target-s390x/kvm.c

   1 /*
   2  * QEMU S390x KVM implementation
   3  *
   4  * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
   5  * Copyright IBM Corp. 2012
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * Contributions after 2012-10-29 are licensed under the terms of the
  18  * GNU GPL, version 2 or (at your option) any later version.
  19  *
  20  * You should have received a copy of the GNU (Lesser) General Public
  21  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  22  */
  23
  24 #include "qemu/osdep.h"
  25 #include <sys/ioctl.h>
  26
  27 #include <linux/kvm.h>
  28 #include <asm/ptrace.h>
  29
  30 #include "qemu-common.h"
  31 #include "cpu.h"
  32 #include "qemu/error-report.h"
  33 #include "qemu/timer.h"
  34 #include "sysemu/sysemu.h"
  35 #include "sysemu/kvm.h"
  36 #include "hw/hw.h"
  37 #include "sysemu/device_tree.h"
  38 #include "qapi/qmp/qjson.h"
  39 #include "exec/gdbstub.h"
  40 #include "exec/address-spaces.h"
  41 #include "trace.h"
  42 #include "qapi-event.h"
  43 #include "hw/s390x/s390-pci-inst.h"
  44 #include "hw/s390x/s390-pci-bus.h"
  45 #include "hw/s390x/ipl.h"
  46 #include "hw/s390x/ebcdic.h"
  47 #include "exec/memattrs.h"
  48 #include "hw/s390x/s390-virtio-ccw.h"
  49
  50 /* #define DEBUG_KVM */
  51
  52 #ifdef DEBUG_KVM
  53 #define DPRINTF(fmt, ...) \
  54     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  55 #else
  56 #define DPRINTF(fmt, ...) \
  57     do { } while (0)
  58 #endif
  59
  60 #define kvm_vm_check_mem_attr(s, attr) \
  61     kvm_vm_check_attr(s, KVM_S390_VM_MEM_CTRL, attr)
  62
  63 #define IPA0_DIAG                       0x8300
  64 #define IPA0_SIGP                       0xae00
  65 #define IPA0_B2                         0xb200
  66 #define IPA0_B9                         0xb900
  67 #define IPA0_EB                         0xeb00
  68 #define IPA0_E3                         0xe300
  69
  70 #define PRIV_B2_SCLP_CALL               0x20
  71 #define PRIV_B2_CSCH                    0x30
  72 #define PRIV_B2_HSCH                    0x31
  73 #define PRIV_B2_MSCH                    0x32
  74 #define PRIV_B2_SSCH                    0x33
  75 #define PRIV_B2_STSCH                   0x34
  76 #define PRIV_B2_TSCH                    0x35
  77 #define PRIV_B2_TPI                     0x36
  78 #define PRIV_B2_SAL                     0x37
  79 #define PRIV_B2_RSCH                    0x38
  80 #define PRIV_B2_STCRW                   0x39
  81 #define PRIV_B2_STCPS                   0x3a
  82 #define PRIV_B2_RCHP                    0x3b
  83 #define PRIV_B2_SCHM                    0x3c
  84 #define PRIV_B2_CHSC                    0x5f
  85 #define PRIV_B2_SIGA                    0x74
  86 #define PRIV_B2_XSCH                    0x76
  87
  88 #define PRIV_EB_SQBS                    0x8a
  89 #define PRIV_EB_PCISTB                  0xd0
  90 #define PRIV_EB_SIC                     0xd1
  91
  92 #define PRIV_B9_EQBS                    0x9c
  93 #define PRIV_B9_CLP                     0xa0
  94 #define PRIV_B9_PCISTG                  0xd0
  95 #define PRIV_B9_PCILG                   0xd2
  96 #define PRIV_B9_RPCIT                   0xd3
  97
  98 #define PRIV_E3_MPCIFC                  0xd0
  99 #define PRIV_E3_STPCIFC                 0xd4
 100
 101 #define DIAG_TIMEREVENT                 0x288
 102 #define DIAG_IPL                        0x308
 103 #define DIAG_KVM_HYPERCALL              0x500
 104 #define DIAG_KVM_BREAKPOINT             0x501
 105
 106 #define ICPT_INSTRUCTION                0x04
 107 #define ICPT_PROGRAM                    0x08
 108 #define ICPT_EXT_INT                    0x14
 109 #define ICPT_WAITPSW                    0x1c
 110 #define ICPT_SOFT_INTERCEPT             0x24
 111 #define ICPT_CPU_STOP                   0x28
 112 #define ICPT_OPEREXC                    0x2c
 113 #define ICPT_IO                         0x40
 114
 115 #define NR_LOCAL_IRQS 32
 116 /*
 117  * Needs to be big enough to contain max_cpus emergency signals
 118  * and in addition NR_LOCAL_IRQS interrupts
 119  */
 120 #define VCPU_IRQ_BUF_SIZE (sizeof(struct kvm_s390_irq) * \
 121                            (max_cpus + NR_LOCAL_IRQS))
 122
 123 static CPUWatchpoint hw_watchpoint;
 124 /*
 125  * We don't use a list because this structure is also used to transmit the
 126  * hardware breakpoints to the kernel.
 127  */
 128 static struct kvm_hw_breakpoint *hw_breakpoints;
 129 static int nb_hw_breakpoints;
 130
 131 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
 132     KVM_CAP_LAST_INFO
 133 };
 134
 135 static QemuMutex qemu_sigp_mutex;
 136
 137 static int cap_sync_regs;
 138 static int cap_async_pf;
 139 static int cap_mem_op;
 140 static int cap_s390_irq;
 141 static int cap_ri;
 142
 143 static void *legacy_s390_alloc(size_t size, uint64_t *align);
 144
 145 static int kvm_s390_query_mem_limit(KVMState *s, uint64_t *memory_limit)
 146 {
 147     struct kvm_device_attr attr = {
 148         .group = KVM_S390_VM_MEM_CTRL,
 149         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
 150         .addr = (uint64_t) memory_limit,
 151     };
 152
 153     return kvm_vm_ioctl(s, KVM_GET_DEVICE_ATTR, &attr);
 154 }
 155
 156 int kvm_s390_set_mem_limit(KVMState *s, uint64_t new_limit, uint64_t *hw_limit)
 157 {
 158     int rc;
 159
 160     struct kvm_device_attr attr = {
 161         .group = KVM_S390_VM_MEM_CTRL,
 162         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
 163         .addr = (uint64_t) &new_limit,
 164     };
 165
 166     if (!kvm_vm_check_mem_attr(s, KVM_S390_VM_MEM_LIMIT_SIZE)) {
 167         return 0;
 168     }
 169
 170     rc = kvm_s390_query_mem_limit(s, hw_limit);
 171     if (rc) {
 172         return rc;
 173     } else if (*hw_limit < new_limit) {
 174         return -E2BIG;
 175     }
 176
 177     return kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 178 }
 179
 180 static bool kvm_s390_cmma_available(void)
 181 {
 182     static bool initialized, value;
 183
 184     if (!initialized) {
 185         initialized = true;
 186         value = kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_ENABLE_CMMA) &&
 187                 kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_CLR_CMMA);
 188     }
 189     return value;
 190 }
 191
 192 void kvm_s390_cmma_reset(void)
 193 {
 194     int rc;
 195     struct kvm_device_attr attr = {
 196         .group = KVM_S390_VM_MEM_CTRL,
 197         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 198     };
 199
 200     if (!mem_path || !kvm_s390_cmma_available()) {
 201         return;
 202     }
 203
 204     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 205     trace_kvm_clear_cmma(rc);
 206 }
 207
 208 static void kvm_s390_enable_cmma(void)
 209 {
 210     int rc;
 211     struct kvm_device_attr attr = {
 212         .group = KVM_S390_VM_MEM_CTRL,
 213         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 214     };
 215
 216     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 217     trace_kvm_enable_cmma(rc);
 218 }
 219
 220 static void kvm_s390_set_attr(uint64_t attr)
 221 {
 222     struct kvm_device_attr attribute = {
 223         .group = KVM_S390_VM_CRYPTO,
 224         .attr  = attr,
 225     };
 226
 227     int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
 228
 229     if (ret) {
 230         error_report("Failed to set crypto device attribute %lu: %s",
 231                      attr, strerror(-ret));
 232     }
 233 }
 234
 235 static void kvm_s390_init_aes_kw(void)
 236 {
 237     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
 238
 239     if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
 240                                  NULL)) {
 241             attr = KVM_S390_VM_CRYPTO_ENABLE_AES_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 static void kvm_s390_init_dea_kw(void)
 250 {
 251     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
 252
 253     if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
 254                                  NULL)) {
 255             attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
 256     }
 257
 258     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
 259             kvm_s390_set_attr(attr);
 260     }
 261 }
 262
 263 void kvm_s390_crypto_reset(void)
 264 {
 265     if (s390_has_feat(S390_FEAT_MSA_EXT_3)) {
 266         kvm_s390_init_aes_kw();
 267         kvm_s390_init_dea_kw();
 268     }
 269 }
 270
 271 int kvm_arch_init(MachineState *ms, KVMState *s)
 272 {
 273     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
 274     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
 275     cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
 276     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
 277
 278     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
 279         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
 280         phys_mem_set_alloc(legacy_s390_alloc);
 281     }
 282
 283     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
 284     kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
 285     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_STSI, 0);
 286     if (ri_allowed()) {
 287         if (kvm_vm_enable_cap(s, KVM_CAP_S390_RI, 0) == 0) {
 288             cap_ri = 1;
 289         }
 290     }
 291
 292     qemu_mutex_init(&qemu_sigp_mutex);
 293
 294     return 0;
 295 }
 296
 297 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
 298 {
 299     return cpu->cpu_index;
 300 }
 301
 302 int kvm_arch_init_vcpu(CPUState *cs)
 303 {
 304     S390CPU *cpu = S390_CPU(cs);
 305     kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
 306     cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE);
 307     return 0;
 308 }
 309
 310 void kvm_s390_reset_vcpu(S390CPU *cpu)
 311 {
 312     CPUState *cs = CPU(cpu);
 313
 314     /* The initial reset call is needed here to reset in-kernel
 315      * vcpu data that we can't access directly from QEMU
 316      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
 317      * Before this ioctl cpu_synchronize_state() is called in common kvm
 318      * code (kvm-all) */
 319     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
 320         error_report("Initial CPU reset failed on CPU %i", cs->cpu_index);
 321     }
 322 }
 323
 324 static int can_sync_regs(CPUState *cs, int regs)
 325 {
 326     return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
 327 }
 328
 329 int kvm_arch_put_registers(CPUState *cs, int level)
 330 {
 331     S390CPU *cpu = S390_CPU(cs);
 332     CPUS390XState *env = &cpu->env;
 333     struct kvm_sregs sregs;
 334     struct kvm_regs regs;
 335     struct kvm_fpu fpu = {};
 336     int r;
 337     int i;
 338
 339     /* always save the PSW  and the GPRS*/
 340     cs->kvm_run->psw_addr = env->psw.addr;
 341     cs->kvm_run->psw_mask = env->psw.mask;
 342
 343     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 344         for (i = 0; i < 16; i++) {
 345             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
 346             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
 347         }
 348     } else {
 349         for (i = 0; i < 16; i++) {
 350             regs.gprs[i] = env->regs[i];
 351         }
 352         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
 353         if (r < 0) {
 354             return r;
 355         }
 356     }
 357
 358     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 359         for (i = 0; i < 32; i++) {
 360             cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
 361             cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
 362         }
 363         cs->kvm_run->s.regs.fpc = env->fpc;
 364         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
 365     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
 366         for (i = 0; i < 16; i++) {
 367             cs->kvm_run->s.regs.fprs[i] = get_freg(env, i)->ll;
 368         }
 369         cs->kvm_run->s.regs.fpc = env->fpc;
 370         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_FPRS;
 371     } else {
 372         /* Floating point */
 373         for (i = 0; i < 16; i++) {
 374             fpu.fprs[i] = get_freg(env, i)->ll;
 375         }
 376         fpu.fpc = env->fpc;
 377
 378         r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
 379         if (r < 0) {
 380             return r;
 381         }
 382     }
 383
 384     /* Do we need to save more than that? */
 385     if (level == KVM_PUT_RUNTIME_STATE) {
 386         return 0;
 387     }
 388
 389     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 390         cs->kvm_run->s.regs.cputm = env->cputm;
 391         cs->kvm_run->s.regs.ckc = env->ckc;
 392         cs->kvm_run->s.regs.todpr = env->todpr;
 393         cs->kvm_run->s.regs.gbea = env->gbea;
 394         cs->kvm_run->s.regs.pp = env->pp;
 395         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
 396     } else {
 397         /*
 398          * These ONE_REGS are not protected by a capability. As they are only
 399          * necessary for migration we just trace a possible error, but don't
 400          * return with an error return code.
