target-s390x/kvm.c

   1 /*
   2  * QEMU S390x KVM implementation
   3  *
   4  * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
   5  * Copyright IBM Corp. 2012
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * Contributions after 2012-10-29 are licensed under the terms of the
  18  * GNU GPL, version 2 or (at your option) any later version.
  19  *
  20  * You should have received a copy of the GNU (Lesser) General Public
  21  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  22  */
  23
  24 #include <sys/types.h>
  25 #include <sys/ioctl.h>
  26 #include <sys/mman.h>
  27
  28 #include <linux/kvm.h>
  29 #include <asm/ptrace.h>
  30
  31 #include "qemu-common.h"
  32 #include "qemu/timer.h"
  33 #include "sysemu/sysemu.h"
  34 #include "sysemu/kvm.h"
  35 #include "hw/hw.h"
  36 #include "cpu.h"
  37 #include "sysemu/device_tree.h"
  38 #include "qapi/qmp/qjson.h"
  39 #include "monitor/monitor.h"
  40 #include "exec/gdbstub.h"
  41 #include "exec/address-spaces.h"
  42 #include "trace.h"
  43 #include "qapi-event.h"
  44 #include "hw/s390x/s390-pci-inst.h"
  45 #include "hw/s390x/s390-pci-bus.h"
  46 #include "hw/s390x/ipl.h"
  47 #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_IPL                        0x308
 102 #define DIAG_KVM_HYPERCALL              0x500
 103 #define DIAG_KVM_BREAKPOINT             0x501
 104
 105 #define ICPT_INSTRUCTION                0x04
 106 #define ICPT_PROGRAM                    0x08
 107 #define ICPT_EXT_INT                    0x14
 108 #define ICPT_WAITPSW                    0x1c
 109 #define ICPT_SOFT_INTERCEPT             0x24
 110 #define ICPT_CPU_STOP                   0x28
 111 #define ICPT_IO                         0x40
 112
 113 #define NR_LOCAL_IRQS 32
 114 /*
 115  * Needs to be big enough to contain max_cpus emergency signals
 116  * and in addition NR_LOCAL_IRQS interrupts
 117  */
 118 #define VCPU_IRQ_BUF_SIZE (sizeof(struct kvm_s390_irq) * \
 119                            (max_cpus + NR_LOCAL_IRQS))
 120
 121 static CPUWatchpoint hw_watchpoint;
 122 /*
 123  * We don't use a list because this structure is also used to transmit the
 124  * hardware breakpoints to the kernel.
 125  */
 126 static struct kvm_hw_breakpoint *hw_breakpoints;
 127 static int nb_hw_breakpoints;
 128
 129 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
 130     KVM_CAP_LAST_INFO
 131 };
 132
 133 static int cap_sync_regs;
 134 static int cap_async_pf;
 135 static int cap_mem_op;
 136 static int cap_s390_irq;
 137
 138 static void *legacy_s390_alloc(size_t size, uint64_t *align);
 139
 140 static int kvm_s390_query_mem_limit(KVMState *s, uint64_t *memory_limit)
 141 {
 142     struct kvm_device_attr attr = {
 143         .group = KVM_S390_VM_MEM_CTRL,
 144         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
 145         .addr = (uint64_t) memory_limit,
 146     };
 147
 148     return kvm_vm_ioctl(s, KVM_GET_DEVICE_ATTR, &attr);
 149 }
 150
 151 int kvm_s390_set_mem_limit(KVMState *s, uint64_t new_limit, uint64_t *hw_limit)
 152 {
 153     int rc;
 154
 155     struct kvm_device_attr attr = {
 156         .group = KVM_S390_VM_MEM_CTRL,
 157         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
 158         .addr = (uint64_t) &new_limit,
 159     };
 160
 161     if (!kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_LIMIT_SIZE)) {
 162         return 0;
 163     }
 164
 165     rc = kvm_s390_query_mem_limit(s, hw_limit);
 166     if (rc) {
 167         return rc;
 168     } else if (*hw_limit < new_limit) {
 169         return -E2BIG;
 170     }
 171
 172     return kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 173 }
 174
 175 void kvm_s390_clear_cmma_callback(void *opaque)
 176 {
 177     int rc;
 178     KVMState *s = opaque;
 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(s, 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     if (!rc) {
 203         qemu_register_reset(kvm_s390_clear_cmma_callback, s);
 204     }
 205     trace_kvm_enable_cmma(rc);
 206 }
 207
 208 static void kvm_s390_set_attr(uint64_t attr)
 209 {
 210     struct kvm_device_attr attribute = {
 211         .group = KVM_S390_VM_CRYPTO,
 212         .attr  = attr,
 213     };
 214
 215     int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
 216
 217     if (ret) {
 218         error_report("Failed to set crypto device attribute %lu: %s",
 219                      attr, strerror(-ret));
 220     }
 221 }
 222
 223 static void kvm_s390_init_aes_kw(void)
 224 {
 225     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
 226
 227     if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
 228                                  NULL)) {
 229             attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
 230     }
 231
 232     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 233             kvm_s390_set_attr(attr);
 234     }
 235 }
 236
 237 static void kvm_s390_init_dea_kw(void)
 238 {
 239     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
 240
 241     if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
 242                                  NULL)) {
 243             attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
 244     }
 245
 246     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 247             kvm_s390_set_attr(attr);
 248     }
 249 }
 250
 251 static void kvm_s390_init_crypto(void)
 252 {
 253     kvm_s390_init_aes_kw();
 254     kvm_s390_init_dea_kw();
 255 }
 256
 257 int kvm_arch_init(MachineState *ms, KVMState *s)
 258 {
 259     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
 260     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
 261     cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
 262     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
 263
 264     kvm_s390_enable_cmma(s);
 265
 266     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
 267         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
 268         phys_mem_set_alloc(legacy_s390_alloc);
 269     }
 270
 271     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 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     kvm_s390_init_crypto();
 304 }
 305
 306 static int can_sync_regs(CPUState *cs, int regs)
 307 {
 308     return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
 309 }
 310
 311 int kvm_arch_put_registers(CPUState *cs, int level)
 312 {
 313     S390CPU *cpu = S390_CPU(cs);
 314     CPUS390XState *env = &cpu->env;
 315     struct kvm_sregs sregs;
 316     struct kvm_regs regs;
 317     struct kvm_fpu fpu = {};
 318     int r;
 319     int i;
 320
 321     /* always save the PSW  and the GPRS*/
 322     cs->kvm_run->psw_addr = env->psw.addr;
 323     cs->kvm_run->psw_mask = env->psw.mask;
 324
 325     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 326         for (i = 0; i < 16; i++) {
 327             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
 328             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
 329         }
 330     } else {
 331         for (i = 0; i < 16; i++) {
 332             regs.gprs[i] = env->regs[i];
 333         }
 334         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
 335         if (r < 0) {
 336             return r;
 337         }
 338     }
 339
 340     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 341         for (i = 0; i < 32; i++) {
 342             cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
 343             cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
 344         }
 345         cs->kvm_run->s.regs.fpc = env->fpc;
 346         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
 347     } else {
 348         /* Floating point */
 349         for (i = 0; i < 16; i++) {
 350             fpu.fprs[i] = get_freg(env, i)->ll;
 351         }
 352         fpu.fpc = env->fpc;
 353
 354         r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
 355         if (r < 0) {
 356             return r;
 357         }
 358     }
 359
 360     /* Do we need to save more than that? */
 361     if (level == KVM_PUT_RUNTIME_STATE) {
 362         return 0;
 363     }
 364
 365     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 366         cs->kvm_run->s.regs.cputm = env->cputm;
 367         cs->kvm_run->s.regs.ckc = env->ckc;
 368         cs->kvm_run->s.regs.todpr = env->todpr;
 369         cs->kvm_run->s.regs.gbea = env->gbea;
 370         cs->kvm_run->s.regs.pp = env->pp;
 371         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
 372     } else {
 373         /*
 374          * These ONE_REGS are not protected by a capability. As they are only
 375          * necessary for migration we just trace a possible error, but don't
 376          * return with an error return code.
