target-s390x/kvm.c

   1 /*
   2  * QEMU S390x KVM implementation
   3  *
   4  * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
   5  * Copyright IBM Corp. 2012
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * Contributions after 2012-10-29 are licensed under the terms of the
  18  * GNU GPL, version 2 or (at your option) any later version.
  19  *
  20  * You should have received a copy of the GNU (Lesser) General Public
  21  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  22  */
  23
  24 #include <sys/types.h>
  25 #include <sys/ioctl.h>
  26 #include <sys/mman.h>
  27
  28 #include <linux/kvm.h>
  29 #include <asm/ptrace.h>
  30
  31 #include "qemu-common.h"
  32 #include "qemu/timer.h"
  33 #include "sysemu/sysemu.h"
  34 #include "sysemu/kvm.h"
  35 #include "hw/hw.h"
  36 #include "cpu.h"
  37 #include "sysemu/device_tree.h"
  38 #include "qapi/qmp/qjson.h"
  39 #include "monitor/monitor.h"
  40 #include "exec/gdbstub.h"
  41 #include "trace.h"
  42
  43 /* #define DEBUG_KVM */
  44
  45 #ifdef DEBUG_KVM
  46 #define DPRINTF(fmt, ...) \
  47     do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
  48 #else
  49 #define DPRINTF(fmt, ...) \
  50     do { } while (0)
  51 #endif
  52
  53 #define IPA0_DIAG                       0x8300
  54 #define IPA0_SIGP                       0xae00
  55 #define IPA0_B2                         0xb200
  56 #define IPA0_B9                         0xb900
  57 #define IPA0_EB                         0xeb00
  58
  59 #define PRIV_B2_SCLP_CALL               0x20
  60 #define PRIV_B2_CSCH                    0x30
  61 #define PRIV_B2_HSCH                    0x31
  62 #define PRIV_B2_MSCH                    0x32
  63 #define PRIV_B2_SSCH                    0x33
  64 #define PRIV_B2_STSCH                   0x34
  65 #define PRIV_B2_TSCH                    0x35
  66 #define PRIV_B2_TPI                     0x36
  67 #define PRIV_B2_SAL                     0x37
  68 #define PRIV_B2_RSCH                    0x38
  69 #define PRIV_B2_STCRW                   0x39
  70 #define PRIV_B2_STCPS                   0x3a
  71 #define PRIV_B2_RCHP                    0x3b
  72 #define PRIV_B2_SCHM                    0x3c
  73 #define PRIV_B2_CHSC                    0x5f
  74 #define PRIV_B2_SIGA                    0x74
  75 #define PRIV_B2_XSCH                    0x76
  76
  77 #define PRIV_EB_SQBS                    0x8a
  78
  79 #define PRIV_B9_EQBS                    0x9c
  80
  81 #define DIAG_IPL                        0x308
  82 #define DIAG_KVM_HYPERCALL              0x500
  83 #define DIAG_KVM_BREAKPOINT             0x501
  84
  85 #define ICPT_INSTRUCTION                0x04
  86 #define ICPT_PROGRAM                    0x08
  87 #define ICPT_EXT_INT                    0x14
  88 #define ICPT_WAITPSW                    0x1c
  89 #define ICPT_SOFT_INTERCEPT             0x24
  90 #define ICPT_CPU_STOP                   0x28
  91 #define ICPT_IO                         0x40
  92
  93 static CPUWatchpoint hw_watchpoint;
  94 /*
  95  * We don't use a list because this structure is also used to transmit the
  96  * hardware breakpoints to the kernel.