 401          */
 402         kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 403         kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 404         kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 405         kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 406         kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 407     }
 408
 409     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
 410         memcpy(cs->kvm_run->s.regs.riccb, env->riccb, 64);
 411         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_RICCB;
 412     }
 413
 414     /* pfault parameters */
 415     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 416         cs->kvm_run->s.regs.pft = env->pfault_token;
 417         cs->kvm_run->s.regs.pfs = env->pfault_select;
 418         cs->kvm_run->s.regs.pfc = env->pfault_compare;
 419         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
 420     } else if (cap_async_pf) {
 421         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 422         if (r < 0) {
 423             return r;
 424         }
 425         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 426         if (r < 0) {
 427             return r;
 428         }
 429         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 430         if (r < 0) {
 431             return r;
 432         }
 433     }
 434
 435     /* access registers and control registers*/
 436     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 437         for (i = 0; i < 16; i++) {
 438             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
 439             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
 440         }
 441         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
 442         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
 443     } else {
 444         for (i = 0; i < 16; i++) {
 445             sregs.acrs[i] = env->aregs[i];
 446             sregs.crs[i] = env->cregs[i];
 447         }
 448         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
 449         if (r < 0) {
 450             return r;
 451         }
 452     }
 453
 454     /* Finally the prefix */
 455     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 456         cs->kvm_run->s.regs.prefix = env->psa;
 457         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
 458     } else {
 459         /* prefix is only supported via sync regs */
 460     }
 461     return 0;
 462 }
 463
 464 int kvm_arch_get_registers(CPUState *cs)
 465 {
 466     S390CPU *cpu = S390_CPU(cs);
 467     CPUS390XState *env = &cpu->env;
 468     struct kvm_sregs sregs;
 469     struct kvm_regs regs;
 470     struct kvm_fpu fpu;
 471     int i, r;
 472
 473     /* get the PSW */
 474     env->psw.addr = cs->kvm_run->psw_addr;
 475     env->psw.mask = cs->kvm_run->psw_mask;
 476
 477     /* the GPRS */
 478     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
 479         for (i = 0; i < 16; i++) {
 480             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
 481         }
 482     } else {
 483         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
 484         if (r < 0) {
 485             return r;
 486         }
 487          for (i = 0; i < 16; i++) {
 488             env->regs[i] = regs.gprs[i];
 489         }
 490     }
 491
 492     /* The ACRS and CRS */
 493     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
 494         for (i = 0; i < 16; i++) {
 495             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
 496             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
 497         }
 498     } else {
 499         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
 500         if (r < 0) {
 501             return r;
 502         }
 503          for (i = 0; i < 16; i++) {
 504             env->aregs[i] = sregs.acrs[i];
 505             env->cregs[i] = sregs.crs[i];
 506         }
 507     }
 508
 509     /* Floating point and vector registers */
 510     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
 511         for (i = 0; i < 32; i++) {
 512             env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
 513             env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
 514         }
 515         env->fpc = cs->kvm_run->s.regs.fpc;
 516     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
 517         for (i = 0; i < 16; i++) {
 518             get_freg(env, i)->ll = cs->kvm_run->s.regs.fprs[i];
 519         }
 520         env->fpc = cs->kvm_run->s.regs.fpc;
 521     } else {
 522         r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
 523         if (r < 0) {
 524             return r;
 525         }
 526         for (i = 0; i < 16; i++) {
 527             get_freg(env, i)->ll = fpu.fprs[i];
 528         }
 529         env->fpc = fpu.fpc;
 530     }
 531
 532     /* The prefix */
 533     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
 534         env->psa = cs->kvm_run->s.regs.prefix;
 535     }
 536
 537     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
 538         env->cputm = cs->kvm_run->s.regs.cputm;
 539         env->ckc = cs->kvm_run->s.regs.ckc;
 540         env->todpr = cs->kvm_run->s.regs.todpr;
 541         env->gbea = cs->kvm_run->s.regs.gbea;
 542         env->pp = cs->kvm_run->s.regs.pp;
 543     } else {
 544         /*
 545          * These ONE_REGS are not protected by a capability. As they are only
 546          * necessary for migration we just trace a possible error, but don't
 547          * return with an error return code.
 548          */
 549         kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 550         kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 551         kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 552         kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 553         kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 554     }
 555
 556     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
 557         memcpy(env->riccb, cs->kvm_run->s.regs.riccb, 64);
 558     }
 559
 560     /* pfault parameters */
 561     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
 562         env->pfault_token = cs->kvm_run->s.regs.pft;
 563         env->pfault_select = cs->kvm_run->s.regs.pfs;
 564         env->pfault_compare = cs->kvm_run->s.regs.pfc;
 565     } else if (cap_async_pf) {
 566         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 567         if (r < 0) {
 568             return r;
 569         }
 570         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 571         if (r < 0) {
 572             return r;
 573         }
 574         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 575         if (r < 0) {
 576             return r;
 577         }
 578     }
 579
 580     return 0;
 581 }
 582
 583 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
 584 {
 585     int r;
 586     struct kvm_device_attr attr = {
 587         .group = KVM_S390_VM_TOD,
 588         .attr = KVM_S390_VM_TOD_LOW,
 589         .addr = (uint64_t)tod_low,
 590     };
 591
 592     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 593     if (r) {
 594         return r;
 595     }
 596
 597     attr.attr = KVM_S390_VM_TOD_HIGH;
 598     attr.addr = (uint64_t)tod_high;
 599     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
 600 }
 601
 602 int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_low)
 603 {
 604     int r;
 605
 606     struct kvm_device_attr attr = {
 607         .group = KVM_S390_VM_TOD,
 608         .attr = KVM_S390_VM_TOD_LOW,
 609         .addr = (uint64_t)tod_low,
 610     };
 611
 612     r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 613     if (r) {
 614         return r;
 615     }
 616
 617     attr.attr = KVM_S390_VM_TOD_HIGH;
 618     attr.addr = (uint64_t)tod_high;
 619     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 620 }
 621
 622 /**
 623  * kvm_s390_mem_op:
 624  * @addr:      the logical start address in guest memory
 625  * @ar:        the access register number
 626  * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
 627  * @len:       length that should be transferred
 628  * @is_write:  true = write, false = read
 629  * Returns:    0 on success, non-zero if an exception or error occurred
 630  *
 631  * Use KVM ioctl to read/write from/to guest memory. An access exception
 632  * is injected into the vCPU in case of translation errors.
 633  */
 634 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
 635                     int len, bool is_write)
 636 {
 637     struct kvm_s390_mem_op mem_op = {
 638         .gaddr = addr,
 639         .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
 640         .size = len,
 641         .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
 642                        : KVM_S390_MEMOP_LOGICAL_READ,
 643         .buf = (uint64_t)hostbuf,
 644         .ar = ar,
 645     };
 646     int ret;
 647
 648     if (!cap_mem_op) {
 649         return -ENOSYS;
 650     }
 651     if (!hostbuf) {
 652         mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
 653     }
 654
 655     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
 656     if (ret < 0) {
 657         error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
 658     }
 659     return ret;
 660 }
 661
 662 /*
 663  * Legacy layout for s390:
 664  * Older S390 KVM requires the topmost vma of the RAM to be
 665  * smaller than an system defined value, which is at least 256GB.
 666  * Larger systems have larger values. We put the guest between
 667  * the end of data segment (system break) and this value. We
 668  * use 32GB as a base to have enough room for the system break
 669  * to grow. We also have to use MAP parameters that avoid
 670  * read-only mapping of guest pages.