 377          */
 378         kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 379         kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 380         kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 381         kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 382         kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 383     }
 384
 385     /* pfault parameters */
 386     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 387         cs->kvm_run->s.regs.pft = env->pfault_token;
 388         cs->kvm_run->s.regs.pfs = env->pfault_select;
 389         cs->kvm_run->s.regs.pfc = env->pfault_compare;
 390         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
 391     } else if (cap_async_pf) {
 392         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 393         if (r < 0) {
 394             return r;
 395         }
 396         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 397         if (r < 0) {
 398             return r;
 399         }
 400         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 401         if (r < 0) {
 402             return r;
 403         }
 404     }
 405
 406     /* access registers and control registers*/
 407     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 408         for (i = 0; i < 16; i++) {
 409             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
 410             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
 411         }
 412         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
 413         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
 414     } else {
 415         for (i = 0; i < 16; i++) {
 416             sregs.acrs[i] = env->aregs[i];
 417             sregs.crs[i] = env->cregs[i];
 418         }
 419         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
 420         if (r < 0) {
 421             return r;
 422         }
 423     }
 424
 425     /* Finally the prefix */
 426     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 427         cs->kvm_run->s.regs.prefix = env->psa;
 428         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
 429     } else {
 430         /* prefix is only supported via sync regs */
 431     }
 432     return 0;
 433 }
 434
 435 int kvm_arch_get_registers(CPUState *cs)
 436 {
 437     S390CPU *cpu = S390_CPU(cs);
 438     CPUS390XState *env = &cpu->env;
 439     struct kvm_sregs sregs;
 440     struct kvm_regs regs;
 441     struct kvm_fpu fpu;
 442     int i, r;
 443
 444     /* get the PSW */
 445     env->psw.addr = cs->kvm_run->psw_addr;
 446     env->psw.mask = cs->kvm_run->psw_mask;
 447
 448     /* the GPRS */
 449     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 450         for (i = 0; i < 16; i++) {
 451             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
 452         }
 453     } else {
 454         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
 455         if (r < 0) {
 456             return r;
 457         }
 458          for (i = 0; i < 16; i++) {
 459             env->regs[i] = regs.gprs[i];
 460         }
 461     }
 462
 463     /* The ACRS and CRS */
 464     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 465         for (i = 0; i < 16; i++) {
 466             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
 467             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
 468         }
 469     } else {
 470         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
 471         if (r < 0) {
 472             return r;
 473         }
 474          for (i = 0; i < 16; i++) {
 475             env->aregs[i] = sregs.acrs[i];
 476             env->cregs[i] = sregs.crs[i];
 477         }
 478     }
 479
 480     /* Floating point and vector registers */
 481     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 482         for (i = 0; i < 32; i++) {
 483             env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
 484             env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
 485         }
 486         env->fpc = cs->kvm_run->s.regs.fpc;
 487     } else {
 488         r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
 489         if (r < 0) {
 490             return r;
 491         }
 492         for (i = 0; i < 16; i++) {
 493             get_freg(env, i)->ll = fpu.fprs[i];
 494         }
 495         env->fpc = fpu.fpc;
 496     }
 497
 498     /* The prefix */
 499     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 500         env->psa = cs->kvm_run->s.regs.prefix;
 501     }
 502
 503     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 504         env->cputm = cs->kvm_run->s.regs.cputm;
 505         env->ckc = cs->kvm_run->s.regs.ckc;
 506         env->todpr = cs->kvm_run->s.regs.todpr;
 507         env->gbea = cs->kvm_run->s.regs.gbea;
 508         env->pp = cs->kvm_run->s.regs.pp;
 509     } else {
 510         /*
 511          * These ONE_REGS are not protected by a capability. As they are only
 512          * necessary for migration we just trace a possible error, but don't
 513          * return with an error return code.
 514          */
 515         kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 516         kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 517         kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 518         kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 519         kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 520     }
 521
 522     /* pfault parameters */
 523     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 524         env->pfault_token = cs->kvm_run->s.regs.pft;
 525         env->pfault_select = cs->kvm_run->s.regs.pfs;
 526         env->pfault_compare = cs->kvm_run->s.regs.pfc;
 527     } else if (cap_async_pf) {
 528         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 529         if (r < 0) {
 530             return r;
 531         }
 532         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 533         if (r < 0) {
 534             return r;
 535         }
 536         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 537         if (r < 0) {
 538             return r;
 539         }
 540     }
 541
 542     return 0;
 543 }
 544
 545 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
 546 {
 547     int r;
 548     struct kvm_device_attr attr = {
 549         .group = KVM_S390_VM_TOD,
 550         .attr = KVM_S390_VM_TOD_LOW,
 551         .addr = (uint64_t)tod_low,
 552     };
 553
 554     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 555     if (r) {
 556         return r;
 557     }
 558
 559     attr.attr = KVM_S390_VM_TOD_HIGH;
 560     attr.addr = (uint64_t)tod_high;
 561     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 562 }
 563
 564 int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_low)
 565 {
 566     int r;
 567
 568     struct kvm_device_attr attr = {
 569         .group = KVM_S390_VM_TOD,
 570         .attr = KVM_S390_VM_TOD_LOW,
 571         .addr = (uint64_t)tod_low,
 572     };
 573
 574     r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 575     if (r) {
 576         return r;
 577     }
 578
 579     attr.attr = KVM_S390_VM_TOD_HIGH;
 580     attr.addr = (uint64_t)tod_high;
 581     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 582 }
 583
 584 /**
 585  * kvm_s390_mem_op:
 586  * @addr:      the logical start address in guest memory
 587  * @ar:        the access register number
 588  * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
 589  * @len:       length that should be transfered
 590  * @is_write:  true = write, false = read
 591  * Returns:    0 on success, non-zero if an exception or error occured
 592  *
 593  * Use KVM ioctl to read/write from/to guest memory. An access exception
 594  * is injected into the vCPU in case of translation errors.
 595  */
 596 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
 597                     int len, bool is_write)
 598 {
 599     struct kvm_s390_mem_op mem_op = {
 600         .gaddr = addr,
 601         .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
 602         .size = len,
 603         .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
 604                        : KVM_S390_MEMOP_LOGICAL_READ,
 605         .buf = (uint64_t)hostbuf,
 606         .ar = ar,
 607     };
 608     int ret;
 609
 610     if (!cap_mem_op) {
 611         return -ENOSYS;
 612     }
 613     if (!hostbuf) {
 614         mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
 615     }
 616
 617     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
 618     if (ret < 0) {
 619         error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
 620     }
 621     return ret;
 622 }
 623
 624 /*
 625  * Legacy layout for s390:
 626  * Older S390 KVM requires the topmost vma of the RAM to be
 627  * smaller than an system defined value, which is at least 256GB.
 628  * Larger systems have larger values. We put the guest between
 629  * the end of data segment (system break) and this value. We
 630  * use 32GB as a base to have enough room for the system break
 631  * to grow. We also have to use MAP parameters that avoid
 632  * read-only mapping of guest pages.
