arch/s390/kvm/kvm-s390.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * hosting IBM Z kernel virtual machines (s390x)
   4  *
   5  * Copyright IBM Corp. 2008, 2017
   6  *
   7  *    Author(s): Carsten Otte <cotte@de.ibm.com>
   8  *               Christian Borntraeger <borntraeger@de.ibm.com>
   9  *               Heiko Carstens <heiko.carstens@de.ibm.com>
  10  *               Christian Ehrhardt <ehrhardt@de.ibm.com>
  11  *               Jason J. Herne <jjherne@us.ibm.com>
  12  */
  13
  14 #include <linux/compiler.h>
  15 #include <linux/err.h>
  16 #include <linux/fs.h>
  17 #include <linux/hrtimer.h>
  18 #include <linux/init.h>
  19 #include <linux/kvm.h>
  20 #include <linux/kvm_host.h>
  21 #include <linux/mman.h>
  22 #include <linux/module.h>
  23 #include <linux/moduleparam.h>
  24 #include <linux/random.h>
  25 #include <linux/slab.h>
  26 #include <linux/timer.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/bitmap.h>
  29 #include <linux/sched/signal.h>
  30 #include <linux/string.h>
  31
  32 #include <asm/asm-offsets.h>
  33 #include <asm/lowcore.h>
  34 #include <asm/stp.h>
  35 #include <asm/pgtable.h>
  36 #include <asm/gmap.h>
  37 #include <asm/nmi.h>
  38 #include <asm/switch_to.h>
  39 #include <asm/isc.h>
  40 #include <asm/sclp.h>
  41 #include <asm/cpacf.h>
  42 #include <asm/timex.h>
  43 #include "kvm-s390.h"
  44 #include "gaccess.h"
  45
  46 #define KMSG_COMPONENT "kvm-s390"
  47 #undef pr_fmt
  48 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  49
  50 #define CREATE_TRACE_POINTS
  51 #include "trace.h"
  52 #include "trace-s390.h"
  53
  54 #define MEM_OP_MAX_SIZE 65536   /* Maximum transfer size for KVM_S390_MEM_OP */
  55 #define LOCAL_IRQS 32
  56 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
  57                            (KVM_MAX_VCPUS + LOCAL_IRQS))
  58
  59 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
  60
  61 struct kvm_stats_debugfs_item debugfs_entries[] = {
  62         { "userspace_handled", VCPU_STAT(exit_userspace) },
  63         { "exit_null", VCPU_STAT(exit_null) },
  64         { "exit_validity", VCPU_STAT(exit_validity) },
  65         { "exit_stop_request", VCPU_STAT(exit_stop_request) },
  66         { "exit_external_request", VCPU_STAT(exit_external_request) },
  67         { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
  68         { "exit_instruction", VCPU_STAT(exit_instruction) },
  69         { "exit_pei", VCPU_STAT(exit_pei) },
  70         { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
  71         { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
  72         { "exit_operation_exception", VCPU_STAT(exit_operation_exception) },
  73         { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
  74         { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
  75         { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
  76         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  77         { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
  78         { "instruction_lctl", VCPU_STAT(instruction_lctl) },
  79         { "instruction_stctl", VCPU_STAT(instruction_stctl) },
  80         { "instruction_stctg", VCPU_STAT(instruction_stctg) },
  81         { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
  82         { "deliver_external_call", VCPU_STAT(deliver_external_call) },
  83         { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
  84         { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) },
  85         { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
  86         { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
  87         { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
  88         { "deliver_program_interruption", VCPU_STAT(deliver_program_int) },
  89         { "exit_wait_state", VCPU_STAT(exit_wait_state) },
  90         { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
  91         { "instruction_stidp", VCPU_STAT(instruction_stidp) },
  92         { "instruction_spx", VCPU_STAT(instruction_spx) },
  93         { "instruction_stpx", VCPU_STAT(instruction_stpx) },
  94         { "instruction_stap", VCPU_STAT(instruction_stap) },
  95         { "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
  96         { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
  97         { "instruction_stsch", VCPU_STAT(instruction_stsch) },
  98         { "instruction_chsc", VCPU_STAT(instruction_chsc) },
  99         { "instruction_essa", VCPU_STAT(instruction_essa) },
 100         { "instruction_stsi", VCPU_STAT(instruction_stsi) },
 101         { "instruction_stfl", VCPU_STAT(instruction_stfl) },
 102         { "instruction_tprot", VCPU_STAT(instruction_tprot) },
 103         { "instruction_sthyi", VCPU_STAT(instruction_sthyi) },
 104         { "instruction_sie", VCPU_STAT(instruction_sie) },
 105         { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
 106         { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
 107         { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
 108         { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
 109         { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
 110         { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
 111         { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
 112         { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
 113         { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
 114         { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
 115         { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
 116         { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
 117         { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
 118         { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
 119         { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
 120         { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
 121         { "diagnose_10", VCPU_STAT(diagnose_10) },
 122         { "diagnose_44", VCPU_STAT(diagnose_44) },
 123         { "diagnose_9c", VCPU_STAT(diagnose_9c) },
 124         { "diagnose_258", VCPU_STAT(diagnose_258) },
 125         { "diagnose_308", VCPU_STAT(diagnose_308) },
 126         { "diagnose_500", VCPU_STAT(diagnose_500) },
 127         { NULL }
 128 };
 129
 130 struct kvm_s390_tod_clock_ext {
 131         __u8 epoch_idx;
 132         __u64 tod;
 133         __u8 reserved[7];
 134 } __packed;
 135
 136 /* allow nested virtualization in KVM (if enabled by user space) */
 137 static int nested;
 138 module_param(nested, int, S_IRUGO);
 139 MODULE_PARM_DESC(nested, "Nested virtualization support");
 140
 141 /* upper facilities limit for kvm */
 142 unsigned long kvm_s390_fac_list_mask[16] = { FACILITIES_KVM };
 143
 144 unsigned long kvm_s390_fac_list_mask_size(void)
 145 {
 146         BUILD_BUG_ON(ARRAY_SIZE(kvm_s390_fac_list_mask) > S390_ARCH_FAC_MASK_SIZE_U64);
 147         return ARRAY_SIZE(kvm_s390_fac_list_mask);
 148 }
 149
 150 /* available cpu features supported by kvm */
 151 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
 152 /* available subfunctions indicated via query / "test bit" */
 153 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
 154
 155 static struct gmap_notifier gmap_notifier;
 156 static struct gmap_notifier vsie_gmap_notifier;
 157 debug_info_t *kvm_s390_dbf;
 158
 159 /* Section: not file related */
 160 int kvm_arch_hardware_enable(void)
 161 {
 162         /* every s390 is virtualization enabled ;-) */
 163         return 0;
 164 }
 165
 166 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
 167                               unsigned long end);
 168
 169 /*
 170  * This callback is executed during stop_machine(). All CPUs are therefore
 171  * temporarily stopped. In order not to change guest behavior, we have to
 172  * disable preemption whenever we touch the epoch of kvm and the VCPUs,
 173  * so a CPU won't be stopped while calculating with the epoch.
 174  */
 175 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
 176                           void *v)
 177 {
 178         struct kvm *kvm;
 179         struct kvm_vcpu *vcpu;
 180         int i;
 181         unsigned long long *delta = v;
 182
 183         list_for_each_entry(kvm, &vm_list, vm_list) {
 184                 kvm->arch.epoch -= *delta;
 185                 kvm_for_each_vcpu(i, vcpu, kvm) {
 186                         vcpu->arch.sie_block->epoch -= *delta;
 187                         if (vcpu->arch.cputm_enabled)
 188                                 vcpu->arch.cputm_start += *delta;
 189                         if (vcpu->arch.vsie_block)
 190                                 vcpu->arch.vsie_block->epoch -= *delta;
 191                 }
 192         }
 193         return NOTIFY_OK;
 194 }
 195
 196 static struct notifier_block kvm_clock_notifier = {
 197         .notifier_call = kvm_clock_sync,
 198 };
 199
 200 int kvm_arch_hardware_setup(void)
 201 {
 202         gmap_notifier.notifier_call = kvm_gmap_notifier;
 203         gmap_register_pte_notifier(&gmap_notifier);
 204         vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
 205         gmap_register_pte_notifier(&vsie_gmap_notifier);
 206         atomic_notifier_chain_register(&s390_epoch_delta_notifier,
 207                                        &kvm_clock_notifier);
 208         return 0;
 209 }
 210
 211 void kvm_arch_hardware_unsetup(void)
 212 {
 213         gmap_unregister_pte_notifier(&gmap_notifier);
 214         gmap_unregister_pte_notifier(&vsie_gmap_notifier);
 215         atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
 216                                          &kvm_clock_notifier);
 217 }
 218
 219 static void allow_cpu_feat(unsigned long nr)
 220 {
 221         set_bit_inv(nr, kvm_s390_available_cpu_feat);
 222 }
 223
 224 static inline int plo_test_bit(unsigned char nr)
 225 {
 226         register unsigned long r0 asm("0") = (unsigned long) nr | 0x100;
 227         int cc;
 228
 229         asm volatile(
 230                 /* Parameter registers are ignored for "test bit" */
 231                 "       plo     0,0,0,0(0)\n"
 232                 "       ipm     %0\n"
 233                 "       srl     %0,28\n"
 234                 : "=d" (cc)
 235                 : "d" (r0)
 236                 : "cc");
 237         return cc == 0;
 238 }
 239
 240 static void kvm_s390_cpu_feat_init(void)
 241 {
 242         int i;
 243
 244         for (i = 0; i < 256; ++i) {
 245                 if (plo_test_bit(i))
 246                         kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
 247         }
 248
 249         if (test_facility(28)) /* TOD-clock steering */
 250                 ptff(kvm_s390_available_subfunc.ptff,
 251                      sizeof(kvm_s390_available_subfunc.ptff),
 252                      PTFF_QAF);
 253
 254         if (test_facility(17)) { /* MSA */
 255                 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
 256                               kvm_s390_available_subfunc.kmac);
 257                 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
 258                               kvm_s390_available_subfunc.kmc);
 259                 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
 260                               kvm_s390_available_subfunc.km);
 261                 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
 262                               kvm_s390_available_subfunc.kimd);
 263                 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
 264                               kvm_s390_available_subfunc.klmd);
 265         }
 266         if (test_facility(76)) /* MSA3 */
 267                 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
 268                               kvm_s390_available_subfunc.pckmo);
 269         if (test_facility(77)) { /* MSA4 */
 270                 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
 271                               kvm_s390_available_subfunc.kmctr);
 272                 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
 273                               kvm_s390_available_subfunc.kmf);
 274                 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
 275                               kvm_s390_available_subfunc.kmo);
 276                 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
 277                               kvm_s390_available_subfunc.pcc);
 278         }
 279         if (test_facility(57)) /* MSA5 */
 280                 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
 281                               kvm_s390_available_subfunc.ppno);
 282
 283         if (test_facility(146)) /* MSA8 */
 284                 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
 285                               kvm_s390_available_subfunc.kma);
 286
 287         if (MACHINE_HAS_ESOP)
 288                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
 289         /*
 290          * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
 291          * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
 292          */
 293         if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
 294             !test_facility(3) || !nested)
 295                 return;
 296         allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
 297         if (sclp.has_64bscao)
 298                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
 299         if (sclp.has_siif)
 300                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
 301         if (sclp.has_gpere)
 302                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
 303         if (sclp.has_gsls)
 304                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
 305         if (sclp.has_ib)
 306                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
 307         if (sclp.has_cei)
 308                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
 309         if (sclp.has_ibs)
 310                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
 311         if (sclp.has_kss)
 312                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
 313         /*
 314          * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
 315          * all skey handling functions read/set the skey from the PGSTE
 316          * instead of the real storage key.
 317          *
 318          * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
 319          * pages being detected as preserved although they are resident.
 320          *
 321          * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
 322          * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
 323          *
 324          * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
 325          * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
 326          * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
 327          *
 328          * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
 329          * cannot easily shadow the SCA because of the ipte lock.