  97  */
  98 static struct kvm_hw_breakpoint *hw_breakpoints;
  99 static int nb_hw_breakpoints;
 100
 101 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
 102     KVM_CAP_LAST_INFO
 103 };
 104
 105 static int cap_sync_regs;
 106 static int cap_async_pf;
 107
 108 static void *legacy_s390_alloc(size_t size);
 109
 110 static int kvm_s390_check_clear_cmma(KVMState *s)
 111 {
 112     struct kvm_device_attr attr = {
 113         .group = KVM_S390_VM_MEM_CTRL,
 114         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 115     };
 116
 117     return kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attr);
 118 }
 119
 120 static int kvm_s390_check_enable_cmma(KVMState *s)
 121 {
 122     struct kvm_device_attr attr = {
 123         .group = KVM_S390_VM_MEM_CTRL,
 124         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 125     };
 126
 127     return kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attr);
 128 }
 129
 130 void kvm_s390_clear_cmma_callback(void *opaque)
 131 {
 132     int rc;
 133     KVMState *s = opaque;
 134     struct kvm_device_attr attr = {
 135         .group = KVM_S390_VM_MEM_CTRL,
 136         .attr = KVM_S390_VM_MEM_CLR_CMMA,
 137     };
 138
 139     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 140     trace_kvm_clear_cmma(rc);
 141 }
 142
 143 static void kvm_s390_enable_cmma(KVMState *s)
 144 {
 145     int rc;
 146     struct kvm_device_attr attr = {
 147         .group = KVM_S390_VM_MEM_CTRL,
 148         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
 149     };
 150
 151     if (kvm_s390_check_enable_cmma(s) || kvm_s390_check_clear_cmma(s)) {
 152         return;
 153     }
 154
 155     rc = kvm_vm_ioctl(s, KVM_SET_DEVICE_ATTR, &attr);
 156     if (!rc) {
 157         qemu_register_reset(kvm_s390_clear_cmma_callback, s);
 158     }
 159     trace_kvm_enable_cmma(rc);
 160 }
 161
 162 int kvm_arch_init(KVMState *s)
 163 {
 164     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
 165     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
 166
 167     if (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES)) {
 168         kvm_s390_enable_cmma(s);
 169     }
 170
 171     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
 172         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
 173         phys_mem_set_alloc(legacy_s390_alloc);
 174     }
 175     return 0;
 176 }
 177
 178 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
 179 {
 180     return cpu->cpu_index;
 181 }
 182
 183 int kvm_arch_init_vcpu(CPUState *cpu)
 184 {
 185     /* nothing todo yet */
 186     return 0;
 187 }
 188
 189 void kvm_s390_reset_vcpu(S390CPU *cpu)
 190 {
 191     CPUState *cs = CPU(cpu);
 192
 193     /* The initial reset call is needed here to reset in-kernel
 194      * vcpu data that we can't access directly from QEMU
 195      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
 196      * Before this ioctl cpu_synchronize_state() is called in common kvm
 197      * code (kvm-all) */
 198     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
 199         perror("Can't reset vcpu\n");
 200     }
 201 }
 202
 203 int kvm_arch_put_registers(CPUState *cs, int level)
 204 {
 205     S390CPU *cpu = S390_CPU(cs);
 206     CPUS390XState *env = &cpu->env;
 207     struct kvm_sregs sregs;
 208     struct kvm_regs regs;
 209     int r;
 210     int i;
 211
 212     /* always save the PSW  and the GPRS*/
 213     cs->kvm_run->psw_addr = env->psw.addr;
 214     cs->kvm_run->psw_mask = env->psw.mask;
 215
 216     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_GPRS) {
 217         for (i = 0; i < 16; i++) {
 218             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
 219             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
 220         }
 221     } else {
 222         for (i = 0; i < 16; i++) {
 223             regs.gprs[i] = env->regs[i];
 224         }
 225         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
 226         if (r < 0) {
 227             return r;
 228         }
 229     }
 230
 231     /* Do we need to save more than that? */
 232     if (level == KVM_PUT_RUNTIME_STATE) {
 233         return 0;
 234     }
 235
 236     /*
 237      * These ONE_REGS are not protected by a capability. As they are only
 238      * necessary for migration we just trace a possible error, but don't
 239      * return with an error return code.
 240      */
 241     kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 242     kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 243     kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 244     kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 245     kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 246
 247     if (cap_async_pf) {
 248         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 249         if (r < 0) {
 250             return r;
 251         }
 252         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 253         if (r < 0) {
 254             return r;
 255         }
 256         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 257         if (r < 0) {
 258             return r;
 259         }
 260     }
 261
 262     if (cap_sync_regs &&
 263         cs->kvm_run->kvm_valid_regs & KVM_SYNC_ACRS &&
 264         cs->kvm_run->kvm_valid_regs & KVM_SYNC_CRS) {
 265         for (i = 0; i < 16; i++) {
 266             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
 267             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
 268         }
 269         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
 270         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
 271     } else {
 272         for (i = 0; i < 16; i++) {
 273             sregs.acrs[i] = env->aregs[i];
 274             sregs.crs[i] = env->cregs[i];
 275         }
 276         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
 277         if (r < 0) {
 278             return r;
 279         }
 280     }
 281
 282     /* Finally the prefix */
 283     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_PREFIX) {
 284         cs->kvm_run->s.regs.prefix = env->psa;
 285         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
 286     } else {
 287         /* prefix is only supported via sync regs */
 288     }
 289     return 0;
 290 }
 291
 292 int kvm_arch_get_registers(CPUState *cs)
 293 {
 294     S390CPU *cpu = S390_CPU(cs);
 295     CPUS390XState *env = &cpu->env;
 296     struct kvm_sregs sregs;
 297     struct kvm_regs regs;
 298     int i, r;
 299
 300     /* get the PSW */
 301     env->psw.addr = cs->kvm_run->psw_addr;
 302     env->psw.mask = cs->kvm_run->psw_mask;
 303
 304     /* the GPRS */
 305     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_GPRS) {
 306         for (i = 0; i < 16; i++) {
 307             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
 308         }
 309     } else {
 310         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
 311         if (r < 0) {
 312             return r;
 313         }
 314          for (i = 0; i < 16; i++) {
 315             env->regs[i] = regs.gprs[i];
 316         }
 317     }
 318
 319     /* The ACRS and CRS */
 320     if (cap_sync_regs &&
 321         cs->kvm_run->kvm_valid_regs & KVM_SYNC_ACRS &&
 322         cs->kvm_run->kvm_valid_regs & KVM_SYNC_CRS) {
 323         for (i = 0; i < 16; i++) {
 324             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
 325             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
 326         }
 327     } else {
 328         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
 329         if (r < 0) {
 330             return r;
 331         }
 332          for (i = 0; i < 16; i++) {
 333             env->aregs[i] = sregs.acrs[i];
 334             env->cregs[i] = sregs.crs[i];
 335         }
 336     }
 337
 338     /* The prefix */
 339     if (cap_sync_regs && cs->kvm_run->kvm_valid_regs & KVM_SYNC_PREFIX) {
 340         env->psa = cs->kvm_run->s.regs.prefix;
 341     }
 342
 343     /*
 344      * These ONE_REGS are not protected by a capability. As they are only
 345      * necessary for migration we just trace a possible error, but don't
 346      * return with an error return code.
