target-s390x/kvm.c

   1 /*
   2  * QEMU S390x KVM implementation
   3  *
   4  * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
   5  * Copyright IBM Corp. 2012
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * Contributions after 2012-10-29 are licensed under the terms of the
  18  * GNU GPL, version 2 or (at your option) any later version.
  19  *
  20  * You should have received a copy of the GNU (Lesser) General Public
  21  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  22  */
  23
  24 #include "qemu/osdep.h"
  25 #include <sys/ioctl.h>
  26
  27 #include <linux/kvm.h>
  28 #include <asm/ptrace.h>
  29
  30 #include "qemu-common.h"
  31 #include "cpu.h"
  32 #include "qemu/error-report.h"
  33 #include "qemu/timer.h"
  34 #include "sysemu/sysemu.h"
  35 #include "sysemu/kvm.h"
  36 #include "hw/hw.h"
  37 #include "sysemu/device_tree.h"
  38 #include "qapi/qmp/qjson.h"
  39 #include "exec/gdbstub.h"
  40 #include "exec/address-spaces.h"
  41 #include "trace.h"
  42 #include "qapi-event.h"
  43 #include "hw/s390x/s390-pci-inst.h"
  44 #include "hw/s390x/s390-pci-bus.h"
  45 #include "hw/s390x/ipl.h"
  46 #include "hw/s390x/ebcdic.h"
  47 #include "exec/memattrs.h"
  48 #include "hw/s390x/s390-virtio-ccw.h"
  49
  50 /* #define DEBUG_KVM */
  51
  52 #ifdef DEBUG_KVM
  53 #define DPRINTF(fmt, ...) \
  54     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  55 #else
  56 #define DPRINTF(fmt, ...) \
  57     do { } while (0)
  58 #endif
  59
  60 #define kvm_vm_check_mem_attr(s, attr) \
  61     kvm_vm_check_attr(s, KVM_S390_VM_MEM_CTRL, attr)
  62
  63 #define IPA0_DIAG                       0x8300
  64 #define IPA0_SIGP                       0xae00
  65 #define IPA0_B2                         0xb200
  66 #define IPA0_B9                         0xb900
  67 #define IPA0_EB                         0xeb00
  68 #define IPA0_E3                         0xe300
  69
  70 #define PRIV_B2_SCLP_CALL               0x20
  71 #define PRIV_B2_CSCH                    0x30
  72 #define PRIV_B2_HSCH                    0x31
  73 #define PRIV_B2_MSCH                    0x32
  74 #define PRIV_B2_SSCH                    0x33
  75 #define PRIV_B2_STSCH                   0x34
  76 #define PRIV_B2_TSCH                    0x35
  77 #define PRIV_B2_TPI                     0x36
  78 #define PRIV_B2_SAL                     0x37
  79 #define PRIV_B2_RSCH                    0x38
  80 #define PRIV_B2_STCRW                   0x39
  81 #define PRIV_B2_STCPS                   0x3a
  82 #define PRIV_B2_RCHP                    0x3b
  83 #define PRIV_B2_SCHM                    0x3c
  84 #define PRIV_B2_CHSC                    0x5f
  85 #define PRIV_B2_SIGA                    0x74
  86 #define PRIV_B2_XSCH                    0x76
  87
  88 #define PRIV_EB_SQBS                    0x8a
  89 #define PRIV_EB_PCISTB                  0xd0
  90 #define PRIV_EB_SIC                     0xd1
  91
  92 #define PRIV_B9_EQBS                    0x9c
  93 #define PRIV_B9_CLP                     0xa0
  94 #define PRIV_B9_PCISTG                  0xd0
  95 #define PRIV_B9_PCILG                   0xd2
  96 #define PRIV_B9_RPCIT                   0xd3
  97
  98 #define PRIV_E3_MPCIFC                  0xd0
  99 #define PRIV_E3_STPCIFC                 0xd4
 100
 101 #define DIAG_TIMEREVENT                 0x288
 102 #define DIAG_IPL                        0x308
 103 #define DIAG_KVM_HYPERCALL              0x500
 104 #define DIAG_KVM_BREAKPOINT             0x501
 105
 106 #define ICPT_INSTRUCTION                0x04
 107 #define ICPT_PROGRAM                    0x08
 108 #define ICPT_EXT_INT                    0x14
 109 #define ICPT_WAITPSW                    0x1c
 110 #define ICPT_SOFT_INTERCEPT             0x24
 111 #define ICPT_CPU_STOP                   0x28
 112 #define ICPT_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 static int cap_ri;
 139
 140 static void *legacy_s390_alloc(size_t size, uint64_t *align);
 141
 142 static int kvm_s390_query_mem_limit(KVMState *s, uint64_t *memory_limit)
 143 {
 144     struct kvm_device_attr attr = {
 145         .group = KVM_S390_VM_MEM_CTRL,
 146         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
 147         .addr = (uint64_t) memory_limit,
 148     };
 149
 150     return kvm_vm_ioctl(s, KVM_GET_DEVICE_ATTR, &attr);
 151 }
 152
 153 int kvm_s390_set_mem_limit(KVMState *s, uint64_t new_limit, uint64_t *hw_limit)
 154 {
 155     int rc;
 156
 157     struct kvm_device_attr attr = {
 158         .group = KVM_S390_VM_MEM_CTRL,
 159         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
 160         .addr = (uint64_t) &new_limit,
 161     };
 162
 163     if (!kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_LIMIT_SIZE)) {
 164         return 0;
 165     }
 166
 167     rc = kvm_s390_query_mem_limit(s, hw_limit);
 168     if (rc) {
 169         return rc;
 170     } else if (*hw_limit < new_limit) {
 171         return -E2BIG;
 172     }
 173
 174     return kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 175 }
 176
 177 void kvm_s390_cmma_reset(void)
 178 {
 179     int rc;
 180     struct kvm_device_attr attr = {
 181         .group = KVM_S390_VM_MEM_CTRL,
 182         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 183     };
 184
 185     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 186     trace_kvm_clear_cmma(rc);
 187 }
 188
 189 static void kvm_s390_enable_cmma(KVMState *s)
 190 {
 191     int rc;
 192     struct kvm_device_attr attr = {
 193         .group = KVM_S390_VM_MEM_CTRL,
 194         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 195     };
 196
 197     if (!kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_ENABLE_CMMA) ||
 198         !kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_CLR_CMMA)) {
 199         return;
 200     }
 201
 202     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 203     trace_kvm_enable_cmma(rc);
 204 }
 205
 206 static void kvm_s390_set_attr(uint64_t attr)
 207 {
 208     struct kvm_device_attr attribute = {
 209         .group = KVM_S390_VM_CRYPTO,
 210         .attr  = attr,
 211     };
 212
 213     int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
 214
 215     if (ret) {
 216         error_report("Failed to set crypto device attribute %lu: %s",
 217                      attr, strerror(-ret));
 218     }
 219 }
 220
 221 static void kvm_s390_init_aes_kw(void)
 222 {
 223     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
 224
 225     if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
 226                                  NULL)) {
 227             attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
 228     }
 229
 230     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 231             kvm_s390_set_attr(attr);
 232     }
 233 }
 234
 235 static void kvm_s390_init_dea_kw(void)
 236 {
 237     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
 238
 239     if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
 240                                  NULL)) {
 241             attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
 242     }
 243
 244     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 245             kvm_s390_set_attr(attr);
 246     }
 247 }
 248
 249 void kvm_s390_crypto_reset(void)
 250 {
 251     kvm_s390_init_aes_kw();
 252     kvm_s390_init_dea_kw();
 253 }
 254
 255 int kvm_arch_init(MachineState *ms, KVMState *s)
 256 {
 257     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
 258     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
 259     cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
 260     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
 261
 262     if (!mem_path) {
 263         kvm_s390_enable_cmma(s);
 264     }
 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_VECTOR_REGISTERS, 0);
 273     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_STSI, 0);
 274     if (ri_allowed()) {
 275         if (kvm_vm_enable_cap(s, KVM_CAP_S390_RI, 0) == 0) {
 276             cap_ri = 1;
 277         }
 278     }
 279
 280     return 0;
 281 }
 282
 283 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
 284 {
 285     return cpu->cpu_index;
 286 }
 287
 288 int kvm_arch_init_vcpu(CPUState *cs)
 289 {
 290     S390CPU *cpu = S390_CPU(cs);
 291     kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
 292     cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE);
 293     return 0;
 294 }
 295
 296 void kvm_s390_reset_vcpu(S390CPU *cpu)
 297 {
 298     CPUState *cs = CPU(cpu);
 299
 300     /* The initial reset call is needed here to reset in-kernel
 301      * vcpu data that we can't access directly from QEMU
 302      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
 303      * Before this ioctl cpu_synchronize_state() is called in common kvm
 304      * code (kvm-all) */
 305     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
 306         error_report("Initial CPU reset failed on CPU %i", cs->cpu_index);
 307     }
 308 }
 309
 310 static int can_sync_regs(CPUState *cs, int regs)
 311 {
 312     return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
 313 }
 314
 315 int kvm_arch_put_registers(CPUState *cs, int level)
 316 {
 317     S390CPU *cpu = S390_CPU(cs);
 318     CPUS390XState *env = &cpu->env;
 319     struct kvm_sregs sregs;
 320     struct kvm_regs regs;
 321     struct kvm_fpu fpu = {};
 322     int r;
 323     int i;
 324
 325     /* always save the PSW  and the GPRS*/
 326     cs->kvm_run->psw_addr = env->psw.addr;
 327     cs->kvm_run->psw_mask = env->psw.mask;
 328
 329     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 330         for (i = 0; i < 16; i++) {
 331             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
 332             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
 333         }
 334     } else {
 335         for (i = 0; i < 16; i++) {
 336             regs.gprs[i] = env->regs[i];
 337         }
 338         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
 339         if (r < 0) {
 340             return r;
 341         }
 342     }
 343
 344     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 345         for (i = 0; i < 32; i++) {
