si);
1663         break;
1664     default:
1665         DPRINTF("KVM: unknown SIGP: 0x%x\n", order);
1666         set_sigp_status(&si, SIGP_STAT_INVALID_ORDER);
1667     }
1668
1669     return si.cc;
1670 }
1671
1672 static int sigp_set_architecture(S390CPU *cpu, uint32_t param,
1673                                  uint64_t *status_reg)
1674 {
1675     CPUState *cur_cs;
1676     S390CPU *cur_cpu;
1677
1678     /* due to the BQL, we are the only active cpu */
1679     CPU_FOREACH(cur_cs) {
1680         cur_cpu = S390_CPU(cur_cs);
1681         if (cur_cpu->env.sigp_order != 0) {
1682             return SIGP_CC_BUSY;
1683         }
1684         cpu_synchronize_state(cur_cs);
1685         /* all but the current one have to be stopped */
1686         if (cur_cpu != cpu &&
1687             s390_cpu_get_state(cur_cpu) != CPU_STATE_STOPPED) {
1688             *status_reg &= 0xffffffff00000000ULL;
1689             *status_reg |= SIGP_STAT_INCORRECT_STATE;
1690             return SIGP_CC_STATUS_STORED;
1691         }
1692     }
1693
1694     switch (param & 0xff) {
1695     case SIGP_MODE_ESA_S390:
1696         /* not supported */
1697         return SIGP_CC_NOT_OPERATIONAL;
1698     case SIGP_MODE_Z_ARCH_TRANS_ALL_PSW:
1699     case SIGP_MODE_Z_ARCH_TRANS_CUR_PSW:
1700         CPU_FOREACH(cur_cs) {
1701             cur_cpu = S390_CPU(cur_cs);
1702             cur_cpu->env.pfault_token = -1UL;
1703         }
1704         break;
1705     default:
1706         *status_reg &= 0xffffffff00000000ULL;
1707         *status_reg |= SIGP_STAT_INVALID_PARAMETER;
1708         return SIGP_CC_STATUS_STORED;
1709     }
1710
1711     return SIGP_CC_ORDER_CODE_ACCEPTED;
1712 }
1713
1714 #define SIGP_ORDER_MASK 0x000000ff
1715
1716 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1717 {
1718     CPUS390XState *env = &cpu->env;
1719     const uint8_t r1 = ipa1 >> 4;
1720     const uint8_t r3 = ipa1 & 0x0f;
1721     int ret;
1722     uint8_t order;
1723     uint64_t *status_reg;
1724     uint64_t param;
1725     S390CPU *dst_cpu = NULL;
1726
1727     cpu_synchronize_state(CPU(cpu));
1728
1729     /* get order code */
1730     order = decode_basedisp_rs(env, run->s390_sieic.ipb, NULL)
1731         & SIGP_ORDER_MASK;
1732     status_reg = &env->regs[r1];
1733     param = (r1 % 2) ? env->regs[r1] : env->regs[r1 + 1];
1734
1735     switch (order) {
1736     case SIGP_SET_ARCH:
1737         ret = sigp_set_architecture(cpu, param, status_reg);
1738         break;
1739     default:
1740         /* all other sigp orders target a single vcpu */
1741         dst_cpu = s390_cpu_addr2state(env->regs[r3]);
1742         ret = handle_sigp_single_dst(dst_cpu, order, param, status_reg);
1743     }
1744
1745     trace_kvm_sigp_finished(order, CPU(cpu)->cpu_index,
1746                             dst_cpu ? CPU(dst_cpu)->cpu_index : -1, ret);
1747
1748     if (ret >= 0) {
1749         setcc(cpu, ret);
1750         return 0;
1751     }
1752
1753     return ret;
1754 }
1755
1756 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1757 {
1758     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1759     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1760     int r = -1;
1761
1762     DPRINTF("handle_instruction 0x%x 0x%x\n",
1763             run->s390_sieic.ipa, run->s390_sieic.ipb);
1764     switch (ipa0) {
1765     case IPA0_B2:
1766         r = handle_b2(cpu, run, ipa1);
1767         break;
1768     case IPA0_B9:
1769         r = handle_b9(cpu, run, ipa1);
1770         break;
1771     case IPA0_EB:
1772         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1773         break;
1774     case IPA0_E3:
1775         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1776         break;
1777     case IPA0_DIAG:
1778         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1779         break;
1780     case IPA0_SIGP:
1781         r = handle_sigp(cpu, run, ipa1);
1782         break;
1783     }
1784
1785     if (r < 0) {
1786         r = 0;
1787         enter_pgmcheck(cpu, 0x0001);
1788     }
1789
1790     return r;
1791 }
1792
1793 static bool is_special_wait_psw(CPUState *cs)
1794 {
1795     /* signal quiesce */
1796     return cs->kvm_run->psw_addr == 0xfffUL;
1797 }
1798
1799 static void guest_panicked(void)
1800 {
1801     qapi_event_send_guest_panicked(GUEST_PANIC_ACTION_PAUSE,
1802                                    &error_abort);
1803     vm_stop(RUN_STATE_GUEST_PANICKED);
1804 }
1805
1806 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1807 {
1808     CPUState *cs = CPU(cpu);
1809
1810     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1811                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1812                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1813     s390_cpu_halt(cpu);