 347      */
 348     kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
 349     kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
 350     kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
 351     kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
 352     kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
 353
 354     if (cap_async_pf) {
 355         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
 356         if (r < 0) {
 357             return r;
 358         }
 359         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
 360         if (r < 0) {
 361             return r;
 362         }
 363         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
 364         if (r < 0) {
 365             return r;
 366         }
 367     }
 368
 369     return 0;
 370 }
 371
 372 /*
 373  * Legacy layout for s390:
 374  * Older S390 KVM requires the topmost vma of the RAM to be
 375  * smaller than an system defined value, which is at least 256GB.
 376  * Larger systems have larger values. We put the guest between
 377  * the end of data segment (system break) and this value. We
 378  * use 32GB as a base to have enough room for the system break
 379  * to grow. We also have to use MAP parameters that avoid
 380  * read-only mapping of guest pages.
 381  */
 382 static void *legacy_s390_alloc(size_t size)
 383 {
 384     void *mem;
 385
 386     mem = mmap((void *) 0x800000000ULL, size,
 387                PROT_EXEC|PROT_READ|PROT_WRITE,
 388                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 389     return mem == MAP_FAILED ? NULL : mem;
 390 }
 391
 392 /* DIAG 501 is used for sw breakpoints */
 393 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
 394
 395 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 396 {
 397
 398     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 399                             sizeof(diag_501), 0) ||
 400         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)diag_501,
 401                             sizeof(diag_501), 1)) {
 402         return -EINVAL;
 403     }
 404     return 0;
 405 }
 406
 407 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
 408 {
 409     uint8_t t[sizeof(diag_501)];
 410
 411     if (cpu_memory_rw_debug(cs, bp->pc, t, sizeof(diag_501), 0)) {
 412         return -EINVAL;
 413     } else if (memcmp(t, diag_501, sizeof(diag_501))) {
 414         return -EINVAL;
 415     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
 416                                    sizeof(diag_501), 1)) {
 417         return -EINVAL;
 418     }
 419
 420     return 0;
 421 }
 422
 423 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
 424                                                     int len, int type)
 425 {
 426     int n;
 427
 428     for (n = 0; n < nb_hw_breakpoints; n++) {
 429         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
 430             (hw_breakpoints[n].len == len || len == -1)) {
 431             return &hw_breakpoints[n];
 432         }
 433     }
 434
 435     return NULL;
 436 }
 437
 438 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
 439 {
 440     int size;
 441
 442     if (find_hw_breakpoint(addr, len, type)) {
 443         return -EEXIST;
 444     }
 445
 446     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
 447
 448     if (!hw_breakpoints) {
 449         nb_hw_breakpoints = 0;
 450         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
 451     } else {
 452         hw_breakpoints =
 453             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
 454     }
 455
 456     if (!hw_breakpoints) {
 457         nb_hw_breakpoints = 0;
 458         return -ENOMEM;
 459     }
 460
 461     hw_breakpoints[nb_hw_breakpoints].addr = addr;
 462     hw_breakpoints[nb_hw_breakpoints].len = len;
 463     hw_breakpoints[nb_hw_breakpoints].type = type;
 464
 465     nb_hw_breakpoints++;
 466
 467     return 0;
 468 }
 469
 470 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
 471                                   target_ulong len, int type)
 472 {
 473     switch (type) {
 474     case GDB_BREAKPOINT_HW:
 475         type = KVM_HW_BP;
 476         break;
 477     case GDB_WATCHPOINT_WRITE:
 478         if (len < 1) {
 479             return -EINVAL;
 480         }
 481         type = KVM_HW_WP_WRITE;
 482         break;
 483     default:
 484         return -ENOSYS;
 485     }
 486     return insert_hw_breakpoint(addr, len, type);
 487 }
 488
 489 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
 490                                   target_ulong len, int type)
 491 {
 492     int size;
 493     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
 494
 495     if (bp == NULL) {
 496         return -ENOENT;
 497     }
 498
 499     nb_hw_breakpoints--;
 500     if (nb_hw_breakpoints > 0) {
 501         /*
 502          * In order to trim the array, move the last element to the position to
 503          * be removed - if necessary.
