target-s390x/kvm.c

   1 /*
   2  * QEMU S390x KVM implementation
   3  *
   4  * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
   5  * Copyright IBM Corp. 2012
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * Contributions after 2012-10-29 are licensed under the terms of the
  18  * GNU GPL, version 2 or (at your option) any later version.
  19  *
  20  * You should have received a copy of the GNU (Lesser) General Public
  21  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  22  */
  23
  24 #include "qemu/osdep.h"
  25 #include <sys/ioctl.h>
  26 #include <sys/mman.h>
  27
  28 #include <linux/kvm.h>
  29 #include <asm/ptrace.h>
  30
  31 #include "qemu-common.h"
  32 #include "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_cmma_reset(void)
 177 {
 178     int rc;
 179     struct kvm_device_attr attr = {
 180         .group = KVM_S390_VM_MEM_CTRL,
 181         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 182     };
 183
 184     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 185     trace_kvm_clear_cmma(rc);
 186 }
 187
 188 static void kvm_s390_enable_cmma(KVMState *s)
 189 {
 190     int rc;
 191     struct kvm_device_attr attr = {
 192         .group = KVM_S390_VM_MEM_CTRL,
 193         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 194     };
 195
 196     if (!kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_ENABLE_CMMA) ||
 197         !kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_CLR_CMMA)) {
 198         return;
 199     }
 200
 201     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 202     trace_kvm_enable_cmma(rc);
 203 }
 204
 205 static void kvm_s390_set_attr(uint64_t attr)
 206 {
 207     struct kvm_device_attr attribute = {
 208         .group = KVM_S390_VM_CRYPTO,
 209         .attr  = attr,
 210     };
 211
 212     int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
 213
 214     if (ret) {
 215         error_report("Failed to set crypto device attribute %lu: %s",
 216                      attr, strerror(-ret));
 217     }
 218 }
 219
 220 static void kvm_s390_init_aes_kw(void)
 221 {
 222     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
 223
 224     if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
 225                                  NULL)) {
 226             attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
 227     }
 228
 229     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 230             kvm_s390_set_attr(attr);
 231     }
 232 }
 233
 234 static void kvm_s390_init_dea_kw(void)
 235 {
 236     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
 237
 238     if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
 239                                  NULL)) {
 240             attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
 241     }
 242
 243     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 244             kvm_s390_set_attr(attr);
 245     }
 246 }
 247
 248 void kvm_s390_crypto_reset(void)
 249 {
 250     kvm_s390_init_aes_kw();
 251     kvm_s390_init_dea_kw();
 252 }
 253
 254 int kvm_arch_init(MachineState *ms, KVMState *s)
 255 {
 256     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
 257     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
 258     cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
 259     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
 260
 261     if (!mem_path) {
 262         kvm_s390_enable_cmma(s);
 263     }
 264
 265     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
 266         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
 267         phys_mem_set_alloc(legacy_s390_alloc);
 268     }
 269
 270     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
 271     kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
 272     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_STSI, 0);
 273
 274     return 0;
 275 }
 276
 277 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
 278 {
 279     return cpu->cpu_index;
 280 }
 281
 282 int kvm_arch_init_vcpu(CPUState *cs)
 283 {
 284     S390CPU *cpu = S390_CPU(cs);
 285     kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
 286     cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE);
 287     return 0;
 288 }
 289
 290 void kvm_s390_reset_vcpu(S390CPU *cpu)
 291 {
 292     CPUState *cs = CPU(cpu);
 293
 294     /* The initial reset call is needed here to reset in-kernel
 295      * vcpu data that we can't access directly from QEMU
 296      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
 297      * Before this ioctl cpu_synchronize_state() is called in common kvm
 298      * code (kvm-all) */
 299     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
 300         error_report("Initial CPU reset failed on CPU %i", cs->cpu_index);
 301     }
 302 }
 303
 304 static int can_sync_regs(CPUState *cs, int regs)
 305 {
 306     return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
 307 }
 308
 309 int kvm_arch_put_registers(CPUState *cs, int level)
 310 {
 311     S390CPU *cpu = S390_CPU(cs);
 312     CPUS390XState *env = &cpu->env;
 313     struct kvm_sregs sregs;
 314     struct kvm_regs regs;
 315     struct kvm_fpu fpu = {};
 316     int r;
 317     int i;
 318
 319     /* always save the PSW  and the GPRS*/
 320     cs->kvm_run->psw_addr = env->psw.addr;
 321     cs->kvm_run->psw_mask = env->psw.mask;
 322
 323     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 324         for (i = 0; i < 16; i++) {
 325             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
 326             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
 327         }
 328     } else {
 329         for (i = 0; i < 16; i++) {
 330             regs.gprs[i] = env->regs[i];
 331         }
 332         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
 333         if (r < 0) {
 334             return r;
 335         }
 336     }
 337
 338     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 339         for (i = 0; i < 32; i++) {
 340             cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
 341             cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
 342         }
 343         cs->kvm_run->s.regs.fpc = env->fpc;
 344         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
 345     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
 346         for (i = 0; i < 16; i++) {
 347             cs->kvm_run->s.regs.fprs[i] = get_freg(env, i)->ll;
 348         }
 349         cs->kvm_run->s.regs.fpc = env->fpc;
 350         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_FPRS;
 351     } else {
 352         /* Floating point */
 353         for (i = 0; i < 16; i++) {
 354             fpu.fprs[i] = get_freg(env, i)->ll;
 355         }
 356         fpu.fpc = env->fpc;
 357
 358         r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
 359         if (r < 0) {
 360             return r;
 361         }
 362     }
 363
 364     /* Do we need to save more than that? */
 365     if (level == KVM_PUT_RUNTIME_STATE) {
 366         return 0;
 367     }
 368
 369     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 370         cs->kvm_run->s.regs.cputm = env->cputm;
 371         cs->kvm_run->s.regs.ckc = env->ckc;
 372         cs->kvm_run->s.regs.todpr = env->todpr;
 373         cs->kvm_run->s.regs.gbea = env->gbea;
 374         cs->kvm_run->s.regs.pp = env->pp;
 375         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
 376     } else {
 377         /*
 378          * These ONE_REGS are not protected by a capability. As they are only
 379          * necessary for migration we just trace a possible error, but don't
 380          * return with an error return code.