 330          */
 331 }
 332
 333 int kvm_arch_init(void *opaque)
 334 {
 335         kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
 336         if (!kvm_s390_dbf)
 337                 return -ENOMEM;
 338
 339         if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view)) {
 340                 debug_unregister(kvm_s390_dbf);
 341                 return -ENOMEM;
 342         }
 343
 344         kvm_s390_cpu_feat_init();
 345
 346         /* Register floating interrupt controller interface. */
 347         return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
 348 }
 349
 350 void kvm_arch_exit(void)
 351 {
 352         debug_unregister(kvm_s390_dbf);
 353 }
 354
 355 /* Section: device related */
 356 long kvm_arch_dev_ioctl(struct file *filp,
 357                         unsigned int ioctl, unsigned long arg)
 358 {
 359         if (ioctl == KVM_S390_ENABLE_SIE)
 360                 return s390_enable_sie();
 361         return -EINVAL;
 362 }
 363
 364 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 365 {
 366         int r;
 367
 368         switch (ext) {
 369         case KVM_CAP_S390_PSW:
 370         case KVM_CAP_S390_GMAP:
 371         case KVM_CAP_SYNC_MMU:
 372 #ifdef CONFIG_KVM_S390_UCONTROL
 373         case KVM_CAP_S390_UCONTROL:
 374 #endif
 375         case KVM_CAP_ASYNC_PF:
 376         case KVM_CAP_SYNC_REGS:
 377         case KVM_CAP_ONE_REG:
 378         case KVM_CAP_ENABLE_CAP:
 379         case KVM_CAP_S390_CSS_SUPPORT:
 380         case KVM_CAP_IOEVENTFD:
 381         case KVM_CAP_DEVICE_CTRL:
 382         case KVM_CAP_ENABLE_CAP_VM:
 383         case KVM_CAP_S390_IRQCHIP:
 384         case KVM_CAP_VM_ATTRIBUTES:
 385         case KVM_CAP_MP_STATE:
 386         case KVM_CAP_IMMEDIATE_EXIT:
 387         case KVM_CAP_S390_INJECT_IRQ:
 388         case KVM_CAP_S390_USER_SIGP:
 389         case KVM_CAP_S390_USER_STSI:
 390         case KVM_CAP_S390_SKEYS:
 391         case KVM_CAP_S390_IRQ_STATE:
 392         case KVM_CAP_S390_USER_INSTR0:
 393         case KVM_CAP_S390_CMMA_MIGRATION:
 394         case KVM_CAP_S390_AIS:
 395         case KVM_CAP_S390_AIS_MIGRATION:
 396                 r = 1;
 397                 break;
 398         case KVM_CAP_S390_MEM_OP:
 399                 r = MEM_OP_MAX_SIZE;
 400                 break;
 401         case KVM_CAP_NR_VCPUS:
 402         case KVM_CAP_MAX_VCPUS:
 403                 r = KVM_S390_BSCA_CPU_SLOTS;
 404                 if (!kvm_s390_use_sca_entries())
 405                         r = KVM_MAX_VCPUS;
 406                 else if (sclp.has_esca && sclp.has_64bscao)
 407                         r = KVM_S390_ESCA_CPU_SLOTS;
 408                 break;
 409         case KVM_CAP_NR_MEMSLOTS:
 410                 r = KVM_USER_MEM_SLOTS;
 411                 break;
 412         case KVM_CAP_S390_COW:
 413                 r = MACHINE_HAS_ESOP;
 414                 break;
 415         case KVM_CAP_S390_VECTOR_REGISTERS:
 416                 r = MACHINE_HAS_VX;
 417                 break;
 418         case KVM_CAP_S390_RI:
 419                 r = test_facility(64);
 420                 break;
 421         case KVM_CAP_S390_GS:
 422                 r = test_facility(133);
 423                 break;
 424         default:
 425                 r = 0;
 426         }
 427         return r;
 428 }
 429
 430 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
 431                                         struct kvm_memory_slot *memslot)
 432 {
 433         gfn_t cur_gfn, last_gfn;
 434         unsigned long address;
 435         struct gmap *gmap = kvm->arch.gmap;
 436
 437         /* Loop over all guest pages */
 438         last_gfn = memslot->base_gfn + memslot->npages;
 439         for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
 440                 address = gfn_to_hva_memslot(memslot, cur_gfn);
 441
 442                 if (test_and_clear_guest_dirty(gmap->mm, address))
 443                         mark_page_dirty(kvm, cur_gfn);
 444                 if (fatal_signal_pending(current))
 445                         return;
 446                 cond_resched();
 447         }
 448 }
 449
 450 /* Section: vm related */
 451 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
 452
 453 /*
 454  * Get (and clear) the dirty memory log for a memory slot.
 455  */
 456 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 457                                struct kvm_dirty_log *log)
 458 {
 459         int r;
 460         unsigned long n;
 461         struct kvm_memslots *slots;
 462         struct kvm_memory_slot *memslot;
 463         int is_dirty = 0;
 464
 465         if (kvm_is_ucontrol(kvm))
 466                 return -EINVAL;
 467
 468         mutex_lock(&kvm->slots_lock);
 469
 470         r = -EINVAL;
 471         if (log->slot >= KVM_USER_MEM_SLOTS)
 472                 goto out;
 473
 474         slots = kvm_memslots(kvm);
 475         memslot = id_to_memslot(slots, log->slot);
 476         r = -ENOENT;
 477         if (!memslot->dirty_bitmap)
 478                 goto out;
 479
 480         kvm_s390_sync_dirty_log(kvm, memslot);
 481         r = kvm_get_dirty_log(kvm, log, &is_dirty);
 482         if (r)
 483                 goto out;
 484
 485         /* Clear the dirty log */
 486         if (is_dirty) {
 487                 n = kvm_dirty_bitmap_bytes(memslot);
 488                 memset(memslot->dirty_bitmap, 0, n);
 489         }
 490         r = 0;
 491 out:
 492         mutex_unlock(&kvm->slots_lock);
 493         return r;
 494 }
 495
 496 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
 497 {
 498         unsigned int i;
 499         struct kvm_vcpu *vcpu;
 500
 501         kvm_for_each_vcpu(i, vcpu, kvm) {
 502                 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
 503         }
 504 }
 505
 506 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
 507 {
 508         int r;
 509
 510         if (cap->flags)
 511                 return -EINVAL;
 512
 513         switch (cap->cap) {
 514         case KVM_CAP_S390_IRQCHIP:
 515                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
 516                 kvm->arch.use_irqchip = 1;
 517                 r = 0;
 518                 break;
 519         case KVM_CAP_S390_USER_SIGP:
 520                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
 521                 kvm->arch.user_sigp = 1;
 522                 r = 0;
 523                 break;
 524         case KVM_CAP_S390_VECTOR_REGISTERS:
 525                 mutex_lock(&kvm->lock);
 526                 if (kvm->created_vcpus) {
 527                         r = -EBUSY;
 528                 } else if (MACHINE_HAS_VX) {
 529                         set_kvm_facility(kvm->arch.model.fac_mask, 129);
 530                         set_kvm_facility(kvm->arch.model.fac_list, 129);
 531                         if (test_facility(134)) {
 532                                 set_kvm_facility(kvm->arch.model.fac_mask, 134);
 533                                 set_kvm_facility(kvm->arch.model.fac_list, 134);
 534                         }
 535                         if (test_facility(135)) {
 536                                 set_kvm_facility(kvm->arch.model.fac_mask, 135);
 537                                 set_kvm_facility(kvm->arch.model.fac_list, 135);
 538                         }
 539                         r = 0;
 540                 } else
 541                         r = -EINVAL;
 542                 mutex_unlock(&kvm->lock);
 543                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
 544                          r ? "(not available)" : "(success)");
 545                 break;
 546         case KVM_CAP_S390_RI:
 547                 r = -EINVAL;
 548                 mutex_lock(&kvm->lock);
 549                 if (kvm->created_vcpus) {
 550                         r = -EBUSY;
 551                 } else if (test_facility(64)) {
 552                         set_kvm_facility(kvm->arch.model.fac_mask, 64);
 553                         set_kvm_facility(kvm->arch.model.fac_list, 64);
 554                         r = 0;
 555                 }
 556                 mutex_unlock(&kvm->lock);
 557                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
 558                          r ? "(not available)" : "(success)");
 559                 break;
 560         case KVM_CAP_S390_AIS:
 561                 mutex_lock(&kvm->lock);
 562                 if (kvm->created_vcpus) {
 563                         r = -EBUSY;
 564                 } else {
 565                         set_kvm_facility(kvm->arch.model.fac_mask, 72);
 566                         set_kvm_facility(kvm->arch.model.fac_list, 72);
 567                         r = 0;
 568                 }
 569                 mutex_unlock(&kvm->lock);
 570                 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
 571                          r ? "(not available)" : "(success)");
 572                 break;
 573         case KVM_CAP_S390_GS:
 574                 r = -EINVAL;
 575                 mutex_lock(&kvm->lock);
 576                 if (atomic_read(&kvm->online_vcpus)) {
 577                         r = -EBUSY;
 578                 } else if (test_facility(133)) {
 579                         set_kvm_facility(kvm->arch.model.fac_mask, 133);
 580                         set_kvm_facility(kvm->arch.model.fac_list, 133);
 581                         r = 0;
 582                 }
 583                 mutex_unlock(&kvm->lock);
 584                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
 585                          r ? "(not available)" : "(success)");
 586                 break;
 587         case KVM_CAP_S390_USER_STSI:
 588                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
 589                 kvm->arch.user_stsi = 1;
 590                 r = 0;
 591                 break;
 592         case KVM_CAP_S390_USER_INSTR0:
 593                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
 594                 kvm->arch.user_instr0 = 1;
 595                 icpt_operexc_on_all_vcpus(kvm);
 596                 r = 0;
 597                 break;
 598         default:
 599                 r = -EINVAL;
 600                 break;
 601         }
 602         return r;
 603 }
 604
 605 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
 606 {
 607         int ret;
 608
 609         switch (attr->attr) {
 610         case KVM_S390_VM_MEM_LIMIT_SIZE:
 611                 ret = 0;
 612                 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
 613                          kvm->arch.mem_limit);
 614                 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
 615                         ret = -EFAULT;
 616                 break;
 617         default:
 618                 ret = -ENXIO;
 619                 break;
 620         }
 621         return ret;
 622 }
 623
 624 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
 625 {
 626         int ret;
 627         unsigned int idx;
 628         switch (attr->attr) {
 629         case KVM_S390_VM_MEM_ENABLE_CMMA:
 630                 ret = -ENXIO;
 631                 if (!sclp.has_cmma)
 632                         break;
 633
 634                 ret = -EBUSY;
 635                 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
 636                 mutex_lock(&kvm->lock);
 637                 if (!kvm->created_vcpus) {
 638                         kvm->arch.use_cmma = 1;
 639                         ret = 0;
 640                 }
 641                 mutex_unlock(&kvm->lock);
 642                 break;
 643         case KVM_S390_VM_MEM_CLR_CMMA:
 644                 ret = -ENXIO;
 645                 if (!sclp.has_cmma)
 646                         break;
 647                 ret = -EINVAL;
 648                 if (!kvm->arch.use_cmma)
 649                         break;
 650
 651                 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
 652                 mutex_lock(&kvm->lock);
 653                 idx = srcu_read_lock(&kvm->srcu);
 654                 s390_reset_cmma(kvm->arch.gmap->mm);
 655                 srcu_read_unlock(&kvm->srcu, idx);
 656                 mutex_unlock(&kvm->lock);
 657                 ret = 0;
 658                 break;
 659         case KVM_S390_VM_MEM_LIMIT_SIZE: {
 660                 unsigned long new_limit;
 661
 662                 if (kvm_is_ucontrol(kvm))
 663                         return -EINVAL;
 664
 665                 if (get_user(new_limit, (u64 __user *)attr->addr))
 666                         return -EFAULT;
 667
 668                 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
 669                     new_limit > kvm->arch.mem_limit)
 670                         return -E2BIG;
 671
 672                 if (!new_limit)
 673                         return -EINVAL;
 674
 675                 /* gmap_create takes last usable address */
 676                 if (new_limit != KVM_S390_NO_MEM_LIMIT)
 677                         new_limit -= 1;
 678
 679                 ret = -EBUSY;
 680                 mutex_lock(&kvm->lock);
 681                 if (!kvm->created_vcpus) {
 682                         /* gmap_create will round the limit up */
 683                         struct gmap *new = gmap_create(current->mm, new_limit);
 684
 685                         if (!new) {
 686                                 ret = -ENOMEM;
 687                         } else {
 688                                 gmap_remove(kvm->arch.gmap);
 689                                 new->private = kvm;
 690                                 kvm->arch.gmap = new;
 691                                 ret = 0;
 692                         }
 693                 }
 694                 mutex_unlock(&kvm->lock);
 695                 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
 696                 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
 697                          (void *) kvm->arch.gmap->asce);
 698                 break;
 699         }
 700         default:
 701                 ret = -ENXIO;
 702                 break;
 703         }
 704         return ret;
 705 }
 706
 707 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
 708
 709 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
 710 {
 711         struct kvm_vcpu *vcpu;
 712         int i;
 713
 714         if (!test_kvm_facility(kvm, 76))
 715                 return -EINVAL;
 716
 717         mutex_lock(&kvm->lock);
 718         switch (attr->attr) {
 719         case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
 720                 get_random_bytes(
 721                         kvm->arch.crypto.crycb->aes_wrapping_key_mask,
 722                         sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
 723                 kvm->arch.crypto.aes_kw = 1;
 724                 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
 725                 break;
 726         case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
 727                 get_random_bytes(
 728                         kvm->arch.crypto.crycb->dea_wrapping_key_mask,
 729                         sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
 730                 kvm->arch.crypto.dea_kw = 1;
 731                 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
 732                 break;
 733         case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
 734                 kvm->arch.crypto.aes_kw = 0;
 735                 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
 736                         sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
 737                 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
 738                 break;
 739         case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
 740                 kvm->arch.crypto.dea_kw = 0;
 741                 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
 742                         sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
 743                 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
 744                 break;
 745         default:
 746                 mutex_unlock(&kvm->lock);
 747                 return -ENXIO;
 748         }
 749
 750         kvm_for_each_vcpu(i, vcpu, kvm) {
 751                 kvm_s390_vcpu_crypto_setup(vcpu);
 752                 exit_sie(vcpu);
 753         }
 754         mutex_unlock(&kvm->lock);
 755         return 0;
 756 }
 757
 758 static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req)
 759 {
 760         int cx;
 761         struct kvm_vcpu *vcpu;
 762
 763         kvm_for_each_vcpu(cx, vcpu, kvm)
 764                 kvm_s390_sync_request(req, vcpu);
 765 }
 766
 767 /*
 768  * Must be called with kvm->srcu held to avoid races on memslots, and with
 769  * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration.