 504          */
 505         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
 506             *bp = hw_breakpoints[nb_hw_breakpoints];
 507         }
 508         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
 509         hw_breakpoints =
 510              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
 511     } else {
 512         g_free(hw_breakpoints);
 513         hw_breakpoints = NULL;
 514     }
 515
 516     return 0;
 517 }
 518
 519 void kvm_arch_remove_all_hw_breakpoints(void)
 520 {
 521     nb_hw_breakpoints = 0;
 522     g_free(hw_breakpoints);
 523     hw_breakpoints = NULL;
 524 }
 525
 526 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
 527 {
 528     int i;
 529
 530     if (nb_hw_breakpoints > 0) {
 531         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
 532         dbg->arch.hw_bp = hw_breakpoints;
 533
 534         for (i = 0; i < nb_hw_breakpoints; ++i) {
 535             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
 536                                                        hw_breakpoints[i].addr);
 537         }
 538         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
 539     } else {
 540         dbg->arch.nr_hw_bp = 0;
 541         dbg->arch.hw_bp = NULL;
 542     }
 543 }
 544
 545 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
 546 {
 547 }
 548
 549 void kvm_arch_post_run(CPUState *cpu, struct kvm_run *run)
 550 {
 551 }
 552
 553 int kvm_arch_process_async_events(CPUState *cs)
 554 {
 555     return cs->halted;
 556 }
 557
 558 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
 559                                      struct kvm_s390_interrupt *interrupt)
 560 {
 561     int r = 0;
 562
 563     interrupt->type = irq->type;
 564     switch (irq->type) {
 565     case KVM_S390_INT_VIRTIO:
 566         interrupt->parm = irq->u.ext.ext_params;
 567         /* fall through */
 568     case KVM_S390_INT_PFAULT_INIT:
 569     case KVM_S390_INT_PFAULT_DONE:
 570         interrupt->parm64 = irq->u.ext.ext_params2;
 571         break;
 572     case KVM_S390_PROGRAM_INT:
 573         interrupt->parm = irq->u.pgm.code;
 574         break;
 575     case KVM_S390_SIGP_SET_PREFIX:
 576         interrupt->parm = irq->u.prefix.address;
 577         break;
 578     case KVM_S390_INT_SERVICE:
 579         interrupt->parm = irq->u.ext.ext_params;
 580         break;
 581     case KVM_S390_MCHK:
 582         interrupt->parm = irq->u.mchk.cr14;
 583         interrupt->parm64 = irq->u.mchk.mcic;
 584         break;
 585     case KVM_S390_INT_EXTERNAL_CALL:
 586         interrupt->parm = irq->u.extcall.code;
 587         break;
 588     case KVM_S390_INT_EMERGENCY:
 589         interrupt->parm = irq->u.emerg.code;
 590         break;
 591     case KVM_S390_SIGP_STOP:
 592     case KVM_S390_RESTART:
 593         break; /* These types have no parameters */
 594     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
 595         interrupt->parm = irq->u.io.subchannel_id << 16;
 596         interrupt->parm |= irq->u.io.subchannel_nr;
 597         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
 598         interrupt->parm64 |= irq->u.io.io_int_word;
 599         break;
 600     default:
 601         r = -EINVAL;
 602         break;
 603     }
 604     return r;
 605 }
 606
 607 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
 608 {
 609     struct kvm_s390_interrupt kvmint = {};
 610     CPUState *cs = CPU(cpu);
 611     int r;
 612
 613     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 614     if (r < 0) {
 615         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 616         exit(1);
 617     }
 618
 619     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
 620     if (r < 0) {
 621         fprintf(stderr, "KVM failed to inject interrupt\n");
 622         exit(1);
 623     }
 624 }
 625
 626 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 627 {
 628     struct kvm_s390_interrupt kvmint = {};
 629     int r;
 630
 631     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
 632     if (r < 0) {
 633         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
 634         exit(1);
 635     }
 636
 637     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
 638     if (r < 0) {
 639         fprintf(stderr, "KVM failed to inject interrupt\n");
 640         exit(1);
 641     }
 642 }
 643
 644 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
 645 {
 646     static bool use_flic = true;
 647     int r;
 648
 649     if (use_flic) {
 650         r = kvm_s390_inject_flic(irq);
 651         if (r == -ENOSYS) {
 652             use_flic = false;
 653         }
 654         if (!r) {
 655             return;
 656         }
 657     }
 658     __kvm_s390_floating_interrupt(irq);
 659 }
 660
 661 void kvm_s390_virtio_irq(int config_change, uint64_t token)
 662 {
 663     struct kvm_s390_irq irq = {
 664         .type = KVM_S390_INT_VIRTIO,
 665         .u.ext.ext_params = config_change,
 666         .u.ext.ext_params2 = token,
 667     };
 668
 669     kvm_s390_floating_interrupt(&irq);
 670 }
 671
 672 void kvm_s390_service_interrupt(uint32_t parm)
 673 {
 674     struct kvm_s390_irq irq = {
 675         .type = KVM_S390_INT_SERVICE,
 676         .u.ext.ext_params = parm,
 677     };
 678
 679     kvm_s390_floating_interrupt(&irq);
 680 }
 681
 682 static void enter_pgmcheck(S390CPU *cpu, uint16_t code)
 683 {
 684     struct kvm_s390_irq irq = {