 381          */
 382         kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 383         kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 384         kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 385         kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 386         kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 387     }
 388
 389     /* pfault parameters */
 390     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 391         cs->kvm_run->s.regs.pft = env->pfault_token;
 392         cs->kvm_run->s.regs.pfs = env->pfault_select;
 393         cs->kvm_run->s.regs.pfc = env->pfault_compare;
 394         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
 395     } else if (cap_async_pf) {
 396         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 397         if (r < 0) {
 398             return r;
 399         }
 400         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 401         if (r < 0) {
 402             return r;
 403         }
 404         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 405         if (r < 0) {
 406             return r;
 407         }
 408     }
 409
 410     /* access registers and control registers*/
 411     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 412         for (i = 0; i < 16; i++) {
 413             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
 414             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
 415         }
 416         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
 417         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
 418     } else {
 419         for (i = 0; i < 16; i++) {
 420             sregs.acrs[i] = env->aregs[i];
 421             sregs.crs[i] = env->cregs[i];
 422         }
 423         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
 424         if (r < 0) {
 425             return r;
 426         }
 427     }
 428
 429     /* Finally the prefix */
 430     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 431         cs->kvm_run->s.regs.prefix = env->psa;
 432         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
 433     } else {
 434         /* prefix is only supported via sync regs */
 435     }
 436     return 0;
 437 }
 438
 439 int kvm_arch_get_registers(CPUState *cs)
 440 {
 441     S390CPU *cpu = S390_CPU(cs);
 442     CPUS390XState *env = &cpu->env;
 443     struct kvm_sregs sregs;
 444     struct kvm_regs regs;
 445     struct kvm_fpu fpu;
 446     int i, r;
 447
 448     /* get the PSW */
 449     env->psw.addr = cs->kvm_run->psw_addr;
 450     env->psw.mask = cs->kvm_run->psw_mask;
 451
 452     /* the GPRS */
 453     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 454         for (i = 0; i < 16; i++) {
 455             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
 456         }
 457     } else {
 458         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
 459         if (r < 0) {
 460             return r;
 461         }
 462          for (i = 0; i < 16; i++) {
 463             env->regs[i] = regs.gprs[i];
 464         }
 465     }
 466
 467     /* The ACRS and CRS */
 468     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 469         for (i = 0; i < 16; i++) {
 470             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
 471             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
 472         }
 473     } else {
 474         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
 475         if (r < 0) {
 476             return r;
 477         }
 478          for (i = 0; i < 16; i++) {
 479             env->aregs[i] = sregs.acrs[i];
 480             env->cregs[i] = sregs.crs[i];
 481         }
 482     }
 483
 484     /* Floating point and vector registers */
 485     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 486         for (i = 0; i < 32; i++) {
 487             env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
 488             env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
 489         }
 490         env->fpc = cs->kvm_run->s.regs.fpc;
 491     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
 492         for (i = 0; i < 16; i++) {
 493             get_freg(env, i)->ll = cs->kvm_run->s.regs.fprs[i];
 494         }
 495         env->fpc = cs->kvm_run->s.regs.fpc;
 496     } else {
 497         r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
 498         if (r < 0) {
 499             return r;
 500         }
 501         for (i = 0; i < 16; i++) {
 502             get_freg(env, i)->ll = fpu.fprs[i];
 503         }
 504         env->fpc = fpu.fpc;
 505     }
 506
 507     /* The prefix */
 508     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 509         env->psa = cs->kvm_run->s.regs.prefix;
 510     }
 511
 512     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 513         env->cputm = cs->kvm_run->s.regs.cputm;
 514         env->ckc = cs->kvm_run->s.regs.ckc;
 515         env->todpr = cs->kvm_run->s.regs.todpr;
 516         env->gbea = cs->kvm_run->s.regs.gbea;
 517         env->pp = cs->kvm_run->s.regs.pp;
 518     } else {
 519         /*
 520          * These ONE_REGS are not protected by a capability. As they are only
 521          * necessary for migration we just trace a possible error, but don't
 522          * return with an error return code.
 523          */
 524         kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 525         kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 526         kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 527         kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 528         kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 529     }
 530
 531     /* pfault parameters */
 532     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 533         env->pfault_token = cs->kvm_run->s.regs.pft;
 534         env->pfault_select = cs->kvm_run->s.regs.pfs;
 535         env->pfault_compare = cs->kvm_run->s.regs.pfc;
 536     } else if (cap_async_pf) {
 537         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 538         if (r < 0) {
 539             return r;
 540         }
 541         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 542         if (r < 0) {
 543             return r;
 544         }
 545         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 546         if (r < 0) {
 547             return r;
 548         }
 549     }
 550
 551     return 0;
 552 }
 553
 554 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
 555 {
 556     int r;
 557     struct kvm_device_attr attr = {
 558         .group = KVM_S390_VM_TOD,
 559         .attr = KVM_S390_VM_TOD_LOW,
 560         .addr = (uint64_t)tod_low,
 561     };
 562
 563     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 564     if (r) {
 565         return r;
 566     }
 567
 568     attr.attr = KVM_S390_VM_TOD_HIGH;
 569     attr.addr = (uint64_t)tod_high;
 570     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 571 }
 572
 573 int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_low)
 574 {
 575     int r;
 576
 577     struct kvm_device_attr attr = {
 578         .group = KVM_S390_VM_TOD,
 579         .attr = KVM_S390_VM_TOD_LOW,
 580         .addr = (uint64_t)tod_low,
 581     };
 582
 583     r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 584     if (r) {
 585         return r;
 586     }
 587
 588     attr.attr = KVM_S390_VM_TOD_HIGH;
 589     attr.addr = (uint64_t)tod_high;
 590     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 591 }
 592
 593 /**
 594  * kvm_s390_mem_op:
 595  * @addr:      the logical start address in guest memory
 596  * @ar:        the access register number
 597  * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
 598  * @len:       length that should be transferred
 599  * @is_write:  true = write, false = read
 600  * Returns:    0 on success, non-zero if an exception or error occurred
 601  *
 602  * Use KVM ioctl to read/write from/to guest memory. An access exception
 603  * is injected into the vCPU in case of translation errors.
 604  */
 605 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
 606                     int len, bool is_write)
 607 {
 608     struct kvm_s390_mem_op mem_op = {
 609         .gaddr = addr,
 610         .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
 611         .size = len,
 612         .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
 613                        : KVM_S390_MEMOP_LOGICAL_READ,
 614         .buf = (uint64_t)hostbuf,
 615         .ar = ar,
 616     };
 617     int ret;
 618
 619     if (!cap_mem_op) {
 620         return -ENOSYS;
 621     }
 622     if (!hostbuf) {
 623         mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
 624     }
 625
 626     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
 627     if (ret < 0) {
 628         error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
 629     }
 630     return ret;
 631 }
 632
 633 /*
 634  * Legacy layout for s390:
 635  * Older S390 KVM requires the topmost vma of the RAM to be
 636  * smaller than an system defined value, which is at least 256GB.
 637  * Larger systems have larger values. We put the guest between
 638  * the end of data segment (system break) and this value. We
 639  * use 32GB as a base to have enough room for the system break
 640  * to grow. We also have to use MAP parameters that avoid
 641  * read-only mapping of guest pages.
 642  */
 643 static void *legacy_s390_alloc(size_t size, uint64_t *align)
 644 {
 645     void *mem;
 646
 647     mem = mmap((void *) 0x800000000ULL, size,
 648                PROT_EXEC|PROT_READ|PROT_WRITE,
 649                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 650     return mem == MAP_FAILED ? NULL : mem;
 651 }
 652
 653 /* DIAG 501 is used for sw breakpoints */
 654 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 655
 656 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 657 {
 658
 659     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 660                             sizeof(diag_501), 0) ||
 661         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501,
 662                             sizeof(diag_501), 1)) {
 663         return -EINVAL;
 664     }
 665     return 0;
 666 }
 667
 668 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 669 {
 670     uint8_t t[sizeof(diag_501)];
 671
 672     if (cpu_memory_rw_debug(cs, bp->pc, t, sizeof(diag_501), 0)) {
 673         return -EINVAL;
 674     } else if (memcmp(t, diag_501, sizeof(diag_501))) {
 675         return -EINVAL;
 676     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 677                                    sizeof(diag_501), 1)) {
 678         return -EINVAL;
 679     }
 680
 681     return 0;
 682 }
 683
 684 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
 685                                                     int len, int type)
 686 {
 687     int n;
 688
 689     for (n = 0; n < nb_hw_breakpoints; n++) {
 690         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
 691             (hw_breakpoints[n].len == len || len == -1)) {
 692             return &hw_breakpoints[n];
 693         }
 694     }
 695
 696     return NULL;
 697 }
 698
 699 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
 700 {
 701     int size;
 702
 703     if (find_hw_breakpoint(addr, len, type)) {
 704         return -EEXIST;
 705     }
 706
 707     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
 708
 709     if (!hw_breakpoints) {
 710         nb_hw_breakpoints = 0;
 711         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
 712     } else {
 713         hw_breakpoints =
 714             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
 715     }
 716
 717     if (!hw_breakpoints) {
 718         nb_hw_breakpoints = 0;
 719         return -ENOMEM;
 720     }
 721
 722     hw_breakpoints[nb_hw_breakpoints].addr = addr;
 723     hw_breakpoints[nb_hw_breakpoints].len = len;
 724     hw_breakpoints[nb_hw_breakpoints].type = type;
 725
 726     nb_hw_breakpoints++;
 727
 728     return 0;
 729 }
 730
 731 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 732                                   target_ulong len, int type)
 733 {
 734     switch (type) {
 735     case GDB_BREAKPOINT_HW:
 736         type = KVM_HW_BP;
 737         break;
 738     case GDB_WATCHPOINT_WRITE:
 739         if (len < 1) {
 740             return -EINVAL;
 741         }
 742         type = KVM_HW_WP_WRITE;
 743         break;
 744     default:
 745         return -ENOSYS;
 746     }
 747     return insert_hw_breakpoint(addr, len, type);
 748 }
 749
 750 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 751                                   target_ulong len, int type)
 752 {
 753     int size;
 754     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
 755
 756     if (bp == NULL) {
 757         return -ENOENT;
 758     }
 759
 760     nb_hw_breakpoints--;
 761     if (nb_hw_breakpoints > 0) {
 762         /*
 763          * In order to trim the array, move the last element to the position to
 764          * be removed - if necessary.