 346             cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
 347             cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
 348         }
 349         cs->kvm_run->s.regs.fpc = env->fpc;
 350         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
 351     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
 352         for (i = 0; i < 16; i++) {
 353             cs->kvm_run->s.regs.fprs[i] = get_freg(env, i)->ll;
 354         }
 355         cs->kvm_run->s.regs.fpc = env->fpc;
 356         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_FPRS;
 357     } else {
 358         /* Floating point */
 359         for (i = 0; i < 16; i++) {
 360             fpu.fprs[i] = get_freg(env, i)->ll;
 361         }
 362         fpu.fpc = env->fpc;
 363
 364         r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
 365         if (r < 0) {
 366             return r;
 367         }
 368     }
 369
 370     /* Do we need to save more than that? */
 371     if (level == KVM_PUT_RUNTIME_STATE) {
 372         return 0;
 373     }
 374
 375     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 376         cs->kvm_run->s.regs.cputm = env->cputm;
 377         cs->kvm_run->s.regs.ckc = env->ckc;
 378         cs->kvm_run->s.regs.todpr = env->todpr;
 379         cs->kvm_run->s.regs.gbea = env->gbea;
 380         cs->kvm_run->s.regs.pp = env->pp;
 381         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
 382     } else {
 383         /*
 384          * These ONE_REGS are not protected by a capability. As they are only
 385          * necessary for migration we just trace a possible error, but don't
 386          * return with an error return code.
 387          */
 388         kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 389         kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 390         kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 391         kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 392         kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 393     }
 394
 395     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
 396         memcpy(cs->kvm_run->s.regs.riccb, env->riccb, 64);
 397         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_RICCB;
 398     }
 399
 400     /* pfault parameters */
 401     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 402         cs->kvm_run->s.regs.pft = env->pfault_token;
 403         cs->kvm_run->s.regs.pfs = env->pfault_select;
 404         cs->kvm_run->s.regs.pfc = env->pfault_compare;
 405         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
 406     } else if (cap_async_pf) {
 407         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 408         if (r < 0) {
 409             return r;
 410         }
 411         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 412         if (r < 0) {
 413             return r;
 414         }
 415         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 416         if (r < 0) {
 417             return r;
 418         }
 419     }
 420
 421     /* access registers and control registers*/
 422     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 423         for (i = 0; i < 16; i++) {
 424             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
 425             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
 426         }
 427         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
 428         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
 429     } else {
 430         for (i = 0; i < 16; i++) {
 431             sregs.acrs[i] = env->aregs[i];
 432             sregs.crs[i] = env->cregs[i];
 433         }
 434         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
 435         if (r < 0) {
 436             return r;
 437         }
 438     }
 439
 440     /* Finally the prefix */
 441     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 442         cs->kvm_run->s.regs.prefix = env->psa;
 443         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
 444     } else {
 445         /* prefix is only supported via sync regs */
 446     }
 447     return 0;
 448 }
 449
 450 int kvm_arch_get_registers(CPUState *cs)
 451 {
 452     S390CPU *cpu = S390_CPU(cs);
 453     CPUS390XState *env = &cpu->env;
 454     struct kvm_sregs sregs;
 455     struct kvm_regs regs;
 456     struct kvm_fpu fpu;
 457     int i, r;
 458
 459     /* get the PSW */
 460     env->psw.addr = cs->kvm_run->psw_addr;
 461     env->psw.mask = cs->kvm_run->psw_mask;
 462
 463     /* the GPRS */
 464     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 465         for (i = 0; i < 16; i++) {
 466             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
 467         }
 468     } else {
 469         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
 470         if (r < 0) {
 471             return r;
 472         }
 473          for (i = 0; i < 16; i++) {
 474             env->regs[i] = regs.gprs[i];
 475         }
 476     }
 477
 478     /* The ACRS and CRS */
 479     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 480         for (i = 0; i < 16; i++) {
 481             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
 482             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
 483         }
 484     } else {
 485         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
 486         if (r < 0) {
 487             return r;
 488         }
 489          for (i = 0; i < 16; i++) {
 490             env->aregs[i] = sregs.acrs[i];
 491             env->cregs[i] = sregs.crs[i];
 492         }
 493     }
 494
 495     /* Floating point and vector registers */
 496     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 497         for (i = 0; i < 32; i++) {
 498             env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
 499             env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
 500         }
 501         env->fpc = cs->kvm_run->s.regs.fpc;
 502     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
 503         for (i = 0; i < 16; i++) {
 504             get_freg(env, i)->ll = cs->kvm_run->s.regs.fprs[i];
 505         }
 506         env->fpc = cs->kvm_run->s.regs.fpc;
 507     } else {
 508         r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
 509         if (r < 0) {
 510             return r;
 511         }
 512         for (i = 0; i < 16; i++) {
 513             get_freg(env, i)->ll = fpu.fprs[i];
 514         }
 515         env->fpc = fpu.fpc;
 516     }
 517
 518     /* The prefix */
 519     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 520         env->psa = cs->kvm_run->s.regs.prefix;
 521     }
 522
 523     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 524         env->cputm = cs->kvm_run->s.regs.cputm;
 525         env->ckc = cs->kvm_run->s.regs.ckc;
 526         env->todpr = cs->kvm_run->s.regs.todpr;
 527         env->gbea = cs->kvm_run->s.regs.gbea;
 528         env->pp = cs->kvm_run->s.regs.pp;
 529     } else {
 530         /*
 531          * These ONE_REGS are not protected by a capability. As they are only
 532          * necessary for migration we just trace a possible error, but don't
 533          * return with an error return code.
 534          */
 535         kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 536         kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 537         kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 538         kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 539         kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 540     }
 541
 542     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
 543         memcpy(env->riccb, cs->kvm_run->s.regs.riccb, 64);
 544     }
 545
 546     /* pfault parameters */
 547     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 548         env->pfault_token = cs->kvm_run->s.regs.pft;
 549         env->pfault_select = cs->kvm_run->s.regs.pfs;
 550         env->pfault_compare = cs->kvm_run->s.regs.pfc;
 551     } else if (cap_async_pf) {
 552         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 553         if (r < 0) {
 554             return r;
 555         }
 556         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 557         if (r < 0) {
 558             return r;
 559         }
 560         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 561         if (r < 0) {
 562             return r;
 563         }
 564     }
 565
 566     return 0;
 567 }
 568
 569 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
 570 {
 571     int r;
 572     struct kvm_device_attr attr = {
 573         .group = KVM_S390_VM_TOD,
 574         .attr = KVM_S390_VM_TOD_LOW,
 575         .addr = (uint64_t)tod_low,
 576     };
 577
 578     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 579     if (r) {
 580         return r;
 581     }
 582
 583     attr.attr = KVM_S390_VM_TOD_HIGH;
 584     attr.addr = (uint64_t)tod_high;
 585     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 586 }
 587
 588 int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_low)
 589 {
 590     int r;
 591
 592     struct kvm_device_attr attr = {
 593         .group = KVM_S390_VM_TOD,
 594         .attr = KVM_S390_VM_TOD_LOW,
 595         .addr = (uint64_t)tod_low,
 596     };
 597
 598     r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 599     if (r) {
 600         return r;
 601     }
 602
 603     attr.attr = KVM_S390_VM_TOD_HIGH;
 604     attr.addr = (uint64_t)tod_high;
 605     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 606 }
 607
 608 /**
 609  * kvm_s390_mem_op:
 610  * @addr:      the logical start address in guest memory
 611  * @ar:        the access register number
 612  * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
 613  * @len:       length that should be transferred
 614  * @is_write:  true = write, false = read
 615  * Returns:    0 on success, non-zero if an exception or error occurred
 616  *
 617  * Use KVM ioctl to read/write from/to guest memory. An access exception
 618  * is injected into the vCPU in case of translation errors.