 685         .type = KVM_S390_PROGRAM_INT,
 686         .u.pgm.code = code,
 687     };
 688
 689     kvm_s390_vcpu_interrupt(cpu, &irq);
 690 }
 691
 692 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
 693                                  uint16_t ipbh0)
 694 {
 695     CPUS390XState *env = &cpu->env;
 696     uint64_t sccb;
 697     uint32_t code;
 698     int r = 0;
 699
 700     cpu_synchronize_state(CPU(cpu));
 701     sccb = env->regs[ipbh0 & 0xf];
 702     code = env->regs[(ipbh0 & 0xf0) >> 4];
 703
 704     r = sclp_service_call(env, sccb, code);
 705     if (r < 0) {
 706         enter_pgmcheck(cpu, -r);
 707     } else {
 708         setcc(cpu, r);
 709     }
 710
 711     return 0;
 712 }
 713
 714 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 715 {
 716     CPUS390XState *env = &cpu->env;
 717     int rc = 0;
 718     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
 719
 720     cpu_synchronize_state(CPU(cpu));
 721
 722     switch (ipa1) {
 723     case PRIV_B2_XSCH:
 724         ioinst_handle_xsch(cpu, env->regs[1]);
 725         break;
 726     case PRIV_B2_CSCH:
 727         ioinst_handle_csch(cpu, env->regs[1]);
 728         break;
 729     case PRIV_B2_HSCH:
 730         ioinst_handle_hsch(cpu, env->regs[1]);
 731         break;
 732     case PRIV_B2_MSCH:
 733         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
 734         break;
 735     case PRIV_B2_SSCH:
 736         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
 737         break;
 738     case PRIV_B2_STCRW:
 739         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
 740         break;
 741     case PRIV_B2_STSCH:
 742         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
 743         break;
 744     case PRIV_B2_TSCH:
 745         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
 746         fprintf(stderr, "Spurious tsch intercept\n");
 747         break;
 748     case PRIV_B2_CHSC:
 749         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
 750         break;
 751     case PRIV_B2_TPI:
 752         /* This should have been handled by kvm already. */
 753         fprintf(stderr, "Spurious tpi intercept\n");
 754         break;
 755     case PRIV_B2_SCHM:
 756         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
 757                            run->s390_sieic.ipb);
 758         break;
 759     case PRIV_B2_RSCH:
 760         ioinst_handle_rsch(cpu, env->regs[1]);
 761         break;
 762     case PRIV_B2_RCHP:
 763         ioinst_handle_rchp(cpu, env->regs[1]);
 764         break;
 765     case PRIV_B2_STCPS:
 766         /* We do not provide this instruction, it is suppressed. */
 767         break;
 768     case PRIV_B2_SAL:
 769         ioinst_handle_sal(cpu, env->regs[1]);
 770         break;
 771     case PRIV_B2_SIGA:
 772         /* Not provided, set CC = 3 for subchannel not operational */
 773         setcc(cpu, 3);
 774         break;
 775     case PRIV_B2_SCLP_CALL:
 776         rc = kvm_sclp_service_call(cpu, run, ipbh0);
 777         break;
 778     default:
 779         rc = -1;
 780         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
 781         break;
 782     }
 783
 784     return rc;
 785 }
 786
 787 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 788 {
 789     int r = 0;
 790
 791     switch (ipa1) {
 792     case PRIV_B9_EQBS:
 793         /* just inject exception */
 794         r = -1;
 795         break;
 796     default:
 797         r = -1;
 798         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
 799         break;
 800     }
 801
 802     return r;
 803 }
 804
 805 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 806 {
 807     int r = 0;
 808
 809     switch (ipa1) {
 810     case PRIV_EB_SQBS:
 811         /* just inject exception */
 812         r = -1;
 813         break;
 814     default:
 815         r = -1;
 816         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipa1);
 817         break;
 818     }
 819
 820     return r;
 821 }
 822
 823 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
 824 {
 825     CPUS390XState *env = &cpu->env;
 826     int ret;
 827
 828     cpu_synchronize_state(CPU(cpu));
 829     ret = s390_virtio_hypercall(env);
 830     if (ret == -EINVAL) {
 831         enter_pgmcheck(cpu, PGM_SPECIFICATION);
 832         return 0;
 833     }
 834
 835     return ret;
 836 }
 837
 838 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
 839 {
 840     uint64_t r1, r3;
 841
 842     cpu_synchronize_state(CPU(cpu));
 843     r1 = (run->s390_sieic.ipa & 0x00f0) >> 8;
 844     r3 = run->s390_sieic.ipa & 0x000f;
 845     handle_diag_308(&cpu->env, r1, r3);
 846 }
 847
 848 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
 849 {
 850     CPUS390XState *env = &cpu->env;
 851     unsigned long pc;
 852
 853     cpu_synchronize_state(CPU(cpu));
 854
 855     pc = env->psw.addr - 4;
 856     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
 857         env->psw.addr = pc;
 858         return EXCP_DEBUG;
 859     }
 860
 861     return -ENOENT;
 862 }
 863
 864 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
 865
 866 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
 867 {
 868     int r = 0;
 869     uint16_t func_code;
 870
 871     /*
 872      * For any diagnose call we support, bits 48-63 of the resulting
 873      * address specify the function code; the remainder is ignored.