 671  */
 672 static void *legacy_s390_alloc(size_t size, uint64_t *align)
 673 {
 674     void *mem;
 675
 676     mem = mmap((void *) 0x800000000ULL, size,
 677                PROT_EXEC|PROT_READ|PROT_WRITE,
 678                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 679     return mem == MAP_FAILED ? NULL : mem;
 680 }
 681
 682 static uint8_t const *sw_bp_inst;
 683 static uint8_t sw_bp_ilen;
 684
 685 static void determine_sw_breakpoint_instr(void)
 686 {
 687         /* DIAG 501 is used for sw breakpoints with old kernels */
 688         static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 689         /* Instruction 0x0000 is used for sw breakpoints with recent kernels */
 690         static const uint8_t instr_0x0000[] = {0x00, 0x00};
 691
 692         if (sw_bp_inst) {
 693             return;
 694         }
 695         if (kvm_vm_enable_cap(kvm_state, KVM_CAP_S390_USER_INSTR0, 0)) {
 696             sw_bp_inst = diag_501;
 697             sw_bp_ilen = sizeof(diag_501);
 698             DPRINTF("KVM: will use 4-byte sw breakpoints.\n");
 699         } else {
 700             sw_bp_inst = instr_0x0000;
 701             sw_bp_ilen = sizeof(instr_0x0000);
 702             DPRINTF("KVM: will use 2-byte sw breakpoints.\n");
 703         }
 704 }
 705
 706 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 707 {
 708     determine_sw_breakpoint_instr();
 709
 710     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 711                             sw_bp_ilen, 0) ||
 712         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)sw_bp_inst, sw_bp_ilen, 1)) {
 713         return -EINVAL;
 714     }
 715     return 0;
 716 }
 717
 718 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 719 {
 720     uint8_t t[MAX_ILEN];
 721
 722     if (cpu_memory_rw_debug(cs, bp->pc, t, sw_bp_ilen, 0)) {
 723         return -EINVAL;
 724     } else if (memcmp(t, sw_bp_inst, sw_bp_ilen)) {
 725         return -EINVAL;
 726     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 727                                    sw_bp_ilen, 1)) {
 728         return -EINVAL;
 729     }
 730
 731     return 0;
 732 }
 733
 734 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
 735                                                     int len, int type)
 736 {
 737     int n;
 738
 739     for (n = 0; n < nb_hw_breakpoints; n++) {
 740         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
 741             (hw_breakpoints[n].len == len || len == -1)) {
 742             return &hw_breakpoints[n];
 743         }
 744     }
 745
 746     return NULL;
 747 }
 748
 749 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
 750 {
 751     int size;
 752
 753     if (find_hw_breakpoint(addr, len, type)) {
 754         return -EEXIST;
 755     }
 756
 757     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
 758
 759     if (!hw_breakpoints) {
 760         nb_hw_breakpoints = 0;
 761         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
 762     } else {
 763         hw_breakpoints =
 764             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
 765     }
 766
 767     if (!hw_breakpoints) {
 768         nb_hw_breakpoints = 0;
 769         return -ENOMEM;
 770     }
 771
 772     hw_breakpoints[nb_hw_breakpoints].addr = addr;
 773     hw_breakpoints[nb_hw_breakpoints].len = len;
 774     hw_breakpoints[nb_hw_breakpoints].type = type;
 775
 776     nb_hw_breakpoints++;
 777
 778     return 0;
 779 }
 780
 781 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 782                                   target_ulong len, int type)
 783 {
 784     switch (type) {
 785     case GDB_BREAKPOINT_HW:
 786         type = KVM_HW_BP;
 787         break;
 788     case GDB_WATCHPOINT_WRITE:
 789         if (len < 1) {
 790             return -EINVAL;
 791         }
 792         type = KVM_HW_WP_WRITE;
 793         break;
 794     default:
 795         return -ENOSYS;
 796     }
 797     return insert_hw_breakpoint(addr, len, type);
 798 }
 799
 800 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 801                                   target_ulong len, int type)
 802 {
 803     int size;
 804     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
 805
 806     if (bp == NULL) {
 807         return -ENOENT;
 808     }
 809
 810     nb_hw_breakpoints--;
 811     if (nb_hw_breakpoints > 0) {
 812         /*
 813          * In order to trim the array, move the last element to the position to
 814          * be removed - if necessary.
 815          */
 816         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
 817             *bp = hw_breakpoints[nb_hw_breakpoints];
 818         }
 819         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
 820         hw_breakpoints =
 821              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
 822     } else {
 823         g_free(hw_breakpoints);
 824         hw_breakpoints = NULL;
 825     }
 826
 827     return 0;
 828 }
 829
 830 void kvm_arch_remove_all_hw_breakpoints(void)
 831 {
 832     nb_hw_breakpoints = 0;
 833     g_free(hw_breakpoints);
 834     hw_breakpoints = NULL;
 835 }
 836
 837 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 838 {
 839     int i;
 840
 841     if (nb_hw_breakpoints > 0) {
 842         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
 843         dbg->arch.hw_bp = hw_breakpoints;
 844
 845         for (i = 0; i < nb_hw_breakpoints; ++i) {
 846             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
 847                                                        hw_breakpoints[i].addr);
 848         }
 849         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 850     } else {
 851         dbg->arch.nr_hw_bp = 0;
 852         dbg->arch.hw_bp = NULL;
 853     }
 854 }
 855
 856 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 857 {
 858 }
 859
 860 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
 861 {
 862     return MEMTXATTRS_UNSPECIFIED;
 863 }
 864
 865 int kvm_arch_process_async_events(CPUState *cs)
 866 {
 867     return cs->halted;
 868 }
 869
 870 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
 871                                      struct kvm_s390_interrupt *interrupt)
 872 {
 873     int r = 0;
 874
 875     interrupt->type = irq->type;
 876     switch (irq->type) {
 877     case KVM_S390_INT_VIRTIO:
 878         interrupt->parm = irq->u.ext.ext_params;
 879         /* fall through */
 880     case KVM_S390_INT_PFAULT_INIT:
 881     case KVM_S390_INT_PFAULT_DONE:
 882         interrupt->parm64 = irq->u.ext.ext_params2;
 883         break;
 884     case KVM_S390_PROGRAM_INT:
 885         interrupt->parm = irq->u.pgm.code;
 886         break;
 887     case KVM_S390_SIGP_SET_PREFIX:
 888         interrupt->parm = irq->u.prefix.address;
 889         break;
 890     case KVM_S390_INT_SERVICE:
 891         interrupt->parm = irq->u.ext.ext_params;
 892         break;
 893     case KVM_S390_MCHK:
 894         interrupt->parm = irq->u.mchk.cr14;
 895         interrupt->parm64 = irq->u.mchk.mcic;
 896         break;
 897     case KVM_S390_INT_EXTERNAL_CALL:
 898         interrupt->parm = irq->u.extcall.code;
 899         break;
 900     case KVM_S390_INT_EMERGENCY:
 901         interrupt->parm = irq->u.emerg.code;
 902         break;
 903     case KVM_S390_SIGP_STOP:
 904     case KVM_S390_RESTART:
 905         break; /* These types have no parameters */
 906     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
 907         interrupt->parm = irq->u.io.subchannel_id << 16;
 908         interrupt->parm |= irq->u.io.subchannel_nr;
 909         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
 910         interrupt->parm64 |= irq->u.io.io_int_word;
 911         break;
 912     default:
 913         r = -EINVAL;
 914         break;
 915     }
 916     return r;
 917 }
 918
 919 static void inject_vcpu_irq_legacy(CPUState *cs, 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_vcpu_ioctl(cs, 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_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
 938 {
 939     CPUState *cs = CPU(cpu);
 940     int r;
 941
 942     if (cap_s390_irq) {
 943         r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
 944         if (!r) {
 945             return;
 946         }
 947         error_report("KVM failed to inject interrupt %llx", irq->type);
 948         exit(1);
 949     }
 950
 951     inject_vcpu_irq_legacy(cs, irq);
 952 }
 953
 954 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 955 {
 956     struct kvm_s390_interrupt kvmint = {};
 957     int r;
 958
 959     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 960     if (r < 0) {
 961         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 962         exit(1);
 963     }
 964
 965     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
 966     if (r < 0) {
 967         fprintf(stderr, "KVM failed to inject interrupt\n");
 968         exit(1);
 969     }
 970 }
 971
 972 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 973 {
 974     static bool use_flic = true;
 975     int r;
 976
 977     if (use_flic) {
 978         r = kvm_s390_inject_flic(irq);
 979         if (r == -ENOSYS) {
 980             use_flic = false;
 981         }
 982         if (!r) {
 983             return;
 984         }
 985     }
 986     __kvm_s390_floating_interrupt(irq);
 987 }
 988
 989 void kvm_s390_service_interrupt(uint32_t parm)
 990 {
 991     struct kvm_s390_irq irq = {
 992         .type = KVM_S390_INT_SERVICE,
 993         .u.ext.ext_params = parm,
 994     };
 995
 996     kvm_s390_floating_interrupt(&irq);
 997 }
 998
 999 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
1000 {
1001     struct kvm_s390_irq irq = {
1002         .type = KVM_S390_PROGRAM_INT,
1003         .u.pgm.code = code,
1004     };
1005
1006     kvm_s390_vcpu_interrupt(cpu, &irq);
1007 }
1008
1009 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
1010 {
1011     struct kvm_s390_irq irq = {
1012         .type = KVM_S390_PROGRAM_INT,
1013         .u.pgm.code = code,
1014         .u.pgm.trans_exc_code = te_code,
1015         .u.pgm.exc_access_id = te_code & 3,
1016     };
1017