 770  */
 771 static int kvm_s390_vm_start_migration(struct kvm *kvm)
 772 {
 773         struct kvm_s390_migration_state *mgs;
 774         struct kvm_memory_slot *ms;
 775         /* should be the only one */
 776         struct kvm_memslots *slots;
 777         unsigned long ram_pages;
 778         int slotnr;
 779
 780         /* migration mode already enabled */
 781         if (kvm->arch.migration_state)
 782                 return 0;
 783
 784         slots = kvm_memslots(kvm);
 785         if (!slots || !slots->used_slots)
 786                 return -EINVAL;
 787
 788         mgs = kzalloc(sizeof(*mgs), GFP_KERNEL);
 789         if (!mgs)
 790                 return -ENOMEM;
 791         kvm->arch.migration_state = mgs;
 792
 793         if (kvm->arch.use_cmma) {
 794                 /*
 795                  * Get the first slot. They are reverse sorted by base_gfn, so
 796                  * the first slot is also the one at the end of the address
 797                  * space. We have verified above that at least one slot is
 798                  * present.
 799                  */
 800                 ms = slots->memslots;
 801                 /* round up so we only use full longs */
 802                 ram_pages = roundup(ms->base_gfn + ms->npages, BITS_PER_LONG);
 803                 /* allocate enough bytes to store all the bits */
 804                 mgs->pgste_bitmap = vmalloc(ram_pages / 8);
 805                 if (!mgs->pgste_bitmap) {
 806                         kfree(mgs);
 807                         kvm->arch.migration_state = NULL;
 808                         return -ENOMEM;
 809                 }
 810
 811                 mgs->bitmap_size = ram_pages;
 812                 atomic64_set(&mgs->dirty_pages, ram_pages);
 813                 /* mark all the pages in active slots as dirty */
 814                 for (slotnr = 0; slotnr < slots->used_slots; slotnr++) {
 815                         ms = slots->memslots + slotnr;
 816                         bitmap_set(mgs->pgste_bitmap, ms->base_gfn, ms->npages);
 817                 }
 818
 819                 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION);
 820         }
 821         return 0;
 822 }
 823
 824 /*
 825  * Must be called with kvm->slots_lock to avoid races with ourselves and
 826  * kvm_s390_vm_start_migration.
 827  */
 828 static int kvm_s390_vm_stop_migration(struct kvm *kvm)
 829 {
 830         struct kvm_s390_migration_state *mgs;
 831
 832         /* migration mode already disabled */
 833         if (!kvm->arch.migration_state)
 834                 return 0;
 835         mgs = kvm->arch.migration_state;
 836         kvm->arch.migration_state = NULL;
 837
 838         if (kvm->arch.use_cmma) {
 839                 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION);
 840                 /* We have to wait for the essa emulation to finish */
 841                 synchronize_srcu(&kvm->srcu);
 842                 vfree(mgs->pgste_bitmap);
 843         }
 844         kfree(mgs);
 845         return 0;
 846 }
 847
 848 static int kvm_s390_vm_set_migration(struct kvm *kvm,
 849                                      struct kvm_device_attr *attr)
 850 {
 851         int res = -ENXIO;
 852
 853         mutex_lock(&kvm->slots_lock);
 854         switch (attr->attr) {
 855         case KVM_S390_VM_MIGRATION_START:
 856                 res = kvm_s390_vm_start_migration(kvm);
 857                 break;
 858         case KVM_S390_VM_MIGRATION_STOP:
 859                 res = kvm_s390_vm_stop_migration(kvm);
 860                 break;
 861         default:
 862                 break;
 863         }
 864         mutex_unlock(&kvm->slots_lock);
 865
 866         return res;
 867 }
 868
 869 static int kvm_s390_vm_get_migration(struct kvm *kvm,
 870                                      struct kvm_device_attr *attr)
 871 {
 872         u64 mig = (kvm->arch.migration_state != NULL);
 873
 874         if (attr->attr != KVM_S390_VM_MIGRATION_STATUS)
 875                 return -ENXIO;
 876
 877         if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig)))
 878                 return -EFAULT;
 879         return 0;
 880 }
 881
 882 static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
 883 {
 884         struct kvm_s390_vm_tod_clock gtod;
 885
 886         if (copy_from_user(&gtod, (void __user *)attr->addr, sizeof(gtod)))
 887                 return -EFAULT;
 888
 889         if (test_kvm_facility(kvm, 139))
 890                 kvm_s390_set_tod_clock_ext(kvm, &gtod);
 891         else if (gtod.epoch_idx == 0)
 892                 kvm_s390_set_tod_clock(kvm, gtod.tod);
 893         else
 894                 return -EINVAL;
 895
 896         VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx",
 897                 gtod.epoch_idx, gtod.tod);
 898
 899         return 0;
 900 }
 901
 902 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
 903 {
 904         u8 gtod_high;
 905
 906         if (copy_from_user(&gtod_high, (void __user *)attr->addr,
 907                                            sizeof(gtod_high)))
 908                 return -EFAULT;
 909
 910         if (gtod_high != 0)
 911                 return -EINVAL;
 912         VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
 913
 914         return 0;
 915 }
 916
 917 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
 918 {
 919         u64 gtod;
 920
 921         if (copy_from_user(&gtod, (void __user *)attr->addr, sizeof(gtod)))
 922                 return -EFAULT;
 923
 924         kvm_s390_set_tod_clock(kvm, gtod);
 925         VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod);
 926         return 0;
 927 }
 928
 929 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
 930 {
 931         int ret;
 932
 933         if (attr->flags)
 934                 return -EINVAL;
 935
 936         switch (attr->attr) {
 937         case KVM_S390_VM_TOD_EXT:
 938                 ret = kvm_s390_set_tod_ext(kvm, attr);
 939                 break;
 940         case KVM_S390_VM_TOD_HIGH:
 941                 ret = kvm_s390_set_tod_high(kvm, attr);
 942                 break;
 943         case KVM_S390_VM_TOD_LOW:
 944                 ret = kvm_s390_set_tod_low(kvm, attr);
 945                 break;
 946         default:
 947                 ret = -ENXIO;
 948                 break;
 949         }
 950         return ret;
 951 }
 952
 953 static void kvm_s390_get_tod_clock_ext(struct kvm *kvm,
 954                                         struct kvm_s390_vm_tod_clock *gtod)
 955 {
 956         struct kvm_s390_tod_clock_ext htod;
 957
 958         preempt_disable();
 959
 960         get_tod_clock_ext((char *)&htod);
 961
 962         gtod->tod = htod.tod + kvm->arch.epoch;
 963         gtod->epoch_idx = htod.epoch_idx + kvm->arch.epdx;
 964
 965         if (gtod->tod < htod.tod)
 966                 gtod->epoch_idx += 1;
 967
 968         preempt_enable();
 969 }
 970
 971 static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
 972 {
 973         struct kvm_s390_vm_tod_clock gtod;
 974
 975         memset(&gtod, 0, sizeof(gtod));
 976
 977         if (test_kvm_facility(kvm, 139))
 978                 kvm_s390_get_tod_clock_ext(kvm, &gtod);
 979         else
 980                 gtod.tod = kvm_s390_get_tod_clock_fast(kvm);
 981
 982         if (copy_to_user((void __user *)attr->addr, &gtod, sizeof(gtod)))
 983                 return -EFAULT;
 984
 985         VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx",
 986                 gtod.epoch_idx, gtod.tod);
 987         return 0;
 988 }
 989
 990 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
 991 {
 992         u8 gtod_high = 0;
 993
 994         if (copy_to_user((void __user *)attr->addr, &gtod_high,
 995                                          sizeof(gtod_high)))
 996                 return -EFAULT;
 997         VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
 998
 999         return 0;
1000 }
1001
1002 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1003 {
1004         u64 gtod;
1005
1006         gtod = kvm_s390_get_tod_clock_fast(kvm);
1007         if (copy_to_user((void __user *)attr->addr, &gtod, sizeof(gtod)))
1008                 return -EFAULT;
1009         VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
1010
1011         return 0;
1012 }
1013
1014 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1015 {
1016         int ret;
1017
1018         if (attr->flags)
1019                 return -EINVAL;
1020
1021         switch (attr->attr) {
1022         case KVM_S390_VM_TOD_EXT:
1023                 ret = kvm_s390_get_tod_ext(kvm, attr);
1024                 break;
1025         case KVM_S390_VM_TOD_HIGH:
1026                 ret = kvm_s390_get_tod_high(kvm, attr);
1027                 break;
1028         case KVM_S390_VM_TOD_LOW:
1029                 ret = kvm_s390_get_tod_low(kvm, attr);
1030                 break;
1031         default:
1032                 ret = -ENXIO;
1033                 break;
1034         }
1035         return ret;
1036 }
1037
1038 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1039 {
1040         struct kvm_s390_vm_cpu_processor *proc;
1041         u16 lowest_ibc, unblocked_ibc;
1042         int ret = 0;
1043
1044         mutex_lock(&kvm->lock);
1045         if (kvm->created_vcpus) {
1046                 ret = -EBUSY;
1047                 goto out;
1048         }
1049         proc = kzalloc(sizeof(*proc), GFP_KERNEL);
1050         if (!proc) {
1051                 ret = -ENOMEM;
1052                 goto out;
1053         }
1054         if (!copy_from_user(proc, (void __user *)attr->addr,
1055                             sizeof(*proc))) {
1056                 kvm->arch.model.cpuid = proc->cpuid;
1057                 lowest_ibc = sclp.ibc >> 16 & 0xfff;
1058                 unblocked_ibc = sclp.ibc & 0xfff;
1059                 if (lowest_ibc && proc->ibc) {
1060                         if (proc->ibc > unblocked_ibc)
1061                                 kvm->arch.model.ibc = unblocked_ibc;
1062                         else if (proc->ibc < lowest_ibc)
1063                                 kvm->arch.model.ibc = lowest_ibc;
1064                         else
1065                                 kvm->arch.model.ibc = proc->ibc;
1066                 }
1067                 memcpy(kvm->arch.model.fac_list, proc->fac_list,
1068                        S390_ARCH_FAC_LIST_SIZE_BYTE);
1069                 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1070                          kvm->arch.model.ibc,
1071                          kvm->arch.model.cpuid);
1072                 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1073                          kvm->arch.model.fac_list[0],
1074                          kvm->arch.model.fac_list[1],
1075                          kvm->arch.model.fac_list[2]);
1076         } else
1077                 ret = -EFAULT;
1078         kfree(proc);
1079 out:
1080         mutex_unlock(&kvm->lock);
1081         return ret;
1082 }
1083
1084 static int kvm_s390_set_processor_feat(struct kvm *kvm,
1085                                        struct kvm_device_attr *attr)
1086 {
1087         struct kvm_s390_vm_cpu_feat data;
1088         int ret = -EBUSY;
1089
1090         if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
1091                 return -EFAULT;
1092         if (!bitmap_subset((unsigned long *) data.feat,
1093                            kvm_s390_available_cpu_feat,
1094                            KVM_S390_VM_CPU_FEAT_NR_BITS))
1095                 return -EINVAL;
1096
1097         mutex_lock(&kvm->lock);
1098         if (!atomic_read(&kvm->online_vcpus)) {
1099                 bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
1100                             KVM_S390_VM_CPU_FEAT_NR_BITS);
1101                 ret = 0;
1102         }
1103         mutex_unlock(&kvm->lock);
1104         return ret;
1105 }
1106
1107 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
1108                                           struct kvm_device_attr *attr)
1109 {
1110         /*
1111          * Once supported by kernel + hw, we have to store the subfunctions
1112          * in kvm->arch and remember that user space configured them.