 874      */
 875     func_code = decode_basedisp_rs(&cpu->env, ipb) & DIAG_KVM_CODE_MASK;
 876     switch (func_code) {
 877     case DIAG_IPL:
 878         kvm_handle_diag_308(cpu, run);
 879         break;
 880     case DIAG_KVM_HYPERCALL:
 881         r = handle_hypercall(cpu, run);
 882         break;
 883     case DIAG_KVM_BREAKPOINT:
 884         r = handle_sw_breakpoint(cpu, run);
 885         break;
 886     default:
 887         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
 888         r = -1;
 889         break;
 890     }
 891
 892     return r;
 893 }
 894
 895 static int kvm_s390_cpu_start(S390CPU *cpu)
 896 {
 897     s390_add_running_cpu(cpu);
 898     qemu_cpu_kick(CPU(cpu));
 899     DPRINTF("DONE: KVM cpu start: %p\n", &cpu->env);
 900     return 0;
 901 }
 902
 903 int kvm_s390_cpu_restart(S390CPU *cpu)
 904 {
 905     struct kvm_s390_irq irq = {
 906         .type = KVM_S390_RESTART,
 907     };
 908
 909     kvm_s390_vcpu_interrupt(cpu, &irq);
 910     s390_add_running_cpu(cpu);
 911     qemu_cpu_kick(CPU(cpu));
 912     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
 913     return 0;
 914 }
 915
 916 static void sigp_initial_cpu_reset(void *arg)
 917 {
 918     CPUState *cpu = arg;
 919     S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
 920
 921     cpu_synchronize_state(cpu);
 922     scc->initial_cpu_reset(cpu);
 923 }
 924
 925 static void sigp_cpu_reset(void *arg)
 926 {
 927     CPUState *cpu = arg;
 928     S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
 929
 930     cpu_synchronize_state(cpu);
 931     scc->cpu_reset(cpu);
 932 }
 933
 934 #define SIGP_ORDER_MASK 0x000000ff
 935
 936 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
 937 {
 938     CPUS390XState *env = &cpu->env;
 939     uint8_t order_code;
 940     uint16_t cpu_addr;
 941     S390CPU *target_cpu;
 942     uint64_t *statusreg = &env->regs[ipa1 >> 4];
 943     int cc;
 944
 945     cpu_synchronize_state(CPU(cpu));
 946
 947     /* get order code */
 948     order_code = decode_basedisp_rs(env, run->s390_sieic.ipb) & SIGP_ORDER_MASK;
 949
 950     cpu_addr = env->regs[ipa1 & 0x0f];
 951     target_cpu = s390_cpu_addr2state(cpu_addr);
 952     if (target_cpu == NULL) {
 953         cc = 3;    /* not operational */
 954         goto out;
 955     }
 956
 957     switch (order_code) {
 958     case SIGP_START:
 959         cc = kvm_s390_cpu_start(target_cpu);
 960         break;
 961     case SIGP_RESTART:
 962         cc = kvm_s390_cpu_restart(target_cpu);
 963         break;
 964     case SIGP_SET_ARCH:
 965         *statusreg &= 0xffffffff00000000UL;
 966         *statusreg |= SIGP_STAT_INVALID_PARAMETER;
 967         cc = 1;   /* status stored */
 968         break;
 969     case SIGP_INITIAL_CPU_RESET:
 970         run_on_cpu(CPU(target_cpu), sigp_initial_cpu_reset, CPU(target_cpu));
 971         cc = 0;
 972         break;
 973     case SIGP_CPU_RESET:
 974         run_on_cpu(CPU(target_cpu), sigp_cpu_reset, CPU(target_cpu));
 975         cc = 0;
 976         break;
 977     default:
 978         DPRINTF("KVM: unknown SIGP: 0x%x\n", order_code);
 979         *statusreg &= 0xffffffff00000000UL;
 980         *statusreg |= SIGP_STAT_INVALID_ORDER;
 981         cc = 1;   /* status stored */
 982         break;
 983     }
 984
 985 out:
 986     setcc(cpu, cc);
 987     return 0;
 988 }
 989
 990 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
 991 {
 992     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
 993     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
 994     int r = -1;
 995
 996     DPRINTF("handle_instruction 0x%x 0x%x\n",
 997             run->s390_sieic.ipa, run->s390_sieic.ipb);
 998     switch (ipa0) {
 999     case IPA0_B2:
1000         r = handle_b2(cpu, run, ipa1);
1001         break;
1002     case IPA0_B9:
1003         r = handle_b9(cpu, run, ipa1);
1004         break;