 619  */
 620 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
 621                     int len, bool is_write)
 622 {
 623     struct kvm_s390_mem_op mem_op = {
 624         .gaddr = addr,
 625         .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
 626         .size = len,
 627         .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
 628                        : KVM_S390_MEMOP_LOGICAL_READ,
 629         .buf = (uint64_t)hostbuf,
 630         .ar = ar,
 631     };
 632     int ret;
 633
 634     if (!cap_mem_op) {
 635         return -ENOSYS;
 636     }
 637     if (!hostbuf) {
 638         mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
 639     }
 640
 641     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
 642     if (ret < 0) {
 643         error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
 644     }
 645     return ret;
 646 }
 647
 648 /*
 649  * Legacy layout for s390:
 650  * Older S390 KVM requires the topmost vma of the RAM to be
 651  * smaller than an system defined value, which is at least 256GB.
 652  * Larger systems have larger values. We put the guest between
 653  * the end of data segment (system break) and this value. We
 654  * use 32GB as a base to have enough room for the system break
 655  * to grow. We also have to use MAP parameters that avoid
 656  * read-only mapping of guest pages.
 657  */
 658 static void *legacy_s390_alloc(size_t size, uint64_t *align)
 659 {
 660     void *mem;
 661
 662     mem = mmap((void *) 0x800000000ULL, size,
 663                PROT_EXEC|PROT_READ|PROT_WRITE,
 664                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 665     return mem == MAP_FAILED ? NULL : mem;
 666 }
 667
 668 /* DIAG 501 is used for sw breakpoints */
 669 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 670
 671 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 672 {
 673
 674     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 675                             sizeof(diag_501), 0) ||
 676         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501,
 677                             sizeof(diag_501), 1)) {
 678         return -EINVAL;
 679     }
 680     return 0;
 681 }
 682
 683 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 684 {
 685     uint8_t t[sizeof(diag_501)];
 686
 687     if (cpu_memory_rw_debug(cs, bp->pc, t, sizeof(diag_501), 0)) {
 688         return -EINVAL;
 689     } else if (memcmp(t, diag_501, sizeof(diag_501))) {
 690         return -EINVAL;
 691     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 692                                    sizeof(diag_501), 1)) {
 693         return -EINVAL;
 694     }
 695
 696     return 0;
 697 }
 698
 699 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
 700                                                     int len, int type)
 701 {
 702     int n;
 703
 704     for (n = 0; n < nb_hw_breakpoints; n++) {
 705         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
 706             (hw_breakpoints[n].len == len || len == -1)) {
 707             return &hw_breakpoints[n];
 708         }
 709     }
 710
 711     return NULL;
 712 }
 713
 714 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
 715 {
 716     int size;
 717
 718     if (find_hw_breakpoint(addr, len, type)) {
 719         return -EEXIST;
 720     }
 721
 722     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
 723
 724     if (!hw_breakpoints) {
 725         nb_hw_breakpoints = 0;
 726         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
 727     } else {
 728         hw_breakpoints =
 729             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
 730     }
 731
 732     if (!hw_breakpoints) {
 733         nb_hw_breakpoints = 0;
 734         return -ENOMEM;
 735     }
 736
 737     hw_breakpoints[nb_hw_breakpoints].addr = addr;
 738     hw_breakpoints[nb_hw_breakpoints].len = len;
 739     hw_breakpoints[nb_hw_breakpoints].type = type;
 740
 741     nb_hw_breakpoints++;
 742
 743     return 0;
 744 }
 745
 746 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 747                                   target_ulong len, int type)
 748 {
 749     switch (type) {
 750     case GDB_BREAKPOINT_HW:
 751         type = KVM_HW_BP;
 752         break;
 753     case GDB_WATCHPOINT_WRITE:
 754         if (len < 1) {
 755             return -EINVAL;
 756         }
 757         type = KVM_HW_WP_WRITE;
 758         break;
 759     default:
 760         return -ENOSYS;
 761     }
 762     return insert_hw_breakpoint(addr, len, type);
 763 }
 764
 765 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 766                                   target_ulong len, int type)
 767 {
 768     int size;
 769     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
 770
 771     if (bp == NULL) {
 772         return -ENOENT;
 773     }
 774
 775     nb_hw_breakpoints--;
 776     if (nb_hw_breakpoints > 0) {
 777         /*
 778          * In order to trim the array, move the last element to the position to
 779          * be removed - if necessary.
 780          */
 781         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
 782             *bp = hw_breakpoints[nb_hw_breakpoints];
 783         }
 784         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
 785         hw_breakpoints =
 786              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
 787     } else {
 788         g_free(hw_breakpoints);
 789         hw_breakpoints = NULL;
 790     }
 791
 792     return 0;
 793 }
 794
 795 void kvm_arch_remove_all_hw_breakpoints(void)
 796 {
 797     nb_hw_breakpoints = 0;
 798     g_free(hw_breakpoints);
 799     hw_breakpoints = NULL;
 800 }
 801
 802 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 803 {
 804     int i;
 805
 806     if (nb_hw_breakpoints > 0) {
 807         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
 808         dbg->arch.hw_bp = hw_breakpoints;
 809
 810         for (i = 0; i < nb_hw_breakpoints; ++i) {
 811             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
 812                                                        hw_breakpoints[i].addr);
 813         }
 814         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 815     } else {
 816         dbg->arch.nr_hw_bp = 0;
 817         dbg->arch.hw_bp = NULL;
 818     }
 819 }
 820
 821 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 822 {
 823 }
 824
 825 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
 826 {
 827     return MEMTXATTRS_UNSPECIFIED;
 828 }
 829
 830 int kvm_arch_process_async_events(CPUState *cs)
 831 {
 832     return cs->halted;
 833 }
 834
 835 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
 836                                      struct kvm_s390_interrupt *interrupt)
 837 {
 838     int r = 0;
 839
 840     interrupt->type = irq->type;
 841     switch (irq->type) {
 842     case KVM_S390_INT_VIRTIO:
 843         interrupt->parm = irq->u.ext.ext_params;
 844         /* fall through */
 845     case KVM_S390_INT_PFAULT_INIT:
 846     case KVM_S390_INT_PFAULT_DONE:
 847         interrupt->parm64 = irq->u.ext.ext_params2;
 848         break;
 849     case KVM_S390_PROGRAM_INT:
 850         interrupt->parm = irq->u.pgm.code;
 851         break;
 852     case KVM_S390_SIGP_SET_PREFIX:
 853         interrupt->parm = irq->u.prefix.address;
 854         break;
 855     case KVM_S390_INT_SERVICE:
 856         interrupt->parm = irq->u.ext.ext_params;
 857         break;
 858     case KVM_S390_MCHK:
 859         interrupt->parm = irq->u.mchk.cr14;
 860         interrupt->parm64 = irq->u.mchk.mcic;
 861         break;
 862     case KVM_S390_INT_EXTERNAL_CALL:
 863         interrupt->parm = irq->u.extcall.code;
 864         break;
 865     case KVM_S390_INT_EMERGENCY:
 866         interrupt->parm = irq->u.emerg.code;
 867         break;
 868     case KVM_S390_SIGP_STOP:
 869     case KVM_S390_RESTART:
 870         break; /* These types have no parameters */
 871     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
 872         interrupt->parm = irq->u.io.subchannel_id << 16;
 873         interrupt->parm |= irq->u.io.subchannel_nr;
 874         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
 875         interrupt->parm64 |= irq->u.io.io_int_word;
 876         break;
 877     default:
 878         r = -EINVAL;