1113          */
1114         return -ENXIO;
1115 }
1116
1117 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1118 {
1119         int ret = -ENXIO;
1120
1121         switch (attr->attr) {
1122         case KVM_S390_VM_CPU_PROCESSOR:
1123                 ret = kvm_s390_set_processor(kvm, attr);
1124                 break;
1125         case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1126                 ret = kvm_s390_set_processor_feat(kvm, attr);
1127                 break;
1128         case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1129                 ret = kvm_s390_set_processor_subfunc(kvm, attr);
1130                 break;
1131         }
1132         return ret;
1133 }
1134
1135 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1136 {
1137         struct kvm_s390_vm_cpu_processor *proc;
1138         int ret = 0;
1139
1140         proc = kzalloc(sizeof(*proc), GFP_KERNEL);
1141         if (!proc) {
1142                 ret = -ENOMEM;
1143                 goto out;
1144         }
1145         proc->cpuid = kvm->arch.model.cpuid;
1146         proc->ibc = kvm->arch.model.ibc;
1147         memcpy(&proc->fac_list, kvm->arch.model.fac_list,
1148                S390_ARCH_FAC_LIST_SIZE_BYTE);
1149         VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1150                  kvm->arch.model.ibc,
1151                  kvm->arch.model.cpuid);
1152         VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1153                  kvm->arch.model.fac_list[0],
1154                  kvm->arch.model.fac_list[1],
1155                  kvm->arch.model.fac_list[2]);
1156         if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
1157                 ret = -EFAULT;
1158         kfree(proc);
1159 out:
1160         return ret;
1161 }
1162
1163 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
1164 {
1165         struct kvm_s390_vm_cpu_machine *mach;
1166         int ret = 0;
1167
1168         mach = kzalloc(sizeof(*mach), GFP_KERNEL);
1169         if (!mach) {
1170                 ret = -ENOMEM;
1171                 goto out;
1172         }
1173         get_cpu_id((struct cpuid *) &mach->cpuid);
1174         mach->ibc = sclp.ibc;
1175         memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
1176                S390_ARCH_FAC_LIST_SIZE_BYTE);
1177         memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
1178                sizeof(S390_lowcore.stfle_fac_list));
1179         VM_EVENT(kvm, 3, "GET: host ibc:  0x%4.4x, host cpuid:  0x%16.16llx",
1180                  kvm->arch.model.ibc,
1181                  kvm->arch.model.cpuid);
1182         VM_EVENT(kvm, 3, "GET: host facmask:  0x%16.16llx.%16.16llx.%16.16llx",
1183                  mach->fac_mask[0],
1184                  mach->fac_mask[1],
1185                  mach->fac_mask[2]);
1186         VM_EVENT(kvm, 3, "GET: host faclist:  0x%16.16llx.%16.16llx.%16.16llx",
1187                  mach->fac_list[0],
1188                  mach->fac_list[1],
1189                  mach->fac_list[2]);
1190         if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1191                 ret = -EFAULT;
1192         kfree(mach);
1193 out:
1194         return ret;
1195 }
1196
1197 static int kvm_s390_get_processor_feat(struct kvm *kvm,
1198                                        struct kvm_device_attr *attr)
1199 {
1200         struct kvm_s390_vm_cpu_feat data;
1201
1202         bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
1203                     KVM_S390_VM_CPU_FEAT_NR_BITS);
1204         if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1205                 return -EFAULT;
1206         return 0;
1207 }
1208
1209 static int kvm_s390_get_machine_feat(struct kvm *kvm,
1210                                      struct kvm_device_attr *attr)
1211 {
1212         struct kvm_s390_vm_cpu_feat data;
1213
1214         bitmap_copy((unsigned long *) data.feat,
1215                     kvm_s390_available_cpu_feat,
1216                     KVM_S390_VM_CPU_FEAT_NR_BITS);
1217         if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1218                 return -EFAULT;
1219         return 0;
1220 }
1221
1222 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1223                                           struct kvm_device_attr *attr)
1224 {
1225         /*
1226          * Once we can actually configure subfunctions (kernel + hw support),
1227          * we have to check if they were already set by user space, if so copy
1228          * them from kvm->arch.
1229          */
1230         return -ENXIO;
1231 }
1232
1233 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1234                                         struct kvm_device_attr *attr)
1235 {
1236         if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1237             sizeof(struct kvm_s390_vm_cpu_subfunc)))
1238                 return -EFAULT;
1239         return 0;
1240 }
1241 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1242 {
1243         int ret = -ENXIO;
1244
1245         switch (attr->attr) {
1246         case KVM_S390_VM_CPU_PROCESSOR:
1247                 ret = kvm_s390_get_processor(kvm, attr);
1248                 break;
1249         case KVM_S390_VM_CPU_MACHINE:
1250                 ret = kvm_s390_get_machine(kvm, attr);
1251                 break;
1252         case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1253                 ret = kvm_s390_get_processor_feat(kvm, attr);
1254                 break;
1255         case KVM_S390_VM_CPU_MACHINE_FEAT:
1256                 ret = kvm_s390_get_machine_feat(kvm, attr);
1257                 break;
1258         case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1259                 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1260                 break;
1261         case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1262                 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1263                 break;
1264         }
1265         return ret;
1266 }
1267
1268 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1269 {
1270         int ret;
1271
1272         switch (attr->group) {
1273         case KVM_S390_VM_MEM_CTRL:
1274                 ret = kvm_s390_set_mem_control(kvm, attr);
1275                 break;
1276         case KVM_S390_VM_TOD:
1277                 ret = kvm_s390_set_tod(kvm, attr);
1278                 break;
1279         case KVM_S390_VM_CPU_MODEL:
1280                 ret = kvm_s390_set_cpu_model(kvm, attr);
1281                 break;
1282         case KVM_S390_VM_CRYPTO:
1283                 ret = kvm_s390_vm_set_crypto(kvm, attr);
1284                 break;
1285         case KVM_S390_VM_MIGRATION:
1286                 ret = kvm_s390_vm_set_migration(kvm, attr);
1287                 break;
1288         default:
1289                 ret = -ENXIO;
1290                 break;
1291         }
1292
1293         return ret;
1294 }
1295
1296 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1297 {
1298         int ret;
1299
1300         switch (attr->group) {
1301         case KVM_S390_VM_MEM_CTRL:
1302                 ret = kvm_s390_get_mem_control(kvm, attr);
1303                 break;
1304         case KVM_S390_VM_TOD:
1305                 ret = kvm_s390_get_tod(kvm, attr);
1306                 break;
1307         case KVM_S390_VM_CPU_MODEL:
1308                 ret = kvm_s390_get_cpu_model(kvm, attr);
1309                 break;
1310         case KVM_S390_VM_MIGRATION:
1311                 ret = kvm_s390_vm_get_migration(kvm, attr);
1312                 break;
1313         default:
1314                 ret = -ENXIO;
1315                 break;
1316         }
1317
1318         return ret;
1319 }
1320
1321 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1322 {
1323         int ret;
1324
1325         switch (attr->group) {
1326         case KVM_S390_VM_MEM_CTRL:
1327                 switch (attr->attr) {
1328                 case KVM_S390_VM_MEM_ENABLE_CMMA:
1329                 case KVM_S390_VM_MEM_CLR_CMMA:
1330                         ret = sclp.has_cmma ? 0 : -ENXIO;
1331                         break;
1332                 case KVM_S390_VM_MEM_LIMIT_SIZE:
1333                         ret = 0;
1334                         break;
1335                 default:
1336                         ret = -ENXIO;
1337                         break;
1338                 }
1339                 break;
1340         case KVM_S390_VM_TOD:
1341                 switch (attr->attr) {
1342                 case KVM_S390_VM_TOD_LOW:
1343                 case KVM_S390_VM_TOD_HIGH:
1344                         ret = 0;
1345                         break;
1346                 default:
1347                         ret = -ENXIO;
1348                         break;
1349                 }
1350                 break;
1351         case KVM_S390_VM_CPU_MODEL:
1352                 switch (attr->attr) {
1353                 case KVM_S390_VM_CPU_PROCESSOR:
1354                 case KVM_S390_VM_CPU_MACHINE:
1355                 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1356                 case KVM_S390_VM_CPU_MACHINE_FEAT:
1357                 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1358                         ret = 0;
1359                         break;
1360                 /* configuring subfunctions is not supported yet */
1361                 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1362                 default:
1363                         ret = -ENXIO;
1364                         break;
1365                 }
1366                 break;
1367         case KVM_S390_VM_CRYPTO:
1368                 switch (attr->attr) {
1369                 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1370                 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1371                 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1372                 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1373                         ret = 0;
1374                         break;
1375                 default:
1376                         ret = -ENXIO;
1377                         break;
1378                 }
1379                 break;
1380         case KVM_S390_VM_MIGRATION:
1381                 ret = 0;
1382                 break;
1383         default:
1384                 ret = -ENXIO;
1385                 break;
1386         }
1387
1388         return ret;
1389 }
1390
1391 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1392 {
1393         uint8_t *keys;
1394         uint64_t hva;
1395         int srcu_idx, i, r = 0;
1396
1397         if (args->flags != 0)
1398                 return -EINVAL;
1399
1400         /* Is this guest using storage keys? */
1401         if (!mm_use_skey(current->mm))
1402                 return KVM_S390_GET_SKEYS_NONE;
1403
1404         /* Enforce sane limit on memory allocation */
1405         if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1406                 return -EINVAL;
1407
1408         keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL);
1409         if (!keys)
1410                 return -ENOMEM;
1411
1412         down_read(&current->mm->mmap_sem);
1413         srcu_idx = srcu_read_lock(&kvm->srcu);
1414         for (i = 0; i < args->count; i++) {
1415                 hva = gfn_to_hva(kvm, args->start_gfn + i);
1416                 if (kvm_is_error_hva(hva)) {
1417                         r = -EFAULT;
1418                         break;
1419                 }
1420
1421                 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1422                 if (r)
1423                         break;
1424         }
1425         srcu_read_unlock(&kvm->srcu, srcu_idx);
1426         up_read(&current->mm->mmap_sem);
1427
1428         if (!r) {
1429                 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1430                                  sizeof(uint8_t) * args->count);
1431                 if (r)
1432                         r = -EFAULT;
1433         }
1434
1435         kvfree(keys);
1436         return r;
1437 }
1438
1439 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1440 {
1441         uint8_t *keys;
1442         uint64_t hva;
1443         int srcu_idx, i, r = 0;
1444
1445         if (args->flags != 0)
1446                 return -EINVAL;
1447
1448         /* Enforce sane limit on memory allocation */
1449         if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1450                 return -EINVAL;
1451
1452         keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL);
1453         if (!keys)
1454                 return -ENOMEM;
1455
1456         r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1457                            sizeof(uint8_t) * args->count);
1458         if (r) {
1459                 r = -EFAULT;
1460                 goto out;
1461         }
1462
1463         /* Enable storage key handling for the guest */
1464         r = s390_enable_skey();
1465         if (r)
1466                 goto out;
1467
1468         down_read(&current->mm->mmap_sem);
1469         srcu_idx = srcu_read_lock(&kvm->srcu);
1470         for (i = 0; i < args->count; i++) {
1471                 hva = gfn_to_hva(kvm, args->start_gfn + i);
1472                 if (kvm_is_error_hva(hva)) {
1473                         r = -EFAULT;
1474                         break;
1475                 }
1476
1477                 /* Lowest order bit is reserved */
1478                 if (keys[i] & 0x01) {
1479                         r = -EINVAL;
1480                         break;
1481                 }
1482
1483                 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
1484                 if (r)
1485                         break;
1486         }
1487         srcu_read_unlock(&kvm->srcu, srcu_idx);
1488         up_read(&current->mm->mmap_sem);
1489 out:
1490         kvfree(keys);
1491         return r;
1492 }
1493
1494 /*
1495  * Base address and length must be sent at the start of each block, therefore
1496  * it's cheaper to send some clean data, as long as it's less than the size of
1497  * two longs.
1498  */
1499 #define KVM_S390_MAX_BIT_DISTANCE (2 * sizeof(void *))
1500 /* for consistency */
1501 #define KVM_S390_CMMA_SIZE_MAX ((u32)KVM_S390_SKEYS_MAX)
1502
1503 /*
1504  * This function searches for the next page with dirty CMMA attributes, and
1505  * saves the attributes in the buffer up to either the end of the buffer or
1506  * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found;
1507  * no trailing clean bytes are saved.
1508  * In case no dirty bits were found, or if CMMA was not enabled or used, the
1509  * output buffer will indicate 0 as length.
1510  */
1511 static int kvm_s390_get_cmma_bits(struct kvm *kvm,
1512                                   struct kvm_s390_cmma_log *args)
1513 {
1514         struct kvm_s390_migration_state *s = kvm->arch.migration_state;
1515         unsigned long bufsize, hva, pgstev, i, next, cur;
1516         int srcu_idx, peek, r = 0, rr;
1517         u8 *res;
1518
1519         cur = args->start_gfn;
1520         i = next = pgstev = 0;
1521
1522         if (unlikely(!kvm->arch.use_cmma))
1523                 return -ENXIO;
1524         /* Invalid/unsupported flags were specified */
1525         if (args->flags & ~KVM_S390_CMMA_PEEK)
1526                 return -EINVAL;
1527         /* Migration mode query, and we are not doing a migration */
1528         peek = !!(args->flags & KVM_S390_CMMA_PEEK);
1529         if (!peek && !s)
1530                 return -EINVAL;
1531         /* CMMA is disabled or was not used, or the buffer has length zero */
1532         bufsize = min(args->count, KVM_S390_CMMA_SIZE_MAX);
1533         if (!bufsize || !kvm->mm->context.use_cmma) {
1534                 memset(args, 0, sizeof(*args));
1535                 return 0;
1536         }
1537
1538         if (!peek) {
1539                 /* We are not peeking, and there are no dirty pages */
1540                 if (!atomic64_read(&s->dirty_pages)) {
1541                         memset(args, 0, sizeof(*args));
1542                         return 0;
1543                 }
1544                 cur = find_next_bit(s->pgste_bitmap, s->bitmap_size,
1545                                     args->start_gfn);
1546                 if (cur >= s->bitmap_size)      /* nothing found, loop back */
1547                         cur = find_next_bit(s->pgste_bitmap, s->bitmap_size, 0);
1548                 if (cur >= s->bitmap_size) {    /* again! (very unlikely) */
1549                         memset(args, 0, sizeof(*args));
1550                         return 0;
1551                 }
1552                 next = find_next_bit(s->pgste_bitmap, s->bitmap_size, cur + 1);
1553         }
1554
1555         res = vmalloc(bufsize);
1556         if (!res)
1557                 return -ENOMEM;
1558
1559         args->start_gfn = cur;
1560
1561         down_read(&kvm->mm->mmap_sem);
1562         srcu_idx = srcu_read_lock(&kvm->srcu);
1563         while (i < bufsize) {
1564                 hva = gfn_to_hva(kvm, cur);
1565                 if (kvm_is_error_hva(hva)) {
1566                         r = -EFAULT;
1567                         break;
1568                 }
1569                 /* decrement only if we actually flipped the bit to 0 */
1570                 if (!peek && test_and_clear_bit(cur, s->pgste_bitmap))
1571                         atomic64_dec(&s->dirty_pages);
1572                 r = get_pgste(kvm->mm, hva, &pgstev);
1573                 if (r < 0)
1574                         pgstev = 0;
1575                 /* save the value */
1576                 res[i++] = (pgstev >> 24) & 0x43;
1577                 /*
1578                  * if the next bit is too far away, stop.