1005     case IPA0_EB:
1006         r = handle_eb(cpu, run, ipa1);
1007         break;
1008     case IPA0_DIAG:
1009         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1010         break;
1011     case IPA0_SIGP:
1012         r = handle_sigp(cpu, run, ipa1);
1013         break;
1014     }
1015
1016     if (r < 0) {
1017         r = 0;
1018         enter_pgmcheck(cpu, 0x0001);
1019     }
1020
1021     return r;
1022 }
1023
1024 static bool is_special_wait_psw(CPUState *cs)
1025 {
1026     /* signal quiesce */
1027     return cs->kvm_run->psw_addr == 0xfffUL;
1028 }
1029
1030 static void guest_panicked(void)
1031 {
1032     QObject *data;
1033
1034     data = qobject_from_jsonf("{ 'action': %s }", "pause");
1035     monitor_protocol_event(QEVENT_GUEST_PANICKED, data);
1036     qobject_decref(data);
1037
1038     vm_stop(RUN_STATE_GUEST_PANICKED);
1039 }
1040
1041 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1042 {
1043     CPUState *cs = CPU(cpu);
1044
1045     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1046                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1047                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1048     s390_del_running_cpu(cpu);
1049     guest_panicked();
1050 }
1051
1052 static int handle_intercept(S390CPU *cpu)
1053 {
1054     CPUState *cs = CPU(cpu);
1055     struct kvm_run *run = cs->kvm_run;
1056     int icpt_code = run->s390_sieic.icptcode;
1057     int r = 0;
1058
1059     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1060             (long)cs->kvm_run->psw_addr);
1061     switch (icpt_code) {
1062         case ICPT_INSTRUCTION:
1063             r = handle_instruction(cpu, run);
1064             break;
1065         case ICPT_PROGRAM:
1066             unmanageable_intercept(cpu, "program interrupt",
1067                                    offsetof(LowCore, program_new_psw));
1068             r = EXCP_HALTED;
1069             break;
1070         case ICPT_EXT_INT:
1071             unmanageable_intercept(cpu, "external interrupt",
1072                                    offsetof(LowCore, external_new_psw));
1073             r = EXCP_HALTED;
1074             break;
1075         case ICPT_WAITPSW:
1076             /* disabled wait, since enabled wait is handled in kernel */
1077             if (s390_del_running_cpu(cpu) == 0) {
1078                 if (is_special_wait_psw(cs)) {
1079                     qemu_system_shutdown_request();
1080                 } else {
1081                     guest_panicked();
1082                 }
1083             }
1084             r = EXCP_HALTED;
1085             break;
1086         case ICPT_CPU_STOP:
1087             if (s390_del_running_cpu(cpu) == 0) {
1088                 qemu_system_shutdown_request();
1089             }
1090             r = EXCP_HALTED;
1091             break;
1092         case ICPT_SOFT_INTERCEPT:
1093             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1094             exit(1);
1095             break;
1096         case ICPT_IO:
1097             fprintf(stderr, "KVM unimplemented icpt IO\n");
1098             exit(1);
1099             break;
1100         default:
1101             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1102             exit(1);
1103             break;
1104     }
1105
1106     return r;
1107 }
1108
1109 static int handle_tsch(S390CPU *cpu)
1110 {
1111     CPUS390XState *env = &cpu->env;
1112     CPUState *cs = CPU(cpu);
1113     struct kvm_run *run = cs->kvm_run;
1114     int ret;
1115
1116     cpu_synchronize_state(cs);
1117
1118     ret = ioinst_handle_tsch(env, env->regs[1], run->s390_tsch.ipb);
1119     if (ret >= 0) {
1120         /* Success; set condition code. */
1121         setcc(cpu, ret);
1122         ret = 0;
1123     } else if (ret < -1) {
1124         /*
1125          * Failure.
1126          * If an I/O interrupt had been dequeued, we have to reinject it.