 633  */
 634 static void *legacy_s390_alloc(size_t size, uint64_t *align)
 635 {
 636     void *mem;
 637
 638     mem = mmap((void *) 0x800000000ULL, size,
 639                PROT_EXEC|PROT_READ|PROT_WRITE,
 640                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 641     return mem == MAP_FAILED ? NULL : mem;
 642 }
 643
 644 /* DIAG 501 is used for sw breakpoints */
 645 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 646
 647 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 648 {
 649
 650     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 651                             sizeof(diag_501), 0) ||
 652         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501,
 653                             sizeof(diag_501), 1)) {
 654         return -EINVAL;
 655     }
 656     return 0;
 657 }
 658
 659 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 660 {
 661     uint8_t t[sizeof(diag_501)];
 662
 663     if (cpu_memory_rw_debug(cs, bp->pc, t, sizeof(diag_501), 0)) {
 664         return -EINVAL;
 665     } else if (memcmp(t, diag_501, sizeof(diag_501))) {
 666         return -EINVAL;
 667     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 668                                    sizeof(diag_501), 1)) {
 669         return -EINVAL;
 670     }
 671
 672     return 0;
 673 }
 674
 675 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
 676                                                     int len, int type)
 677 {
 678     int n;
 679
 680     for (n = 0; n < nb_hw_breakpoints; n++) {
 681         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
 682             (hw_breakpoints[n].len == len || len == -1)) {
 683             return &hw_breakpoints[n];
 684         }
 685     }
 686
 687     return NULL;
 688 }
 689
 690 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
 691 {
 692     int size;
 693
 694     if (find_hw_breakpoint(addr, len, type)) {
 695         return -EEXIST;
 696     }
 697
 698     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
 699
 700     if (!hw_breakpoints) {
 701         nb_hw_breakpoints = 0;
 702         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
 703     } else {
 704         hw_breakpoints =
 705             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
 706     }
 707
 708     if (!hw_breakpoints) {
 709         nb_hw_breakpoints = 0;
 710         return -ENOMEM;
 711     }
 712
 713     hw_breakpoints[nb_hw_breakpoints].addr = addr;
 714     hw_breakpoints[nb_hw_breakpoints].len = len;
 715     hw_breakpoints[nb_hw_breakpoints].type = type;
 716
 717     nb_hw_breakpoints++;
 718
 719     return 0;
 720 }
 721
 722 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 723                                   target_ulong len, int type)
 724 {
 725     switch (type) {
 726     case GDB_BREAKPOINT_HW:
 727         type = KVM_HW_BP;
 728         break;
 729     case GDB_WATCHPOINT_WRITE:
 730         if (len < 1) {
 731             return -EINVAL;
 732         }
 733         type = KVM_HW_WP_WRITE;
 734         break;
 735     default:
 736         return -ENOSYS;
 737     }
 738     return insert_hw_breakpoint(addr, len, type);
 739 }
 740
 741 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 742                                   target_ulong len, int type)
 743 {
 744     int size;
 745     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
 746
 747     if (bp == NULL) {
 748         return -ENOENT;
 749     }
 750
 751     nb_hw_breakpoints--;
 752     if (nb_hw_breakpoints > 0) {
 753         /*
 754          * In order to trim the array, move the last element to the position to
 755          * be removed - if necessary.
 756          */
 757         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
 758             *bp = hw_breakpoints[nb_hw_breakpoints];
 759         }
 760         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
 761         hw_breakpoints =
 762              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
 763     } else {
 764         g_free(hw_breakpoints);
 765         hw_breakpoints = NULL;
 766     }
 767
 768     return 0;
 769 }
 770
 771 void kvm_arch_remove_all_hw_breakpoints(void)
 772 {
 773     nb_hw_breakpoints = 0;
 774     g_free(hw_breakpoints);
 775     hw_breakpoints = NULL;
 776 }
 777
 778 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 779 {
 780     int i;
 781
 782     if (nb_hw_breakpoints > 0) {
 783         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
 784         dbg->arch.hw_bp = hw_breakpoints;
 785
 786         for (i = 0; i < nb_hw_breakpoints; ++i) {
 787             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
 788                                                        hw_breakpoints[i].addr);
 789         }
 790         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 791     } else {
 792         dbg->arch.nr_hw_bp = 0;
 793         dbg->arch.hw_bp = NULL;
 794     }
 795 }
 796
 797 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 798 {
 799 }
 800
 801 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
 802 {
 803     return MEMTXATTRS_UNSPECIFIED;
 804 }
 805
 806 int kvm_arch_process_async_events(CPUState *cs)
 807 {
 808     return cs->halted;
 809 }
 810
 811 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
 812                                      struct kvm_s390_interrupt *interrupt)
 813 {
 814     int r = 0;
 815
 816     interrupt->type = irq->type;
 817     switch (irq->type) {
 818     case KVM_S390_INT_VIRTIO:
 819         interrupt->parm = irq->u.ext.ext_params;
 820         /* fall through */
 821     case KVM_S390_INT_PFAULT_INIT:
 822     case KVM_S390_INT_PFAULT_DONE:
 823         interrupt->parm64 = irq->u.ext.ext_params2;
 824         break;
 825     case KVM_S390_PROGRAM_INT:
 826         interrupt->parm = irq->u.pgm.code;
 827         break;
 828     case KVM_S390_SIGP_SET_PREFIX:
 829         interrupt->parm = irq->u.prefix.address;
 830         break;
 831     case KVM_S390_INT_SERVICE:
 832         interrupt->parm = irq->u.ext.ext_params;
 833         break;
 834     case KVM_S390_MCHK:
 835         interrupt->parm = irq->u.mchk.cr14;
 836         interrupt->parm64 = irq->u.mchk.mcic;
 837         break;
 838     case KVM_S390_INT_EXTERNAL_CALL:
 839         interrupt->parm = irq->u.extcall.code;
 840         break;
 841     case KVM_S390_INT_EMERGENCY:
 842         interrupt->parm = irq->u.emerg.code;
 843         break;
 844     case KVM_S390_SIGP_STOP:
 845     case KVM_S390_RESTART:
 846         break; /* These types have no parameters */
 847     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
 848         interrupt->parm = irq->u.io.subchannel_id << 16;
 849         interrupt->parm |= irq->u.io.subchannel_nr;
 850         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
 851         interrupt->parm64 |= irq->u.io.io_int_word;
 852         break;
 853     default:
 854         r = -EINVAL;
 855         break;
 856     }
 857     return r;
 858 }
 859
 860 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
 861 {
 862     struct kvm_s390_interrupt kvmint = {};
 863     int r;
 864
 865     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 866     if (r < 0) {
 867         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 868         exit(1);
 869     }
 870
 871     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
 872     if (r < 0) {
 873         fprintf(stderr, "KVM failed to inject interrupt\n");
 874         exit(1);
 875     }
 876 }
 877
 878 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
 879 {
 880     CPUState *cs = CPU(cpu);
 881     int r;
 882
 883     if (cap_s390_irq) {
 884         r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
 885         if (!r) {
 886             return;
 887         }
 888         error_report("KVM failed to inject interrupt %llx", irq->type);
 889         exit(1);
 890     }
 891
 892     inject_vcpu_irq_legacy(cs, irq);
 893 }
 894
 895 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 896 {
 897     struct kvm_s390_interrupt kvmint = {};
 898     int r;
 899
 900     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 901     if (r < 0) {
 902         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 903         exit(1);
 904     }
 905
 906     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
 907     if (r < 0) {
 908         fprintf(stderr, "KVM failed to inject interrupt\n");
 909         exit(1);
 910     }
 911 }
 912
 913 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 914 {
 915     static bool use_flic = true;
 916     int r;
 917
 918     if (use_flic) {
 919         r = kvm_s390_inject_flic(irq);
 920         if (r == -ENOSYS) {
 921             use_flic = false;
 922         }
 923         if (!r) {
 924             return;
 925         }
 926     }
 927     __kvm_s390_floating_interrupt(irq);
 928 }
 929
 930 void kvm_s390_virtio_irq(int config_change, uint64_t token)
 931 {
 932     struct kvm_s390_irq irq = {
 933         .type = KVM_S390_INT_VIRTIO,
 934         .u.ext.ext_params = config_change,
 935         .u.ext.ext_params2 = token,
 936     };
 937
 938     kvm_s390_floating_interrupt(&irq);
 939 }
 940
 941 void kvm_s390_service_interrupt(uint32_t parm)
 942 {
 943     struct kvm_s390_irq irq = {
 944         .type = KVM_S390_INT_SERVICE,
 945         .u.ext.ext_params = parm,
 946     };
 947
 948     kvm_s390_floating_interrupt(&irq);
 949 }
 950
 951 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
 952 {
 953     struct kvm_s390_irq irq = {
 954         .type = KVM_S390_PROGRAM_INT,
 955         .u.pgm.code = code,
 956     };
 957
 958     kvm_s390_vcpu_interrupt(cpu, &irq);
 959 }
 960
 961 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
 962 {
 963     struct kvm_s390_irq irq = {
 964         .type = KVM_S390_PROGRAM_INT,
 965         .u.pgm.code = code,