 765          */
 766         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
 767             *bp = hw_breakpoints[nb_hw_breakpoints];
 768         }
 769         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
 770         hw_breakpoints =
 771              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
 772     } else {
 773         g_free(hw_breakpoints);
 774         hw_breakpoints = NULL;
 775     }
 776
 777     return 0;
 778 }
 779
 780 void kvm_arch_remove_all_hw_breakpoints(void)
 781 {
 782     nb_hw_breakpoints = 0;
 783     g_free(hw_breakpoints);
 784     hw_breakpoints = NULL;
 785 }
 786
 787 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 788 {
 789     int i;
 790
 791     if (nb_hw_breakpoints > 0) {
 792         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
 793         dbg->arch.hw_bp = hw_breakpoints;
 794
 795         for (i = 0; i < nb_hw_breakpoints; ++i) {
 796             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
 797                                                        hw_breakpoints[i].addr);
 798         }
 799         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 800     } else {
 801         dbg->arch.nr_hw_bp = 0;
 802         dbg->arch.hw_bp = NULL;
 803     }
 804 }
 805
 806 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 807 {
 808 }
 809
 810 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
 811 {
 812     return MEMTXATTRS_UNSPECIFIED;
 813 }
 814
 815 int kvm_arch_process_async_events(CPUState *cs)
 816 {
 817     return cs->halted;
 818 }
 819
 820 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
 821                                      struct kvm_s390_interrupt *interrupt)
 822 {
 823     int r = 0;
 824
 825     interrupt->type = irq->type;
 826     switch (irq->type) {
 827     case KVM_S390_INT_VIRTIO:
 828         interrupt->parm = irq->u.ext.ext_params;
 829         /* fall through */
 830     case KVM_S390_INT_PFAULT_INIT:
 831     case KVM_S390_INT_PFAULT_DONE:
 832         interrupt->parm64 = irq->u.ext.ext_params2;
 833         break;
 834     case KVM_S390_PROGRAM_INT:
 835         interrupt->parm = irq->u.pgm.code;
 836         break;
 837     case KVM_S390_SIGP_SET_PREFIX:
 838         interrupt->parm = irq->u.prefix.address;
 839         break;
 840     case KVM_S390_INT_SERVICE:
 841         interrupt->parm = irq->u.ext.ext_params;
 842         break;
 843     case KVM_S390_MCHK:
 844         interrupt->parm = irq->u.mchk.cr14;
 845         interrupt->parm64 = irq->u.mchk.mcic;
 846         break;
 847     case KVM_S390_INT_EXTERNAL_CALL:
 848         interrupt->parm = irq->u.extcall.code;
 849         break;
 850     case KVM_S390_INT_EMERGENCY:
 851         interrupt->parm = irq->u.emerg.code;
 852         break;
 853     case KVM_S390_SIGP_STOP:
 854     case KVM_S390_RESTART:
 855         break; /* These types have no parameters */
 856     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
 857         interrupt->parm = irq->u.io.subchannel_id << 16;
 858         interrupt->parm |= irq->u.io.subchannel_nr;
 859         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
 860         interrupt->parm64 |= irq->u.io.io_int_word;
 861         break;
 862     default:
 863         r = -EINVAL;
 864         break;
 865     }
 866     return r;
 867 }
 868
 869 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
 870 {
 871     struct kvm_s390_interrupt kvmint = {};
 872     int r;
 873
 874     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 875     if (r < 0) {
 876         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 877         exit(1);
 878     }
 879
 880     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
 881     if (r < 0) {
 882         fprintf(stderr, "KVM failed to inject interrupt\n");
 883         exit(1);
 884     }
 885 }
 886
 887 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
 888 {
 889     CPUState *cs = CPU(cpu);
 890     int r;
 891
 892     if (cap_s390_irq) {
 893         r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
 894         if (!r) {
 895             return;
 896         }
 897         error_report("KVM failed to inject interrupt %llx", irq->type);
 898         exit(1);
 899     }
 900
 901     inject_vcpu_irq_legacy(cs, irq);
 902 }
 903
 904 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 905 {
 906     struct kvm_s390_interrupt kvmint = {};
 907     int r;
 908
 909     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 910     if (r < 0) {
 911         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 912         exit(1);
 913     }
 914
 915     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
 916     if (r < 0) {
 917         fprintf(stderr, "KVM failed to inject interrupt\n");
 918         exit(1);
 919     }
 920 }
 921
 922 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 923 {
 924     static bool use_flic = true;
 925     int r;
 926
 927     if (use_flic) {
 928         r = kvm_s390_inject_flic(irq);
 929         if (r == -ENOSYS) {
 930             use_flic = false;
 931         }
 932         if (!r) {
 933             return;
 934         }
 935     }
 936     __kvm_s390_floating_interrupt(irq);
 937 }
 938
 939 void kvm_s390_service_interrupt(uint32_t parm)
 940 {
 941     struct kvm_s390_irq irq = {
 942         .type = KVM_S390_INT_SERVICE,
 943         .u.ext.ext_params = parm,
 944     };
 945
 946     kvm_s390_floating_interrupt(&irq);
 947 }
 948
 949 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
 950 {
 951     struct kvm_s390_irq irq = {
 952         .type = KVM_S390_PROGRAM_INT,
 953         .u.pgm.code = code,
 954     };
 955
 956     kvm_s390_vcpu_interrupt(cpu, &irq);
 957 }
 958
 959 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
 960 {
 961     struct kvm_s390_irq irq = {
 962         .type = KVM_S390_PROGRAM_INT,
 963         .u.pgm.code = code,
 964         .u.pgm.trans_exc_code = te_code,
 965         .u.pgm.exc_access_id = te_code & 3,
 966     };
 967
 968     kvm_s390_vcpu_interrupt(cpu, &irq);
 969 }
 970
 971 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
 972                                  uint16_t ipbh0)
 973 {
 974     CPUS390XState *env = &cpu->env;
 975     uint64_t sccb;
 976     uint32_t code;
 977     int r = 0;
 978
 979     cpu_synchronize_state(CPU(cpu));
 980     sccb = env->regs[ipbh0 & 0xf];
 981     code = env->regs[(ipbh0 & 0xf0) >> 4];
 982
 983     r = sclp_service_call(env, sccb, code);
 984     if (r < 0) {
 985         enter_pgmcheck(cpu, -r);
 986     } else {
 987         setcc(cpu, r);
 988     }
 989
 990     return 0;
 991 }
 992
 993 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 994 {
 995     CPUS390XState *env = &cpu->env;
 996     int rc = 0;
 997     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
 998
 999     cpu_synchronize_state(CPU(cpu));
1000
1001     switch (ipa1) {
1002     case PRIV_B2_XSCH:
1003         ioinst_handle_xsch(cpu, env->regs[1]);
1004         break;
1005     case PRIV_B2_CSCH:
1006         ioinst_handle_csch(cpu, env->regs[1]);
1007         break;
1008     case PRIV_B2_HSCH:
1009         ioinst_handle_hsch(cpu, env->regs[1]);
1010         break;
1011     case PRIV_B2_MSCH:
1012         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
1013         break;
1014     case PRIV_B2_SSCH:
1015         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
1016         break;
1017     case PRIV_B2_STCRW:
1018         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
1019         break;
1020     case PRIV_B2_STSCH:
1021         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
1022         break;
1023     case PRIV_B2_TSCH:
1024         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1025         fprintf(stderr, "Spurious tsch intercept\n");
1026         break;
1027     case PRIV_B2_CHSC:
1028         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
1029         break;
1030     case PRIV_B2_TPI:
1031         /* This should have been handled by kvm already. */
1032         fprintf(stderr, "Spurious tpi intercept\n");
1033         break;
