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, 2018
   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 #define KMSG_COMPONENT "kvm-s390"
  15 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  16
  17 #include <linux/compiler.h>
  18 #include <linux/err.h>
  19 #include <linux/fs.h>
  20 #include <linux/hrtimer.h>
  21 #include <linux/init.h>
  22 #include <linux/kvm.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/mman.h>
  25 #include <linux/module.h>
  26 #include <linux/moduleparam.h>
  27 #include <linux/random.h>
  28 #include <linux/slab.h>
  29 #include <linux/timer.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/bitmap.h>
  32 #include <linux/sched/signal.h>
  33 #include <linux/string.h>
  34
  35 #include <asm/asm-offsets.h>
  36 #include <asm/lowcore.h>
  37 #include <asm/stp.h>
  38 #include <asm/pgtable.h>
  39 #include <asm/gmap.h>
  40 #include <asm/nmi.h>
  41 #include <asm/switch_to.h>
  42 #include <asm/isc.h>
  43 #include <asm/sclp.h>
  44 #include <asm/cpacf.h>
  45 #include <asm/timex.h>
  46 #include <asm/ap.h>
  47 #include "kvm-s390.h"
  48 #include "gaccess.h"
  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 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
  61
  62 struct kvm_stats_debugfs_item debugfs_entries[] = {
  63         { "userspace_handled", VCPU_STAT(exit_userspace) },
  64         { "exit_null", VCPU_STAT(exit_null) },
  65         { "exit_validity", VCPU_STAT(exit_validity) },
  66         { "exit_stop_request", VCPU_STAT(exit_stop_request) },
  67         { "exit_external_request", VCPU_STAT(exit_external_request) },
  68         { "exit_io_request", VCPU_STAT(exit_io_request) },
  69         { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
  70         { "exit_instruction", VCPU_STAT(exit_instruction) },
  71         { "exit_pei", VCPU_STAT(exit_pei) },
  72         { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
  73         { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
  74         { "exit_operation_exception", VCPU_STAT(exit_operation_exception) },
  75         { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
  76         { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
  77         { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
  78         { "halt_no_poll_steal", VCPU_STAT(halt_no_poll_steal) },
  79         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
  80         { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
  81         { "instruction_lctl", VCPU_STAT(instruction_lctl) },
  82         { "instruction_stctl", VCPU_STAT(instruction_stctl) },
  83         { "instruction_stctg", VCPU_STAT(instruction_stctg) },
  84         { "deliver_ckc", VCPU_STAT(deliver_ckc) },
  85         { "deliver_cputm", VCPU_STAT(deliver_cputm) },
  86         { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
  87         { "deliver_external_call", VCPU_STAT(deliver_external_call) },
  88         { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
  89         { "deliver_virtio", VCPU_STAT(deliver_virtio) },
  90         { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
  91         { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
  92         { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
  93         { "deliver_program", VCPU_STAT(deliver_program) },
  94         { "deliver_io", VCPU_STAT(deliver_io) },
  95         { "deliver_machine_check", VCPU_STAT(deliver_machine_check) },
  96         { "exit_wait_state", VCPU_STAT(exit_wait_state) },
  97         { "inject_ckc", VCPU_STAT(inject_ckc) },
  98         { "inject_cputm", VCPU_STAT(inject_cputm) },
  99         { "inject_external_call", VCPU_STAT(inject_external_call) },
 100         { "inject_float_mchk", VM_STAT(inject_float_mchk) },
 101         { "inject_emergency_signal", VCPU_STAT(inject_emergency_signal) },
 102         { "inject_io", VM_STAT(inject_io) },
 103         { "inject_mchk", VCPU_STAT(inject_mchk) },
 104         { "inject_pfault_done", VM_STAT(inject_pfault_done) },
 105         { "inject_program", VCPU_STAT(inject_program) },
 106         { "inject_restart", VCPU_STAT(inject_restart) },
 107         { "inject_service_signal", VM_STAT(inject_service_signal) },
 108         { "inject_set_prefix", VCPU_STAT(inject_set_prefix) },
 109         { "inject_stop_signal", VCPU_STAT(inject_stop_signal) },
 110         { "inject_pfault_init", VCPU_STAT(inject_pfault_init) },
 111         { "inject_virtio", VM_STAT(inject_virtio) },
 112         { "instruction_epsw", VCPU_STAT(instruction_epsw) },
 113         { "instruction_gs", VCPU_STAT(instruction_gs) },