 966         .u.pgm.trans_exc_code = te_code,
 967         .u.pgm.exc_access_id = te_code & 3,
 968     };
 969
 970     kvm_s390_vcpu_interrupt(cpu, &irq);
 971 }
 972
 973 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
 974                                  uint16_t ipbh0)
 975 {
 976     CPUS390XState *env = &cpu->env;
 977     uint64_t sccb;
 978     uint32_t code;
 979     int r = 0;
 980
 981     cpu_synchronize_state(CPU(cpu));
 982     sccb = env->regs[ipbh0 & 0xf];
 983     code = env->regs[(ipbh0 & 0xf0) >> 4];
 984
 985     r = sclp_service_call(env, sccb, code);
 986     if (r < 0) {
 987         enter_pgmcheck(cpu, -r);
 988     } else {
 989         setcc(cpu, r);
 990     }
 991
 992     return 0;
 993 }
 994
 995 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 996 {
 997     CPUS390XState *env = &cpu->env;
 998     int rc = 0;
 999     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1000
1001     cpu_synchronize_state(CPU(cpu));
1002
1003     switch (ipa1) {
1004     case PRIV_B2_XSCH:
1005         ioinst_handle_xsch(cpu, env->regs[1]);
1006         break;
1007     case PRIV_B2_CSCH:
1008         ioinst_handle_csch(cpu, env->regs[1]);
1009         break;
1010     case PRIV_B2_HSCH:
1011         ioinst_handle_hsch(cpu, env->regs[1]);
1012         break;
1013     case PRIV_B2_MSCH:
1014         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
1015         break;
1016     case PRIV_B2_SSCH:
1017         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
1018         break;
1019     case PRIV_B2_STCRW:
1020         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
1021         break;
1022     case PRIV_B2_STSCH:
1023         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
1024         break;
1025     case PRIV_B2_TSCH:
1026         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1027         fprintf(stderr, "Spurious tsch intercept\n");
1028         break;
1029     case PRIV_B2_CHSC:
1030         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
1031         break;
1032     case PRIV_B2_TPI:
1033         /* This should have been handled by kvm already. */
1034         fprintf(stderr, "Spurious tpi intercept\n");
1035         break;
1036     case PRIV_B2_SCHM:
1037         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1038                            run->s390_sieic.ipb);
1039         break;
1040     case PRIV_B2_RSCH:
1041         ioinst_handle_rsch(cpu, env->regs[1]);
1042         break;
1043     case PRIV_B2_RCHP:
1044         ioinst_handle_rchp(cpu, env->regs[1]);
1045         break;
1046     case PRIV_B2_STCPS:
1047         /* We do not provide this instruction, it is suppressed. */
1048         break;
1049     case PRIV_B2_SAL:
1050         ioinst_handle_sal(cpu, env->regs[1]);
1051         break;
1052     case PRIV_B2_SIGA:
1053         /* Not provided, set CC = 3 for subchannel not operational */
1054         setcc(cpu, 3);
1055         break;
1056     case PRIV_B2_SCLP_CALL:
1057         rc = kvm_sclp_service_call(cpu, run, ipbh0);
1058         break;
1059     default:
1060         rc = -1;
1061         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1062         break;
1063     }
1064
1065     return rc;
1066 }
1067
1068 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1069                                   uint8_t *ar)
1070 {
1071     CPUS390XState *env = &cpu->env;
1072     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1073     uint32_t base2 = run->s390_sieic.ipb >> 28;
1074     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1075                      ((run->s390_sieic.ipb & 0xff00) << 4);
1076
1077     if (disp2 & 0x80000) {
1078         disp2 += 0xfff00000;
1079     }
1080     if (ar) {
1081         *ar = base2;
1082     }
1083
1084     return (base2 ? env->regs[base2] : 0) +
1085            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1086 }
1087
1088 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1089                                   uint8_t *ar)
1090 {
1091     CPUS390XState *env = &cpu->env;
1092     uint32_t base2 = run->s390_sieic.ipb >> 28;
1093     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1094                      ((run->s390_sieic.ipb & 0xff00) << 4);
1095
1096     if (disp2 & 0x80000) {
1097         disp2 += 0xfff00000;
1098     }
1099     if (ar) {
1100         *ar = base2;
1101     }
1102
1103     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1104 }
1105
1106 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1107 {
1108     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1109
1110     return clp_service_call(cpu, r2);
1111 }
1112
1113 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1114 {
1115     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1116     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1117
1118     return pcilg_service_call(cpu, r1, r2);
1119 }
1120
1121 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1122 {
1123     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1124     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1125
1126     return pcistg_service_call(cpu, r1, r2);
1127 }
1128
1129 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1130 {
1131     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1132     uint64_t fiba;
1133     uint8_t ar;
1134
1135     cpu_synchronize_state(CPU(cpu));
1136     fiba = get_base_disp_rxy(cpu, run, &ar);
1137
1138     return stpcifc_service_call(cpu, r1, fiba, ar);
1139 }
1140
1141 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1142 {
1143     /* NOOP */
1144     return 0;
1145 }
1146
1147 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1148 {
1149     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1150     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1151
1152     return rpcit_service_call(cpu, r1, r2);
1153 }
1154
1155 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1156 {
1157     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1158     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1159     uint64_t gaddr;
1160     uint8_t ar;
1161
1162     cpu_synchronize_state(CPU(cpu));
1163     gaddr = get_base_disp_rsy(cpu, run, &ar);
1164
1165     return pcistb_service_call(cpu, r1, r3, gaddr, ar);
1166 }
1167
1168 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1169 {
1170     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1171     uint64_t fiba;
1172     uint8_t ar;
1173
1174     cpu_synchronize_state(CPU(cpu));
1175     fiba = get_base_disp_rxy(cpu, run, &ar);
1176
1177     return mpcifc_service_call(cpu, r1, fiba, ar);
1178 }
1179
1180 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1181 {
1182     int r = 0;
1183
1184     switch (ipa1) {
1185     case PRIV_B9_CLP:
1186         r = kvm_clp_service_call(cpu, run);
1187         break;
1188     case PRIV_B9_PCISTG:
1189         r = kvm_pcistg_service_call(cpu, run);
1190         break;
1191     case PRIV_B9_PCILG:
1192         r = kvm_pcilg_service_call(cpu, run);
1193         break;
1194     case PRIV_B9_RPCIT:
1195         r = kvm_rpcit_service_call(cpu, run);
1196         break;
1197     case PRIV_B9_EQBS:
1198         /* just inject exception */
1199         r = -1;
1200         break;
1201     default:
1202         r = -1;
1203         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1204         break;
1205     }
1206
1207     return r;
1208 }
1209
1210 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1211 {
1212     int r = 0;
1213
1214     switch (ipbl) {
1215     case PRIV_EB_PCISTB:
1216         r = kvm_pcistb_service_call(cpu, run);
1217         break;
1218     case PRIV_EB_SIC:
1219         r = kvm_sic_service_call(cpu, run);
1220         break;
1221     case PRIV_EB_SQBS:
1222         /* just inject exception */
1223         r = -1;
1224         break;
1225     default:
1226         r = -1;
1227         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1228         break;
1229     }
1230
1231     return r;
1232 }
1233
1234 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1235 {
1236     int r = 0;
1237
1238     switch (ipbl) {
1239     case PRIV_E3_MPCIFC:
1240         r = kvm_mpcifc_service_call(cpu, run);
1241         break;
1242     case PRIV_E3_STPCIFC:
1243         r = kvm_stpcifc_service_call(cpu, run);
1244         break;
1245     default:
1246         r = -1;
1247         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1248         break;
1249     }
1250
1251     return r;
1252 }
1253
1254 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1255 {
1256     CPUS390XState *env = &cpu->env;
1257     int ret;
1258
1259     cpu_synchronize_state(CPU(cpu));
1260     ret = s390_virtio_hypercall(env);
1261     if (ret == -EINVAL) {
1262         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1263         return 0;
1264     }
1265
1266     return ret;
1267 }
1268
1269 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1270 {
1271     uint64_t r1, r3;
1272
1273     cpu_synchronize_state(CPU(cpu));
1274     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1275     r3 = run->s390_sieic.ipa & 0x000f;
1276     handle_diag_308(&cpu->env, r1, r3);
1277 }
1278
1279 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1280 {
1281     CPUS390XState *env = &cpu->env;
1282     unsigned long pc;
1283
1284     cpu_synchronize_state(CPU(cpu));
1285
1286     pc = env->psw.addr - 4;
1287     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1288         env->psw.addr = pc;
1289         return EXCP_DEBUG;
1290     }
1291
1292     return -ENOENT;
1293 }
1294
1295 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1296
1297 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1298 {
1299     int r = 0;
1300     uint16_t func_code;
1301
1302     /*
1303      * For any diagnose call we support, bits 48-63 of the resulting
1304      * address specify the function code; the remainder is ignored.