1579                  * if we reached the previous "next", find the next one
1580                  */
1581                 if (!peek) {
1582                         if (next > cur + KVM_S390_MAX_BIT_DISTANCE)
1583                                 break;
1584                         if (cur == next)
1585                                 next = find_next_bit(s->pgste_bitmap,
1586                                                      s->bitmap_size, cur + 1);
1587                 /* reached the end of the bitmap or of the buffer, stop */
1588                         if ((next >= s->bitmap_size) ||
1589                             (next >= args->start_gfn + bufsize))
1590                                 break;
1591                 }
1592                 cur++;
1593         }
1594         srcu_read_unlock(&kvm->srcu, srcu_idx);
1595         up_read(&kvm->mm->mmap_sem);
1596         args->count = i;
1597         args->remaining = s ? atomic64_read(&s->dirty_pages) : 0;
1598
1599         rr = copy_to_user((void __user *)args->values, res, args->count);
1600         if (rr)
1601                 r = -EFAULT;
1602
1603         vfree(res);
1604         return r;
1605 }
1606
1607 /*
1608  * This function sets the CMMA attributes for the given pages. If the input
1609  * buffer has zero length, no action is taken, otherwise the attributes are
1610  * set and the mm->context.use_cmma flag is set.
1611  */
1612 static int kvm_s390_set_cmma_bits(struct kvm *kvm,
1613                                   const struct kvm_s390_cmma_log *args)
1614 {
1615         unsigned long hva, mask, pgstev, i;
1616         uint8_t *bits;
1617         int srcu_idx, r = 0;
1618
1619         mask = args->mask;
1620
1621         if (!kvm->arch.use_cmma)
1622                 return -ENXIO;
1623         /* invalid/unsupported flags */
1624         if (args->flags != 0)
1625                 return -EINVAL;
1626         /* Enforce sane limit on memory allocation */
1627         if (args->count > KVM_S390_CMMA_SIZE_MAX)
1628                 return -EINVAL;
1629         /* Nothing to do */
1630         if (args->count == 0)
1631                 return 0;
1632
1633         bits = vmalloc(sizeof(*bits) * args->count);
1634         if (!bits)
1635                 return -ENOMEM;
1636
1637         r = copy_from_user(bits, (void __user *)args->values, args->count);
1638         if (r) {
1639                 r = -EFAULT;
1640                 goto out;
1641         }
1642
1643         down_read(&kvm->mm->mmap_sem);
1644         srcu_idx = srcu_read_lock(&kvm->srcu);
1645         for (i = 0; i < args->count; i++) {
1646                 hva = gfn_to_hva(kvm, args->start_gfn + i);
1647                 if (kvm_is_error_hva(hva)) {
1648                         r = -EFAULT;
1649                         break;
1650                 }
1651
1652                 pgstev = bits[i];
1653                 pgstev = pgstev << 24;
1654                 mask &= _PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT;
1655                 set_pgste_bits(kvm->mm, hva, mask, pgstev);
1656         }
1657         srcu_read_unlock(&kvm->srcu, srcu_idx);
1658         up_read(&kvm->mm->mmap_sem);
1659
1660         if (!kvm->mm->context.use_cmma) {
1661                 down_write(&kvm->mm->mmap_sem);
1662                 kvm->mm->context.use_cmma = 1;
1663                 up_write(&kvm->mm->mmap_sem);
1664         }
1665 out:
1666         vfree(bits);
1667         return r;
1668 }
1669
1670 long kvm_arch_vm_ioctl(struct file *filp,
1671                        unsigned int ioctl, unsigned long arg)
1672 {
1673         struct kvm *kvm = filp->private_data;
1674         void __user *argp = (void __user *)arg;
1675         struct kvm_device_attr attr;
1676         int r;
1677
1678         switch (ioctl) {
1679         case KVM_S390_INTERRUPT: {
1680                 struct kvm_s390_interrupt s390int;
1681
1682                 r = -EFAULT;
1683                 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1684                         break;
1685                 r = kvm_s390_inject_vm(kvm, &s390int);
1686                 break;
1687         }
1688         case KVM_ENABLE_CAP: {
1689                 struct kvm_enable_cap cap;
1690                 r = -EFAULT;
1691                 if (copy_from_user(&cap, argp, sizeof(cap)))
1692                         break;
1693                 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1694                 break;
1695         }
1696         case KVM_CREATE_IRQCHIP: {
1697                 struct kvm_irq_routing_entry routing;
1698
1699                 r = -EINVAL;
1700                 if (kvm->arch.use_irqchip) {
1701                         /* Set up dummy routing. */
1702                         memset(&routing, 0, sizeof(routing));
1703                         r = kvm_set_irq_routing(kvm, &routing, 0, 0);
1704                 }
1705                 break;
1706         }
1707         case KVM_SET_DEVICE_ATTR: {
1708                 r = -EFAULT;
1709                 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1710                         break;
1711                 r = kvm_s390_vm_set_attr(kvm, &attr);
1712                 break;
1713         }
1714         case KVM_GET_DEVICE_ATTR: {
1715                 r = -EFAULT;
1716                 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1717                         break;
1718                 r = kvm_s390_vm_get_attr(kvm, &attr);
1719                 break;
1720         }
1721         case KVM_HAS_DEVICE_ATTR: {
1722                 r = -EFAULT;
1723                 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1724                         break;
1725                 r = kvm_s390_vm_has_attr(kvm, &attr);
1726                 break;
1727         }
1728         case KVM_S390_GET_SKEYS: {
1729                 struct kvm_s390_skeys args;
1730
1731                 r = -EFAULT;
1732                 if (copy_from_user(&args, argp,
1733                                    sizeof(struct kvm_s390_skeys)))
1734                         break;
1735                 r = kvm_s390_get_skeys(kvm, &args);
1736                 break;
1737         }
1738         case KVM_S390_SET_SKEYS: {
1739                 struct kvm_s390_skeys args;
1740
1741                 r = -EFAULT;
1742                 if (copy_from_user(&args, argp,
1743                                    sizeof(struct kvm_s390_skeys)))
1744                         break;
1745                 r = kvm_s390_set_skeys(kvm, &args);
1746                 break;
1747         }
1748         case KVM_S390_GET_CMMA_BITS: {
1749                 struct kvm_s390_cmma_log args;
1750
1751                 r = -EFAULT;
1752                 if (copy_from_user(&args, argp, sizeof(args)))
1753                         break;
1754                 mutex_lock(&kvm->slots_lock);
1755                 r = kvm_s390_get_cmma_bits(kvm, &args);
1756                 mutex_unlock(&kvm->slots_lock);
1757                 if (!r) {
1758                         r = copy_to_user(argp, &args, sizeof(args));
1759                         if (r)
1760                                 r = -EFAULT;
1761                 }
1762                 break;
1763         }
1764         case KVM_S390_SET_CMMA_BITS: {
1765                 struct kvm_s390_cmma_log args;
1766
1767                 r = -EFAULT;
1768                 if (copy_from_user(&args, argp, sizeof(args)))
1769                         break;
1770                 mutex_lock(&kvm->slots_lock);
1771                 r = kvm_s390_set_cmma_bits(kvm, &args);
1772                 mutex_unlock(&kvm->slots_lock);
1773                 break;
1774         }
1775         default:
1776                 r = -ENOTTY;
1777         }
1778
1779         return r;
1780 }
1781
1782 static int kvm_s390_query_ap_config(u8 *config)
1783 {
1784         u32 fcn_code = 0x04000000UL;
1785         u32 cc = 0;
1786
1787         memset(config, 0, 128);
1788         asm volatile(
1789                 "lgr 0,%1\n"
1790                 "lgr 2,%2\n"
1791                 ".long 0xb2af0000\n"            /* PQAP(QCI) */
1792                 "0: ipm %0\n"
1793                 "srl %0,28\n"
1794                 "1:\n"
1795                 EX_TABLE(0b, 1b)
1796                 : "+r" (cc)
1797                 : "r" (fcn_code), "r" (config)
1798                 : "cc", "0", "2", "memory"
1799         );
1800
1801         return cc;
1802 }
1803
1804 static int kvm_s390_apxa_installed(void)
1805 {
1806         u8 config[128];
1807         int cc;
1808
1809         if (test_facility(12)) {
1810                 cc = kvm_s390_query_ap_config(config);
1811
1812                 if (cc)
1813                         pr_err("PQAP(QCI) failed with cc=%d", cc);
1814                 else
1815                         return config[0] & 0x40;
1816         }
1817
1818         return 0;
1819 }
1820
1821 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1822 {
1823         kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1824
1825         if (kvm_s390_apxa_installed())
1826                 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1827         else
1828                 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1829 }
1830
1831 static u64 kvm_s390_get_initial_cpuid(void)
1832 {
1833         struct cpuid cpuid;
1834
1835         get_cpu_id(&cpuid);
1836         cpuid.version = 0xff;
1837         return *((u64 *) &cpuid);
1838 }
1839
1840 static void kvm_s390_crypto_init(struct kvm *kvm)
1841 {
1842         if (!test_kvm_facility(kvm, 76))
1843                 return;
1844
1845         kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
1846         kvm_s390_set_crycb_format(kvm);
1847
1848         /* Enable AES/DEA protected key functions by default */
1849         kvm->arch.crypto.aes_kw = 1;
1850         kvm->arch.crypto.dea_kw = 1;
1851         get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1852                          sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1853         get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1854                          sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1855 }
1856
1857 static void sca_dispose(struct kvm *kvm)
1858 {
1859         if (kvm->arch.use_esca)
1860                 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
1861         else
1862                 free_page((unsigned long)(kvm->arch.sca));
1863         kvm->arch.sca = NULL;
1864 }
1865
1866 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1867 {
1868         gfp_t alloc_flags = GFP_KERNEL;
1869         int i, rc;
1870         char debug_name[16];
1871         static unsigned long sca_offset;
1872
1873         rc = -EINVAL;
1874 #ifdef CONFIG_KVM_S390_UCONTROL
1875         if (type & ~KVM_VM_S390_UCONTROL)
1876                 goto out_err;
1877         if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1878                 goto out_err;
1879 #else
1880         if (type)
1881                 goto out_err;
1882 #endif
1883
1884         rc = s390_enable_sie();
1885         if (rc)
1886                 goto out_err;
1887
1888         rc = -ENOMEM;
1889
1890         kvm->arch.use_esca = 0; /* start with basic SCA */
1891         if (!sclp.has_64bscao)
1892                 alloc_flags |= GFP_DMA;
1893         rwlock_init(&kvm->arch.sca_lock);
1894         kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
1895         if (!kvm->arch.sca)
1896                 goto out_err;
1897         spin_lock(&kvm_lock);
1898         sca_offset += 16;
1899         if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
1900                 sca_offset = 0;
1901         kvm->arch.sca = (struct bsca_block *)
1902                         ((char *) kvm->arch.sca + sca_offset);
1903         spin_unlock(&kvm_lock);
1904
1905         sprintf(debug_name, "kvm-%u", current->pid);
1906
1907         kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
1908         if (!kvm->arch.dbf)
1909                 goto out_err;
1910
1911         kvm->arch.sie_page2 =
1912              (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1913         if (!kvm->arch.sie_page2)
1914                 goto out_err;
1915
1916         /* Populate the facility mask initially. */
1917         memcpy(kvm->arch.model.fac_mask, S390_lowcore.stfle_fac_list,
1918                sizeof(S390_lowcore.stfle_fac_list));
1919         for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) {
1920                 if (i < kvm_s390_fac_list_mask_size())
1921                         kvm->arch.model.fac_mask[i] &= kvm_s390_fac_list_mask[i];
1922                 else
1923                         kvm->arch.model.fac_mask[i] = 0UL;
1924         }
1925
1926         /* Populate the facility list initially. */
1927         kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
1928         memcpy(kvm->arch.model.fac_list, kvm->arch.model.fac_mask,
1929                S390_ARCH_FAC_LIST_SIZE_BYTE);
1930
1931         /* we are always in czam mode - even on pre z14 machines */
1932         set_kvm_facility(kvm->arch.model.fac_mask, 138);
1933         set_kvm_facility(kvm->arch.model.fac_list, 138);
1934         /* we emulate STHYI in kvm */
1935         set_kvm_facility(kvm->arch.model.fac_mask, 74);
1936         set_kvm_facility(kvm->arch.model.fac_list, 74);
1937         if (MACHINE_HAS_TLB_GUEST) {
1938                 set_kvm_facility(kvm->arch.model.fac_mask, 147);
1939                 set_kvm_facility(kvm->arch.model.fac_list, 147);
1940         }
1941
1942         kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
1943         kvm->arch.model.ibc = sclp.ibc & 0x0fff;