 879         break;
 880     }
 881     return r;
 882 }
 883
 884 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
 885 {
 886     struct kvm_s390_interrupt kvmint = {};
 887     int r;
 888
 889     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 890     if (r < 0) {
 891         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 892         exit(1);
 893     }
 894
 895     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
 896     if (r < 0) {
 897         fprintf(stderr, "KVM failed to inject interrupt\n");
 898         exit(1);
 899     }
 900 }
 901
 902 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
 903 {
 904     CPUState *cs = CPU(cpu);
 905     int r;
 906
 907     if (cap_s390_irq) {
 908         r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
 909         if (!r) {
 910             return;
 911         }
 912         error_report("KVM failed to inject interrupt %llx", irq->type);
 913         exit(1);
 914     }
 915
 916     inject_vcpu_irq_legacy(cs, irq);
 917 }
 918
 919 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 920 {
 921     struct kvm_s390_interrupt kvmint = {};
 922     int r;
 923
 924     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 925     if (r < 0) {
 926         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 927         exit(1);
 928     }
 929
 930     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
 931     if (r < 0) {
 932         fprintf(stderr, "KVM failed to inject interrupt\n");
 933         exit(1);
 934     }
 935 }
 936
 937 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 938 {
 939     static bool use_flic = true;
 940     int r;
 941
 942     if (use_flic) {
 943         r = kvm_s390_inject_flic(irq);
 944         if (r == -ENOSYS) {
 945             use_flic = false;
 946         }
 947         if (!r) {
 948             return;
 949         }
 950     }
 951     __kvm_s390_floating_interrupt(irq);
 952 }
 953
 954 void kvm_s390_service_interrupt(uint32_t parm)
 955 {
 956     struct kvm_s390_irq irq = {
 957         .type = KVM_S390_INT_SERVICE,
 958         .u.ext.ext_params = parm,
 959     };
 960
 961     kvm_s390_floating_interrupt(&irq);
 962 }
 963
 964 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
 965 {
 966     struct kvm_s390_irq irq = {
 967         .type = KVM_S390_PROGRAM_INT,
 968         .u.pgm.code = code,
 969     };
 970
 971     kvm_s390_vcpu_interrupt(cpu, &irq);
 972 }
 973
 974 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
 975 {
 976     struct kvm_s390_irq irq = {
 977         .type = KVM_S390_PROGRAM_INT,
 978         .u.pgm.code = code,
 979         .u.pgm.trans_exc_code = te_code,
 980         .u.pgm.exc_access_id = te_code & 3,
 981     };
 982
 983     kvm_s390_vcpu_interrupt(cpu, &irq);
 984 }
 985
 986 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
 987                                  uint16_t ipbh0)
 988 {
 989     CPUS390XState *env = &cpu->env;
 990     uint64_t sccb;
 991     uint32_t code;
 992     int r = 0;
 993
 994     cpu_synchronize_state(CPU(cpu));
 995     sccb = env->regs[ipbh0 & 0xf];
 996     code = env->regs[(ipbh0 & 0xf0) >> 4];
 997
 998     r = sclp_service_call(env, sccb, code);
 999     if (r < 0) {
1000         enter_pgmcheck(cpu, -r);
1001     } else {
1002         setcc(cpu, r);
1003     }
1004
1005     return 0;
1006 }
1007
1008 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1009 {
1010     CPUS390XState *env = &cpu->env;
1011     int rc = 0;
1012     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1013
1014     cpu_synchronize_state(CPU(cpu));
1015
1016     switch (ipa1) {
1017     case PRIV_B2_XSCH:
1018         ioinst_handle_xsch(cpu, env->regs[1]);
1019         break;
1020     case PRIV_B2_CSCH:
1021         ioinst_handle_csch(cpu, env->regs[1]);
1022         break;
1023     case PRIV_B2_HSCH:
1024         ioinst_handle_hsch(cpu, env->regs[1]);
1025         break;
1026     case PRIV_B2_MSCH:
1027         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
1028         break;
1029     case PRIV_B2_SSCH:
1030         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
1031         break;
1032     case PRIV_B2_STCRW:
1033         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
1034         break;
1035     case PRIV_B2_STSCH:
1036         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
1037         break;
1038     case PRIV_B2_TSCH:
1039         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1040         fprintf(stderr, "Spurious tsch intercept\n");
1041         break;
1042     case PRIV_B2_CHSC:
1043         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
1044         break;
1045     case PRIV_B2_TPI:
1046         /* This should have been handled by kvm already. */
1047         fprintf(stderr, "Spurious tpi intercept\n");
1048         break;
1049     case PRIV_B2_SCHM:
1050         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1051                            run->s390_sieic.ipb);
1052         break;
1053     case PRIV_B2_RSCH:
1054         ioinst_handle_rsch(cpu, env->regs[1]);
1055         break;
1056     case PRIV_B2_RCHP:
1057         ioinst_handle_rchp(cpu, env->regs[1]);
1058         break;
1059     case PRIV_B2_STCPS:
1060         /* We do not provide this instruction, it is suppressed. */
1061         break;
1062     case PRIV_B2_SAL:
1063         ioinst_handle_sal(cpu, env->regs[1]);
1064         break;
1065     case PRIV_B2_SIGA:
1066         /* Not provided, set CC = 3 for subchannel not operational */
1067         setcc(cpu, 3);
1068         break;
1069     case PRIV_B2_SCLP_CALL:
1070         rc = kvm_sclp_service_call(cpu, run, ipbh0);
1071         break;
1072     default:
1073         rc = -1;
1074         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1075         break;
1076     }
1077
1078     return rc;
1079 }
1080
1081 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1082                                   uint8_t *ar)
1083 {
1084     CPUS390XState *env = &cpu->env;
1085     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1086     uint32_t base2 = run->s390_sieic.ipb >> 28;
1087     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1088                      ((run->s390_sieic.ipb & 0xff00) << 4);
1089
1090     if (disp2 & 0x80000) {
1091         disp2 += 0xfff00000;
1092     }
1093     if (ar) {
1094         *ar = base2;
1095     }
1096
1097     return (base2 ? env->regs[base2] : 0) +
1098            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1099 }
1100
1101 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1102                                   uint8_t *ar)
1103 {
1104     CPUS390XState *env = &cpu->env;
1105     uint32_t base2 = run->s390_sieic.ipb >> 28;
1106     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1107                      ((run->s390_sieic.ipb & 0xff00) << 4);
1108
1109     if (disp2 & 0x80000) {
1110         disp2 += 0xfff00000;
1111     }
1112     if (ar) {
1113         *ar = base2;
1114     }
1115
1116     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1117 }
1118
1119 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1120 {
1121     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1122
1123     return clp_service_call(cpu, r2);
1124 }
1125
1126 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1127 {
1128     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1129     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1130
1131     return pcilg_service_call(cpu, r1, r2);
1132 }
1133
1134 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1135 {
1136     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1137     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1138
1139     return pcistg_service_call(cpu, r1, r2);
1140 }
1141
1142 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1143 {
1144     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1145     uint64_t fiba;
1146     uint8_t ar;
1147
1148     cpu_synchronize_state(CPU(cpu));
1149     fiba = get_base_disp_rxy(cpu, run, &ar);
1150
1151     return stpcifc_service_call(cpu, r1, fiba, ar);
1152 }
1153
1154 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1155 {
1156     /* NOOP */
1157     return 0;
1158 }
1159
1160 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1161 {