1814     guest_panicked();
1815 }
1816
1817 static int handle_intercept(S390CPU *cpu)
1818 {
1819     CPUState *cs = CPU(cpu);
1820     struct kvm_run *run = cs->kvm_run;
1821     int icpt_code = run->s390_sieic.icptcode;
1822     int r = 0;
1823
1824     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1825             (long)cs->kvm_run->psw_addr);
1826     switch (icpt_code) {
1827         case ICPT_INSTRUCTION:
1828             r = handle_instruction(cpu, run);
1829             break;
1830         case ICPT_PROGRAM:
1831             unmanageable_intercept(cpu, "program interrupt",
1832                                    offsetof(LowCore, program_new_psw));
1833             r = EXCP_HALTED;
1834             break;
1835         case ICPT_EXT_INT:
1836             unmanageable_intercept(cpu, "external interrupt",
1837                                    offsetof(LowCore, external_new_psw));
1838             r = EXCP_HALTED;
1839             break;
1840         case ICPT_WAITPSW:
1841             /* disabled wait, since enabled wait is handled in kernel */
1842             cpu_synchronize_state(cs);
1843             if (s390_cpu_halt(cpu) == 0) {
1844                 if (is_special_wait_psw(cs)) {
1845                     qemu_system_shutdown_request();
1846                 } else {
1847                     guest_panicked();
1848                 }
1849             }
1850             r = EXCP_HALTED;
1851             break;
1852         case ICPT_CPU_STOP:
1853             if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
1854                 qemu_system_shutdown_request();
1855             }
1856             if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
1857                 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR,
1858                                       true);
1859             }
1860             cpu->env.sigp_order = 0;
1861             r = EXCP_HALTED;
1862             break;
1863         case ICPT_SOFT_INTERCEPT:
1864             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1865             exit(1);
1866             break;
1867         case ICPT_IO:
1868             fprintf(stderr, "KVM unimplemented icpt IO\n");
1869             exit(1);
1870             break;
1871         default:
1872             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1873             exit(1);
1874             break;
1875     }
1876
1877     return r;
1878 }
1879
1880 static int handle_tsch(S390CPU *cpu)
1881 {
1882     CPUState *cs = CPU(cpu);
1883     struct kvm_run *run = cs->kvm_run;
1884     int ret;
1885
1886     cpu_synchronize_state(cs);
1887
1888     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb);
1889     if (ret < 0) {
1890         /*
1891          * Failure.
1892          * If an I/O interrupt had been dequeued, we have to reinject it.
1893          */
1894         if (run->s390_tsch.dequeued) {
1895             kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1896                                   run->s390_tsch.subchannel_nr,
1897                                   run->s390_tsch.io_int_parm,
1898                                   run->s390_tsch.io_int_word);
1899         }
1900         ret = 0;
1901     }
1902     return ret;
1903 }
1904
1905 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1906 {
1907     struct sysib_322 sysib;
1908     int del;
1909
1910     if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1911         return;
1912     }
1913     /* Shift the stack of Extended Names to prepare for our own data */
1914     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1915             sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1916     /* First virt level, that doesn't provide Ext Names delimits stack. It is
1917      * assumed it's not capable of managing Extended Names for lower levels.
1918      */
1919     for (del = 1; del < sysib.count; del++) {
1920         if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1921             break;
1922         }
1923     }
1924     if (del < sysib.count) {
1925         memset(sysib.ext_names[del], 0,
1926                sizeof(sysib.ext_names[0]) * (sysib.count - del));
1927     }
1928     /* Insert short machine name in EBCDIC, padded with blanks */
1929     if (qemu_name) {
1930         memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1931         ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1932                                                     strlen(qemu_name)));
1933     }
1934     sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1935     memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1936     /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1937      * considered by s390 as not capable of providing any Extended Name.