1034     case PRIV_B2_SCHM:
1035         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1036                            run->s390_sieic.ipb);
1037         break;
1038     case PRIV_B2_RSCH:
1039         ioinst_handle_rsch(cpu, env->regs[1]);
1040         break;
1041     case PRIV_B2_RCHP:
1042         ioinst_handle_rchp(cpu, env->regs[1]);
1043         break;
1044     case PRIV_B2_STCPS:
1045         /* We do not provide this instruction, it is suppressed. */
1046         break;
1047     case PRIV_B2_SAL:
1048         ioinst_handle_sal(cpu, env->regs[1]);
1049         break;
1050     case PRIV_B2_SIGA:
1051         /* Not provided, set CC = 3 for subchannel not operational */
1052         setcc(cpu, 3);
1053         break;
1054     case PRIV_B2_SCLP_CALL:
1055         rc = kvm_sclp_service_call(cpu, run, ipbh0);
1056         break;
1057     default:
1058         rc = -1;
1059         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1060         break;
1061     }
1062
1063     return rc;
1064 }
1065
1066 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1067                                   uint8_t *ar)
1068 {
1069     CPUS390XState *env = &cpu->env;
1070     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1071     uint32_t base2 = run->s390_sieic.ipb >> 28;
1072     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1073                      ((run->s390_sieic.ipb & 0xff00) << 4);
1074
1075     if (disp2 & 0x80000) {
1076         disp2 += 0xfff00000;
1077     }
1078     if (ar) {
1079         *ar = base2;
1080     }
1081
1082     return (base2 ? env->regs[base2] : 0) +
1083            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1084 }
1085
1086 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1087                                   uint8_t *ar)
1088 {
1089     CPUS390XState *env = &cpu->env;
1090     uint32_t base2 = run->s390_sieic.ipb >> 28;
1091     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1092                      ((run->s390_sieic.ipb & 0xff00) << 4);
1093
1094     if (disp2 & 0x80000) {
1095         disp2 += 0xfff00000;
1096     }
1097     if (ar) {
1098         *ar = base2;
1099     }
1100
1101     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1102 }
1103
1104 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1105 {
1106     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1107
1108     return clp_service_call(cpu, r2);
1109 }
1110
1111 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1112 {
1113     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1114     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1115
1116     return pcilg_service_call(cpu, r1, r2);
1117 }
1118
1119 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1120 {
1121     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1122     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1123
1124     return pcistg_service_call(cpu, r1, r2);
1125 }
1126
1127 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1128 {
1129     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1130     uint64_t fiba;
1131     uint8_t ar;
1132
1133     cpu_synchronize_state(CPU(cpu));
1134     fiba = get_base_disp_rxy(cpu, run, &ar);
1135
1136     return stpcifc_service_call(cpu, r1, fiba, ar);
1137 }
1138
1139 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1140 {
1141     /* NOOP */
1142     return 0;
1143 }
1144
1145 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1146 {
1147     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1148     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1149
1150     return rpcit_service_call(cpu, r1, r2);
1151 }
1152
1153 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1154 {
1155     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1156     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1157     uint64_t gaddr;
1158     uint8_t ar;
1159
1160     cpu_synchronize_state(CPU(cpu));
1161     gaddr = get_base_disp_rsy(cpu, run, &ar);
1162
1163     return pcistb_service_call(cpu, r1, r3, gaddr, ar);
1164 }
1165
1166 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1167 {
1168     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1169     uint64_t fiba;
1170     uint8_t ar;
1171
1172     cpu_synchronize_state(CPU(cpu));
1173     fiba = get_base_disp_rxy(cpu, run, &ar);
1174
1175     return mpcifc_service_call(cpu, r1, fiba, ar);
1176 }
1177
1178 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1179 {
1180     int r = 0;
1181
1182     switch (ipa1) {
1183     case PRIV_B9_CLP:
1184         r = kvm_clp_service_call(cpu, run);
1185         break;
1186     case PRIV_B9_PCISTG:
1187         r = kvm_pcistg_service_call(cpu, run);
1188         break;
1189     case PRIV_B9_PCILG:
1190         r = kvm_pcilg_service_call(cpu, run);
1191         break;
1192     case PRIV_B9_RPCIT:
1193         r = kvm_rpcit_service_call(cpu, run);
1194         break;
1195     case PRIV_B9_EQBS:
1196         /* just inject exception */
1197         r = -1;
1198         break;
1199     default:
1200         r = -1;
1201         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1202         break;
1203     }
1204
1205     return r;
1206 }
1207
1208 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1209 {
1210     int r = 0;
1211
1212     switch (ipbl) {
1213     case PRIV_EB_PCISTB:
1214         r = kvm_pcistb_service_call(cpu, run);
1215         break;
1216     case PRIV_EB_SIC:
1217         r = kvm_sic_service_call(cpu, run);
1218         break;
1219     case PRIV_EB_SQBS:
1220         /* just inject exception */
1221         r = -1;
1222         break;
1223     default:
1224         r = -1;
1225         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1226         break;
1227     }
1228
1229     return r;
1230 }
1231
1232 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1233 {
1234     int r = 0;
1235
1236     switch (ipbl) {
1237     case PRIV_E3_MPCIFC:
1238         r = kvm_mpcifc_service_call(cpu, run);
1239         break;
1240     case PRIV_E3_STPCIFC:
1241         r = kvm_stpcifc_service_call(cpu, run);
1242         break;
1243     default:
1244         r = -1;
1245         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1246         break;
1247     }
1248
1249     return r;
1250 }
1251
1252 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1253 {
1254     CPUS390XState *env = &cpu->env;
1255     int ret;
1256
1257     cpu_synchronize_state(CPU(cpu));
1258     ret = s390_virtio_hypercall(env);
1259     if (ret == -EINVAL) {
1260         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1261         return 0;
1262     }
1263
1264     return ret;
1265 }
1266
1267 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1268 {
1269     uint64_t r1, r3;
1270     int rc;
1271
1272     cpu_synchronize_state(CPU(cpu));
1273     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1274     r3 = run->s390_sieic.ipa & 0x000f;
1275     rc = handle_diag_288(&cpu->env, r1, r3);
1276     if (rc) {
1277         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1278     }
1279 }
1280
1281 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1282 {
1283     uint64_t r1, r3;
1284
1285     cpu_synchronize_state(CPU(cpu));
1286     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1287     r3 = run->s390_sieic.ipa & 0x000f;
1288     handle_diag_308(&cpu->env, r1, r3);
1289 }
1290
1291 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1292 {
1293     CPUS390XState *env = &cpu->env;
1294     unsigned long pc;
1295
1296     cpu_synchronize_state(CPU(cpu));
1297
1298     pc = env->psw.addr - 4;
1299     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1300         env->psw.addr = pc;
1301         return EXCP_DEBUG;
1302     }
1303
1304     return -ENOENT;
1305 }
1306
1307 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1308
1309 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1310 {
1311     int r = 0;
1312     uint16_t func_code;
1313
1314     /*
1315      * For any diagnose call we support, bits 48-63 of the resulting
1316      * address specify the function code; the remainder is ignored.