1162     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1163     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1164
1165     return rpcit_service_call(cpu, r1, r2);
1166 }
1167
1168 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1169 {
1170     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1171     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1172     uint64_t gaddr;
1173     uint8_t ar;
1174
1175     cpu_synchronize_state(CPU(cpu));
1176     gaddr = get_base_disp_rsy(cpu, run, &ar);
1177
1178     return pcistb_service_call(cpu, r1, r3, gaddr, ar);
1179 }
1180
1181 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1182 {
1183     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1184     uint64_t fiba;
1185     uint8_t ar;
1186
1187     cpu_synchronize_state(CPU(cpu));
1188     fiba = get_base_disp_rxy(cpu, run, &ar);
1189
1190     return mpcifc_service_call(cpu, r1, fiba, ar);
1191 }
1192
1193 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1194 {
1195     int r = 0;
1196
1197     switch (ipa1) {
1198     case PRIV_B9_CLP:
1199         r = kvm_clp_service_call(cpu, run);
1200         break;
1201     case PRIV_B9_PCISTG:
1202         r = kvm_pcistg_service_call(cpu, run);
1203         break;
1204     case PRIV_B9_PCILG:
1205         r = kvm_pcilg_service_call(cpu, run);
1206         break;
1207     case PRIV_B9_RPCIT:
1208         r = kvm_rpcit_service_call(cpu, run);
1209         break;
1210     case PRIV_B9_EQBS:
1211         /* just inject exception */
1212         r = -1;
1213         break;
1214     default:
1215         r = -1;
1216         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1217         break;
1218     }
1219
1220     return r;
1221 }
1222
1223 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1224 {
1225     int r = 0;
1226
1227     switch (ipbl) {
1228     case PRIV_EB_PCISTB:
1229         r = kvm_pcistb_service_call(cpu, run);
1230         break;
1231     case PRIV_EB_SIC:
1232         r = kvm_sic_service_call(cpu, run);
1233         break;
1234     case PRIV_EB_SQBS:
1235         /* just inject exception */
1236         r = -1;
1237         break;
1238     default:
1239         r = -1;
1240         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1241         break;
1242     }
1243
1244     return r;
1245 }
1246
1247 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1248 {
1249     int r = 0;
1250
1251     switch (ipbl) {
1252     case PRIV_E3_MPCIFC:
1253         r = kvm_mpcifc_service_call(cpu, run);
1254         break;
1255     case PRIV_E3_STPCIFC:
1256         r = kvm_stpcifc_service_call(cpu, run);
1257         break;
1258     default:
1259         r = -1;
1260         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1261         break;
1262     }
1263
1264     return r;
1265 }
1266
1267 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1268 {
1269     CPUS390XState *env = &cpu->env;
1270     int ret;
1271
1272     cpu_synchronize_state(CPU(cpu));
1273     ret = s390_virtio_hypercall(env);
1274     if (ret == -EINVAL) {
1275         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1276         return 0;
1277     }
1278
1279     return ret;
1280 }
1281
1282 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1283 {
1284     uint64_t r1, r3;
1285     int rc;
1286
1287     cpu_synchronize_state(CPU(cpu));
1288     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1289     r3 = run->s390_sieic.ipa & 0x000f;
1290     rc = handle_diag_288(&cpu->env, r1, r3);
1291     if (rc) {
1292         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1293     }
1294 }
1295
1296 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1297 {
1298     uint64_t r1, r3;
1299
1300     cpu_synchronize_state(CPU(cpu));
1301     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1302     r3 = run->s390_sieic.ipa & 0x000f;
1303     handle_diag_308(&cpu->env, r1, r3);
1304 }
1305
1306 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1307 {
1308     CPUS390XState *env = &cpu->env;
1309     unsigned long pc;
1310
1311     cpu_synchronize_state(CPU(cpu));
1312
1313     pc = env->psw.addr - 4;
1314     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1315         env->psw.addr = pc;
1316         return EXCP_DEBUG;
1317     }
1318
1319     return -ENOENT;
1320 }
1321
1322 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1323
1324 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1325 {
1326     int r = 0;
1327     uint16_t func_code;
1328
1329     /*
1330      * For any diagnose call we support, bits 48-63 of the resulting
1331      * address specify the function code; the remainder is ignored.
1332      */
1333     func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1334     switch (func_code) {
1335     case DIAG_TIMEREVENT:
1336         kvm_handle_diag_288(cpu, run);
1337         break;
1338     case DIAG_IPL:
1339         kvm_handle_diag_308(cpu, run);
1340         break;
1341     case DIAG_KVM_HYPERCALL:
1342         r = handle_hypercall(cpu, run);
1343         break;
1344     case DIAG_KVM_BREAKPOINT:
1345         r = handle_sw_breakpoint(cpu, run);
1346         break;
1347     default:
1348         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1349         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1350         break;
1351     }
1352
1353     return r;
1354 }
1355
1356 typedef struct SigpInfo {
1357     S390CPU *cpu;
1358     uint64_t param;
1359     int cc;
1360     uint64_t *status_reg;
1361 } SigpInfo;
1362
1363 static void set_sigp_status(SigpInfo *si, uint64_t status)
1364 {
1365     *si->status_reg &= 0xffffffff00000000ULL;
1366     *si->status_reg |= status;
1367     si->cc = SIGP_CC_STATUS_STORED;
1368 }
1369
1370 static void sigp_start(void *arg)
1371 {
1372     SigpInfo *si = arg;
1373
1374     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1375         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1376         return;
1377     }
1378
1379     s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1380     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1381 }
1382
1383 static void sigp_stop(void *arg)
1384 {
1385     SigpInfo *si = arg;
1386     struct kvm_s390_irq irq = {
1387         .type = KVM_S390_SIGP_STOP,
1388     };
1389
1390     if (s390_cpu_get_state(si->cpu) != CPU_STATE_OPERATING) {
1391         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1392         return;
1393     }
1394
1395     /* disabled wait - sleeping in user space */
1396     if (CPU(si->cpu)->halted) {
1397         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1398     } else {
1399         /* execute the stop function */
1400         si->cpu->env.sigp_order = SIGP_STOP;
1401         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1402     }
1403     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1404 }
1405
1406 #define ADTL_SAVE_AREA_SIZE 1024
1407 static int kvm_s390_store_adtl_status(S390CPU *cpu, hwaddr addr)
1408 {
1409     void *mem;
1410     hwaddr len = ADTL_SAVE_AREA_SIZE;
1411
1412     mem = cpu_physical_memory_map(addr, &len, 1);
1413     if (!mem) {
1414         return -EFAULT;
1415     }
1416     if (len != ADTL_SAVE_AREA_SIZE) {
1417         cpu_physical_memory_unmap(mem, len, 1, 0);
1418         return -EFAULT;
1419     }
1420
1421     memcpy(mem, &cpu->env.vregs, 512);
1422
1423     cpu_physical_memory_unmap(mem, len, 1, len);
1424
1425     return 0;
1426 }
1427
1428 #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, floating_pt_save_area)
1429 #define SAVE_AREA_SIZE 512
1430 static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
1431 {
1432     static const uint8_t ar_id = 1;
1433     uint64_t ckc = cpu->env.ckc >> 8;
1434     void *mem;
1435     int i;
1436     hwaddr len = SAVE_AREA_SIZE;
1437
1438     mem = cpu_physical_memory_map(addr, &len, 1);
1439     if (!mem) {
1440         return -EFAULT;
1441     }
1442     if (len != SAVE_AREA_SIZE) {
1443         cpu_physical_memory_unmap(mem, len, 1, 0);
1444         return -EFAULT;
1445     }
1446
1447     if (store_arch) {
1448         cpu_physical_memory_write(offsetof(LowCore, ar_access_id), &ar_id, 1);
1449     }
1450     for (i = 0; i < 16; ++i) {
1451         *((uint64_t *)mem + i) = get_freg(&cpu->env, i)->ll;
1452     }
1453     memcpy(mem + 128, &cpu->env.regs, 128);