1018     kvm_s390_vcpu_interrupt(cpu, &irq);
1019 }
1020
1021 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
1022                                  uint16_t ipbh0)
1023 {
1024     CPUS390XState *env = &cpu->env;
1025     uint64_t sccb;
1026     uint32_t code;
1027     int r = 0;
1028
1029     cpu_synchronize_state(CPU(cpu));
1030     sccb = env->regs[ipbh0 & 0xf];
1031     code = env->regs[(ipbh0 & 0xf0) >> 4];
1032
1033     r = sclp_service_call(env, sccb, code);
1034     if (r < 0) {
1035         enter_pgmcheck(cpu, -r);
1036     } else {
1037         setcc(cpu, r);
1038     }
1039
1040     return 0;
1041 }
1042
1043 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1044 {
1045     CPUS390XState *env = &cpu->env;
1046     int rc = 0;
1047     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1048
1049     cpu_synchronize_state(CPU(cpu));
1050
1051     switch (ipa1) {
1052     case PRIV_B2_XSCH:
1053         ioinst_handle_xsch(cpu, env->regs[1]);
1054         break;
1055     case PRIV_B2_CSCH:
1056         ioinst_handle_csch(cpu, env->regs[1]);
1057         break;
1058     case PRIV_B2_HSCH:
1059         ioinst_handle_hsch(cpu, env->regs[1]);
1060         break;
1061     case PRIV_B2_MSCH:
1062         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
1063         break;
1064     case PRIV_B2_SSCH:
1065         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
1066         break;
1067     case PRIV_B2_STCRW:
1068         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
1069         break;
1070     case PRIV_B2_STSCH:
1071         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
1072         break;
1073     case PRIV_B2_TSCH:
1074         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1075         fprintf(stderr, "Spurious tsch intercept\n");
1076         break;
1077     case PRIV_B2_CHSC:
1078         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
1079         break;
1080     case PRIV_B2_TPI:
1081         /* This should have been handled by kvm already. */
1082         fprintf(stderr, "Spurious tpi intercept\n");
1083         break;
1084     case PRIV_B2_SCHM:
1085         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1086                            run->s390_sieic.ipb);
1087         break;
1088     case PRIV_B2_RSCH:
1089         ioinst_handle_rsch(cpu, env->regs[1]);
1090         break;
1091     case PRIV_B2_RCHP:
1092         ioinst_handle_rchp(cpu, env->regs[1]);
1093         break;
1094     case PRIV_B2_STCPS:
1095         /* We do not provide this instruction, it is suppressed. */
1096         break;
1097     case PRIV_B2_SAL:
1098         ioinst_handle_sal(cpu, env->regs[1]);
1099         break;
1100     case PRIV_B2_SIGA:
1101         /* Not provided, set CC = 3 for subchannel not operational */
1102         setcc(cpu, 3);
1103         break;
1104     case PRIV_B2_SCLP_CALL:
1105         rc = kvm_sclp_service_call(cpu, run, ipbh0);
1106         break;
1107     default:
1108         rc = -1;
1109         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1110         break;
1111     }
1112
1113     return rc;
1114 }
1115
1116 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1117                                   uint8_t *ar)
1118 {
1119     CPUS390XState *env = &cpu->env;
1120     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1121     uint32_t base2 = run->s390_sieic.ipb >> 28;
1122     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1123                      ((run->s390_sieic.ipb & 0xff00) << 4);
1124
1125     if (disp2 & 0x80000) {
1126         disp2 += 0xfff00000;
1127     }
1128     if (ar) {
1129         *ar = base2;
1130     }
1131
1132     return (base2 ? env->regs[base2] : 0) +
1133            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1134 }
1135
1136 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1137                                   uint8_t *ar)
1138 {
1139     CPUS390XState *env = &cpu->env;
1140     uint32_t base2 = run->s390_sieic.ipb >> 28;
1141     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1142                      ((run->s390_sieic.ipb & 0xff00) << 4);
1143
1144     if (disp2 & 0x80000) {
1145         disp2 += 0xfff00000;
1146     }
1147     if (ar) {
1148         *ar = base2;
1149     }
1150
1151     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1152 }
1153
1154 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1155 {
1156     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1157
1158     return clp_service_call(cpu, r2);
1159 }
1160
1161 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1162 {
1163     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1164     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1165
1166     return pcilg_service_call(cpu, r1, r2);
1167 }
1168
1169 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1170 {
1171     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1172     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1173
1174     return pcistg_service_call(cpu, r1, r2);
1175 }
1176
1177 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1178 {
1179     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1180     uint64_t fiba;
1181     uint8_t ar;
1182
1183     cpu_synchronize_state(CPU(cpu));
1184     fiba = get_base_disp_rxy(cpu, run, &ar);
1185
1186     return stpcifc_service_call(cpu, r1, fiba, ar);
1187 }
1188
1189 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1190 {
1191     /* NOOP */
1192     return 0;
1193 }
1194
1195 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1196 {
1197     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1198     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1199
1200     return rpcit_service_call(cpu, r1, r2);
1201 }
1202
1203 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1204 {
1205     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1206     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1207     uint64_t gaddr;
1208     uint8_t ar;
1209
1210     cpu_synchronize_state(CPU(cpu));
1211     gaddr = get_base_disp_rsy(cpu, run, &ar);
1212
1213     return pcistb_service_call(cpu, r1, r3, gaddr, ar);
1214 }
1215
1216 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1217 {
1218     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1219     uint64_t fiba;
1220     uint8_t ar;
1221
1222     cpu_synchronize_state(CPU(cpu));
1223     fiba = get_base_disp_rxy(cpu, run, &ar);
1224
1225     return mpcifc_service_call(cpu, r1, fiba, ar);
1226 }
1227
1228 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1229 {
1230     int r = 0;
1231
1232     switch (ipa1) {
1233     case PRIV_B9_CLP:
1234         r = kvm_clp_service_call(cpu, run);
1235         break;
1236     case PRIV_B9_PCISTG:
1237         r = kvm_pcistg_service_call(cpu, run);
1238         break;
1239     case PRIV_B9_PCILG:
1240         r = kvm_pcilg_service_call(cpu, run);
1241         break;
1242     case PRIV_B9_RPCIT:
1243         r = kvm_rpcit_service_call(cpu, run);
1244         break;
1245     case PRIV_B9_EQBS:
1246         /* just inject exception */
1247         r = -1;
1248         break;
1249     default:
1250         r = -1;
1251         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1252         break;
1253     }
1254
1255     return r;
1256 }
1257
1258 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1259 {
1260     int r = 0;
1261
1262     switch (ipbl) {
1263     case PRIV_EB_PCISTB:
1264         r = kvm_pcistb_service_call(cpu, run);
1265         break;
1266     case PRIV_EB_SIC:
1267         r = kvm_sic_service_call(cpu, run);
1268         break;
1269     case PRIV_EB_SQBS:
1270         /* just inject exception */
1271         r = -1;
1272         break;
1273     default:
1274         r = -1;
1275         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1276         break;
1277     }
1278
1279     return r;
1280 }
1281
1282 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1283 {
1284     int r = 0;
1285
1286     switch (ipbl) {
1287     case PRIV_E3_MPCIFC:
1288         r = kvm_mpcifc_service_call(cpu, run);
1289         break;
1290     case PRIV_E3_STPCIFC:
1291         r = kvm_stpcifc_service_call(cpu, run);
1292         break;
1293     default:
1294         r = -1;
1295         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1296         break;
1297     }
1298
1299     return r;
1300 }
1301
1302 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1303 {
1304     CPUS390XState *env = &cpu->env;
1305     int ret;
1306
1307     cpu_synchronize_state(CPU(cpu));
1308     ret = s390_virtio_hypercall(env);
1309     if (ret == -EINVAL) {
1310         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1311         return 0;
1312     }
1313
1314     return ret;
1315 }
1316
1317 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1318 {
1319     uint64_t r1, r3;
1320     int rc;
1321
1322     cpu_synchronize_state(CPU(cpu));
1323     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1324     r3 = run->s390_sieic.ipa & 0x000f;
1325     rc = handle_diag_288(&cpu->env, r1, r3);
1326     if (rc) {
1327         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1328     }
1329 }
1330
1331 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1332 {
1333     uint64_t r1, r3;
1334
1335     cpu_synchronize_state(CPU(cpu));
1336     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1337     r3 = run->s390_sieic.ipa & 0x000f;
1338     handle_diag_308(&cpu->env, r1, r3);
1339 }
1340
1341 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1342 {
1343     CPUS390XState *env = &cpu->env;
1344     unsigned long pc;
1345
1346     cpu_synchronize_state(CPU(cpu));
1347
1348     pc = env->psw.addr - sw_bp_ilen;
1349     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1350         env->psw.addr = pc;
1351         return EXCP_DEBUG;
1352     }
1353
1354     return -ENOENT;
1355 }
1356
1357 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1358
1359 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1360 {
1361     int r = 0;
1362     uint16_t func_code;
1363
1364     /*
1365      * For any diagnose call we support, bits 48-63 of the resulting
1366      * address specify the function code; the remainder is ignored.