1127          */
1128         if (run->s390_tsch.dequeued) {
1129             kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
1130                                   run->s390_tsch.subchannel_nr,
1131                                   run->s390_tsch.io_int_parm,
1132                                   run->s390_tsch.io_int_word);
1133         }
1134         ret = 0;
1135     }
1136     return ret;
1137 }
1138
1139 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1140 {
1141     CPUState *cs = CPU(cpu);
1142     struct kvm_run *run = cs->kvm_run;
1143
1144     int ret = 0;
1145     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1146
1147     switch (arch_info->type) {
1148     case KVM_HW_WP_WRITE:
1149         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1150             cs->watchpoint_hit = &hw_watchpoint;
1151             hw_watchpoint.vaddr = arch_info->addr;
1152             hw_watchpoint.flags = BP_MEM_WRITE;
1153             ret = EXCP_DEBUG;
1154         }
1155         break;
1156     case KVM_HW_BP:
1157         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1158             ret = EXCP_DEBUG;
1159         }
1160         break;
1161     case KVM_SINGLESTEP:
1162         if (cs->singlestep_enabled) {
1163             ret = EXCP_DEBUG;
1164         }
1165         break;
1166     default:
1167         ret = -ENOSYS;
1168     }
1169
1170     return ret;
1171 }
1172
1173 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1174 {
1175     S390CPU *cpu = S390_CPU(cs);
1176     int ret = 0;
1177
1178     switch (run->exit_reason) {
1179         case KVM_EXIT_S390_SIEIC:
1180             ret = handle_intercept(cpu);
1181             break;
1182         case KVM_EXIT_S390_RESET:
1183             qemu_system_reset_request();
1184             break;
1185         case KVM_EXIT_S390_TSCH:
1186             ret = handle_tsch(cpu);
1187             break;
1188         case KVM_EXIT_DEBUG:
1189             ret = kvm_arch_handle_debug_exit(cpu);
1190             break;
1191         default:
1192             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
1193             break;
1194     }
1195
1196     if (ret == 0) {
1197         ret = EXCP_INTERRUPT;
1198     }
1199     return ret;
1200 }
1201
1202 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
1203 {
1204     return true;
1205 }
1206
1207 int kvm_arch_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
1208 {
1209     return 1;
1210 }
1211
1212 int kvm_arch_on_sigbus(int code, void *addr)
1213 {
1214     return 1;
1215 }
1216
1217 void kvm_s390_io_interrupt(uint16_t subchannel_id,
1218                            uint16_t subchannel_nr, uint32_t io_int_parm,
1219                            uint32_t io_int_word)
1220 {
1221     struct kvm_s390_irq irq = {
1222         .u.io.subchannel_id = subchannel_id,
1223         .u.io.subchannel_nr = subchannel_nr,
1224         .u.io.io_int_parm = io_int_parm,
1225         .u.io.io_int_word = io_int_word,
1226     };
1227
1228     if (io_int_word & IO_INT_WORD_AI) {
1229         irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
1230     } else {
1231         irq.type = ((subchannel_id & 0xff00) << 24) |
1232             ((subchannel_id & 0x00060) << 22) | (subchannel_nr << 16);
1233     }
1234     kvm_s390_floating_interrupt(&irq);
1235 }
1236
1237 void kvm_s390_crw_mchk(void)
1238 {
1239     struct kvm_s390_irq irq = {
1240         .type = KVM_S390_MCHK,
1241         .u.mchk.cr14 = 1 << 28,
1242         .u.mchk.mcic = 0x00400f1d40330000,
1243     };
1244     kvm_s390_floating_interrupt(&irq);
1245 }
1246
1247 void kvm_s390_enable_css_support(S390CPU *cpu)
1248 {
1249     int r;
1250
1251     /* Activate host kernel channel subsystem support. */
1252     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
1253     assert(r == 0);
1254 }
1255
1256 void kvm_arch_init_irq_routing(KVMState *s)
1257 {
1258     /*
1259      * Note that while irqchip capabilities generally imply that cpustates
1260      * are handled in-kernel, it is not true for s390 (yet); therefore, we
1261      * have to override the common code kvm_halt_in_kernel_allowed setting.
1262      */
1263     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
1264         kvm_irqfds_allowed = true;
1265         kvm_gsi_routing_allowed = true;
1266         kvm_halt_in_kernel_allowed = false;
1267     }
1268 }
1269
1270 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
1271                                     int vq, bool assign)
1272 {
1273     struct kvm_ioeventfd kick = {
1274         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
1275         KVM_IOEVENTFD_FLAG_DATAMATCH,
1276         .fd = event_notifier_get_fd(notifier),
1277         .datamatch = vq,
1278         .addr = sch,
1279         .len = 8,
1280     };
1281     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
1282         return -ENOSYS;
1283     }
1284     if (!assign) {
1285         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
1286     }
1287     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
1288 }