1305      */
1306     func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1307     switch (func_code) {
1308     case DIAG_IPL:
1309         kvm_handle_diag_308(cpu, run);
1310         break;
1311     case DIAG_KVM_HYPERCALL:
1312         r = handle_hypercall(cpu, run);
1313         break;
1314     case DIAG_KVM_BREAKPOINT:
1315         r = handle_sw_breakpoint(cpu, run);
1316         break;
1317     default:
1318         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1319         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1320         break;
1321     }
1322
1323     return r;
1324 }
1325
1326 typedef struct SigpInfo {
1327     S390CPU *cpu;
1328     uint64_t param;
1329     int cc;
1330     uint64_t *status_reg;
1331 } SigpInfo;
1332
1333 static void set_sigp_status(SigpInfo *si, uint64_t status)
1334 {
1335     *si->status_reg &= 0xffffffff00000000ULL;
1336     *si->status_reg |= status;
1337     si->cc = SIGP_CC_STATUS_STORED;
1338 }
1339
1340 static void sigp_start(void *arg)
1341 {
1342     SigpInfo *si = arg;
1343
1344     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1345         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1346         return;
1347     }
1348
1349     s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1350     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1351 }
1352
1353 static void sigp_stop(void *arg)
1354 {
1355     SigpInfo *si = arg;
1356     struct kvm_s390_irq irq = {
1357         .type = KVM_S390_SIGP_STOP,
1358     };
1359
1360     if (s390_cpu_get_state(si->cpu) != CPU_STATE_OPERATING) {
1361         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1362         return;
1363     }
1364
1365     /* disabled wait - sleeping in user space */
1366     if (CPU(si->cpu)->halted) {
1367         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1368     } else {
1369         /* execute the stop function */
1370         si->cpu->env.sigp_order = SIGP_STOP;
1371         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1372     }
1373     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1374 }
1375
1376 #define ADTL_SAVE_AREA_SIZE 1024
1377 static int kvm_s390_store_adtl_status(S390CPU *cpu, hwaddr addr)
1378 {
1379     void *mem;
1380     hwaddr len = ADTL_SAVE_AREA_SIZE;
1381
1382     mem = cpu_physical_memory_map(addr, &len, 1);
1383     if (!mem) {
1384         return -EFAULT;
1385     }
1386     if (len != ADTL_SAVE_AREA_SIZE) {
1387         cpu_physical_memory_unmap(mem, len, 1, 0);
1388         return -EFAULT;
1389     }
1390
1391     memcpy(mem, &cpu->env.vregs, 512);
1392
1393     cpu_physical_memory_unmap(mem, len, 1, len);
1394
1395     return 0;
1396 }
1397
1398 #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, floating_pt_save_area)
1399 #define SAVE_AREA_SIZE 512
1400 static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
1401 {
1402     static const uint8_t ar_id = 1;
1403     uint64_t ckc = cpu->env.ckc >> 8;
1404     void *mem;
1405     int i;
1406     hwaddr len = SAVE_AREA_SIZE;
1407
1408     mem = cpu_physical_memory_map(addr, &len, 1);
1409     if (!mem) {
1410         return -EFAULT;
1411     }
1412     if (len != SAVE_AREA_SIZE) {
1413         cpu_physical_memory_unmap(mem, len, 1, 0);
1414         return -EFAULT;
1415     }
1416
1417     if (store_arch) {
1418         cpu_physical_memory_write(offsetof(LowCore, ar_access_id), &ar_id, 1);
1419     }
1420     for (i = 0; i < 16; ++i) {
1421         *((uint64 *)mem + i) = get_freg(&cpu->env, i)->ll;
1422     }
1423     memcpy(mem + 128, &cpu->env.regs, 128);
1424     memcpy(mem + 256, &cpu->env.psw, 16);
1425     memcpy(mem + 280, &cpu->env.psa, 4);
1426     memcpy(mem + 284, &cpu->env.fpc, 4);
1427     memcpy(mem + 292, &cpu->env.todpr, 4);
1428     memcpy(mem + 296, &cpu->env.cputm, 8);
1429     memcpy(mem + 304, &ckc, 8);
1430     memcpy(mem + 320, &cpu->env.aregs, 64);
1431     memcpy(mem + 384, &cpu->env.cregs, 128);
1432
1433     cpu_physical_memory_unmap(mem, len, 1, len);
1434
1435     return 0;
1436 }
1437
1438 static void sigp_stop_and_store_status(void *arg)
1439 {
1440     SigpInfo *si = arg;
1441     struct kvm_s390_irq irq = {
1442         .type = KVM_S390_SIGP_STOP,
1443     };
1444
1445     /* disabled wait - sleeping in user space */
1446     if (s390_cpu_get_state(si->cpu) == CPU_STATE_OPERATING &&
1447         CPU(si->cpu)->halted) {
1448         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1449     }
1450
1451     switch (s390_cpu_get_state(si->cpu)) {
1452     case CPU_STATE_OPERATING:
1453         si->cpu->env.sigp_order = SIGP_STOP_STORE_STATUS;
1454         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1455         /* store will be performed when handling the stop intercept */
1456         break;
1457     case CPU_STATE_STOPPED:
1458         /* already stopped, just store the status */
1459         cpu_synchronize_state(CPU(si->cpu));
1460         kvm_s390_store_status(si->cpu, KVM_S390_STORE_STATUS_DEF_ADDR, true);
1461         break;
1462     }
1463     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1464 }
1465
1466 static void sigp_store_status_at_address(void *arg)
1467 {
1468     SigpInfo *si = arg;
1469     uint32_t address = si->param & 0x7ffffe00u;
1470
1471     /* cpu has to be stopped */
1472     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1473         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1474         return;
1475     }
1476
1477     cpu_synchronize_state(CPU(si->cpu));
1478
1479     if (kvm_s390_store_status(si->cpu, address, false)) {
1480         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1481         return;
1482     }
1483     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1484 }
1485
1486 static void sigp_store_adtl_status(void *arg)
1487 {
1488     SigpInfo *si = arg;
1489
1490     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_VECTOR_REGISTERS)) {
1491         set_sigp_status(si, SIGP_STAT_INVALID_ORDER);
1492         return;
1493     }
1494
1495     /* cpu has to be stopped */
1496     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1497         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1498         return;
1499     }
1500
1501     /* parameter must be aligned to 1024-byte boundary */
1502     if (si->param & 0x3ff) {
1503         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1504         return;
1505     }
1506
1507     cpu_synchronize_state(CPU(si->cpu));
1508
1509     if (kvm_s390_store_adtl_status(si->cpu, si->param)) {
1510         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1511         return;
1512     }
1513     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1514 }
1515
1516 static void sigp_restart(void *arg)
1517 {
1518     SigpInfo *si = arg;
1519     struct kvm_s390_irq irq = {
1520         .type = KVM_S390_RESTART,
1521     };
1522