1367      */
1368     func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1369     switch (func_code) {
1370     case DIAG_TIMEREVENT:
1371         kvm_handle_diag_288(cpu, run);
1372         break;
1373     case DIAG_IPL:
1374         kvm_handle_diag_308(cpu, run);
1375         break;
1376     case DIAG_KVM_HYPERCALL:
1377         r = handle_hypercall(cpu, run);
1378         break;
1379     case DIAG_KVM_BREAKPOINT:
1380         r = handle_sw_breakpoint(cpu, run);
1381         break;
1382     default:
1383         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1384         enter_pgmcheck(cpu, PGM_SPECIFICATION);
1385         break;
1386     }
1387
1388     return r;
1389 }
1390
1391 typedef struct SigpInfo {
1392     S390CPU *cpu;
1393     uint64_t param;
1394     int cc;
1395     uint64_t *status_reg;
1396 } SigpInfo;
1397
1398 static void set_sigp_status(SigpInfo *si, uint64_t status)
1399 {
1400     *si->status_reg &= 0xffffffff00000000ULL;
1401     *si->status_reg |= status;
1402     si->cc = SIGP_CC_STATUS_STORED;
1403 }
1404
1405 static void sigp_start(void *arg)
1406 {
1407     SigpInfo *si = arg;
1408
1409     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1410         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1411         return;
1412     }
1413
1414     s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1415     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1416 }
1417
1418 static void sigp_stop(void *arg)
1419 {
1420     SigpInfo *si = arg;
1421     struct kvm_s390_irq irq = {
1422         .type = KVM_S390_SIGP_STOP,
1423     };
1424
1425     if (s390_cpu_get_state(si->cpu) != CPU_STATE_OPERATING) {
1426         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1427         return;
1428     }
1429
1430     /* disabled wait - sleeping in user space */
1431     if (CPU(si->cpu)->halted) {
1432         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1433     } else {
1434         /* execute the stop function */
1435         si->cpu->env.sigp_order = SIGP_STOP;
1436         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1437     }
1438     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1439 }
1440
1441 #define ADTL_SAVE_AREA_SIZE 1024
1442 static int kvm_s390_store_adtl_status(S390CPU *cpu, hwaddr addr)
1443 {
1444     void *mem;
1445     hwaddr len = ADTL_SAVE_AREA_SIZE;
1446
1447     mem = cpu_physical_memory_map(addr, &len, 1);
1448     if (!mem) {
1449         return -EFAULT;
1450     }
1451     if (len != ADTL_SAVE_AREA_SIZE) {
1452         cpu_physical_memory_unmap(mem, len, 1, 0);
1453         return -EFAULT;
1454     }
1455
1456     memcpy(mem, &cpu->env.vregs, 512);
1457
1458     cpu_physical_memory_unmap(mem, len, 1, len);
1459
1460     return 0;
1461 }
1462
1463 #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, floating_pt_save_area)
1464 #define SAVE_AREA_SIZE 512
1465 static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
1466 {
1467     static const uint8_t ar_id = 1;
1468     uint64_t ckc = cpu->env.ckc >> 8;
1469     void *mem;
1470     int i;
1471     hwaddr len = SAVE_AREA_SIZE;
1472
1473     mem = cpu_physical_memory_map(addr, &len, 1);
1474     if (!mem) {
1475         return -EFAULT;
1476     }
1477     if (len != SAVE_AREA_SIZE) {
1478         cpu_physical_memory_unmap(mem, len, 1, 0);
1479         return -EFAULT;
1480     }
1481
1482     if (store_arch) {
1483         cpu_physical_memory_write(offsetof(LowCore, ar_access_id), &ar_id, 1);
1484     }
1485     for (i = 0; i < 16; ++i) {
1486         *((uint64_t *)mem + i) = get_freg(&cpu->env, i)->ll;
1487     }
1488     memcpy(mem + 128, &cpu->env.regs, 128);
1489     memcpy(mem + 256, &cpu->env.psw, 16);
1490     memcpy(mem + 280, &cpu->env.psa, 4);
1491     memcpy(mem + 284, &cpu->env.fpc, 4);
1492     memcpy(mem + 292, &cpu->env.todpr, 4);
1493     memcpy(mem + 296, &cpu->env.cputm, 8);
1494     memcpy(mem + 304, &ckc, 8);
1495     memcpy(mem + 320, &cpu->env.aregs, 64);
1496     memcpy(mem + 384, &cpu->env.cregs, 128);
1497
1498     cpu_physical_memory_unmap(mem, len, 1, len);
1499
1500     return 0;
1501 }
1502
1503 static void sigp_stop_and_store_status(void *arg)
1504 {
1505     SigpInfo *si = arg;
1506     struct kvm_s390_irq irq = {
1507         .type = KVM_S390_SIGP_STOP,
1508     };
1509
1510     /* disabled wait - sleeping in user space */
1511     if (s390_cpu_get_state(si->cpu) == CPU_STATE_OPERATING &&
1512         CPU(si->cpu)->halted) {
1513         s390_cpu_set_state(CPU_STATE_STOPPED, si->cpu);
1514     }
1515
1516     switch (s390_cpu_get_state(si->cpu)) {
1517     case CPU_STATE_OPERATING:
1518         si->cpu->env.sigp_order = SIGP_STOP_STORE_STATUS;
1519         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1520         /* store will be performed when handling the stop intercept */
1521         break;
1522     case CPU_STATE_STOPPED:
1523         /* already stopped, just store the status */
1524         cpu_synchronize_state(CPU(si->cpu));
1525         kvm_s390_store_status(si->cpu, KVM_S390_STORE_STATUS_DEF_ADDR, true);
1526         break;
1527     }
1528     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1529 }
1530
1531 static void sigp_store_status_at_address(void *arg)
1532 {
1533     SigpInfo *si = arg;
1534     uint32_t address = si->param & 0x7ffffe00u;
1535
1536     /* cpu has to be stopped */
1537     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1538         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1539         return;
1540     }
1541
1542     cpu_synchronize_state(CPU(si->cpu));
1543
1544     if (kvm_s390_store_status(si->cpu, address, false)) {
1545         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1546         return;
1547     }
1548     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1549 }
1550
1551 static void sigp_store_adtl_status(void *arg)
1552 {
1553     SigpInfo *si = arg;
1554
1555     if (!s390_has_feat(S390_FEAT_VECTOR)) {
1556         set_sigp_status(si, SIGP_STAT_INVALID_ORDER);
1557         return;
1558     }
1559
1560     /* cpu has to be stopped */
1561     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1562         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1563         return;
1564     }
1565
1566     /* parameter must be aligned to 1024-byte boundary */
1567     if (si->param & 0x3ff) {
1568         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1569         return;
1570     }
1571
1572     cpu_synchronize_state(CPU(si->cpu));
1573
1574     if (kvm_s390_store_adtl_status(si->cpu, si->param)) {
1575         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1576         return;
1577     }
1578     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1579 }
1580
1581 static void sigp_restart(void *arg)
1582 {
1583     SigpInfo *si = arg;
1584     struct kvm_s390_irq irq = {
1585         .type = KVM_S390_RESTART,
1586     };
1587
1588     switch (s390_cpu_get_state(si->cpu)) {
1589     case CPU_STATE_STOPPED:
1590         /* the restart irq has to be delivered prior to any other pending irq */
1591         cpu_synchronize_state(CPU(si->cpu));
1592         do_restart_interrupt(&si->cpu->env);
1593         s390_cpu_set_state(CPU_STATE_OPERATING, si->cpu);
1594         break;
1595     case CPU_STATE_OPERATING:
1596         kvm_s390_vcpu_interrupt(si->cpu, &irq);
1597         break;
1598     }
1599     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1600 }
1601
1602 int kvm_s390_cpu_restart(S390CPU *cpu)
1603 {
1604     SigpInfo si = {
1605         .cpu = cpu,
1606     };
1607
1608     run_on_cpu(CPU(cpu), sigp_restart, &si);
1609     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1610     return 0;
1611 }
1612
1613 static void sigp_initial_cpu_reset(void *arg)
1614 {
1615     SigpInfo *si = arg;
1616     CPUState *cs = CPU(si->cpu);
1617     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1618
1619     cpu_synchronize_state(cs);
1620     scc->initial_cpu_reset(cs);
1621     cpu_synchronize_post_reset(cs);
1622     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1623 }
1624
1625 static void sigp_cpu_reset(void *arg)
1626 {
1627     SigpInfo *si = arg;
1628     CPUState *cs = CPU(si->cpu);
1629     S390CPUClass *scc = S390_CPU_GET_CLASS(si->cpu);
1630
1631     cpu_synchronize_state(cs);
1632     scc->cpu_reset(cs);
1633     cpu_synchronize_post_reset(cs);
1634     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1635 }
1636
1637 static void sigp_set_prefix(void *arg)
1638 {
1639     SigpInfo *si = arg;
1640     uint32_t addr = si->param & 0x7fffe000u;
1641
1642     cpu_synchronize_state(CPU(si->cpu));
1643
1644     if (!address_space_access_valid(&address_space_memory, addr,
1645                                     sizeof(struct LowCore), false)) {
1646         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1647         return;
1648     }
1649
1650     /* cpu has to be stopped */
1651     if (s390_cpu_get_state(si->cpu) != CPU_STATE_STOPPED) {
1652         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1653         return;
1654     }
1655
1656     si->cpu->env.psa = addr;
1657     cpu_synchronize_post_init(CPU(si->cpu));
1658     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1659 }
1660
1661 static int handle_sigp_single_dst(S390CPU *dst_cpu, uint8_t order,
1662                                   uint64_t param, uint64_t *status_reg)
1663 {
1664     SigpInfo si = {
1665         .cpu = dst_cpu,
1666         .param = param,
1667         .status_reg = status_reg,
1668     };
1669
1670     /* cpu available? */
1671     if (dst_cpu == NULL) {
1672         return SIGP_CC_NOT_OPERATIONAL;
1673     }
1674
1675     /* only resets can break pending orders */
1676     if (dst_cpu->env.sigp_order != 0 &&
1677         order != SIGP_CPU_RESET &&
1678         order != SIGP_INITIAL_CPU_RESET) {
1679         return SIGP_CC_BUSY;
1680     }
1681
1682     switch (order) {
1683     case SIGP_START:
1684         run_on_cpu(CPU(dst_cpu), sigp_start, &si);
1685         break;
1686     case SIGP_STOP:
1687         run_on_cpu(CPU(dst_cpu), sigp_stop, &si);
1688         break;
1689     case SIGP_RESTART:
1690         run_on_cpu(CPU(dst_cpu), sigp_restart, &si);
1691         break;
1692     case SIGP_STOP_STORE_STATUS:
1693         run_on_cpu(CPU(dst_cpu), sigp_stop_and_store_status, &si);
1694         break;
1695     case SIGP_STORE_STATUS_ADDR:
1696         run_on_cpu(CPU(dst_cpu), sigp_store_status_at_address, &si);