1938      * Therefore if no name was specified on qemu invocation, we go with the
1939      * same "KVMguest" default, which KVM has filled into short name field.
1940      */
1941     if (qemu_name) {
1942         strncpy((char *)sysib.ext_names[0], qemu_name,
1943                 sizeof(sysib.ext_names[0]));
1944     } else {
1945         strcpy((char *)sysib.ext_names[0], "KVMguest");
1946     }
1947     /* Insert UUID */
1948     memcpy(sysib.vm[0].uuid, qemu_uuid, sizeof(sysib.vm[0].uuid));
1949
1950     s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1951 }
1952
1953 static int handle_stsi(S390CPU *cpu)
1954 {
1955     CPUState *cs = CPU(cpu);
1956     struct kvm_run *run = cs->kvm_run;
1957
1958     switch (run->s390_stsi.fc) {
1959     case 3:
1960         if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
1961             return 0;
1962         }
1963         /* Only sysib 3.2.2 needs post-handling for now. */
1964         insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
1965         return 0;
1966     default:
1967         return 0;
1968     }
1969 }
1970
1971 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1972 {
1973     CPUState *cs = CPU(cpu);
1974     struct kvm_run *run = cs->kvm_run;
1975
1976     int ret = 0;
1977     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1978
1979     switch (arch_info->type) {
1980     case KVM_HW_WP_WRITE:
1981         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1982             cs->watchpoint_hit = &hw_watchpoint;
1983             hw_watchpoint.vaddr = arch_info->addr;
1984             hw_watchpoint.flags = BP_MEM_WRITE;
1985             ret = EXCP_DEBUG;
1986         }
1987         break;
1988     case KVM_HW_BP:
1989         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1990             ret = EXCP_DEBUG;
1991         }
1992         break;
1993     case KVM_SINGLESTEP:
1994         if (cs->singlestep_enabled) {
1995             ret = EXCP_DEBUG;
1996         }
1997         break;
1998     default:
1999         ret = -ENOSYS;
2000     }
2001
2002     return ret;
2003 }
2004
2005 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
2006 {
2007     S390CPU *cpu = S390_CPU(cs);
2008     int ret = 0;
2009
2010     qemu_mutex_lock_iothread();
2011
2012     switch (run->exit_reason) {
2013         case KVM_EXIT_S390_SIEIC:
2014             ret = handle_intercept(cpu);
2015             break;
2016         case KVM_EXIT_S390_RESET:
2017             s390_reipl_request();
2018             break;
2019         case KVM_EXIT_S390_TSCH:
2020             ret = handle_tsch(cpu);
2021             break;
2022         case KVM_EXIT_S390_STSI:
2023             ret = handle_stsi(cpu);
2024             break;
2025         case KVM_EXIT_DEBUG:
2026             ret = kvm_arch_handle_debug_exit(cpu);
2027             break;
2028         default:
2029             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
2030             break;
2031     }
2032     qemu_mutex_unlock_iothread();
2033
2034     if (ret == 0) {
2035         ret = EXCP_INTERRUPT;
2036     }
2037     return ret;
2038 }
2039
2040 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2041 {
2042     return true;
2043 }
2044
2045 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2046 {
2047     return 1;
2048 }
2049
2050 int kvm_arch_on_sigbus(int code, void *addr)
2051 {
2052     return 1;
2053 }
2054
2055 void kvm_s390_io_interrupt(uint16_t subchannel_id,
2056                            uint16_t subchannel_nr, uint32_t io_int_parm,
2057                            uint32_t io_int_word)
2058 {
2059     struct kvm_s390_irq irq = {
2060         .u.io.subchannel_id = subchannel_id,
2061         .u.io.subchannel_nr = subchannel_nr,
2062         .u.io.io_int_parm = io_int_parm,
2063         .u.io.io_int_word = io_int_word,
2064     };
2065
2066     if (io_int_word & IO_INT_WORD_AI) {
2067         irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
2068     } else {
2069         irq.type = ((subchannel_id & 0xff00) << 24) |
2070             ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
2071     }
2072     kvm_s390_floating_interrupt(&irq);
2073 }
2074
2075 void kvm_s390_crw_mchk(void)
2076 {
2077     struct kvm_s390_irq irq = {
2078         .type = KVM_S390_MCHK,
2079         .u.mchk.cr14 = 1 << 28,
2080         .u.mchk.mcic = 0x00400f1d40330000ULL,
2081     };
2082     kvm_s390_floating_interrupt(&irq);
2083 }
2084
2085 void kvm_s390_enable_css_support(S390CPU *cpu)
2086 {
2087     int r;
2088
2089     /* Activate host kernel channel subsystem support. */
2090     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
2091     assert(r == 0);
2092 }
2093
2094 void kvm_arch_init_irq_routing(KVMState *s)
2095 {
2096     /*
2097      * Note that while irqchip capabilities generally imply that cpustates
2098      * are handled in-kernel, it is not true for s390 (yet); therefore, we
2099      * have to override the common code kvm_halt_in_kernel_allowed setting.