1523     switch (s390_cpu_get_state(si->cpu)) {
1524     case CPU_STATE_STOPPED:
1525         /* the restart irq has to be delivered prior to any other pending irq */
1526         cpu_synchronize_state(CPU(si->cpu));
1527         do_restart_interrupt(&si->cpu->env);
1528         s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1529         break;
1530     case CPU_STATE_OPERATING:
1531         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1532         break;
1533     }
1534     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1535 }
1536
1537 int kvm_s390_cpu_restart(S390CPU *cpu)
1538 {
1539     SigpInfo si = {
1540         .cpu = cpu,
1541     };
1542
1543     run_on_cpu(CPU(cpu), sigp_restart, &si);
1544     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1545     return 0;
1546 }
1547
1548 static void sigp_initial_cpu_reset(void *arg)
1549 {
1550     SigpInfo *si = arg;
1551     CPUState *cs = CPU(si->cpu);
1552     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1553
1554     cpu_synchronize_state(cs);
1555     scc->initial_cpu_reset(cs);
1556     cpu_synchronize_post_reset(cs);
1557     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1558 }
1559
1560 static void sigp_cpu_reset(void *arg)
1561 {
1562     SigpInfo *si = arg;
1563     CPUState *cs = CPU(si->cpu);
1564     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1565
1566     cpu_synchronize_state(cs);
1567     scc->cpu_reset(cs);
1568     cpu_synchronize_post_reset(cs);
1569     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1570 }
1571
1572 static void sigp_set_prefix(void *arg)
1573 {
1574     SigpInfo *si = arg;
1575     uint32_t addr = si->param & 0x7fffe000u;
1576
1577     cpu_synchronize_state(CPU(si->cpu));
1578
1579     if (!address_space_access_valid(&address_space_memory, addr,
1580                                     sizeof(struct LowCore), false)) {
1581         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1582         return;
1583     }
1584
1585     /* cpu has to be stopped */
1586     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1587         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1588         return;
1589     }
1590
1591     si->cpu->env.psa = addr;
1592     cpu_synchronize_post_init(CPU(si->cpu));
1593     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1594 }
1595
1596 static int handle_sigp_single_dst(S390CPU *dst_cpu, uint8_t order,
1597                                   uint64_t param, uint64_t *status_reg)
1598 {
1599     SigpInfo si = {
1600         .cpu = dst_cpu,
1601         .param = param,
1602         .status_reg = status_reg,
1603     };
1604
1605     /* cpu available? */
1606     if (dst_cpu == NULL) {
1607         return SIGP_CC_NOT_OPERATIONAL;
1608     }
1609
1610     /* only resets can break pending orders */
1611     if (dst_cpu->env.sigp_order != 0 &&
1612         order != SIGP_CPU_RESET &&
1613         order != SIGP_INITIAL_CPU_RESET) {
1614         return SIGP_CC_BUSY;
1615     }
1616
1617     switch (order) {
1618     case SIGP_START:
1619         run_on_cpu(CPU(dst_cpu), sigp_start, &si);
1620         break;
1621     case SIGP_STOP:
1622         run_on_cpu(CPU(dst_cpu), sigp_stop, &si);
1623         break;
1624     case SIGP_RESTART:
1625         run_on_cpu(CPU(dst_cpu), sigp_restart, &si);
1626         break;
1627     case SIGP_STOP_STORE_STATUS:
1628         run_on_cpu(CPU(dst_cpu), sigp_stop_and_store_status, &si);
1629         break;
1630     case SIGP_STORE_STATUS_ADDR:
1631         run_on_cpu(CPU(dst_cpu), sigp_store_status_at_address, &si);
1632         break;
1633     case SIGP_STORE_ADTL_STATUS:
1634         run_on_cpu(CPU(dst_cpu), sigp_store_adtl_status, &si);
1635         break;
1636     case SIGP_SET_PREFIX:
1637         run_on_cpu(CPU(dst_cpu), sigp_set_prefix, &si);
1638         break;
1639     case SIGP_INITIAL_CPU_RESET:
1640         run_on_cpu(CPU(dst_cpu), sigp_initial_cpu_reset, &si);
1641         break;
1642     case SIGP_CPU_RESET:
1643         run_on_cpu(CPU(dst_cpu), sigp_cpu_reset, &si);
1644         break;
1645     default:
1646         DPRINTF("KVM: unknown SIGP: 0x%x\n", order);
1647         set_sigp_status(&si, SIGP_STAT_INVALID_ORDER);
1648     }
1649
1650     return si.cc;
1651 }
1652
1653 static int sigp_set_architecture(S390CPU *cpu, uint32_t param,
1654                                  uint64_t *status_reg)
1655 {
1656     CPUState *cur_cs;
1657     S390CPU *cur_cpu;
1658
1659     /* due to the BQL, we are the only active cpu */
1660     CPU_FOREACH(cur_cs) {
1661         cur_cpu = S390_CPU(cur_cs);
1662         if (cur_cpu->env.sigp_order != 0) {
1663             return SIGP_CC_BUSY;
1664         }
1665         cpu_synchronize_state(cur_cs);
1666         /* all but the current one have to be stopped */
1667         if (cur_cpu != cpu &&
1668             s390_cpu_get_state(cur_cpu) != CPU_STATE_STOPPED) {
1669             *status_reg &= 0xffffffff00000000ULL;
1670             *status_reg |= SIGP_STAT_INCORRECT_STATE;
1671             return SIGP_CC_STATUS_STORED;
1672         }
1673     }
1674
1675     switch (param & 0xff) {
1676     case SIGP_MODE_ESA_S390:
1677         /* not supported */
1678         return SIGP_CC_NOT_OPERATIONAL;
1679     case SIGP_MODE_Z_ARCH_TRANS_ALL_PSW:
1680     case SIGP_MODE_Z_ARCH_TRANS_CUR_PSW:
1681         CPU_FOREACH(cur_cs) {
1682             cur_cpu = S390_CPU(cur_cs);
1683             cur_cpu->env.pfault_token = -1UL;
1684         }
1685         break;
1686     default:
1687         *status_reg &= 0xffffffff00000000ULL;
1688         *status_reg |= SIGP_STAT_INVALID_PARAMETER;
1689         return SIGP_CC_STATUS_STORED;
1690     }
1691
1692     return SIGP_CC_ORDER_CODE_ACCEPTED;
1693 }
1694
1695 #define SIGP_ORDER_MASK 0x000000ff
1696
1697 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1698 {
1699     CPUS390XState *env = &cpu->env;
1700     const uint8_t r1 = ipa1 >> 4;
1701     const uint8_t r3 = ipa1 & 0x0f;
1702     int ret;
1703     uint8_t order;
1704     uint64_t *status_reg;
1705     uint64_t param;
1706     S390CPU *dst_cpu = NULL;
1707
1708     cpu_synchronize_state(CPU(cpu));
1709
1710     /* get order code */
1711     order = decode_basedisp_rs(env, run->s390_sieic.ipb, NULL)
1712         & SIGP_ORDER_MASK;
1713     status_reg = &env->regs[r1];
1714     param = (r1 % 2) ? env->regs[r1] : env->regs[r1 + 1];
1715
1716     switch (order) {
1717     case SIGP_SET_ARCH:
1718         ret = sigp_set_architecture(cpu, param, status_reg);
1719         break;
1720     default:
1721         /* all other sigp orders target a single vcpu */
1722         dst_cpu = s390_cpu_addr2state(env->regs[r3]);
1723         ret = handle_sigp_single_dst(dst_cpu, order, param, status_reg);
1724     }
1725
1726     trace_kvm_sigp_finished(order, CPU(cpu)->cpu_index,
1727                             dst_cpu ? CPU(dst_cpu)->cpu_index : -1, ret);