1697         break;
1698     case SIGP_STORE_ADTL_STATUS:
1699         run_on_cpu(CPU(dst_cpu), sigp_store_adtl_status, &si);
1700         break;
1701     case SIGP_SET_PREFIX:
1702         run_on_cpu(CPU(dst_cpu), sigp_set_prefix, &si);
1703         break;
1704     case SIGP_INITIAL_CPU_RESET:
1705         run_on_cpu(CPU(dst_cpu), sigp_initial_cpu_reset, &si);
1706         break;
1707     case SIGP_CPU_RESET:
1708         run_on_cpu(CPU(dst_cpu), sigp_cpu_reset, &si);
1709         break;
1710     default:
1711         DPRINTF("KVM: unknown SIGP: 0x%x\n", order);
1712         set_sigp_status(&si, SIGP_STAT_INVALID_ORDER);
1713     }
1714
1715     return si.cc;
1716 }
1717
1718 static int sigp_set_architecture(S390CPU *cpu, uint32_t param,
1719                                  uint64_t *status_reg)
1720 {
1721     CPUState *cur_cs;
1722     S390CPU *cur_cpu;
1723
1724     /* due to the BQL, we are the only active cpu */
1725     CPU_FOREACH(cur_cs) {
1726         cur_cpu = S390_CPU(cur_cs);
1727         if (cur_cpu->env.sigp_order != 0) {
1728             return SIGP_CC_BUSY;
1729         }
1730         cpu_synchronize_state(cur_cs);
1731         /* all but the current one have to be stopped */
1732         if (cur_cpu != cpu &&
1733             s390_cpu_get_state(cur_cpu) != CPU_STATE_STOPPED) {
1734             *status_reg &= 0xffffffff00000000ULL;
1735             *status_reg |= SIGP_STAT_INCORRECT_STATE;
1736             return SIGP_CC_STATUS_STORED;
1737         }
1738     }
1739
1740     switch (param & 0xff) {
1741     case SIGP_MODE_ESA_S390:
1742         /* not supported */
1743         return SIGP_CC_NOT_OPERATIONAL;
1744     case SIGP_MODE_Z_ARCH_TRANS_ALL_PSW:
1745     case SIGP_MODE_Z_ARCH_TRANS_CUR_PSW:
1746         CPU_FOREACH(cur_cs) {
1747             cur_cpu = S390_CPU(cur_cs);
1748             cur_cpu->env.pfault_token = -1UL;
1749         }
1750         break;
1751     default:
1752         *status_reg &= 0xffffffff00000000ULL;
1753         *status_reg |= SIGP_STAT_INVALID_PARAMETER;
1754         return SIGP_CC_STATUS_STORED;
1755     }
1756
1757     return SIGP_CC_ORDER_CODE_ACCEPTED;
1758 }
1759
1760 #define SIGP_ORDER_MASK 0x000000ff
1761
1762 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1763 {
1764     CPUS390XState *env = &cpu->env;
1765     const uint8_t r1 = ipa1 >> 4;
1766     const uint8_t r3 = ipa1 & 0x0f;
1767     int ret;
1768     uint8_t order;
1769     uint64_t *status_reg;
1770     uint64_t param;
1771     S390CPU *dst_cpu = NULL;
1772
1773     cpu_synchronize_state(CPU(cpu));
1774
1775     /* get order code */
1776     order = decode_basedisp_rs(env, run->s390_sieic.ipb, NULL)
1777         & SIGP_ORDER_MASK;
1778     status_reg = &env->regs[r1];
1779     param = (r1 % 2) ? env->regs[r1] : env->regs[r1 + 1];
1780
1781     if (qemu_mutex_trylock(&qemu_sigp_mutex)) {
1782         ret = SIGP_CC_BUSY;
1783         goto out;
1784     }
1785
1786     switch (order) {
1787     case SIGP_SET_ARCH:
1788         ret = sigp_set_architecture(cpu, param, status_reg);
1789         break;
1790     default:
1791         /* all other sigp orders target a single vcpu */
1792         dst_cpu = s390_cpu_addr2state(env->regs[r3]);
1793         ret = handle_sigp_single_dst(dst_cpu, order, param, status_reg);
1794     }
1795     qemu_mutex_unlock(&qemu_sigp_mutex);
1796
1797 out:
1798     trace_kvm_sigp_finished(order, CPU(cpu)->cpu_index,
1799                             dst_cpu ? CPU(dst_cpu)->cpu_index : -1, ret);
1800
1801     if (ret >= 0) {
1802         setcc(cpu, ret);
1803         return 0;
1804     }
1805
1806     return ret;
1807 }
1808
1809 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1810 {
1811     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1812     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1813     int r = -1;
1814
1815     DPRINTF("handle_instruction 0x%x 0x%x\n",
1816             run->s390_sieic.ipa, run->s390_sieic.ipb);
1817     switch (ipa0) {
1818     case IPA0_B2:
1819         r = handle_b2(cpu, run, ipa1);
1820         break;
1821     case IPA0_B9:
1822         r = handle_b9(cpu, run, ipa1);
1823         break;
1824     case IPA0_EB:
1825         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1826         break;
1827     case IPA0_E3:
1828         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1829         break;
1830     case IPA0_DIAG:
1831         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1832         break;
1833     case IPA0_SIGP:
1834         r = handle_sigp(cpu, run, ipa1);
1835         break;
1836     }
1837
1838     if (r < 0) {
1839         r = 0;
1840         enter_pgmcheck(cpu, 0x0001);
1841     }
1842
1843     return r;
1844 }
1845
1846 static bool is_special_wait_psw(CPUState *cs)
1847 {
1848     /* signal quiesce */
1849     return cs->kvm_run->psw_addr == 0xfffUL;
1850 }
1851
1852 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1853 {
1854     CPUState *cs = CPU(cpu);
1855
1856     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1857                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1858                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1859     s390_cpu_halt(cpu);
1860     qemu_system_guest_panicked();
1861 }
1862
1863 static int handle_intercept(S390CPU *cpu)
1864 {
1865     CPUState *cs = CPU(cpu);
1866     struct kvm_run *run = cs->kvm_run;
1867     int icpt_code = run->s390_sieic.icptcode;
1868     int r = 0;
1869
1870     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1871             (long)cs->kvm_run->psw_addr);
1872     switch (icpt_code) {
1873         case ICPT_INSTRUCTION:
1874             r = handle_instruction(cpu, run);
1875             break;
1876         case ICPT_PROGRAM:
1877             unmanageable_intercept(cpu, "program interrupt",
1878                                    offsetof(LowCore, program_new_psw));
1879             r = EXCP_HALTED;
1880             break;
1881         case ICPT_EXT_INT:
1882             unmanageable_intercept(cpu, "external interrupt",
1883                                    offsetof(LowCore, external_new_psw));
1884             r = EXCP_HALTED;
1885             break;
1886         case ICPT_WAITPSW:
1887             /* disabled wait, since enabled wait is handled in kernel */
1888             cpu_synchronize_state(cs);
1889             if (s390_cpu_halt(cpu) == 0) {
1890                 if (is_special_wait_psw(cs)) {
1891                     qemu_system_shutdown_request();
1892                 } else {
1893                     qemu_system_guest_panicked();
1894                 }
1895             }
1896             r = EXCP_HALTED;
1897             break;
1898         case ICPT_CPU_STOP:
1899             if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
1900                 qemu_system_shutdown_request();
1901             }
1902             if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
1903                 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR,
1904                                       true);
1905             }
1906             cpu->env.sigp_order = 0;
1907             r = EXCP_HALTED;
1908             break;
1909         case ICPT_OPEREXC:
1910             /* currently only instr 0x0000 after enabled via capability */
1911             r = handle_sw_breakpoint(cpu, run);
1912             if (r == -ENOENT) {
1913                 enter_pgmcheck(cpu, PGM_OPERATION);
1914                 r = 0;
1915             }
1916             break;
1917         case ICPT_SOFT_INTERCEPT:
1918             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1919             exit(1);
1920             break;
1921         case ICPT_IO:
1922             fprintf(stderr, "KVM unimplemented icpt IO\n");
1923             exit(1);
1924             break;
1925         default:
1926             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1927             exit(1);
1928             break;
1929     }
1930
1931     return r;
1932 }
1933
1934 static int handle_tsch(S390CPU *cpu)
1935 {
1936     CPUState *cs = CPU(cpu);
1937     struct kvm_run *run = cs->kvm_run;
1938     int ret;
1939
1940     cpu_synchronize_state(cs);
1941
1942     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb);
1943     if (ret < 0) {
1944         /*
1945          * Failure.
1946          * If an I/O interrupt had been dequeued, we have to reinject it.
1947          */
1948         if (run->s390_tsch.dequeued) {
1949             kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1950                                   run->s390_tsch.subchannel_nr,
1951                                   run->s390_tsch.io_int_parm,
1952                                   run->s390_tsch.io_int_word);
1953         }
1954         ret = 0;
1955     }
1956     return ret;
1957 }
1958
1959 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1960 {
1961     struct sysib_322 sysib;
1962     int del;
1963
1964     if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1965         return;
1966     }
1967     /* Shift the stack of Extended Names to prepare for our own data */
1968     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1969             sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1970     /* First virt level, that doesn't provide Ext Names delimits stack. It is
1971      * assumed it's not capable of managing Extended Names for lower levels.
1972      */
1973     for (del = 1; del < sysib.count; del++) {
1974         if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1975             break;
1976         }
1977     }
1978     if (del < sysib.count) {
1979         memset(sysib.ext_names[del], 0,
1980                sizeof(sysib.ext_names[0]) * (sysib.count - del));
1981     }
1982     /* Insert short machine name in EBCDIC, padded with blanks */
1983     if (qemu_name) {
1984         memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1985         ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1986                                                     strlen(qemu_name)));
1987     }
1988     sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1989     memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1990     /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1991      * considered by s390 as not capable of providing any Extended Name.
1992      * Therefore if no name was specified on qemu invocation, we go with the
1993      * same "KVMguest" default, which KVM has filled into short name field.