1454     memcpy(mem + 256, &cpu->env.psw, 16);
1455     memcpy(mem + 280, &cpu->env.psa, 4);
1456     memcpy(mem + 284, &cpu->env.fpc, 4);
1457     memcpy(mem + 292, &cpu->env.todpr, 4);
1458     memcpy(mem + 296, &cpu->env.cputm, 8);
1459     memcpy(mem + 304, &ckc, 8);
1460     memcpy(mem + 320, &cpu->env.aregs, 64);
1461     memcpy(mem + 384, &cpu->env.cregs, 128);
1462
1463     cpu_physical_memory_unmap(mem, len, 1, len);
1464
1465     return 0;
1466 }
1467
1468 static void sigp_stop_and_store_status(void *arg)
1469 {
1470     SigpInfo *si = arg;
1471     struct kvm_s390_irq irq = {
1472         .type = KVM_S390_SIGP_STOP,
1473     };
1474
1475     /* disabled wait - sleeping in user space */
1476     if (s390_cpu_get_state(si->cpu) == CPU_STATE_OPERATING &&
1477         CPU(si->cpu)->halted) {
1478         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1479     }
1480
1481     switch (s390_cpu_get_state(si->cpu)) {
1482     case CPU_STATE_OPERATING:
1483         si->cpu->env.sigp_order = SIGP_STOP_STORE_STATUS;
1484         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1485         /* store will be performed when handling the stop intercept */
1486         break;
1487     case CPU_STATE_STOPPED:
1488         /* already stopped, just store the status */
1489         cpu_synchronize_state(CPU(si->cpu));
1490         kvm_s390_store_status(si->cpu, KVM_S390_STORE_STATUS_DEF_ADDR, true);
1491         break;
1492     }
1493     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1494 }
1495
1496 static void sigp_store_status_at_address(void *arg)
1497 {
1498     SigpInfo *si = arg;
1499     uint32_t address = si->param & 0x7ffffe00u;
1500
1501     /* cpu has to be stopped */
1502     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1503         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1504         return;
1505     }
1506
1507     cpu_synchronize_state(CPU(si->cpu));
1508
1509     if (kvm_s390_store_status(si->cpu, address, false)) {
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_store_adtl_status(void *arg)
1517 {
1518     SigpInfo *si = arg;
1519
1520     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_VECTOR_REGISTERS)) {
1521         set_sigp_status(si, SIGP_STAT_INVALID_ORDER);
1522         return;
1523     }
1524
1525     /* cpu has to be stopped */
1526     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1527         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1528         return;
1529     }
1530
1531     /* parameter must be aligned to 1024-byte boundary */
1532     if (si->param & 0x3ff) {
1533         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1534         return;
1535     }
1536
1537     cpu_synchronize_state(CPU(si->cpu));
1538
1539     if (kvm_s390_store_adtl_status(si->cpu, si->param)) {
1540         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1541         return;
1542     }
1543     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1544 }
1545
1546 static void sigp_restart(void *arg)
1547 {
1548     SigpInfo *si = arg;
1549     struct kvm_s390_irq irq = {
1550         .type = KVM_S390_RESTART,
1551     };
1552
1553     switch (s390_cpu_get_state(si->cpu)) {
1554     case CPU_STATE_STOPPED:
1555         /* the restart irq has to be delivered prior to any other pending irq */
1556         cpu_synchronize_state(CPU(si->cpu));
1557         do_restart_interrupt(&si->cpu->env);
1558         s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1559         break;
1560     case CPU_STATE_OPERATING:
1561         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1562         break;
1563     }
1564     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1565 }
1566
1567 int kvm_s390_cpu_restart(S390CPU *cpu)
1568 {
1569     SigpInfo si = {
1570         .cpu = cpu,
1571     };
1572
1573     run_on_cpu(CPU(cpu), sigp_restart, &si);
1574     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1575     return 0;
1576 }
1577
1578 static void sigp_initial_cpu_reset(void *arg)
1579 {
1580     SigpInfo *si = arg;
1581     CPUState *cs = CPU(si->cpu);
1582     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1583
1584     cpu_synchronize_state(cs);
1585     scc->initial_cpu_reset(cs);
1586     cpu_synchronize_post_reset(cs);
1587     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1588 }
1589
1590 static void sigp_cpu_reset(void *arg)
1591 {
1592     SigpInfo *si = arg;
1593     CPUState *cs = CPU(si->cpu);
1594     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1595
1596     cpu_synchronize_state(cs);
1597     scc->cpu_reset(cs);
1598     cpu_synchronize_post_reset(cs);
1599     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1600 }
1601
1602 static void sigp_set_prefix(void *arg)
1603 {
1604     SigpInfo *si = arg;
1605     uint32_t addr = si->param & 0x7fffe000u;
1606
1607     cpu_synchronize_state(CPU(si->cpu));
1608
1609     if (!address_space_access_valid(&address_space_memory, addr,
1610                                     sizeof(struct LowCore), false)) {
1611         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1612         return;
1613     }
1614
1615     /* cpu has to be stopped */
1616     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1617         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1618         return;
1619     }
1620
1621     si->cpu->env.psa = addr;
1622     cpu_synchronize_post_init(CPU(si->cpu));
1623     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1624 }
1625
1626 static int handle_sigp_single_dst(S390CPU *dst_cpu, uint8_t order,
1627                                   uint64_t param, uint64_t *status_reg)
1628 {
1629     SigpInfo si = {
1630         .cpu = dst_cpu,
1631         .param = param,
1632         .status_reg = status_reg,
1633     };
1634
1635     /* cpu available? */
1636     if (dst_cpu == NULL) {
1637         return SIGP_CC_NOT_OPERATIONAL;
1638     }
1639
1640     /* only resets can break pending orders */
1641     if (dst_cpu->env.sigp_order != 0 &&
1642         order != SIGP_CPU_RESET &&
1643         order != SIGP_INITIAL_CPU_RESET) {
1644         return SIGP_CC_BUSY;
1645     }
1646
1647     switch (order) {
1648     case SIGP_START:
1649         run_on_cpu(CPU(dst_cpu), sigp_start, &si);
1650         break;
1651     case SIGP_STOP:
1652         run_on_cpu(CPU(dst_cpu), sigp_stop, &si);
1653         break;
1654     case SIGP_RESTART:
1655         run_on_cpu(CPU(dst_cpu), sigp_restart, &si);
1656         break;
1657     case SIGP_STOP_STORE_STATUS:
1658         run_on_cpu(CPU(dst_cpu), sigp_stop_and_store_status, &si);
1659         break;
1660     case SIGP_STORE_STATUS_ADDR:
1661         run_on_cpu(CPU(dst_cpu), sigp_store_status_at_address, &si);
1662         break;
1663     case SIGP_STORE_ADTL_STATUS:
1664         run_on_cpu(CPU(dst_cpu), sigp_store_adtl_status, &si);
1665         break;
1666     case SIGP_SET_PREFIX:
1667         run_on_cpu(CPU(dst_cpu), sigp_set_prefix, &si);
1668         break;
1669     case SIGP_INITIAL_CPU_RESET:
1670         run_on_cpu(CPU(dst_cpu), sigp_initial_cpu_reset, &si);
1671         break;
1672     case SIGP_CPU_RESET:
1673         run_on_cpu(CPU(dst_cpu), sigp_cpu_reset, &si);
1674         break;
1675     default:
1676         DPRINTF("KVM: unknown SIGP: 0x%x\n", order);
1677         set_sigp_status(&si, SIGP_STAT_INVALID_ORDER);
1678     }
1679
1680     return si.cc;
1681 }
1682
1683 static int sigp_set_architecture(S390CPU *cpu, uint32_t param,
1684                                  uint64_t *status_reg)
1685 {
1686     CPUState *cur_cs;
1687     S390CPU *cur_cpu;
1688
1689     /* due to the BQL, we are the only active cpu */
1690     CPU_FOREACH(cur_cs) {
1691         cur_cpu = S390_CPU(cur_cs);
1692         if (cur_cpu->env.sigp_order != 0) {
1693             return SIGP_CC_BUSY;
1694         }
1695         cpu_synchronize_state(cur_cs);
1696         /* all but the current one have to be stopped */
1697         if (cur_cpu != cpu &&
1698             s390_cpu_get_state(cur_cpu) != CPU_STATE_STOPPED) {
1699             *status_reg &= 0xffffffff00000000ULL;
1700             *status_reg |= SIGP_STAT_INCORRECT_STATE;
1701             return SIGP_CC_STATUS_STORED;
1702         }
1703     }
1704
1705     switch (param & 0xff) {
1706     case SIGP_MODE_ESA_S390:
1707         /* not supported */
1708         return SIGP_CC_NOT_OPERATIONAL;