1728
1729     if (ret >= 0) {
1730         setcc(cpu, ret);
1731         return 0;
1732     }
1733
1734     return ret;
1735 }
1736
1737 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1738 {
1739     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1740     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1741     int r = -1;
1742
1743     DPRINTF("handle_instruction 0x%x 0x%x\n",
1744             run->s390_sieic.ipa, run->s390_sieic.ipb);
1745     switch (ipa0) {
1746     case IPA0_B2:
1747         r = handle_b2(cpu, run, ipa1);
1748         break;
1749     case IPA0_B9:
1750         r = handle_b9(cpu, run, ipa1);
1751         break;
1752     case IPA0_EB:
1753         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1754         break;
1755     case IPA0_E3:
1756         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1757         break;
1758     case IPA0_DIAG:
1759         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1760         break;
1761     case IPA0_SIGP:
1762         r = handle_sigp(cpu, run, ipa1);
1763         break;
1764     }
1765
1766     if (r < 0) {
1767         r = 0;
1768         enter_pgmcheck(cpu, 0x0001);
1769     }
1770
1771     return r;
1772 }
1773
1774 static bool is_special_wait_psw(CPUState *cs)
1775 {
1776     /* signal quiesce */
1777     return cs->kvm_run->psw_addr == 0xfffUL;
1778 }
1779
1780 static void guest_panicked(void)
1781 {
1782     qapi_event_send_guest_panicked(GUEST_PANIC_ACTION_PAUSE,
1783                                    &error_abort);
1784     vm_stop(RUN_STATE_GUEST_PANICKED);
1785 }
1786
1787 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1788 {
1789     CPUState *cs = CPU(cpu);
1790
1791     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1792                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1793                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1794     s390_cpu_halt(cpu);
1795     guest_panicked();
1796 }
1797
1798 static int handle_intercept(S390CPU *cpu)
1799 {
1800     CPUState *cs = CPU(cpu);
1801     struct kvm_run *run = cs->kvm_run;
1802     int icpt_code = run->s390_sieic.icptcode;
1803     int r = 0;
1804
1805     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1806             (long)cs->kvm_run->psw_addr);
1807     switch (icpt_code) {
1808         case ICPT_INSTRUCTION:
1809             r = handle_instruction(cpu, run);
1810             break;
1811         case ICPT_PROGRAM:
1812             unmanageable_intercept(cpu, "program interrupt",
1813                                    offsetof(LowCore, program_new_psw));
1814             r = EXCP_HALTED;
1815             break;
1816         case ICPT_EXT_INT:
1817             unmanageable_intercept(cpu, "external interrupt",
1818                                    offsetof(LowCore, external_new_psw));
1819             r = EXCP_HALTED;
1820             break;
1821         case ICPT_WAITPSW:
1822             /* disabled wait, since enabled wait is handled in kernel */
1823             cpu_synchronize_state(cs);
1824             if (s390_cpu_halt(cpu) == 0) {
1825                 if (is_special_wait_psw(cs)) {
1826                     qemu_system_shutdown_request();
1827                 } else {
1828                     guest_panicked();
1829                 }
1830             }
1831             r = EXCP_HALTED;
1832             break;
1833         case ICPT_CPU_STOP:
1834             if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
1835                 qemu_system_shutdown_request();
1836             }
1837             if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
1838                 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR,
1839                                       true);
1840             }
1841             cpu->env.sigp_order = 0;
1842             r = EXCP_HALTED;
1843             break;
1844         case ICPT_SOFT_INTERCEPT:
1845             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1846             exit(1);
1847             break;
1848         case ICPT_IO:
1849             fprintf(stderr, "KVM unimplemented icpt IO\n");
1850             exit(1);
1851             break;
1852         default:
1853             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1854             exit(1);
1855             break;
1856     }
1857
1858     return r;
1859 }
1860
1861 static int handle_tsch(S390CPU *cpu)
1862 {
1863     CPUState *cs = CPU(cpu);
1864     struct kvm_run *run = cs->kvm_run;
1865     int ret;
1866
1867     cpu_synchronize_state(cs);
1868
1869     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb);
1870     if (ret < 0) {
1871         /*
1872          * Failure.
1873          * If an I/O interrupt had been dequeued, we have to reinject it.
1874          */
1875         if (run->s390_tsch.dequeued) {
1876             kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1877                                   run->s390_tsch.subchannel_nr,
1878                                   run->s390_tsch.io_int_parm,
1879                                   run->s390_tsch.io_int_word);
1880         }
1881         ret = 0;
1882     }
1883     return ret;
1884 }
1885
1886 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1887 {
1888     struct sysib_322 sysib;
1889     int del;
1890
1891     if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1892         return;
1893     }
1894     /* Shift the stack of Extended Names to prepare for our own data */
1895     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1896             sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1897     /* First virt level, that doesn't provide Ext Names delimits stack. It is
1898      * assumed it's not capable of managing Extended Names for lower levels.
1899      */
1900     for (del = 1; del < sysib.count; del++) {
1901         if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1902             break;
1903         }
1904     }
1905     if (del < sysib.count) {
1906         memset(sysib.ext_names[del], 0,
1907                sizeof(sysib.ext_names[0]) * (sysib.count - del));
1908     }
1909     /* Insert short machine name in EBCDIC, padded with blanks */
1910     if (qemu_name) {
1911         memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1912         ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1913                                                     strlen(qemu_name)));
1914     }
1915     sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1916     memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1917     /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1918      * considered by s390 as not capable of providing any Extended Name.
1919      * Therefore if no name was specified on qemu invocation, we go with the
1920      * same "KVMguest" default, which KVM has filled into short name field.