1994      */
1995     if (qemu_name) {
1996         strncpy((char *)sysib.ext_names[0], qemu_name,
1997                 sizeof(sysib.ext_names[0]));
1998     } else {
1999         strcpy((char *)sysib.ext_names[0], "KVMguest");
2000     }
2001     /* Insert UUID */
2002     memcpy(sysib.vm[0].uuid, qemu_uuid, sizeof(sysib.vm[0].uuid));
2003
2004     s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
2005 }
2006
2007 static int handle_stsi(S390CPU *cpu)
2008 {
2009     CPUState *cs = CPU(cpu);
2010     struct kvm_run *run = cs->kvm_run;
2011
2012     switch (run->s390_stsi.fc) {
2013     case 3:
2014         if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
2015             return 0;
2016         }
2017         /* Only sysib 3.2.2 needs post-handling for now. */
2018         insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
2019         return 0;
2020     default:
2021         return 0;
2022     }
2023 }
2024
2025 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
2026 {
2027     CPUState *cs = CPU(cpu);
2028     struct kvm_run *run = cs->kvm_run;
2029
2030     int ret = 0;
2031     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
2032
2033     switch (arch_info->type) {
2034     case KVM_HW_WP_WRITE:
2035         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
2036             cs->watchpoint_hit = &hw_watchpoint;
2037             hw_watchpoint.vaddr = arch_info->addr;
2038             hw_watchpoint.flags = BP_MEM_WRITE;
2039             ret = EXCP_DEBUG;
2040         }
2041         break;
2042     case KVM_HW_BP:
2043         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
2044             ret = EXCP_DEBUG;
2045         }
2046         break;
2047     case KVM_SINGLESTEP:
2048         if (cs->singlestep_enabled) {
2049             ret = EXCP_DEBUG;
2050         }
2051         break;
2052     default:
2053         ret = -ENOSYS;
2054     }
2055
2056     return ret;
2057 }
2058
2059 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
2060 {
2061     S390CPU *cpu = S390_CPU(cs);
2062     int ret = 0;
2063
2064     qemu_mutex_lock_iothread();
2065
2066     switch (run->exit_reason) {
2067         case KVM_EXIT_S390_SIEIC:
2068             ret = handle_intercept(cpu);
2069             break;
2070         case KVM_EXIT_S390_RESET:
2071             s390_reipl_request();
2072             break;
2073         case KVM_EXIT_S390_TSCH:
2074             ret = handle_tsch(cpu);
2075             break;
2076         case KVM_EXIT_S390_STSI:
2077             ret = handle_stsi(cpu);
2078             break;
2079         case KVM_EXIT_DEBUG:
2080             ret = kvm_arch_handle_debug_exit(cpu);
2081             break;
2082         default:
2083             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
2084             break;
2085     }
2086     qemu_mutex_unlock_iothread();
2087
2088     if (ret == 0) {
2089         ret = EXCP_INTERRUPT;
2090     }
2091     return ret;
2092 }
2093
2094 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2095 {
2096     return true;
2097 }
2098
2099 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2100 {
2101     return 1;
2102 }
2103
2104 int kvm_arch_on_sigbus(int code, void *addr)
2105 {
2106     return 1;
2107 }
2108
2109 void kvm_s390_io_interrupt(uint16_t subchannel_id,
2110                            uint16_t subchannel_nr, uint32_t io_int_parm,
2111                            uint32_t io_int_word)
2112 {
2113     struct kvm_s390_irq irq = {
2114         .u.io.subchannel_id = subchannel_id,
2115         .u.io.subchannel_nr = subchannel_nr,
2116         .u.io.io_int_parm = io_int_parm,
2117         .u.io.io_int_word = io_int_word,
2118     };
2119
2120     if (io_int_word & IO_INT_WORD_AI) {
2121         irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
2122     } else {
2123         irq.type = KVM_S390_INT_IO(0, (subchannel_id & 0xff00) >> 8,
2124                                       (subchannel_id & 0x0006),
2125                                       subchannel_nr);
2126     }
2127     kvm_s390_floating_interrupt(&irq);
2128 }
2129
2130 static uint64_t build_channel_report_mcic(void)
2131 {
2132     uint64_t mcic;
2133
2134     /* subclass: indicate channel report pending */
2135     mcic = MCIC_SC_CP |
2136     /* subclass modifiers: none */
2137     /* storage errors: none */
2138     /* validity bits: no damage */
2139         MCIC_VB_WP | MCIC_VB_MS | MCIC_VB_PM | MCIC_VB_IA | MCIC_VB_FP |
2140         MCIC_VB_GR | MCIC_VB_CR | MCIC_VB_ST | MCIC_VB_AR | MCIC_VB_PR |
2141         MCIC_VB_FC | MCIC_VB_CT | MCIC_VB_CC;
2142     if (s390_has_feat(S390_FEAT_VECTOR)) {
2143         mcic |= MCIC_VB_VR;
2144     }
2145     return mcic;
2146 }
2147
2148 void kvm_s390_crw_mchk(void)
2149 {
2150     struct kvm_s390_irq irq = {
2151         .type = KVM_S390_MCHK,
2152         .u.mchk.cr14 = 1 << 28,
2153         .u.mchk.mcic = build_channel_report_mcic(),
2154     };
2155     kvm_s390_floating_interrupt(&irq);
2156 }
2157
2158 void kvm_s390_enable_css_support(S390CPU *cpu)
2159 {
2160     int r;
2161
2162     /* Activate host kernel channel subsystem support. */
2163     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
2164     assert(r == 0);
2165 }
2166
2167 void kvm_arch_init_irq_routing(KVMState *s)
2168 {
2169     /*
2170      * Note that while irqchip capabilities generally imply that cpustates
2171      * are handled in-kernel, it is not true for s390 (yet); therefore, we
2172      * have to override the common code kvm_halt_in_kernel_allowed setting.
2173      */
2174     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2175         kvm_gsi_routing_allowed = true;
2176         kvm_halt_in_kernel_allowed = false;
2177     }
2178 }
2179
2180 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
2181                                     int vq, bool assign)
2182 {
2183     struct kvm_ioeventfd kick = {
2184         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
2185         KVM_IOEVENTFD_FLAG_DATAMATCH,
2186         .fd = event_notifier_get_fd(notifier),
2187         .datamatch = vq,
2188         .addr = sch,
2189         .len = 8,
2190     };
2191     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
2192         return -ENOSYS;
2193     }
2194     if (!assign) {
2195         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
2196     }
2197     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
2198 }
2199
2200 int kvm_s390_get_memslot_count(KVMState *s)
2201 {
2202     return kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
2203 }
2204
2205 int kvm_s390_get_ri(void)
2206 {
2207     return cap_ri;
2208 }
2209
2210 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
2211 {
2212     struct kvm_mp_state mp_state = {};
2213     int ret;
2214
2215     /* the kvm part might not have been initialized yet */
2216     if (CPU(cpu)->kvm_state == NULL) {
2217         return 0;
2218     }
2219
2220     switch (cpu_state) {
2221     case CPU_STATE_STOPPED:
2222         mp_state.mp_state = KVM_MP_STATE_STOPPED;
2223         break;
2224     case CPU_STATE_CHECK_STOP:
2225         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
2226         break;
2227     case CPU_STATE_OPERATING:
2228         mp_state.mp_state = KVM_MP_STATE_OPERATING;
2229         break;
2230     case CPU_STATE_LOAD:
2231         mp_state.mp_state = KVM_MP_STATE_LOAD;
2232         break;
2233     default:
2234         error_report("Requested CPU state is not a valid S390 CPU state: %u",
2235                      cpu_state);
2236         exit(1);
2237     }
2238
2239     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2240     if (ret) {
2241         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
2242                                        strerror(-ret));
2243     }
2244
2245     return ret;
2246 }
2247
2248 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
2249 {
2250     struct kvm_s390_irq_state irq_state;
2251     CPUState *cs = CPU(cpu);
2252     int32_t bytes;
2253
2254     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2255         return;
2256     }
2257
2258     irq_state.buf = (uint64_t) cpu->irqstate;
2259     irq_state.len = VCPU_IRQ_BUF_SIZE;
2260
2261     bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
2262     if (bytes < 0) {
2263         cpu->irqstate_saved_size = 0;
2264         error_report("Migration of interrupt state failed");
2265         return;
2266     }
2267
2268     cpu->irqstate_saved_size = bytes;
2269 }
2270
2271 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
2272 {
2273     CPUState *cs = CPU(cpu);
2274     struct kvm_s390_irq_state irq_state;
2275     int r;
2276
2277     if (cpu->irqstate_saved_size == 0) {
2278         return 0;
2279     }
2280
2281     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2282         return -ENOSYS;
2283     }
2284
2285     irq_state.buf = (uint64_t) cpu->irqstate;
2286     irq_state.len = cpu->irqstate_saved_size;
2287
2288     r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
2289     if (r) {
2290         error_report("Setting interrupt state failed %d", r);
2291     }
2292     return r;
2293 }
2294
2295 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2296                              uint64_t address, uint32_t data, PCIDevice *dev)
2297 {
2298     S390PCIBusDevice *pbdev;
2299     uint32_t idx = data >> ZPCI_MSI_VEC_BITS;
2300     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
2301
2302     pbdev = s390_pci_find_dev_by_idx(idx);
2303     if (!pbdev) {
2304         DPRINTF("add_msi_route no dev\n");
2305         return -ENODEV;
2306     }
2307
2308     pbdev->routes.adapter.ind_offset = vec;
2309
2310     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2311     route->flags = 0;
2312     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2313     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2314     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2315     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset;
2316     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2317     return 0;
2318 }
2319
2320 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2321                                 int vector, PCIDevice *dev)
2322 {
2323     return 0;
2324 }
2325
2326 int kvm_arch_release_virq_post(int virq)
2327 {
2328     return 0;
2329 }
2330
2331 int kvm_arch_msi_data_to_gsi(uint32_t data)
2332 {
2333     abort();
2334 }
2335
2336 static inline int test_bit_inv(long nr, const unsigned long *addr)
2337 {
2338     return test_bit(BE_BIT_NR(nr), addr);
2339 }
2340
2341 static inline void set_bit_inv(long nr, unsigned long *addr)
2342 {
2343     set_bit(BE_BIT_NR(nr), addr);
2344 }
2345
2346 static int query_cpu_subfunc(S390FeatBitmap features)
2347 {
2348     struct kvm_s390_vm_cpu_subfunc prop;
2349     struct kvm_device_attr attr = {
2350         .group = KVM_S390_VM_CPU_MODEL,
2351         .attr = KVM_S390_VM_CPU_MACHINE_SUBFUNC,
2352         .addr = (uint64_t) &prop,
2353     };
2354     int rc;
2355
2356     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2357     if (rc) {
2358         return  rc;
2359     }
2360
2361     /*
2362      * We're going to add all subfunctions now, if the corresponding feature
2363      * is available that unlocks the query functions.