1944
1945         kvm_s390_crypto_init(kvm);
1946
1947         mutex_init(&kvm->arch.float_int.ais_lock);
1948         kvm->arch.float_int.simm = 0;
1949         kvm->arch.float_int.nimm = 0;
1950         spin_lock_init(&kvm->arch.float_int.lock);
1951         for (i = 0; i < FIRQ_LIST_COUNT; i++)
1952                 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
1953         init_waitqueue_head(&kvm->arch.ipte_wq);
1954         mutex_init(&kvm->arch.ipte_mutex);
1955
1956         debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
1957         VM_EVENT(kvm, 3, "vm created with type %lu", type);
1958
1959         if (type & KVM_VM_S390_UCONTROL) {
1960                 kvm->arch.gmap = NULL;
1961                 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
1962         } else {
1963                 if (sclp.hamax == U64_MAX)
1964                         kvm->arch.mem_limit = TASK_SIZE_MAX;
1965                 else
1966                         kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
1967                                                     sclp.hamax + 1);
1968                 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
1969                 if (!kvm->arch.gmap)
1970                         goto out_err;
1971                 kvm->arch.gmap->private = kvm;
1972                 kvm->arch.gmap->pfault_enabled = 0;
1973         }
1974
1975         kvm->arch.css_support = 0;
1976         kvm->arch.use_irqchip = 0;
1977         kvm->arch.epoch = 0;
1978
1979         spin_lock_init(&kvm->arch.start_stop_lock);
1980         kvm_s390_vsie_init(kvm);
1981         KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
1982
1983         return 0;
1984 out_err:
1985         free_page((unsigned long)kvm->arch.sie_page2);
1986         debug_unregister(kvm->arch.dbf);
1987         sca_dispose(kvm);
1988         KVM_EVENT(3, "creation of vm failed: %d", rc);
1989         return rc;
1990 }
1991
1992 bool kvm_arch_has_vcpu_debugfs(void)
1993 {
1994         return false;
1995 }
1996
1997 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
1998 {
1999         return 0;
2000 }
2001
2002 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
2003 {
2004         VCPU_EVENT(vcpu, 3, "%s", "free cpu");
2005         trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
2006         kvm_s390_clear_local_irqs(vcpu);
2007         kvm_clear_async_pf_completion_queue(vcpu);
2008         if (!kvm_is_ucontrol(vcpu->kvm))
2009                 sca_del_vcpu(vcpu);
2010
2011         if (kvm_is_ucontrol(vcpu->kvm))
2012                 gmap_remove(vcpu->arch.gmap);
2013
2014         if (vcpu->kvm->arch.use_cmma)
2015                 kvm_s390_vcpu_unsetup_cmma(vcpu);
2016         free_page((unsigned long)(vcpu->arch.sie_block));
2017
2018         kvm_vcpu_uninit(vcpu);
2019         kmem_cache_free(kvm_vcpu_cache, vcpu);
2020 }
2021
2022 static void kvm_free_vcpus(struct kvm *kvm)
2023 {
2024         unsigned int i;
2025         struct kvm_vcpu *vcpu;
2026
2027         kvm_for_each_vcpu(i, vcpu, kvm)
2028                 kvm_arch_vcpu_destroy(vcpu);
2029
2030         mutex_lock(&kvm->lock);
2031         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
2032                 kvm->vcpus[i] = NULL;
2033
2034         atomic_set(&kvm->online_vcpus, 0);
2035         mutex_unlock(&kvm->lock);
2036 }
2037
2038 void kvm_arch_destroy_vm(struct kvm *kvm)
2039 {
2040         kvm_free_vcpus(kvm);
2041         sca_dispose(kvm);
2042         debug_unregister(kvm->arch.dbf);
2043         free_page((unsigned long)kvm->arch.sie_page2);
2044         if (!kvm_is_ucontrol(kvm))
2045                 gmap_remove(kvm->arch.gmap);
2046         kvm_s390_destroy_adapters(kvm);
2047         kvm_s390_clear_float_irqs(kvm);
2048         kvm_s390_vsie_destroy(kvm);
2049         if (kvm->arch.migration_state) {
2050                 vfree(kvm->arch.migration_state->pgste_bitmap);
2051                 kfree(kvm->arch.migration_state);
2052         }
2053         KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
2054 }
2055
2056 /* Section: vcpu related */
2057 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
2058 {
2059         vcpu->arch.gmap = gmap_create(current->mm, -1UL);
2060         if (!vcpu->arch.gmap)
2061                 return -ENOMEM;
2062         vcpu->arch.gmap->private = vcpu->kvm;
2063
2064         return 0;
2065 }
2066
2067 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
2068 {
2069         if (!kvm_s390_use_sca_entries())
2070                 return;
2071         read_lock(&vcpu->kvm->arch.sca_lock);
2072         if (vcpu->kvm->arch.use_esca) {
2073                 struct esca_block *sca = vcpu->kvm->arch.sca;
2074
2075                 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
2076                 sca->cpu[vcpu->vcpu_id].sda = 0;
2077         } else {
2078                 struct bsca_block *sca = vcpu->kvm->arch.sca;
2079
2080                 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
2081                 sca->cpu[vcpu->vcpu_id].sda = 0;
2082         }
2083         read_unlock(&vcpu->kvm->arch.sca_lock);
2084 }
2085
2086 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
2087 {
2088         if (!kvm_s390_use_sca_entries()) {
2089                 struct bsca_block *sca = vcpu->kvm->arch.sca;
2090
2091                 /* we still need the basic sca for the ipte control */
2092                 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2093                 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
2094         }
2095         read_lock(&vcpu->kvm->arch.sca_lock);
2096         if (vcpu->kvm->arch.use_esca) {
2097                 struct esca_block *sca = vcpu->kvm->arch.sca;
2098
2099                 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
2100                 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2101                 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
2102                 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
2103                 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
2104         } else {
2105                 struct bsca_block *sca = vcpu->kvm->arch.sca;
2106
2107                 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
2108                 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2109                 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
2110                 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
2111         }
2112         read_unlock(&vcpu->kvm->arch.sca_lock);
2113 }
2114
2115 /* Basic SCA to Extended SCA data copy routines */
2116 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
2117 {
2118         d->sda = s->sda;
2119         d->sigp_ctrl.c = s->sigp_ctrl.c;
2120         d->sigp_ctrl.scn = s->sigp_ctrl.scn;
2121 }
2122
2123 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
2124 {
2125         int i;
2126
2127         d->ipte_control = s->ipte_control;
2128         d->mcn[0] = s->mcn;
2129         for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
2130                 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
2131 }
2132
2133 static int sca_switch_to_extended(struct kvm *kvm)
2134 {
2135         struct bsca_block *old_sca = kvm->arch.sca;
2136         struct esca_block *new_sca;
2137         struct kvm_vcpu *vcpu;
2138         unsigned int vcpu_idx;
2139         u32 scaol, scaoh;
2140
2141         new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
2142         if (!new_sca)
2143                 return -ENOMEM;
2144
2145         scaoh = (u32)((u64)(new_sca) >> 32);
2146         scaol = (u32)(u64)(new_sca) & ~0x3fU;
2147
2148         kvm_s390_vcpu_block_all(kvm);
2149         write_lock(&kvm->arch.sca_lock);
2150
2151         sca_copy_b_to_e(new_sca, old_sca);
2152
2153         kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
2154                 vcpu->arch.sie_block->scaoh = scaoh;
2155                 vcpu->arch.sie_block->scaol = scaol;
2156                 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
2157         }
2158         kvm->arch.sca = new_sca;
2159         kvm->arch.use_esca = 1;
2160
2161         write_unlock(&kvm->arch.sca_lock);
2162         kvm_s390_vcpu_unblock_all(kvm);
2163
2164         free_page((unsigned long)old_sca);
2165
2166         VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
2167                  old_sca, kvm->arch.sca);
2168         return 0;
2169 }
2170
2171 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
2172 {
2173         int rc;
2174
2175         if (!kvm_s390_use_sca_entries()) {
2176                 if (id < KVM_MAX_VCPUS)
2177                         return true;
2178                 return false;
2179         }
2180         if (id < KVM_S390_BSCA_CPU_SLOTS)
2181                 return true;
2182         if (!sclp.has_esca || !sclp.has_64bscao)
2183                 return false;
2184
2185         mutex_lock(&kvm->lock);
2186         rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
2187         mutex_unlock(&kvm->lock);
2188
2189         return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
2190 }
2191
2192 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2193 {
2194         vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
2195         kvm_clear_async_pf_completion_queue(vcpu);
2196         vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
2197                                     KVM_SYNC_GPRS |
2198                                     KVM_SYNC_ACRS |
2199                                     KVM_SYNC_CRS |
2200                                     KVM_SYNC_ARCH0 |
2201                                     KVM_SYNC_PFAULT;
2202         kvm_s390_set_prefix(vcpu, 0);
2203         if (test_kvm_facility(vcpu->kvm, 64))
2204                 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
2205         if (test_kvm_facility(vcpu->kvm, 133))
2206                 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
2207         /* fprs can be synchronized via vrs, even if the guest has no vx. With
2208          * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
2209          */
2210         if (MACHINE_HAS_VX)
2211                 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
2212         else
2213                 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
2214
2215         if (kvm_is_ucontrol(vcpu->kvm))
2216                 return __kvm_ucontrol_vcpu_init(vcpu);
2217
2218         return 0;
2219 }
2220
2221 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2222 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2223 {
2224         WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
2225         raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2226         vcpu->arch.cputm_start = get_tod_clock_fast();
2227         raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2228 }
2229
2230 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2231 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2232 {
2233         WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
2234         raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2235         vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
2236         vcpu->arch.cputm_start = 0;
2237         raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2238 }
2239
2240 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2241 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2242 {
2243         WARN_ON_ONCE(vcpu->arch.cputm_enabled);
2244         vcpu->arch.cputm_enabled = true;
2245         __start_cpu_timer_accounting(vcpu);
2246 }
2247
2248 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2249 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2250 {
2251         WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
2252         __stop_cpu_timer_accounting(vcpu);
2253         vcpu->arch.cputm_enabled = false;
2254 }
2255
2256 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2257 {
2258         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2259         __enable_cpu_timer_accounting(vcpu);
2260         preempt_enable();
2261 }
2262
2263 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2264 {
2265         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2266         __disable_cpu_timer_accounting(vcpu);
2267         preempt_enable();
2268 }
2269
2270 /* set the cpu timer - may only be called from the VCPU thread itself */
2271 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
2272 {
2273         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2274         raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2275         if (vcpu->arch.cputm_enabled)
2276                 vcpu->arch.cputm_start = get_tod_clock_fast();
2277         vcpu->arch.sie_block->cputm = cputm;
2278         raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2279         preempt_enable();
2280 }
2281
2282 /* update and get the cpu timer - can also be called from other VCPU threads */
2283 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
2284 {
2285         unsigned int seq;
2286         __u64 value;
2287
2288         if (unlikely(!vcpu->arch.cputm_enabled))
2289                 return vcpu->arch.sie_block->cputm;
2290
2291         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2292         do {
2293                 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
2294                 /*
2295                  * If the writer would ever execute a read in the critical
2296                  * section, e.g. in irq context, we have a deadlock.