1921      */
1922     if (qemu_name) {
1923         strncpy((char *)sysib.ext_names[0], qemu_name,
1924                 sizeof(sysib.ext_names[0]));
1925     } else {
1926         strcpy((char *)sysib.ext_names[0], "KVMguest");
1927     }
1928     /* Insert UUID */
1929     memcpy(sysib.vm[0].uuid, qemu_uuid, sizeof(sysib.vm[0].uuid));
1930
1931     s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1932 }
1933
1934 static int handle_stsi(S390CPU *cpu)
1935 {
1936     CPUState *cs = CPU(cpu);
1937     struct kvm_run *run = cs->kvm_run;
1938
1939     switch (run->s390_stsi.fc) {
1940     case 3:
1941         if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
1942             return 0;
1943         }
1944         /* Only sysib 3.2.2 needs post-handling for now. */
1945         insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
1946         return 0;
1947     default:
1948         return 0;
1949     }
1950 }
1951
1952 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1953 {
1954     CPUState *cs = CPU(cpu);
1955     struct kvm_run *run = cs->kvm_run;
1956
1957     int ret = 0;
1958     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1959
1960     switch (arch_info->type) {
1961     case KVM_HW_WP_WRITE:
1962         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1963             cs->watchpoint_hit = &hw_watchpoint;
1964             hw_watchpoint.vaddr = arch_info->addr;
1965             hw_watchpoint.flags = BP_MEM_WRITE;
1966             ret = EXCP_DEBUG;
1967         }
1968         break;
1969     case KVM_HW_BP:
1970         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1971             ret = EXCP_DEBUG;
1972         }
1973         break;
1974     case KVM_SINGLESTEP:
1975         if (cs->singlestep_enabled) {
1976             ret = EXCP_DEBUG;
1977         }
1978         break;
1979     default:
1980         ret = -ENOSYS;
1981     }
1982
1983     return ret;
1984 }
1985
1986 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1987 {
1988     S390CPU *cpu = S390_CPU(cs);
1989     int ret = 0;
1990
1991     switch (run->exit_reason) {
1992         case KVM_EXIT_S390_SIEIC:
1993             ret = handle_intercept(cpu);
1994             break;
1995         case KVM_EXIT_S390_RESET:
1996             s390_reipl_request();
1997             break;
1998         case KVM_EXIT_S390_TSCH:
1999             ret = handle_tsch(cpu);
2000             break;
2001         case KVM_EXIT_S390_STSI:
2002             ret = handle_stsi(cpu);
2003             break;
2004         case KVM_EXIT_DEBUG:
2005             ret = kvm_arch_handle_debug_exit(cpu);
2006             break;
2007         default:
2008             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
2009             break;
2010     }
2011
2012     if (ret == 0) {
2013         ret = EXCP_INTERRUPT;
2014     }
2015     return ret;
2016 }
2017
2018 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2019 {
2020     return true;
2021 }
2022
2023 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2024 {
2025     return 1;
2026 }
2027
2028 int kvm_arch_on_sigbus(int code, void *addr)
2029 {
2030     return 1;
2031 }
2032
2033 void kvm_s390_io_interrupt(uint16_t subchannel_id,
2034                            uint16_t subchannel_nr, uint32_t io_int_parm,
2035                            uint32_t io_int_word)
2036 {
2037     struct kvm_s390_irq irq = {
2038         .u.io.subchannel_id = subchannel_id,
2039         .u.io.subchannel_nr = subchannel_nr,
2040         .u.io.io_int_parm = io_int_parm,
2041         .u.io.io_int_word = io_int_word,
2042     };
2043
2044     if (io_int_word & IO_INT_WORD_AI) {
2045         irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
2046     } else {
2047         irq.type = ((subchannel_id & 0xff00) << 24) |
2048             ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
2049     }
2050     kvm_s390_floating_interrupt(&irq);
2051 }
2052
2053 void kvm_s390_crw_mchk(void)
2054 {
2055     struct kvm_s390_irq irq = {
2056         .type = KVM_S390_MCHK,
2057         .u.mchk.cr14 = 1 << 28,
2058         .u.mchk.mcic = 0x00400f1d40330000ULL,
2059     };
2060     kvm_s390_floating_interrupt(&irq);
2061 }
2062
2063 void kvm_s390_enable_css_support(S390CPU *cpu)
2064 {
2065     int r;
2066
2067     /* Activate host kernel channel subsystem support. */
2068     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
2069     assert(r == 0);
2070 }
2071
2072 void kvm_arch_init_irq_routing(KVMState *s)
2073 {
2074     /*
2075      * Note that while irqchip capabilities generally imply that cpustates
2076      * are handled in-kernel, it is not true for s390 (yet); therefore, we
2077      * have to override the common code kvm_halt_in_kernel_allowed setting.
2078      */
2079     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2080         kvm_gsi_routing_allowed = true;
2081         kvm_halt_in_kernel_allowed = false;
2082     }
2083 }
2084
2085 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
2086                                     int vq, bool assign)
2087 {
2088     struct kvm_ioeventfd kick = {
2089         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
2090         KVM_IOEVENTFD_FLAG_DATAMATCH,
2091         .fd = event_notifier_get_fd(notifier),
2092         .datamatch = vq,
2093         .addr = sch,
2094         .len = 8,
2095     };
2096     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
2097         return -ENOSYS;
2098     }
2099     if (!assign) {
2100         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
2101     }
2102     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
2103 }
2104
2105 int kvm_s390_get_memslot_count(KVMState *s)
2106 {
2107     return kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
2108 }
2109
2110 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
2111 {
2112     struct kvm_mp_state mp_state = {};
2113     int ret;
2114
2115     /* the kvm part might not have been initialized yet */
2116     if (CPU(cpu)->kvm_state == NULL) {
2117         return 0;
2118     }
2119
2120     switch (cpu_state) {
2121     case CPU_STATE_STOPPED:
2122         mp_state.mp_state = KVM_MP_STATE_STOPPED;
2123         break;
2124     case CPU_STATE_CHECK_STOP:
2125         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
2126         break;
2127     case CPU_STATE_OPERATING:
2128         mp_state.mp_state = KVM_MP_STATE_OPERATING;
2129         break;
2130     case CPU_STATE_LOAD:
2131         mp_state.mp_state = KVM_MP_STATE_LOAD;
2132         break;
2133     default:
2134         error_report("Requested CPU state is not a valid S390 CPU state: %u",
2135                      cpu_state);
2136         exit(1);
2137     }
2138
2139     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2140     if (ret) {
2141         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
2142                                        strerror(-ret));
2143     }
2144
2145     return ret;
2146 }
2147
2148 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
2149 {
2150     struct kvm_s390_irq_state irq_state;
2151     CPUState *cs = CPU(cpu);
2152     int32_t bytes;
2153
2154     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2155         return;
2156     }
2157
2158     irq_state.buf = (uint64_t) cpu->irqstate;
2159     irq_state.len = VCPU_IRQ_BUF_SIZE;
2160
2161     bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
2162     if (bytes < 0) {
2163         cpu->irqstate_saved_size = 0;
2164         error_report("Migration of interrupt state failed");
2165         return;
2166     }
2167
2168     cpu->irqstate_saved_size = bytes;
2169 }
2170
2171 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
2172 {
2173     CPUState *cs = CPU(cpu);
2174     struct kvm_s390_irq_state irq_state;
2175     int r;
2176
2177     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2178         return -ENOSYS;
2179     }
2180
2181     if (cpu->irqstate_saved_size == 0) {
2182         return 0;
2183     }
2184     irq_state.buf = (uint64_t) cpu->irqstate;
2185     irq_state.len = cpu->irqstate_saved_size;
2186
2187     r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
2188     if (r) {
2189         error_report("Setting interrupt state failed %d", r);
2190     }
2191     return r;
2192 }
2193
2194 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2195                               uint64_t address, uint32_t data)
2196 {
2197     S390PCIBusDevice *pbdev;
2198     uint32_t fid = data >> ZPCI_MSI_VEC_BITS;
2199     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
2200
2201     pbdev = s390_pci_find_dev_by_fid(fid);
2202     if (!pbdev) {
2203         DPRINTF("add_msi_route no dev\n");
2204         return -ENODEV;
2205     }
2206
2207     pbdev->routes.adapter.ind_offset = vec;
2208
2209     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2210     route->flags = 0;
2211     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2212     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2213     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2214     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset;
2215     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2216     return 0;
2217 }