1317      */
1318     func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1319     switch (func_code) {
1320     case DIAG_TIMEREVENT:
1321         kvm_handle_diag_288(cpu, run);
1322         break;
1323     case DIAG_IPL:
1324         kvm_handle_diag_308(cpu, run);
1325         break;
1326     case DIAG_KVM_HYPERCALL:
1327         r = handle_hypercall(cpu, run);
1328         break;
1329     case DIAG_KVM_BREAKPOINT:
1330         r = handle_sw_breakpoint(cpu, run);
1331         break;
1332     default:
1333         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1334         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1335         break;
1336     }
1337
1338     return r;
1339 }
1340
1341 typedef struct SigpInfo {
1342     S390CPU *cpu;
1343     uint64_t param;
1344     int cc;
1345     uint64_t *status_reg;
1346 } SigpInfo;
1347
1348 static void set_sigp_status(SigpInfo *si, uint64_t status)
1349 {
1350     *si->status_reg &= 0xffffffff00000000ULL;
1351     *si->status_reg |= status;
1352     si->cc = SIGP_CC_STATUS_STORED;
1353 }
1354
1355 static void sigp_start(void *arg)
1356 {
1357     SigpInfo *si = arg;
1358
1359     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1360         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1361         return;
1362     }
1363
1364     s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1365     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1366 }
1367
1368 static void sigp_stop(void *arg)
1369 {
1370     SigpInfo *si = arg;
1371     struct kvm_s390_irq irq = {
1372         .type = KVM_S390_SIGP_STOP,
1373     };
1374
1375     if (s390_cpu_get_state(si->cpu) != CPU_STATE_OPERATING) {
1376         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1377         return;
1378     }
1379
1380     /* disabled wait - sleeping in user space */
1381     if (CPU(si->cpu)->halted) {
1382         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1383     } else {
1384         /* execute the stop function */
1385         si->cpu->env.sigp_order = SIGP_STOP;
1386         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1387     }
1388     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1389 }
1390
1391 #define ADTL_SAVE_AREA_SIZE 1024
1392 static int kvm_s390_store_adtl_status(S390CPU *cpu, hwaddr addr)
1393 {
1394     void *mem;
1395     hwaddr len = ADTL_SAVE_AREA_SIZE;
1396
1397     mem = cpu_physical_memory_map(addr, &len, 1);
1398     if (!mem) {
1399         return -EFAULT;
1400     }
1401     if (len != ADTL_SAVE_AREA_SIZE) {
1402         cpu_physical_memory_unmap(mem, len, 1, 0);
1403         return -EFAULT;
1404     }
1405
1406     memcpy(mem, &cpu->env.vregs, 512);
1407
1408     cpu_physical_memory_unmap(mem, len, 1, len);
1409
1410     return 0;
1411 }
1412
1413 #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, floating_pt_save_area)
1414 #define SAVE_AREA_SIZE 512
1415 static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
1416 {
1417     static const uint8_t ar_id = 1;
1418     uint64_t ckc = cpu->env.ckc >> 8;
1419     void *mem;
1420     int i;
1421     hwaddr len = SAVE_AREA_SIZE;
1422
1423     mem = cpu_physical_memory_map(addr, &len, 1);
1424     if (!mem) {
1425         return -EFAULT;
1426     }
1427     if (len != SAVE_AREA_SIZE) {
1428         cpu_physical_memory_unmap(mem, len, 1, 0);
1429         return -EFAULT;
1430     }
1431
1432     if (store_arch) {
1433         cpu_physical_memory_write(offsetof(LowCore, ar_access_id), &ar_id, 1);
1434     }
1435     for (i = 0; i < 16; ++i) {
1436         *((uint64_t *)mem + i) = get_freg(&cpu->env, i)->ll;
1437     }
1438     memcpy(mem + 128, &cpu->env.regs, 128);
1439     memcpy(mem + 256, &cpu->env.psw, 16);
1440     memcpy(mem + 280, &cpu->env.psa, 4);
1441     memcpy(mem + 284, &cpu->env.fpc, 4);
1442     memcpy(mem + 292, &cpu->env.todpr, 4);
1443     memcpy(mem + 296, &cpu->env.cputm, 8);
1444     memcpy(mem + 304, &ckc, 8);
1445     memcpy(mem + 320, &cpu->env.aregs, 64);
1446     memcpy(mem + 384, &cpu->env.cregs, 128);
1447
1448     cpu_physical_memory_unmap(mem, len, 1, len);
1449
1450     return 0;
1451 }
1452
1453 static void sigp_stop_and_store_status(void *arg)
1454 {
1455     SigpInfo *si = arg;
1456     struct kvm_s390_irq irq = {
1457         .type = KVM_S390_SIGP_STOP,
1458     };
1459
1460     /* disabled wait - sleeping in user space */
1461     if (s390_cpu_get_state(si->cpu) == CPU_STATE_OPERATING &&
1462         CPU(si->cpu)->halted) {
1463         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1464     }
1465
1466     switch (s390_cpu_get_state(si->cpu)) {
1467     case CPU_STATE_OPERATING:
1468         si->cpu->env.sigp_order = SIGP_STOP_STORE_STATUS;
1469         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1470         /* store will be performed when handling the stop intercept */
1471         break;
1472     case CPU_STATE_STOPPED:
1473         /* already stopped, just store the status */
1474         cpu_synchronize_state(CPU(si->cpu));
1475         kvm_s390_store_status(si->cpu, KVM_S390_STORE_STATUS_DEF_ADDR, true);
1476         break;
1477     }
1478     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1479 }
1480
1481 static void sigp_store_status_at_address(void *arg)
1482 {
1483     SigpInfo *si = arg;
1484     uint32_t address = si->param & 0x7ffffe00u;
1485
1486     /* cpu has to be stopped */
1487     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1488         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1489         return;
1490     }
1491
1492     cpu_synchronize_state(CPU(si->cpu));
1493
1494     if (kvm_s390_store_status(si->cpu, address, false)) {
1495         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1496         return;
1497     }
1498     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1499 }
1500
1501 static void sigp_store_adtl_status(void *arg)
1502 {
1503     SigpInfo *si = arg;
1504
1505     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_VECTOR_REGISTERS)) {
1506         set_sigp_status(si, SIGP_STAT_INVALID_ORDER);
1507         return;
1508     }
1509
1510     /* cpu has to be stopped */
1511     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1512         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1513         return;
1514     }
1515
1516     /* parameter must be aligned to 1024-byte boundary */
1517     if (si->param & 0x3ff) {
1518         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1519         return;
1520     }
1521
1522     cpu_synchronize_state(CPU(si->cpu));
1523
1524     if (kvm_s390_store_adtl_status(si->cpu, si->param)) {
1525         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1526         return;
1527     }
1528     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1529 }
1530
1531 static void sigp_restart(void *arg)