1709     case SIGP_MODE_Z_ARCH_TRANS_ALL_PSW:
1710     case SIGP_MODE_Z_ARCH_TRANS_CUR_PSW:
1711         CPU_FOREACH(cur_cs) {
1712             cur_cpu = S390_CPU(cur_cs);
1713             cur_cpu->env.pfault_token = -1UL;
1714         }
1715         break;
1716     default:
1717         *status_reg &= 0xffffffff00000000ULL;
1718         *status_reg |= SIGP_STAT_INVALID_PARAMETER;
1719         return SIGP_CC_STATUS_STORED;
1720     }
1721
1722     return SIGP_CC_ORDER_CODE_ACCEPTED;
1723 }
1724
1725 #define SIGP_ORDER_MASK 0x000000ff
1726
1727 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1728 {
1729     CPUS390XState *env = &cpu->env;
1730     const uint8_t r1 = ipa1 >> 4;
1731     const uint8_t r3 = ipa1 & 0x0f;
1732     int ret;
1733     uint8_t order;
1734     uint64_t *status_reg;
1735     uint64_t param;
1736     S390CPU *dst_cpu = NULL;
1737
1738     cpu_synchronize_state(CPU(cpu));
1739
1740     /* get order code */
1741     order = decode_basedisp_rs(env, run->s390_sieic.ipb, NULL)
1742         & SIGP_ORDER_MASK;
1743     status_reg = &env->regs[r1];
1744     param = (r1 % 2) ? env->regs[r1] : env->regs[r1 + 1];
1745
1746     switch (order) {
1747     case SIGP_SET_ARCH:
1748         ret = sigp_set_architecture(cpu, param, status_reg);
1749         break;
1750     default:
1751         /* all other sigp orders target a single vcpu */
1752         dst_cpu = s390_cpu_addr2state(env->regs[r3]);
1753         ret = handle_sigp_single_dst(dst_cpu, order, param, status_reg);
1754     }
1755
1756     trace_kvm_sigp_finished(order, CPU(cpu)->cpu_index,
1757                             dst_cpu ? CPU(dst_cpu)->cpu_index : -1, ret);
1758
1759     if (ret >= 0) {
1760         setcc(cpu, ret);
1761         return 0;
1762     }
1763
1764     return ret;
1765 }
1766
1767 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1768 {
1769     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1770     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1771     int r = -1;
1772
1773     DPRINTF("handle_instruction 0x%x 0x%x\n",
1774             run->s390_sieic.ipa, run->s390_sieic.ipb);
1775     switch (ipa0) {
1776     case IPA0_B2:
1777         r = handle_b2(cpu, run, ipa1);
1778         break;
1779     case IPA0_B9:
1780         r = handle_b9(cpu, run, ipa1);
1781         break;
1782     case IPA0_EB:
1783         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1784         break;
1785     case IPA0_E3:
1786         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1787         break;
1788     case IPA0_DIAG:
1789         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1790         break;
1791     case IPA0_SIGP:
1792         r = handle_sigp(cpu, run, ipa1);
1793         break;
1794     }
1795
1796     if (r < 0) {
1797         r = 0;
1798         enter_pgmcheck(cpu, 0x0001);
1799     }
1800
1801     return r;
1802 }
1803
1804 static bool is_special_wait_psw(CPUState *cs)
1805 {
1806     /* signal quiesce */
1807     return cs->kvm_run->psw_addr == 0xfffUL;
1808 }
1809
1810 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1811 {
1812     CPUState *cs = CPU(cpu);
1813
1814     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1815                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1816                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1817     s390_cpu_halt(cpu);
1818     qemu_system_guest_panicked();
1819 }
1820
1821 static int handle_intercept(S390CPU *cpu)
1822 {
1823     CPUState *cs = CPU(cpu);
1824     struct kvm_run *run = cs->kvm_run;
1825     int icpt_code = run->s390_sieic.icptcode;
1826     int r = 0;
1827
1828     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1829             (long)cs->kvm_run->psw_addr);
1830     switch (icpt_code) {
1831         case ICPT_INSTRUCTION:
1832             r = handle_instruction(cpu, run);
1833             break;
1834         case ICPT_PROGRAM:
1835             unmanageable_intercept(cpu, "program interrupt",
1836                                    offsetof(LowCore, program_new_psw));
1837             r = EXCP_HALTED;
1838             break;
1839         case ICPT_EXT_INT:
1840             unmanageable_intercept(cpu, "external interrupt",
1841                                    offsetof(LowCore, external_new_psw));
1842             r = EXCP_HALTED;
1843             break;
1844         case ICPT_WAITPSW:
1845             /* disabled wait, since enabled wait is handled in kernel */
1846             cpu_synchronize_state(cs);
1847             if (s390_cpu_halt(cpu) == 0) {
1848                 if (is_special_wait_psw(cs)) {
1849                     qemu_system_shutdown_request();
1850                 } else {
1851                     qemu_system_guest_panicked();
1852                 }
1853             }
1854             r = EXCP_HALTED;
1855             break;
1856         case ICPT_CPU_STOP:
1857             if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
1858                 qemu_system_shutdown_request();
1859             }
1860             if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
1861                 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR,
1862                                       true);
1863             }
1864             cpu->env.sigp_order = 0;
1865             r = EXCP_HALTED;
1866             break;
1867         case ICPT_SOFT_INTERCEPT:
1868             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1869             exit(1);
1870             break;
1871         case ICPT_IO:
1872             fprintf(stderr, "KVM unimplemented icpt IO\n");
1873             exit(1);
1874             break;
1875         default:
1876             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1877             exit(1);
1878             break;
1879     }
1880
1881     return r;
1882 }
1883
1884 static int handle_tsch(S390CPU *cpu)
1885 {
1886     CPUState *cs = CPU(cpu);
1887     struct kvm_run *run = cs->kvm_run;
1888     int ret;
1889
1890     cpu_synchronize_state(cs);
1891
1892     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb);
1893     if (ret < 0) {
1894         /*
1895          * Failure.
1896          * If an I/O interrupt had been dequeued, we have to reinject it.
1897          */
1898         if (run->s390_tsch.dequeued) {
1899             kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1900                                   run->s390_tsch.subchannel_nr,
1901                                   run->s390_tsch.io_int_parm,
1902                                   run->s390_tsch.io_int_word);
1903         }
1904         ret = 0;
1905     }
1906     return ret;
1907 }
1908
1909 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1910 {
1911     struct sysib_322 sysib;
1912     int del;
1913
1914     if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1915         return;
1916     }
1917     /* Shift the stack of Extended Names to prepare for our own data */
1918     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1919             sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1920     /* First virt level, that doesn't provide Ext Names delimits stack. It is
1921      * assumed it's not capable of managing Extended Names for lower levels.
1922      */
1923     for (del = 1; del < sysib.count; del++) {
1924         if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1925             break;
1926         }
1927     }
1928     if (del < sysib.count) {
1929         memset(sysib.ext_names[del], 0,
1930                sizeof(sysib.ext_names[0]) * (sysib.count - del));
1931     }
1932     /* Insert short machine name in EBCDIC, padded with blanks */
1933     if (qemu_name) {
1934         memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1935         ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1936                                                     strlen(qemu_name)));
1937     }
1938     sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1939     memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1940     /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1941      * considered by s390 as not capable of providing any Extended Name.
1942      * Therefore if no name was specified on qemu invocation, we go with the
1943      * same "KVMguest" default, which KVM has filled into short name field.