2364      */
2365     s390_add_from_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2366     if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2367         s390_add_from_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2368     }
2369     if (test_bit(S390_FEAT_MSA, features)) {
2370         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2371         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2372         s390_add_from_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2373         s390_add_from_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2374         s390_add_from_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2375     }
2376     if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2377         s390_add_from_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2378     }
2379     if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2380         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2381         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2382         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2383         s390_add_from_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2384     }
2385     if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2386         s390_add_from_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2387     }
2388     return 0;
2389 }
2390
2391 static int configure_cpu_subfunc(const S390FeatBitmap features)
2392 {
2393     struct kvm_s390_vm_cpu_subfunc prop = {};
2394     struct kvm_device_attr attr = {
2395         .group = KVM_S390_VM_CPU_MODEL,
2396         .attr = KVM_S390_VM_CPU_PROCESSOR_SUBFUNC,
2397         .addr = (uint64_t) &prop,
2398     };
2399
2400     if (!kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2401                            KVM_S390_VM_CPU_PROCESSOR_SUBFUNC)) {
2402         /* hardware support might be missing, IBC will handle most of this */
2403         return 0;
2404     }
2405
2406     s390_fill_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2407     if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2408         s390_fill_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2409         prop.ptff[0] |= 0x80; /* query is always available */
2410     }
2411     if (test_bit(S390_FEAT_MSA, features)) {
2412         s390_fill_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2413         prop.kmac[0] |= 0x80; /* query is always available */
2414         s390_fill_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2415         prop.kmc[0] |= 0x80; /* query is always available */
2416         s390_fill_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2417         prop.km[0] |= 0x80; /* query is always available */
2418         s390_fill_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2419         prop.kimd[0] |= 0x80; /* query is always available */
2420         s390_fill_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2421         prop.klmd[0] |= 0x80; /* query is always available */
2422     }
2423     if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2424         s390_fill_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2425         prop.pckmo[0] |= 0x80; /* query is always available */
2426     }
2427     if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2428         s390_fill_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2429         prop.kmctr[0] |= 0x80; /* query is always available */
2430         s390_fill_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2431         prop.kmf[0] |= 0x80; /* query is always available */
2432         s390_fill_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2433         prop.kmo[0] |= 0x80; /* query is always available */
2434         s390_fill_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2435         prop.pcc[0] |= 0x80; /* query is always available */
2436     }
2437     if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2438         s390_fill_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2439         prop.ppno[0] |= 0x80; /* query is always available */
2440     }
2441     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2442 }
2443
2444 static int kvm_to_feat[][2] = {
2445     { KVM_S390_VM_CPU_FEAT_ESOP, S390_FEAT_ESOP },
2446     { KVM_S390_VM_CPU_FEAT_SIEF2, S390_FEAT_SIE_F2 },
2447     { KVM_S390_VM_CPU_FEAT_64BSCAO , S390_FEAT_SIE_64BSCAO },
2448     { KVM_S390_VM_CPU_FEAT_SIIF, S390_FEAT_SIE_SIIF },
2449     { KVM_S390_VM_CPU_FEAT_GPERE, S390_FEAT_SIE_GPERE },
2450     { KVM_S390_VM_CPU_FEAT_GSLS, S390_FEAT_SIE_GSLS },
2451     { KVM_S390_VM_CPU_FEAT_IB, S390_FEAT_SIE_IB },
2452     { KVM_S390_VM_CPU_FEAT_CEI, S390_FEAT_SIE_CEI },
2453     { KVM_S390_VM_CPU_FEAT_IBS, S390_FEAT_SIE_IBS },
2454     { KVM_S390_VM_CPU_FEAT_SKEY, S390_FEAT_SIE_SKEY },
2455     { KVM_S390_VM_CPU_FEAT_CMMA, S390_FEAT_SIE_CMMA },
2456     { KVM_S390_VM_CPU_FEAT_PFMFI, S390_FEAT_SIE_PFMFI},
2457     { KVM_S390_VM_CPU_FEAT_SIGPIF, S390_FEAT_SIE_SIGPIF},
2458 };
2459
2460 static int query_cpu_feat(S390FeatBitmap features)
2461 {
2462     struct kvm_s390_vm_cpu_feat prop;
2463     struct kvm_device_attr attr = {
2464         .group = KVM_S390_VM_CPU_MODEL,
2465         .attr = KVM_S390_VM_CPU_MACHINE_FEAT,
2466         .addr = (uint64_t) &prop,
2467     };
2468     int rc;
2469     int i;
2470
2471     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2472     if (rc) {
2473         return  rc;
2474     }
2475
2476     for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2477         if (test_bit_inv(kvm_to_feat[i][0], (unsigned long *)prop.feat)) {
2478             set_bit(kvm_to_feat[i][1], features);
2479         }
2480     }
2481     return 0;
2482 }
2483
2484 static int configure_cpu_feat(const S390FeatBitmap features)
2485 {
2486     struct kvm_s390_vm_cpu_feat prop = {};
2487     struct kvm_device_attr attr = {
2488         .group = KVM_S390_VM_CPU_MODEL,
2489         .attr = KVM_S390_VM_CPU_PROCESSOR_FEAT,
2490         .addr = (uint64_t) &prop,
2491     };
2492     int i;
2493
2494     for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2495         if (test_bit(kvm_to_feat[i][1], features)) {
2496             set_bit_inv(kvm_to_feat[i][0], (unsigned long *)prop.feat);
2497         }
2498     }
2499     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2500 }
2501
2502 bool kvm_s390_cpu_models_supported(void)
2503 {
2504     if (!cpu_model_allowed()) {
2505         /* compatibility machines interfere with the cpu model */
2506         return false;
2507     }
2508     return kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2509                              KVM_S390_VM_CPU_MACHINE) &&
2510            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2511                              KVM_S390_VM_CPU_PROCESSOR) &&
2512            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2513                              KVM_S390_VM_CPU_MACHINE_FEAT) &&
2514            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2515                              KVM_S390_VM_CPU_PROCESSOR_FEAT) &&
2516            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2517                              KVM_S390_VM_CPU_MACHINE_SUBFUNC);
2518 }
2519
2520 void kvm_s390_get_host_cpu_model(S390CPUModel *model, Error **errp)
2521 {
2522     struct kvm_s390_vm_cpu_machine prop = {};
2523     struct kvm_device_attr attr = {
2524         .group = KVM_S390_VM_CPU_MODEL,
2525         .attr = KVM_S390_VM_CPU_MACHINE,
2526         .addr = (uint64_t) &prop,
2527     };
2528     uint16_t unblocked_ibc = 0, cpu_type = 0;
2529     int rc;
2530
2531     memset(model, 0, sizeof(*model));
2532
2533     if (!kvm_s390_cpu_models_supported()) {
2534         error_setg(errp, "KVM doesn't support CPU models");
2535         return;
2536     }
2537
2538     /* query the basic cpu model properties */
2539     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2540     if (rc) {
2541         error_setg(errp, "KVM: Error querying host CPU model: %d", rc);
2542         return;
2543     }
2544
2545     cpu_type = cpuid_type(prop.cpuid);
2546     if (has_ibc(prop.ibc)) {
2547         model->lowest_ibc = lowest_ibc(prop.ibc);
2548         unblocked_ibc = unblocked_ibc(prop.ibc);
2549     }
2550     model->cpu_id = cpuid_id(prop.cpuid);
2551     model->cpu_ver = 0xff;
2552
2553     /* get supported cpu features indicated via STFL(E) */
2554     s390_add_from_feat_block(model->features, S390_FEAT_TYPE_STFL,
2555                              (uint8_t *) prop.fac_mask);
2556     /* dat-enhancement facility 2 has no bit but was introduced with stfle */
2557     if (test_bit(S390_FEAT_STFLE, model->features)) {
2558         set_bit(S390_FEAT_DAT_ENH_2, model->features);
2559     }
2560     /* get supported cpu features indicated e.g. via SCLP */
2561     rc = query_cpu_feat(model->features);
2562     if (rc) {
2563         error_setg(errp, "KVM: Error querying CPU features: %d", rc);
2564         return;
2565     }
2566     /* get supported cpu subfunctions indicated via query / test bit */
2567     rc = query_cpu_subfunc(model->features);
2568     if (rc) {
2569         error_setg(errp, "KVM: Error querying CPU subfunctions: %d", rc);
2570         return;
2571     }
2572
2573     /* with cpu model support, CMM is only indicated if really available */
2574     if (kvm_s390_cmma_available()) {
2575         set_bit(S390_FEAT_CMM, model->features);
2576     }
2577
2578     if (s390_known_cpu_type(cpu_type)) {
2579         /* we want the exact model, even if some features are missing */
2580         model->def = s390_find_cpu_def(cpu_type, ibc_gen(unblocked_ibc),
2581                                        ibc_ec_ga(unblocked_ibc), NULL);
2582     } else {
2583         /* model unknown, e.g. too new - search using features */
2584         model->def = s390_find_cpu_def(0, ibc_gen(unblocked_ibc),
2585                                        ibc_ec_ga(unblocked_ibc),
2586                                        model->features);
2587     }
2588     if (!model->def) {
2589         error_setg(errp, "KVM: host CPU model could not be identified");
2590         return;
2591     }
2592     /* strip of features that are not part of the maximum model */
2593     bitmap_and(model->features, model->features, model->def->full_feat,
2594                S390_FEAT_MAX);
2595 }
2596
2597 void kvm_s390_apply_cpu_model(const S390CPUModel *model, Error **errp)
2598 {
2599     struct kvm_s390_vm_cpu_processor prop  = {
2600         .fac_list = { 0 },
2601     };
2602     struct kvm_device_attr attr = {
2603         .group = KVM_S390_VM_CPU_MODEL,
2604         .attr = KVM_S390_VM_CPU_PROCESSOR,
2605         .addr = (uint64_t) &prop,
2606     };
2607     int rc;
2608
2609     if (!model) {
2610         /* compatibility handling if cpu models are disabled */
2611         if (kvm_s390_cmma_available() && !mem_path) {
2612             kvm_s390_enable_cmma();
2613         }
2614         return;
2615     }
2616     if (!kvm_s390_cpu_models_supported()) {
2617         error_setg(errp, "KVM doesn't support CPU models");
2618         return;
2619     }
2620     prop.cpuid = s390_cpuid_from_cpu_model(model);
2621     prop.ibc = s390_ibc_from_cpu_model(model);
2622     /* configure cpu features indicated via STFL(e) */
2623     s390_fill_feat_block(model->features, S390_FEAT_TYPE_STFL,
2624                          (uint8_t *) prop.fac_list);
2625     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2626     if (rc) {
2627         error_setg(errp, "KVM: Error configuring the CPU model: %d", rc);
2628         return;
2629     }
2630     /* configure cpu features indicated e.g. via SCLP */
2631     rc = configure_cpu_feat(model->features);
2632     if (rc) {
2633         error_setg(errp, "KVM: Error configuring CPU features: %d", rc);
2634         return;
2635     }
2636     /* configure cpu subfunctions indicated via query / test bit */
2637     rc = configure_cpu_subfunc(model->features);
2638     if (rc) {
2639         error_setg(errp, "KVM: Error configuring CPU subfunctions: %d", rc);
2640         return;
2641     }
2642     /* enable CMM via CMMA - disable on hugetlbfs */
2643     if (test_bit(S390_FEAT_CMM, model->features)) {
2644         if (mem_path) {
2645             error_report("Warning: CMM will not be enabled because it is not "
2646                          "compatible to hugetlbfs.");
2647         } else {
2648             kvm_s390_enable_cmma();
2649         }
2650     }
2651 }