2100      */
2101     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2102         kvm_gsi_routing_allowed = true;
2103         kvm_halt_in_kernel_allowed = false;
2104     }
2105 }
2106
2107 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
2108                                     int vq, bool assign)
2109 {
2110     struct kvm_ioeventfd kick = {
2111         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
2112         KVM_IOEVENTFD_FLAG_DATAMATCH,
2113         .fd = event_notifier_get_fd(notifier),
2114         .datamatch = vq,
2115         .addr = sch,
2116         .len = 8,
2117     };
2118     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
2119         return -ENOSYS;
2120     }
2121     if (!assign) {
2122         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
2123     }
2124     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
2125 }
2126
2127 int kvm_s390_get_memslot_count(KVMState *s)
2128 {
2129     return kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
2130 }
2131
2132 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
2133 {
2134     struct kvm_mp_state mp_state = {};
2135     int ret;
2136
2137     /* the kvm part might not have been initialized yet */
2138     if (CPU(cpu)->kvm_state == NULL) {
2139         return 0;
2140     }
2141
2142     switch (cpu_state) {
2143     case CPU_STATE_STOPPED:
2144         mp_state.mp_state = KVM_MP_STATE_STOPPED;
2145         break;
2146     case CPU_STATE_CHECK_STOP:
2147         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
2148         break;
2149     case CPU_STATE_OPERATING:
2150         mp_state.mp_state = KVM_MP_STATE_OPERATING;
2151         break;
2152     case CPU_STATE_LOAD:
2153         mp_state.mp_state = KVM_MP_STATE_LOAD;
2154         break;
2155     default:
2156         error_report("Requested CPU state is not a valid S390 CPU state: %u",
2157                      cpu_state);
2158         exit(1);
2159     }
2160
2161     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2162     if (ret) {
2163         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
2164                                        strerror(-ret));
2165     }
2166
2167     return ret;
2168 }
2169
2170 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
2171 {
2172     struct kvm_s390_irq_state irq_state;
2173     CPUState *cs = CPU(cpu);
2174     int32_t bytes;
2175
2176     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2177         return;
2178     }
2179
2180     irq_state.buf = (uint64_t) cpu->irqstate;
2181     irq_state.len = VCPU_IRQ_BUF_SIZE;
2182
2183     bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
2184     if (bytes < 0) {
2185         cpu->irqstate_saved_size = 0;
2186         error_report("Migration of interrupt state failed");
2187         return;
2188     }
2189
2190     cpu->irqstate_saved_size = bytes;
2191 }
2192
2193 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
2194 {
2195     CPUState *cs = CPU(cpu);
2196     struct kvm_s390_irq_state irq_state;
2197     int r;
2198
2199     if (cpu->irqstate_saved_size == 0) {
2200         return 0;
2201     }
2202
2203     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2204         return -ENOSYS;
2205     }
2206
2207     irq_state.buf = (uint64_t) cpu->irqstate;
2208     irq_state.len = cpu->irqstate_saved_size;
2209
2210     r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
2211     if (r) {
2212         error_report("Setting interrupt state failed %d", r);
2213     }
2214     return r;
2215 }
2216
2217 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2218                               uint64_t address, uint32_t data)
2219 {
2220     S390PCIBusDevice *pbdev;
2221     uint32_t fid = data >> ZPCI_MSI_VEC_BITS;
2222     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
2223
2224     pbdev = s390_pci_find_dev_by_fid(fid);
2225     if (!pbdev) {
2226         DPRINTF("add_msi_route no dev\n");
2227         return -ENODEV;
2228     }
2229
2230     pbdev->routes.adapter.ind_offset = vec;
2231
2232     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2233     route->flags = 0;
2234     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2235     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2236     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2237     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset;
2238     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2239     return 0;
2240 }
2241
2242 int kvm_arch_msi_data_to_gsi(uint32_t data)
2243 {
2244     abort();
2245 }