2297                  */
2298                 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
2299                 value = vcpu->arch.sie_block->cputm;
2300                 /* if cputm_start is 0, accounting is being started/stopped */
2301                 if (likely(vcpu->arch.cputm_start))
2302                         value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
2303         } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
2304         preempt_enable();
2305         return value;
2306 }
2307
2308 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2309 {
2310
2311         gmap_enable(vcpu->arch.enabled_gmap);
2312         atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
2313         if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
2314                 __start_cpu_timer_accounting(vcpu);
2315         vcpu->cpu = cpu;
2316 }
2317
2318 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2319 {
2320         vcpu->cpu = -1;
2321         if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
2322                 __stop_cpu_timer_accounting(vcpu);
2323         atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
2324         vcpu->arch.enabled_gmap = gmap_get_enabled();
2325         gmap_disable(vcpu->arch.enabled_gmap);
2326
2327 }
2328
2329 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
2330 {
2331         /* this equals initial cpu reset in pop, but we don't switch to ESA */
2332         vcpu->arch.sie_block->gpsw.mask = 0UL;
2333         vcpu->arch.sie_block->gpsw.addr = 0UL;
2334         kvm_s390_set_prefix(vcpu, 0);
2335         kvm_s390_set_cpu_timer(vcpu, 0);
2336         vcpu->arch.sie_block->ckc       = 0UL;
2337         vcpu->arch.sie_block->todpr     = 0;
2338         memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
2339         vcpu->arch.sie_block->gcr[0]  = 0xE0UL;
2340         vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
2341         /* make sure the new fpc will be lazily loaded */
2342         save_fpu_regs();
2343         current->thread.fpu.fpc = 0;
2344         vcpu->arch.sie_block->gbea = 1;
2345         vcpu->arch.sie_block->pp = 0;
2346         vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
2347         kvm_clear_async_pf_completion_queue(vcpu);
2348         if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
2349                 kvm_s390_vcpu_stop(vcpu);
2350         kvm_s390_clear_local_irqs(vcpu);
2351 }
2352
2353 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
2354 {
2355         mutex_lock(&vcpu->kvm->lock);
2356         preempt_disable();
2357         vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
2358         preempt_enable();
2359         mutex_unlock(&vcpu->kvm->lock);
2360         if (!kvm_is_ucontrol(vcpu->kvm)) {
2361                 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
2362                 sca_add_vcpu(vcpu);
2363         }
2364         if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
2365                 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2366         /* make vcpu_load load the right gmap on the first trigger */
2367         vcpu->arch.enabled_gmap = vcpu->arch.gmap;
2368 }
2369
2370 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
2371 {
2372         if (!test_kvm_facility(vcpu->kvm, 76))
2373                 return;
2374
2375         vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
2376
2377         if (vcpu->kvm->arch.crypto.aes_kw)
2378                 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
2379         if (vcpu->kvm->arch.crypto.dea_kw)
2380                 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
2381
2382         vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
2383 }
2384
2385 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
2386 {
2387         free_page(vcpu->arch.sie_block->cbrlo);
2388         vcpu->arch.sie_block->cbrlo = 0;
2389 }
2390
2391 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
2392 {
2393         vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
2394         if (!vcpu->arch.sie_block->cbrlo)
2395                 return -ENOMEM;
2396
2397         vcpu->arch.sie_block->ecb2 &= ~ECB2_PFMFI;
2398         return 0;
2399 }
2400
2401 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
2402 {
2403         struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
2404
2405         vcpu->arch.sie_block->ibc = model->ibc;
2406         if (test_kvm_facility(vcpu->kvm, 7))
2407                 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
2408 }
2409
2410 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
2411 {
2412         int rc = 0;
2413
2414         atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
2415                                                     CPUSTAT_SM |
2416                                                     CPUSTAT_STOPPED);
2417
2418         if (test_kvm_facility(vcpu->kvm, 78))
2419                 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
2420         else if (test_kvm_facility(vcpu->kvm, 8))
2421                 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
2422
2423         kvm_s390_vcpu_setup_model(vcpu);
2424
2425         /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
2426         if (MACHINE_HAS_ESOP)
2427                 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
2428         if (test_kvm_facility(vcpu->kvm, 9))
2429                 vcpu->arch.sie_block->ecb |= ECB_SRSI;
2430         if (test_kvm_facility(vcpu->kvm, 73))
2431                 vcpu->arch.sie_block->ecb |= ECB_TE;
2432
2433         if (test_kvm_facility(vcpu->kvm, 8) && sclp.has_pfmfi)
2434                 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
2435         if (test_kvm_facility(vcpu->kvm, 130))
2436                 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
2437         vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
2438         if (sclp.has_cei)
2439                 vcpu->arch.sie_block->eca |= ECA_CEI;
2440         if (sclp.has_ib)
2441                 vcpu->arch.sie_block->eca |= ECA_IB;
2442         if (sclp.has_siif)
2443                 vcpu->arch.sie_block->eca |= ECA_SII;
2444         if (sclp.has_sigpif)
2445                 vcpu->arch.sie_block->eca |= ECA_SIGPI;
2446         if (test_kvm_facility(vcpu->kvm, 129)) {
2447                 vcpu->arch.sie_block->eca |= ECA_VX;
2448                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
2449         }
2450         if (test_kvm_facility(vcpu->kvm, 139))
2451                 vcpu->arch.sie_block->ecd |= ECD_MEF;
2452
2453         vcpu->arch.sie_block->sdnxo = ((unsigned long) &vcpu->run->s.regs.sdnx)
2454                                         | SDNXC;
2455         vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
2456
2457         if (sclp.has_kss)
2458                 atomic_or(CPUSTAT_KSS, &vcpu->arch.sie_block->cpuflags);
2459         else
2460                 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
2461
2462         if (vcpu->kvm->arch.use_cmma) {
2463                 rc = kvm_s390_vcpu_setup_cmma(vcpu);
2464                 if (rc)
2465                         return rc;
2466         }
2467         hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2468         vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
2469
2470         kvm_s390_vcpu_crypto_setup(vcpu);
2471
2472         return rc;
2473 }
2474
2475 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2476                                       unsigned int id)
2477 {
2478         struct kvm_vcpu *vcpu;
2479         struct sie_page *sie_page;
2480         int rc = -EINVAL;
2481
2482         if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
2483                 goto out;
2484
2485         rc = -ENOMEM;
2486
2487         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2488         if (!vcpu)
2489                 goto out;
2490
2491         BUILD_BUG_ON(sizeof(struct sie_page) != 4096);
2492         sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
2493         if (!sie_page)
2494                 goto out_free_cpu;
2495
2496         vcpu->arch.sie_block = &sie_page->sie_block;
2497         vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
2498
2499         /* the real guest size will always be smaller than msl */
2500         vcpu->arch.sie_block->mso = 0;
2501         vcpu->arch.sie_block->msl = sclp.hamax;
2502
2503         vcpu->arch.sie_block->icpua = id;
2504         spin_lock_init(&vcpu->arch.local_int.lock);
2505         vcpu->arch.local_int.float_int = &kvm->arch.float_int;
2506         vcpu->arch.local_int.wq = &vcpu->wq;
2507         vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
2508         seqcount_init(&vcpu->arch.cputm_seqcount);
2509
2510         rc = kvm_vcpu_init(vcpu, kvm, id);
2511         if (rc)
2512                 goto out_free_sie_block;
2513         VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
2514                  vcpu->arch.sie_block);
2515         trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
2516
2517         return vcpu;
2518 out_free_sie_block:
2519         free_page((unsigned long)(vcpu->arch.sie_block));
2520 out_free_cpu:
2521         kmem_cache_free(kvm_vcpu_cache, vcpu);
2522 out:
2523         return ERR_PTR(rc);
2524 }
2525
2526 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
2527 {
2528         return kvm_s390_vcpu_has_irq(vcpu, 0);
2529 }
2530
2531 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
2532 {
2533         return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE);
2534 }
2535
2536 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
2537 {
2538         atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2539         exit_sie(vcpu);
2540 }
2541
2542 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
2543 {
2544         atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2545 }
2546
2547 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
2548 {
2549         atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2550         exit_sie(vcpu);
2551 }
2552
2553 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
2554 {
2555         atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2556 }
2557
2558 /*
2559  * Kick a guest cpu out of SIE and wait until SIE is not running.
2560  * If the CPU is not running (e.g. waiting as idle) the function will
2561  * return immediately. */
2562 void exit_sie(struct kvm_vcpu *vcpu)
2563 {
2564         atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
2565         while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
2566                 cpu_relax();
2567 }
2568
2569 /* Kick a guest cpu out of SIE to process a request synchronously */
2570 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
2571 {
2572         kvm_make_request(req, vcpu);
2573         kvm_s390_vcpu_request(vcpu);
2574 }
2575
2576 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
2577                               unsigned long end)
2578 {
2579         struct kvm *kvm = gmap->private;
2580         struct kvm_vcpu *vcpu;
2581         unsigned long prefix;
2582         int i;
2583
2584         if (gmap_is_shadow(gmap))
2585                 return;
2586         if (start >= 1UL << 31)
2587                 /* We are only interested in prefix pages */
2588                 return;
2589         kvm_for_each_vcpu(i, vcpu, kvm) {
2590                 /* match against both prefix pages */
2591                 prefix = kvm_s390_get_prefix(vcpu);
2592                 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
2593                         VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
2594                                    start, end);
2595                         kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
2596                 }
2597         }
2598 }
2599
2600 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
2601 {
2602         /* kvm common code refers to this, but never calls it */
2603         BUG();
2604         return 0;
2605 }
2606
2607 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
2608                                            struct kvm_one_reg *reg)
2609 {
2610         int r = -EINVAL;
2611
2612         switch (reg->id) {
2613         case KVM_REG_S390_TODPR:
2614                 r = put_user(vcpu->arch.sie_block->todpr,
2615                              (u32 __user *)reg->addr);
2616                 break;
2617         case KVM_REG_S390_EPOCHDIFF:
2618                 r = put_user(vcpu->arch.sie_block->epoch,
2619                              (u64 __user *)reg->addr);
2620                 break;
2621         case KVM_REG_S390_CPU_TIMER:
2622                 r = put_user(kvm_s390_get_cpu_timer(vcpu),
2623                              (u64 __user *)reg->addr);
2624                 break;
2625         case KVM_REG_S390_CLOCK_COMP:
2626                 r = put_user(vcpu->arch.sie_block->ckc,
2627                              (u64 __user *)reg->addr);
2628                 break;
2629         case KVM_REG_S390_PFTOKEN:
2630                 r = put_user(vcpu->arch.pfault_token,
2631                              (u64 __user *)reg->addr);
2632                 break;
2633         case KVM_REG_S390_PFCOMPARE:
2634                 r = put_user(vcpu->arch.pfault_compare,
2635                              (u64 __user *)reg->addr);
2636                 break;
2637         case KVM_REG_S390_PFSELECT:
2638                 r = put_user(vcpu->arch.pfault_select,
2639                              (u64 __user *)reg->addr);
2640                 break;
2641         case KVM_REG_S390_PP:
2642                 r = put_user(vcpu->arch.sie_block->pp,
2643                              (u64 __user *)reg->addr);
2644                 break;
2645         case KVM_REG_S390_GBEA:
2646                 r = put_user(vcpu->arch.sie_block->gbea,
2647                              (u64 __user *)reg->addr);
2648                 break;
2649         default:
2650                 break;
2651         }
2652
2653         return r;
2654 }
2655
2656 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
2657                                            struct kvm_one_reg *reg)
2658 {
2659         int r = -EINVAL;
2660         __u64 val;
2661
2662         switch (reg->id) {
2663         case KVM_REG_S390_TODPR:
2664                 r = get_user(vcpu->arch.sie_block->todpr,
2665                              (u32 __user *)reg->addr);
2666                 break;
2667         case KVM_REG_S390_EPOCHDIFF:
2668                 r = get_user(vcpu->arch.sie_block->epoch,
2669                              (u64 __user *)reg->addr);
2670                 break;
2671         case KVM_REG_S390_CPU_TIMER:
2672                 r = get_user(val, (u64 __user *)reg->addr);
2673                 if (!r)
2674                         kvm_s390_set_cpu_timer(vcpu, val);
2675                 break;
2676         case KVM_REG_S390_CLOCK_COMP:
2677                 r = get_user(vcpu->arch.sie_block->ckc,
2678                              (u64 __user *)reg->addr);
2679                 break;
2680         case KVM_REG_S390_PFTOKEN:
2681                 r = get_user(vcpu->arch.pfault_token,
2682                              (u64 __user *)reg->addr);
2683                 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2684                         kvm_clear_async_pf_completion_queue(vcpu);
2685                 break;
2686         case KVM_REG_S390_PFCOMPARE:
2687                 r = get_user(vcpu->arch.pfault_compare,
2688                              (u64 __user *)reg->addr);
2689                 break;
2690         case KVM_REG_S390_PFSELECT:
2691                 r = get_user(vcpu->arch.pfault_select,
2692                              (u64 __user *)reg->addr);
2693                 break;
2694         case KVM_REG_S390_PP:
2695                 r = get_user(vcpu->arch.sie_block->pp,
2696                              (u64 __user *)reg->addr);
2697                 break;
2698         case KVM_REG_S390_GBEA:
2699                 r = get_user(vcpu->arch.sie_block->gbea,
2700                              (u64 __user *)reg->addr);
2701                 break;
2702         default:
2703                 break;
2704         }
2705
2706         return r;
2707 }
2708
2709 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
2710 {
2711         kvm_s390_vcpu_initial_reset(vcpu);
2712         return 0;
2713 }
2714
2715 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2716 {
2717         memcpy(&vcpu->run->s.regs.gprs, &regs->gprs, sizeof(regs->gprs));
2718         return 0;
2719 }
2720
2721 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2722 {
2723         memcpy(&regs->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
2724         return 0;
2725 }
2726
2727 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2728                                   struct kvm_sregs *sregs)
2729 {
2730         memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
2731         memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
2732         return 0;
2733 }
2734
2735 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2736                                   struct kvm_sregs *sregs)
2737 {
2738         memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
2739         memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
2740         return 0;
2741 }
2742
2743 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2744 {
2745         if (test_fp_ctl(fpu->fpc))
2746                 return -EINVAL;
2747         vcpu->run->s.regs.fpc = fpu->fpc;
2748         if (MACHINE_HAS_VX)
2749                 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
2750                                  (freg_t *) fpu->fprs);
2751         else
2752                 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
2753         return 0;
2754 }
2755
2756 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2757 {
2758         /* make sure we have the latest values */
2759         save_fpu_regs();
2760         if (MACHINE_HAS_VX)
2761                 convert_vx_to_fp((freg_t *) fpu->fprs,
2762                                  (__vector128 *) vcpu->run->s.regs.vrs);
2763         else
2764                 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
2765         fpu->fpc = vcpu->run->s.regs.fpc;
2766         return 0;
2767 }
2768
2769 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
2770 {
2771         int rc = 0;
2772
2773         if (!is_vcpu_stopped(vcpu))
2774                 rc = -EBUSY;
2775         else {
2776                 vcpu->run->psw_mask = psw.mask;
2777                 vcpu->run->psw_addr = psw.addr;
2778         }
2779         return rc;
2780 }
2781
2782 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2783                                   struct kvm_translation *tr)
2784 {
2785         return -EINVAL; /* not implemented yet */
2786 }
2787
2788 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
2789                               KVM_GUESTDBG_USE_HW_BP | \
2790                               KVM_GUESTDBG_ENABLE)
2791
2792 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
2793                                         struct kvm_guest_debug *dbg)
2794 {
2795         int rc = 0;
2796
2797         vcpu->guest_debug = 0;
2798         kvm_s390_clear_bp_data(vcpu);
2799
2800         if (dbg->control & ~VALID_GUESTDBG_FLAGS)
2801                 return -EINVAL;
2802         if (!sclp.has_gpere)
2803                 return -EINVAL;
2804
2805         if (dbg->control & KVM_GUESTDBG_ENABLE) {
2806                 vcpu->guest_debug = dbg->control;
2807                 /* enforce guest PER */
2808                 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2809
2810                 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
2811                         rc = kvm_s390_import_bp_data(vcpu, dbg);
2812         } else {
2813                 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2814                 vcpu->arch.guestdbg.last_bp = 0;
2815         }
2816
2817         if (rc) {
2818                 vcpu->guest_debug = 0;
2819                 kvm_s390_clear_bp_data(vcpu);
2820                 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2821         }
2822
2823         return rc;
2824 }
2825
2826 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2827                                     struct kvm_mp_state *mp_state)
2828 {
2829         /* CHECK_STOP and LOAD are not supported yet */
2830         return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
2831                                        KVM_MP_STATE_OPERATING;
2832 }
2833
2834 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2835                                     struct kvm_mp_state *mp_state)
2836 {
2837         int rc = 0;
2838
2839         /* user space knows about this interface - let it control the state */
2840         vcpu->kvm->arch.user_cpu_state_ctrl = 1;
2841
2842         switch (mp_state->mp_state) {
2843         case KVM_MP_STATE_STOPPED:
2844                 kvm_s390_vcpu_stop(vcpu);
2845                 break;
2846         case KVM_MP_STATE_OPERATING:
2847                 kvm_s390_vcpu_start(vcpu);
2848                 break;
2849         case KVM_MP_STATE_LOAD:
2850         case KVM_MP_STATE_CHECK_STOP:
2851                 /* fall through - CHECK_STOP and LOAD are not supported yet */
2852         default:
2853                 rc = -ENXIO;
2854         }
2855
2856         return rc;
2857 }
2858
2859 static bool ibs_enabled(struct kvm_vcpu *vcpu)
2860 {
2861         return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
2862 }
2863
2864 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
2865 {
2866 retry:
2867         kvm_s390_vcpu_request_handled(vcpu);
2868         if (!kvm_request_pending(vcpu))
2869                 return 0;
2870         /*
2871          * We use MMU_RELOAD just to re-arm the ipte notifier for the
2872          * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
2873          * This ensures that the ipte instruction for this request has
2874          * already finished. We might race against a second unmapper that
2875          * wants to set the blocking bit. Lets just retry the request loop.