1532 {
1533     SigpInfo *si = arg;
1534     struct kvm_s390_irq irq = {
1535         .type = KVM_S390_RESTART,
1536     };
1537
1538     switch (s390_cpu_get_state(si->cpu)) {
1539     case CPU_STATE_STOPPED:
1540         /* the restart irq has to be delivered prior to any other pending irq */
1541         cpu_synchronize_state(CPU(si->cpu));
1542         do_restart_interrupt(&si->cpu->env);
1543         s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1544         break;
1545     case CPU_STATE_OPERATING:
1546         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1547         break;
1548     }
1549     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1550 }
1551
1552 int kvm_s390_cpu_restart(S390CPU *cpu)
1553 {
1554     SigpInfo si = {
1555         .cpu = cpu,
1556     };
1557
1558     run_on_cpu(CPU(cpu), sigp_restart, &si);
1559     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1560     return 0;
1561 }
1562
1563 static void sigp_initial_cpu_reset(void *arg)
1564 {
1565     SigpInfo *si = arg;
1566     CPUState *cs = CPU(si->cpu);
1567     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1568
1569     cpu_synchronize_state(cs);
1570     scc->initial_cpu_reset(cs);
1571     cpu_synchronize_post_reset(cs);
1572     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1573 }
1574
1575 static void sigp_cpu_reset(void *arg)
1576 {
1577     SigpInfo *si = arg;
1578     CPUState *cs = CPU(si->cpu);
1579     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1580
1581     cpu_synchronize_state(cs);
1582     scc->cpu_reset(cs);
1583     cpu_synchronize_post_reset(cs);
1584     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1585 }
1586
1587 static void sigp_set_prefix(void *arg)
1588 {
1589     SigpInfo *si = arg;
1590     uint32_t addr = si->param & 0x7fffe000u;
1591
1592     cpu_synchronize_state(CPU(si->cpu));
1593
1594     if (!address_space_access_valid(&address_space_memory, addr,
1595                                     sizeof(struct LowCore), false)) {
1596         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1597         return;
1598     }
1599
1600     /* cpu has to be stopped */
1601     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1602         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1603         return;
1604     }
1605
1606     si->cpu->env.psa = addr;
1607     cpu_synchronize_post_init(CPU(si->cpu));
1608     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1609 }
1610
1611 static int handle_sigp_single_dst(S390CPU *dst_cpu, uint8_t order,
1612                                   uint64_t param, uint64_t *status_reg)
1613 {
1614     SigpInfo si = {
1615         .cpu = dst_cpu,
1616         .param = param,
1617         .status_reg = status_reg,
1618     };
1619
1620     /* cpu available? */
1621     if (dst_cpu == NULL) {
1622         return SIGP_CC_NOT_OPERATIONAL;
1623     }
1624
1625     /* only resets can break pending orders */
1626     if (dst_cpu->env.sigp_order != 0 &&
1627         order != SIGP_CPU_RESET &&
1628         order != SIGP_INITIAL_CPU_RESET) {
1629         return SIGP_CC_BUSY;
1630     }
1631
1632     switch (order) {
1633     case SIGP_START:
1634         run_on_cpu(CPU(dst_cpu), sigp_start, &si);
1635         break;
1636     case SIGP_STOP:
1637         run_on_cpu(CPU(dst_cpu), sigp_stop, &si);
1638         break;
1639     case SIGP_RESTART:
1640         run_on_cpu(CPU(dst_cpu), sigp_restart, &si);
1641         break;
1642     case SIGP_STOP_STORE_STATUS:
1643         run_on_cpu(CPU(dst_cpu), sigp_stop_and_store_status, &si);
1644         break;
1645     case SIGP_STORE_STATUS_ADDR:
1646         run_on_cpu(CPU(dst_cpu), sigp_store_status_at_address, &si);
1647         break;
1648     case SIGP_STORE_ADTL_STATUS:
1649         run_on_cpu(CPU(dst_cpu), sigp_store_adtl_status, &si);
1650         break;
1651     case SIGP_SET_PREFIX:
1652         run_on_cpu(CPU(dst_cpu), sigp_set_prefix, &si);
1653         break;
1654     case SIGP_INITIAL_CPU_RESET:
1655         run_on_cpu(CPU(dst_cpu), sigp_initial_cpu_reset, &si);
1656         break;
1657     case SIGP_CPU_RESET:
1658         run_on_cpu(CPU(dst_cpu), sigp_cpu_reset, &si);
1659         break;
1660     default:
1661         DPRINTF("KVM: unknown SIGP: 0x%x\n", order);
1662         set_sigp_status(&si, SIGP_STAT_INVALID_ORDER);
1663     }
1664
1665     return si.cc;
1666 }
1667
1668 static int sigp_set_architecture(S390CPU *cpu, uint32_t param,
1669                                  uint64_t *status_reg)
1670 {
1671     CPUState *cur_cs;
1672     S390CPU *cur_cpu;
1673
1674     /* due to the BQL, we are the only active cpu */
1675     CPU_FOREACH(cur_cs) {
1676         cur_cpu = S390_CPU(cur_cs);
1677         if (cur_cpu->env.sigp_order != 0) {
1678             return SIGP_CC_BUSY;
1679         }
1680         cpu_synchronize_state(cur_cs);
1681         /* all but the current one have to be stopped */
1682         if (cur_cpu != cpu &&
1683             s390_cpu_get_state(cur_cpu) != CPU_STATE_STOPPED) {
1684             *status_reg &= 0xffffffff00000000ULL;
1685             *status_reg |= SIGP_STAT_INCORRECT_STATE;
1686             return SIGP_CC_STATUS_STORED;
1687         }
1688     }
1689
1690     switch (param & 0xff) {
1691     case SIGP_MODE_ESA_S390:
1692         /* not supported */
1693         return SIGP_CC_NOT_OPERATIONAL;
1694     case SIGP_MODE_Z_ARCH_TRANS_ALL_PSW:
1695     case SIGP_MODE_Z_ARCH_TRANS_CUR_PSW:
1696         CPU_FOREACH(cur_cs) {
1697             cur_cpu = S390_CPU(cur_cs);
1698             cur_cpu->env.pfault_token = -1UL;
1699         }
1700         break;
1701     default:
1702         *status_reg &= 0xffffffff00000000ULL;
1703         *status_reg |= SIGP_STAT_INVALID_PARAMETER;
1704         return SIGP_CC_STATUS_STORED;
1705     }
1706
1707     return SIGP_CC_ORDER_CODE_ACCEPTED;
1708 }
1709
1710 #define SIGP_ORDER_MASK 0x000000ff
1711
1712 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1713 {
1714     CPUS390XState *env = &cpu->env;
1715     const uint8_t r1 = ipa1 >> 4;
1716     const uint8_t r3 = ipa1 & 0x0f;
1717     int ret;
1718     uint8_t order;
1719     uint64_t *status_reg;
1720     uint64_t param;
1721     S390CPU *dst_cpu = NULL;
1722
1723     cpu_synchronize_state(CPU(cpu));
1724
1725     /* get order code */
1726     order = decode_basedisp_rs(env, run->s390_sieic.ipb, NULL)
1727         & SIGP_ORDER_MASK;
1728     status_reg = &env->regs[r1];
1729     param = (r1 % 2) ? env->regs[r1] : env->regs[r1 + 1];
1730
1731     switch (order) {
1732     case SIGP_SET_ARCH:
1733         ret = sigp_set_architecture(cpu, param, status_reg);
1734         break;
1735     default:
1736         /* all other sigp orders target a single vcpu */
1737         dst_cpu = s390_cpu_addr2state(env->regs[r3]);
1738         ret = handle_sigp_single_dst(dst_cpu, order, param, status_reg);