1944      */
1945     if (qemu_name) {
1946         strncpy((char *)sysib.ext_names[0], qemu_name,
1947                 sizeof(sysib.ext_names[0]));
1948     } else {
1949         strcpy((char *)sysib.ext_names[0], "KVMguest");
1950     }
1951     /* Insert UUID */
1952     memcpy(sysib.vm[0].uuid, qemu_uuid, sizeof(sysib.vm[0].uuid));
1953
1954     s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1955 }
1956
1957 static int handle_stsi(S390CPU *cpu)
1958 {
1959     CPUState *cs = CPU(cpu);
1960     struct kvm_run *run = cs->kvm_run;
1961
1962     switch (run->s390_stsi.fc) {
1963     case 3:
1964         if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
1965             return 0;
1966         }
1967         /* Only sysib 3.2.2 needs post-handling for now. */
1968         insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
1969         return 0;
1970     default:
1971         return 0;
1972     }
1973 }
1974
1975 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1976 {
1977     CPUState *cs = CPU(cpu);
1978     struct kvm_run *run = cs->kvm_run;
1979
1980     int ret = 0;
1981     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1982
1983     switch (arch_info->type) {
1984     case KVM_HW_WP_WRITE:
1985         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1986             cs->watchpoint_hit = &hw_watchpoint;
1987             hw_watchpoint.vaddr = arch_info->addr;
1988             hw_watchpoint.flags = BP_MEM_WRITE;
1989             ret = EXCP_DEBUG;
1990         }
1991         break;
1992     case KVM_HW_BP:
1993         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1994             ret = EXCP_DEBUG;
1995         }
1996         break;
1997     case KVM_SINGLESTEP:
1998         if (cs->singlestep_enabled) {
1999             ret = EXCP_DEBUG;
2000         }
2001         break;
2002     default:
2003         ret = -ENOSYS;
2004     }
2005
2006     return ret;
2007 }
2008
2009 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
2010 {
2011     S390CPU *cpu = S390_CPU(cs);
2012     int ret = 0;
2013
2014     qemu_mutex_lock_iothread();
2015
2016     switch (run->exit_reason) {
2017         case KVM_EXIT_S390_SIEIC:
2018             ret = handle_intercept(cpu);
2019             break;
2020         case KVM_EXIT_S390_RESET:
2021             s390_reipl_request();
2022             break;
2023         case KVM_EXIT_S390_TSCH:
2024             ret = handle_tsch(cpu);
2025             break;
2026         case KVM_EXIT_S390_STSI:
2027             ret = handle_stsi(cpu);
2028             break;
2029         case KVM_EXIT_DEBUG:
2030             ret = kvm_arch_handle_debug_exit(cpu);
2031             break;
2032         default:
2033             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
2034             break;
2035     }
2036     qemu_mutex_unlock_iothread();
2037
2038     if (ret == 0) {
2039         ret = EXCP_INTERRUPT;
2040     }
2041     return ret;
2042 }
2043
2044 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2045 {
2046     return true;
2047 }
2048
2049 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2050 {
2051     return 1;
2052 }
2053
2054 int kvm_arch_on_sigbus(int code, void *addr)
2055 {
2056     return 1;
2057 }
2058
2059 void kvm_s390_io_interrupt(uint16_t subchannel_id,
2060                            uint16_t subchannel_nr, uint32_t io_int_parm,
2061                            uint32_t io_int_word)
2062 {
2063     struct kvm_s390_irq irq = {
2064         .u.io.subchannel_id = subchannel_id,
2065         .u.io.subchannel_nr = subchannel_nr,
2066         .u.io.io_int_parm = io_int_parm,
2067         .u.io.io_int_word = io_int_word,
2068     };
2069
2070     if (io_int_word & IO_INT_WORD_AI) {
2071         irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
2072     } else {
2073         irq.type = KVM_S390_INT_IO(0, (subchannel_id & 0xff00) >> 8,
2074                                       (subchannel_id & 0x0006),
2075                                       subchannel_nr);
2076     }
2077     kvm_s390_floating_interrupt(&irq);
2078 }
2079
2080 static uint64_t build_channel_report_mcic(void)
2081 {
2082     uint64_t mcic;
2083
2084     /* subclass: indicate channel report pending */
2085     mcic = MCIC_SC_CP |
2086     /* subclass modifiers: none */
2087     /* storage errors: none */
2088     /* validity bits: no damage */
2089         MCIC_VB_WP | MCIC_VB_MS | MCIC_VB_PM | MCIC_VB_IA | MCIC_VB_FP |
2090         MCIC_VB_GR | MCIC_VB_CR | MCIC_VB_ST | MCIC_VB_AR | MCIC_VB_PR |
2091         MCIC_VB_FC | MCIC_VB_CT | MCIC_VB_CC;
2092     if (kvm_check_extension(kvm_state, KVM_CAP_S390_VECTOR_REGISTERS)) {
2093         mcic |= MCIC_VB_VR;
2094     }
2095     return mcic;
2096 }
2097
2098 void kvm_s390_crw_mchk(void)
2099 {
2100     struct kvm_s390_irq irq = {
2101         .type = KVM_S390_MCHK,
2102         .u.mchk.cr14 = 1 << 28,
2103         .u.mchk.mcic = build_channel_report_mcic(),
2104     };
2105     kvm_s390_floating_interrupt(&irq);
2106 }
2107
2108 void kvm_s390_enable_css_support(S390CPU *cpu)
2109 {
2110     int r;
2111
2112     /* Activate host kernel channel subsystem support. */
2113     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
2114     assert(r == 0);
2115 }
2116
2117 void kvm_arch_init_irq_routing(KVMState *s)
2118 {
2119     /*
2120      * Note that while irqchip capabilities generally imply that cpustates
2121      * are handled in-kernel, it is not true for s390 (yet); therefore, we
2122      * have to override the common code kvm_halt_in_kernel_allowed setting.
2123      */
2124     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2125         kvm_gsi_routing_allowed = true;
2126         kvm_halt_in_kernel_allowed = false;
2127     }
2128 }
2129
2130 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
2131                                     int vq, bool assign)
2132 {
2133     struct kvm_ioeventfd kick = {
2134         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
2135         KVM_IOEVENTFD_FLAG_DATAMATCH,
2136         .fd = event_notifier_get_fd(notifier),
2137         .datamatch = vq,
2138         .addr = sch,
2139         .len = 8,
2140     };
2141     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
2142         return -ENOSYS;
2143     }
2144     if (!assign) {
2145         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
2146     }
2147     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
2148 }
2149
2150 int kvm_s390_get_memslot_count(KVMState *s)
2151 {
2152     return kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
2153 }
2154
2155 int kvm_s390_get_ri(void)
2156 {
2157     return cap_ri;
2158 }
2159
2160 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
2161 {
2162     struct kvm_mp_state mp_state = {};
2163     int ret;
2164
2165     /* the kvm part might not have been initialized yet */
2166     if (CPU(cpu)->kvm_state == NULL) {
2167         return 0;
2168     }
2169
2170     switch (cpu_state) {
2171     case CPU_STATE_STOPPED:
2172         mp_state.mp_state = KVM_MP_STATE_STOPPED;
2173         break;
2174     case CPU_STATE_CHECK_STOP:
2175         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
2176         break;
2177     case CPU_STATE_OPERATING:
2178         mp_state.mp_state = KVM_MP_STATE_OPERATING;
2179         break;
2180     case CPU_STATE_LOAD:
2181         mp_state.mp_state = KVM_MP_STATE_LOAD;
2182         break;
2183     default:
2184         error_report("Requested CPU state is not a valid S390 CPU state: %u",
2185                      cpu_state);
2186         exit(1);
2187     }
2188
2189     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2190     if (ret) {
2191         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
2192                                        strerror(-ret));
2193     }
2194
2195     return ret;
2196 }
2197
2198 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
2199 {
2200     struct kvm_s390_irq_state irq_state;
2201     CPUState *cs = CPU(cpu);
2202     int32_t bytes;
2203
2204     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2205         return;
2206     }
2207
2208     irq_state.buf = (uint64_t) cpu->irqstate;
2209     irq_state.len = VCPU_IRQ_BUF_SIZE;
2210
2211     bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
2212     if (bytes < 0) {
2213         cpu->irqstate_saved_size = 0;
2214         error_report("Migration of interrupt state failed");
2215         return;
2216     }
2217
2218     cpu->irqstate_saved_size = bytes;
2219 }
2220
2221 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
2222 {
2223     CPUState *cs = CPU(cpu);
2224     struct kvm_s390_irq_state irq_state;
2225     int r;
2226
2227     if (cpu->irqstate_saved_size == 0) {
2228         return 0;
2229     }
2230
2231     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2232         return -ENOSYS;
2233     }
2234
2235     irq_state.buf = (uint64_t) cpu->irqstate;
2236     irq_state.len = cpu->irqstate_saved_size;
2237
2238     r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
2239     if (r) {
2240         error_report("Setting interrupt state failed %d", r);
2241     }
2242     return r;
2243 }
2244
2245 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2246                              uint64_t address, uint32_t data, PCIDevice *dev)
2247 {
2248     S390PCIBusDevice *pbdev;
2249     uint32_t fid = data >> ZPCI_MSI_VEC_BITS;
2250     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
2251
2252     pbdev = s390_pci_find_dev_by_fid(fid);
2253     if (!pbdev) {
2254         DPRINTF("add_msi_route no dev\n");
2255         return -ENODEV;
2256     }
2257
2258     pbdev->routes.adapter.ind_offset = vec;
2259
2260     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2261     route->flags = 0;
2262     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2263     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2264     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2265     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset;
2266     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2267     return 0;
2268 }
2269
2270 int kvm_arch_msi_data_to_gsi(uint32_t data)
2271 {
2272     abort();
2273 }