2876          */
2877         if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
2878                 int rc;
2879                 rc = gmap_mprotect_notify(vcpu->arch.gmap,
2880                                           kvm_s390_get_prefix(vcpu),
2881                                           PAGE_SIZE * 2, PROT_WRITE);
2882                 if (rc) {
2883                         kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
2884                         return rc;
2885                 }
2886                 goto retry;
2887         }
2888
2889         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2890                 vcpu->arch.sie_block->ihcpu = 0xffff;
2891                 goto retry;
2892         }
2893
2894         if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
2895                 if (!ibs_enabled(vcpu)) {
2896                         trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
2897                         atomic_or(CPUSTAT_IBS,
2898                                         &vcpu->arch.sie_block->cpuflags);
2899                 }
2900                 goto retry;
2901         }
2902
2903         if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
2904                 if (ibs_enabled(vcpu)) {
2905                         trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
2906                         atomic_andnot(CPUSTAT_IBS,
2907                                           &vcpu->arch.sie_block->cpuflags);
2908                 }
2909                 goto retry;
2910         }
2911
2912         if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
2913                 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2914                 goto retry;
2915         }
2916
2917         if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) {
2918                 /*
2919                  * Disable CMMA virtualization; we will emulate the ESSA
2920                  * instruction manually, in order to provide additional
2921                  * functionalities needed for live migration.
2922                  */
2923                 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA;
2924                 goto retry;
2925         }
2926
2927         if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) {
2928                 /*
2929                  * Re-enable CMMA virtualization if CMMA is available and
2930                  * was used.
2931                  */
2932                 if ((vcpu->kvm->arch.use_cmma) &&
2933                     (vcpu->kvm->mm->context.use_cmma))
2934                         vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
2935                 goto retry;
2936         }
2937
2938         /* nothing to do, just clear the request */
2939         kvm_clear_request(KVM_REQ_UNHALT, vcpu);
2940
2941         return 0;
2942 }
2943
2944 void kvm_s390_set_tod_clock_ext(struct kvm *kvm,
2945                                  const struct kvm_s390_vm_tod_clock *gtod)
2946 {
2947         struct kvm_vcpu *vcpu;
2948         struct kvm_s390_tod_clock_ext htod;
2949         int i;
2950
2951         mutex_lock(&kvm->lock);
2952         preempt_disable();
2953
2954         get_tod_clock_ext((char *)&htod);
2955
2956         kvm->arch.epoch = gtod->tod - htod.tod;
2957         kvm->arch.epdx = gtod->epoch_idx - htod.epoch_idx;
2958
2959         if (kvm->arch.epoch > gtod->tod)
2960                 kvm->arch.epdx -= 1;
2961
2962         kvm_s390_vcpu_block_all(kvm);
2963         kvm_for_each_vcpu(i, vcpu, kvm) {
2964                 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
2965                 vcpu->arch.sie_block->epdx  = kvm->arch.epdx;
2966         }
2967
2968         kvm_s390_vcpu_unblock_all(kvm);
2969         preempt_enable();
2970         mutex_unlock(&kvm->lock);
2971 }
2972
2973 void kvm_s390_set_tod_clock(struct kvm *kvm, u64 tod)
2974 {
2975         struct kvm_vcpu *vcpu;
2976         int i;
2977
2978         mutex_lock(&kvm->lock);
2979         preempt_disable();
2980         kvm->arch.epoch = tod - get_tod_clock();
2981         kvm_s390_vcpu_block_all(kvm);
2982         kvm_for_each_vcpu(i, vcpu, kvm)
2983                 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
2984         kvm_s390_vcpu_unblock_all(kvm);
2985         preempt_enable();
2986         mutex_unlock(&kvm->lock);
2987 }
2988
2989 /**
2990  * kvm_arch_fault_in_page - fault-in guest page if necessary
2991  * @vcpu: The corresponding virtual cpu
2992  * @gpa: Guest physical address
2993  * @writable: Whether the page should be writable or not
2994  *
2995  * Make sure that a guest page has been faulted-in on the host.
2996  *
2997  * Return: Zero on success, negative error code otherwise.
2998  */
2999 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
3000 {
3001         return gmap_fault(vcpu->arch.gmap, gpa,
3002                           writable ? FAULT_FLAG_WRITE : 0);
3003 }
3004
3005 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
3006                                       unsigned long token)
3007 {
3008         struct kvm_s390_interrupt inti;
3009         struct kvm_s390_irq irq;
3010
3011         if (start_token) {
3012                 irq.u.ext.ext_params2 = token;
3013                 irq.type = KVM_S390_INT_PFAULT_INIT;
3014                 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
3015         } else {
3016                 inti.type = KVM_S390_INT_PFAULT_DONE;
3017                 inti.parm64 = token;
3018                 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
3019         }
3020 }
3021
3022 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
3023                                      struct kvm_async_pf *work)
3024 {
3025         trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
3026         __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
3027 }
3028
3029 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
3030                                  struct kvm_async_pf *work)
3031 {
3032         trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
3033         __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
3034 }
3035
3036 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
3037                                struct kvm_async_pf *work)
3038 {
3039         /* s390 will always inject the page directly */
3040 }
3041
3042 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
3043 {
3044         /*
3045          * s390 will always inject the page directly,
3046          * but we still want check_async_completion to cleanup
3047          */
3048         return true;
3049 }
3050
3051 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
3052 {
3053         hva_t hva;
3054         struct kvm_arch_async_pf arch;
3055         int rc;
3056
3057         if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
3058                 return 0;
3059         if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
3060             vcpu->arch.pfault_compare)
3061                 return 0;
3062         if (psw_extint_disabled(vcpu))
3063                 return 0;
3064         if (kvm_s390_vcpu_has_irq(vcpu, 0))
3065                 return 0;
3066         if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
3067                 return 0;
3068         if (!vcpu->arch.gmap->pfault_enabled)
3069                 return 0;
3070
3071         hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
3072         hva += current->thread.gmap_addr & ~PAGE_MASK;
3073         if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
3074                 return 0;
3075
3076         rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
3077         return rc;
3078 }
3079
3080 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
3081 {
3082         int rc, cpuflags;
3083
3084         /*
3085          * On s390 notifications for arriving pages will be delivered directly
3086          * to the guest but the house keeping for completed pfaults is
3087          * handled outside the worker.
3088          */
3089         kvm_check_async_pf_completion(vcpu);
3090
3091         vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
3092         vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
3093
3094         if (need_resched())
3095                 schedule();
3096
3097         if (test_cpu_flag(CIF_MCCK_PENDING))
3098                 s390_handle_mcck();
3099
3100         if (!kvm_is_ucontrol(vcpu->kvm)) {
3101                 rc = kvm_s390_deliver_pending_interrupts(vcpu);
3102                 if (rc)
3103                         return rc;
3104         }
3105
3106         rc = kvm_s390_handle_requests(vcpu);
3107         if (rc)
3108                 return rc;
3109
3110         if (guestdbg_enabled(vcpu)) {
3111                 kvm_s390_backup_guest_per_regs(vcpu);
3112                 kvm_s390_patch_guest_per_regs(vcpu);
3113         }
3114
3115         vcpu->arch.sie_block->icptcode = 0;
3116         cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
3117         VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
3118         trace_kvm_s390_sie_enter(vcpu, cpuflags);
3119
3120         return 0;
3121 }
3122
3123 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
3124 {
3125         struct kvm_s390_pgm_info pgm_info = {
3126                 .code = PGM_ADDRESSING,
3127         };
3128         u8 opcode, ilen;
3129         int rc;
3130
3131         VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
3132         trace_kvm_s390_sie_fault(vcpu);
3133
3134         /*
3135          * We want to inject an addressing exception, which is defined as a
3136          * suppressing or terminating exception. However, since we came here
3137          * by a DAT access exception, the PSW still points to the faulting
3138          * instruction since DAT exceptions are nullifying. So we've got
3139          * to look up the current opcode to get the length of the instruction
3140          * to be able to forward the PSW.
3141          */
3142         rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
3143         ilen = insn_length(opcode);
3144         if (rc < 0) {
3145                 return rc;
3146         } else if (rc) {
3147                 /* Instruction-Fetching Exceptions - we can't detect the ilen.
3148                  * Forward by arbitrary ilc, injection will take care of
3149                  * nullification if necessary.
3150                  */
3151                 pgm_info = vcpu->arch.pgm;
3152                 ilen = 4;
3153         }
3154         pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
3155         kvm_s390_forward_psw(vcpu, ilen);
3156         return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
3157 }
3158
3159 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
3160 {
3161         struct mcck_volatile_info *mcck_info;
3162         struct sie_page *sie_page;
3163
3164         VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
3165                    vcpu->arch.sie_block->icptcode);
3166         trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
3167
3168         if (guestdbg_enabled(vcpu))
3169                 kvm_s390_restore_guest_per_regs(vcpu);
3170
3171         vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
3172         vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
3173
3174         if (exit_reason == -EINTR) {
3175                 VCPU_EVENT(vcpu, 3, "%s", "machine check");
3176                 sie_page = container_of(vcpu->arch.sie_block,
3177                                         struct sie_page, sie_block);
3178                 mcck_info = &sie_page->mcck_info;
3179                 kvm_s390_reinject_machine_check(vcpu, mcck_info);
3180                 return 0;
3181         }
3182
3183         if (vcpu->arch.sie_block->icptcode > 0) {
3184                 int rc = kvm_handle_sie_intercept(vcpu);
3185
3186                 if (rc != -EOPNOTSUPP)
3187                         return rc;
3188                 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
3189                 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
3190                 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
3191                 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
3192                 return -EREMOTE;
3193         } else if (exit_reason != -EFAULT) {
3194                 vcpu->stat.exit_null++;
3195                 return 0;
3196         } else if (kvm_is_ucontrol(vcpu->kvm)) {
3197                 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
3198                 vcpu->run->s390_ucontrol.trans_exc_code =
3199                                                 current->thread.gmap_addr;
3200                 vcpu->run->s390_ucontrol.pgm_code = 0x10;
3201                 return -EREMOTE;
3202         } else if (current->thread.gmap_pfault) {
3203                 trace_kvm_s390_major_guest_pfault(vcpu);
3204                 current->thread.gmap_pfault = 0;
3205                 if (kvm_arch_setup_async_pf(vcpu))
3206                         return 0;
3207                 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
3208         }
3209         return vcpu_post_run_fault_in_sie(vcpu);
3210 }
3211
3212 static int __vcpu_run(struct kvm_vcpu *vcpu)
3213 {
3214         int rc, exit_reason;
3215
3216         /*
3217          * We try to hold kvm->srcu during most of vcpu_run (except when run-
3218          * ning the guest), so that memslots (and other stuff) are protected
3219          */
3220         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3221
3222         do {
3223                 rc = vcpu_pre_run(vcpu);
3224                 if (rc)
3225                         break;
3226
3227                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3228                 /*
3229                  * As PF_VCPU will be used in fault handler, between
3230                  * guest_enter and guest_exit should be no uaccess.
3231                  */
3232                 local_irq_disable();
3233                 guest_enter_irqoff();
3234                 __disable_cpu_timer_accounting(vcpu);
3235                 local_irq_enable();
3236                 exit_reason = sie64a(vcpu->arch.sie_block,
3237                                      vcpu->run->s.regs.gprs);
3238                 local_irq_disable();
3239                 __enable_cpu_timer_accounting(vcpu);
3240                 guest_exit_irqoff();
3241                 local_irq_enable();
3242                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3243
3244                 rc = vcpu_post_run(vcpu, exit_reason);
3245         } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
3246