1739     }
1740
1741     trace_kvm_sigp_finished(order, CPU(cpu)->cpu_index,
1742                             dst_cpu ? CPU(dst_cpu)->cpu_index : -1, ret);
1743
1744     if (ret >= 0) {
1745         setcc(cpu, ret);
1746         return 0;
1747     }
1748
1749     return ret;
1750 }
1751
1752 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1753 {
1754     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1755     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1756     int r = -1;
1757
1758     DPRINTF("handle_instruction 0x%x 0x%x\n",
1759             run->s390_sieic.ipa, run->s390_sieic.ipb);
1760     switch (ipa0) {
1761     case IPA0_B2:
1762         r = handle_b2(cpu, run, ipa1);
1763         break;
1764     case IPA0_B9:
1765         r = handle_b9(cpu, run, ipa1);
1766         break;
1767     case IPA0_EB:
1768         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1769         break;
1770     case IPA0_E3:
1771         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1772         break;
1773     case IPA0_DIAG:
1774         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1775         break;
1776     case IPA0_SIGP:
1777         r = handle_sigp(cpu, run, ipa1);
1778         break;
1779     }
1780
1781     if (r < 0) {
1782         r = 0;
1783         enter_pgmcheck(cpu, 0x0001);
1784     }
1785
1786     return r;
1787 }
1788
1789 static bool is_special_wait_psw(CPUState *cs)
1790 {
1791     /* signal quiesce */
1792     return cs->kvm_run->psw_addr == 0xfffUL;
1793 }
1794
1795 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1796 {
1797     CPUState *cs = CPU(cpu);
1798
1799     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1800                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1801                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1802     s390_cpu_halt(cpu);
1803     qemu_system_guest_panicked();
1804 }
1805
1806 static int handle_intercept(S390CPU *cpu)
1807 {
1808     CPUState *cs = CPU(cpu);
1809     struct kvm_run *run = cs->kvm_run;
1810     int icpt_code = run->s390_sieic.icptcode;
1811     int r = 0;
1812
1813     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1814             (long)cs->kvm_run->psw_addr);
1815     switch (icpt_code) {
1816         case ICPT_INSTRUCTION:
1817             r = handle_instruction(cpu, run);
1818             break;
1819         case ICPT_PROGRAM:
1820             unmanageable_intercept(cpu, "program interrupt",
1821                                    offsetof(LowCore, program_new_psw));
1822             r = EXCP_HALTED;
1823             break;
1824         case ICPT_EXT_INT:
1825             unmanageable_intercept(cpu, "external interrupt",
1826                                    offsetof(LowCore, external_new_psw));
1827             r = EXCP_HALTED;
1828             break;
1829         case ICPT_WAITPSW:
1830             /* disabled wait, since enabled wait is handled in kernel */
1831             cpu_synchronize_state(cs);
1832             if (s390_cpu_halt(cpu) == 0) {
1833                 if (is_special_wait_psw(cs)) {
1834                     qemu_system_shutdown_request();
1835                 } else {
1836                     qemu_system_guest_panicked();
1837                 }
1838             }
1839             r = EXCP_HALTED;
1840             break;
1841         case ICPT_CPU_STOP:
1842             if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
1843                 qemu_system_shutdown_request();
1844             }
1845             if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
1846                 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR,
1847                                       true);
1848             }
1849             cpu->env.sigp_order = 0;
1850             r = EXCP_HALTED;
1851             break;
1852         case ICPT_SOFT_INTERCEPT:
1853             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1854             exit(1);
1855             break;
1856         case ICPT_IO:
1857             fprintf(stderr, "KVM unimplemented icpt IO\n");
1858             exit(1);
1859             break;
1860         default:
1861             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1862             exit(1);
1863             break;
1864     }
1865
1866     return r;
1867 }
1868
1869 static int handle_tsch(S390CPU *cpu)
1870 {
1871     CPUState *cs = CPU(cpu);
1872     struct kvm_run *run = cs->kvm_run;
1873     int ret;
1874
1875     cpu_synchronize_state(cs);
1876
1877     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb);
1878     if (ret < 0) {
1879         /*
1880          * Failure.
1881          * If an I/O interrupt had been dequeued, we have to reinject it.
1882          */
1883         if (run->s390_tsch.dequeued) {
1884             kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1885                                   run->s390_tsch.subchannel_nr,
1886                                   run->s390_tsch.io_int_parm,
1887                                   run->s390_tsch.io_int_word);
1888         }
1889         ret = 0;
1890     }
1891     return ret;
1892 }
1893
1894 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1895 {
1896     struct sysib_322 sysib;
1897     int del;
1898
1899     if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1900         return;
1901     }
1902     /* Shift the stack of Extended Names to prepare for our own data */
1903     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1904             sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1905     /* First virt level, that doesn't provide Ext Names delimits stack. It is
1906      * assumed it's not capable of managing Extended Names for lower levels.
1907      */
1908     for (del = 1; del < sysib.count; del++) {
1909         if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1910             break;
1911         }
1912     }
1913     if (del < sysib.count) {
1914         memset(sysib.ext_names[del], 0,
1915                sizeof(sysib.ext_names[0]) * (sysib.count - del));
1916     }
1917     /* Insert short machine name in EBCDIC, padded with blanks */
1918     if (qemu_name) {
1919         memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1920         ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1921                                                     strlen(qemu_name)));
1922     }
1923     sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1924     memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1925     /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1926      * considered by s390 as not capable of providing any Extended Name.
1927      * Therefore if no name was specified on qemu invocation, we go with the
1928      * same "KVMguest" default, which KVM has filled into short name field.
1929      */
1930     if (qemu_name) {
1931         strncpy((char *)sysib.ext_names[0], qemu_name,
1932                 sizeof(sysib.ext_names[0]));
1933     } else {
1934         strcpy((char *)sysib.ext_names[0], "KVMguest");
1935     }
1936     /* Insert UUID */
1937     memcpy(sysib.vm[0].uuid, qemu_uuid, sizeof(sysib.vm[0].uuid));
1938
1939     s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1940 }
1941
1942 static int handle_stsi(S390CPU *cpu)
1943 {
1944     CPUState *cs = CPU(cpu);
1945     struct kvm_run *run = cs->kvm_run;
1946
1947     switch (run->s390_stsi.fc) {
1948     case 3:
1949         if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
1950             return 0;
1951         }
1952         /* Only sysib 3.2.2 needs post-handling for now. */
1953         insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
1954         return 0;
1955     default:
1956         return 0;
1957     }
1958 }
1959
1960 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1961 {
1962     CPUState *cs = CPU(cpu);
1963     struct kvm_run *run = cs->kvm_run;
1964
1965     int ret = 0;
1966     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1967
1968     switch (arch_info->type) {
1969     case KVM_HW_WP_WRITE:
1970         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1971             cs->watchpoint_hit = &hw_watchpoint;
1972             hw_watchpoint.vaddr = arch_info->addr;
1973             hw_watchpoint.flags = BP_MEM_WRITE;
1974             ret = EXCP_DEBUG;
1975         }
1976         break;
1977     case KVM_HW_BP:
1978         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1979             ret = EXCP_DEBUG;
1980         }
1981         break;
1982     case KVM_SINGLESTEP:
1983         if (cs->singlestep_enabled) {
1984             ret = EXCP_DEBUG;
1985         }
1986         break;
1987     default:
1988         ret = -ENOSYS;
1989     }
1990
1991     return ret;
1992 }
1993
1994 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1995 {
1996     S390CPU *cpu = S390_CPU(cs);
1997     int ret = 0;
1998
1999     qemu_mutex_lock_iothread();
2000
2001     switch (run->exit_reason) {
2002         case KVM_EXIT_S390_SIEIC:
2003             ret = handle_intercept(cpu);
2004             break;
2005         case KVM_EXIT_S390_RESET:
2006             s390_reipl_request();
2007             break;
2008         case KVM_EXIT_S390_TSCH:
2009             ret = handle_tsch(cpu);
2010             break;
2011         case KVM_EXIT_S390_STSI:
2012             ret = handle_stsi(cpu);
2013             break;
2014         case KVM_EXIT_DEBUG:
2015             ret = kvm_arch_handle_debug_exit(cpu);
2016             break;
2017         default:
2018             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
2019             break;
2020     }
2021     qemu_mutex_unlock_iothread();
2022
2023     if (ret == 0) {
2024         ret = EXCP_INTERRUPT;
2025     }
2026     return ret;
2027 }
2028
2029 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2030 {
2031     return true;
2032 }
2033
2034 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2035 {
2036     return 1;
2037 }
2038
2039 int kvm_arch_on_sigbus(int code, void *addr)
2040 {
2041     return 1;
2042 }
2043
2044 void kvm_s390_io_interrupt(uint16_t subchannel_id,
2045                            uint16_t subchannel_nr, uint32_t io_int_parm,
2046                            uint32_t io_int_word)
2047 {
2048     struct kvm_s390_irq irq = {
2049         .u.io.subchannel_id = subchannel_id,
2050         .u.io.subchannel_nr = subchannel_nr,
2051         .u.io.io_int_parm = io_int_parm,
2052         .u.io.io_int_word = io_int_word,
2053     };
2054
2055     if (io_int_word & IO_INT_WORD_AI) {
2056         irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
2057     } else {
2058         irq.type = ((subchannel_id & 0xff00) << 24) |
2059             ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
2060     }
2061     kvm_s390_floating_interrupt(&irq);
2062 }
2063
2064 static uint64_t build_channel_report_mcic(void)
2065 {
2066     uint64_t mcic;
2067
2068     /* subclass: indicate channel report pending */
2069     mcic = MCIC_SC_CP |
2070     /* subclass modifiers: none */
2071     /* storage errors: none */
2072     /* validity bits: no damage */
2073         MCIC_VB_WP | MCIC_VB_MS | MCIC_VB_PM | MCIC_VB_IA | MCIC_VB_FP |
2074         MCIC_VB_GR | MCIC_VB_CR | MCIC_VB_ST | MCIC_VB_AR | MCIC_VB_PR |
2075         MCIC_VB_FC | MCIC_VB_CT | MCIC_VB_CC;
2076     if (kvm_check_extension(kvm_state, KVM_CAP_S390_VECTOR_REGISTERS)) {
2077         mcic |= MCIC_VB_VR;
2078     }
2079     return mcic;
2080 }
2081
2082 void kvm_s390_crw_mchk(void)
2083 {
2084     struct kvm_s390_irq irq = {
2085         .type = KVM_S390_MCHK,
2086         .u.mchk.cr14 = 1 << 28,
2087         .u.mchk.mcic = build_channel_report_mcic(),
2088     };
2089     kvm_s390_floating_interrupt(&irq);
2090 }
2091
2092 void kvm_s390_enable_css_support(S390CPU *cpu)
2093 {
2094     int r;
2095
2096     /* Activate host kernel channel subsystem support. */
2097     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
2098     assert(r == 0);
2099 }
2100
2101 void kvm_arch_init_irq_routing(KVMState *s)
2102 {
2103     /*
2104      * Note that while irqchip capabilities generally imply that cpustates
2105      * are handled in-kernel, it is not true for s390 (yet); therefore, we
2106      * have to override the common code kvm_halt_in_kernel_allowed setting.
2107      */
2108     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2109         kvm_gsi_routing_allowed = true;
2110         kvm_halt_in_kernel_allowed = false;
2111     }
2112 }
2113
2114 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
2115                                     int vq, bool assign)
2116 {
2117     struct kvm_ioeventfd kick = {
2118         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
2119         KVM_IOEVENTFD_FLAG_DATAMATCH,
2120         .fd = event_notifier_get_fd(notifier),
2121         .datamatch = vq,
2122         .addr = sch,
2123         .len = 8,
2124     };
2125     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
2126         return -ENOSYS;
2127     }
2128     if (!assign) {
2129         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
2130     }
2131     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
2132 }
2133
2134 int kvm_s390_get_memslot_count(KVMState *s)
2135 {
2136     return kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
2137 }
2138
2139 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
2140 {
2141     struct kvm_mp_state mp_state = {};
2142     int ret;
2143
2144     /* the kvm part might not have been initialized yet */
2145     if (CPU(cpu)->kvm_state == NULL) {
2146         return 0;
2147     }
2148
2149     switch (cpu_state) {
2150     case CPU_STATE_STOPPED:
2151         mp_state.mp_state = KVM_MP_STATE_STOPPED;
2152         break;
2153     case CPU_STATE_CHECK_STOP:
2154         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
2155         break;
2156     case CPU_STATE_OPERATING:
2157         mp_state.mp_state = KVM_MP_STATE_OPERATING;
2158         break;
2159     case CPU_STATE_LOAD:
2160         mp_state.mp_state = KVM_MP_STATE_LOAD;
2161         break;
2162     default:
2163         error_report("Requested CPU state is not a valid S390 CPU state: %u",
2164                      cpu_state);
2165         exit(1);
2166     }
2167
2168     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2169     if (ret) {
2170         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
2171                                        strerror(-ret));
2172     }
2173
2174     return ret;
2175 }
2176
2177 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
2178 {
2179     struct kvm_s390_irq_state irq_state;
2180     CPUState *cs = CPU(cpu);
2181     int32_t bytes;
2182
2183     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2184         return;
2185     }
2186
2187     irq_state.buf = (uint64_t) cpu->irqstate;
2188     irq_state.len = VCPU_IRQ_BUF_SIZE;
2189
2190     bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
2191     if (bytes < 0) {
2192         cpu->irqstate_saved_size = 0;
2193         error_report("Migration of interrupt state failed");
2194         return;
2195     }
2196
2197     cpu->irqstate_saved_size = bytes;
2198 }
2199
2200 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
2201 {
2202     CPUState *cs = CPU(cpu);
2203     struct kvm_s390_irq_state irq_state;
2204     int r;
2205
2206     if (cpu->irqstate_saved_size == 0) {
2207         return 0;
2208     }
2209
2210     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2211         return -ENOSYS;
2212     }
2213
2214     irq_state.buf = (uint64_t) cpu->irqstate;
2215     irq_state.len = cpu->irqstate_saved_size;
2216
2217     r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
2218     if (r) {
2219         error_report("Setting interrupt state failed %d", r);
2220     }
2221     return r;
2222 }
2223
2224 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2225                              uint64_t address, uint32_t data, PCIDevice *dev)
2226 {
2227     S390PCIBusDevice *pbdev;
2228     uint32_t fid = data >> ZPCI_MSI_VEC_BITS;
2229     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
2230
2231     pbdev = s390_pci_find_dev_by_fid(fid);
2232     if (!pbdev) {
2233         DPRINTF("add_msi_route no dev\n");
2234         return -ENODEV;
2235     }
2236
2237     pbdev->routes.adapter.ind_offset = vec;
2238
2239     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2240     route->flags = 0;
2241     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2242     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2243     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2244     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset;
2245     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2246     return 0;
2247 }
2248
2249 int kvm_arch_msi_data_to_gsi(uint32_t data)
2250 {
2251     abort();
2252 }