 114         { "instruction_io_other", VCPU_STAT(instruction_io_other) },
 115         { "instruction_lpsw", VCPU_STAT(instruction_lpsw) },
 116         { "instruction_lpswe", VCPU_STAT(instruction_lpswe) },
 117         { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
 118         { "instruction_ptff", VCPU_STAT(instruction_ptff) },
 119         { "instruction_stidp", VCPU_STAT(instruction_stidp) },
 120         { "instruction_sck", VCPU_STAT(instruction_sck) },
 121         { "instruction_sckpf", VCPU_STAT(instruction_sckpf) },
 122         { "instruction_spx", VCPU_STAT(instruction_spx) },
 123         { "instruction_stpx", VCPU_STAT(instruction_stpx) },
 124         { "instruction_stap", VCPU_STAT(instruction_stap) },
 125         { "instruction_iske", VCPU_STAT(instruction_iske) },
 126         { "instruction_ri", VCPU_STAT(instruction_ri) },
 127         { "instruction_rrbe", VCPU_STAT(instruction_rrbe) },
 128         { "instruction_sske", VCPU_STAT(instruction_sske) },
 129         { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
 130         { "instruction_essa", VCPU_STAT(instruction_essa) },
 131         { "instruction_stsi", VCPU_STAT(instruction_stsi) },
 132         { "instruction_stfl", VCPU_STAT(instruction_stfl) },
 133         { "instruction_tb", VCPU_STAT(instruction_tb) },
 134         { "instruction_tpi", VCPU_STAT(instruction_tpi) },
 135         { "instruction_tprot", VCPU_STAT(instruction_tprot) },
 136         { "instruction_tsch", VCPU_STAT(instruction_tsch) },
 137         { "instruction_sthyi", VCPU_STAT(instruction_sthyi) },
 138         { "instruction_sie", VCPU_STAT(instruction_sie) },
 139         { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
 140         { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
 141         { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
 142         { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
 143         { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
 144         { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
 145         { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
 146         { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
 147         { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
 148         { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
 149         { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
 150         { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
 151         { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
 152         { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
 153         { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
 154         { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
 155         { "instruction_diag_10", VCPU_STAT(diagnose_10) },
 156         { "instruction_diag_44", VCPU_STAT(diagnose_44) },
 157         { "instruction_diag_9c", VCPU_STAT(diagnose_9c) },
 158         { "diag_9c_ignored", VCPU_STAT(diagnose_9c_ignored) },
 159         { "instruction_diag_258", VCPU_STAT(diagnose_258) },
 160         { "instruction_diag_308", VCPU_STAT(diagnose_308) },
 161         { "instruction_diag_500", VCPU_STAT(diagnose_500) },
 162         { "instruction_diag_other", VCPU_STAT(diagnose_other) },
 163         { NULL }
 164 };
 165
 166 struct kvm_s390_tod_clock_ext {
 167         __u8 epoch_idx;
 168         __u64 tod;
 169         __u8 reserved[7];
 170 } __packed;
 171
 172 /* allow nested virtualization in KVM (if enabled by user space) */
 173 static int nested;
 174 module_param(nested, int, S_IRUGO);
 175 MODULE_PARM_DESC(nested, "Nested virtualization support");
 176
 177 /* allow 1m huge page guest backing, if !nested */
 178 static int hpage;
 179 module_param(hpage, int, 0444);
 180 MODULE_PARM_DESC(hpage, "1m huge page backing support");
 181
 182 /* maximum percentage of steal time for polling.  >100 is treated like 100 */
 183 static u8 halt_poll_max_steal = 10;
 184 module_param(halt_poll_max_steal, byte, 0644);
 185 MODULE_PARM_DESC(halt_poll_max_steal, "Maximum percentage of steal time to allow polling");
 186
 187 /*
 188  * For now we handle at most 16 double words as this is what the s390 base
 189  * kernel handles and stores in the prefix page. If we ever need to go beyond
 190  * this, this requires changes to code, but the external uapi can stay.
 191  */
 192 #define SIZE_INTERNAL 16
 193
 194 /*
 195  * Base feature mask that defines default mask for facilities. Consists of the
 196  * defines in FACILITIES_KVM and the non-hypervisor managed bits.
 197  */
 198 static unsigned long kvm_s390_fac_base[SIZE_INTERNAL] = { FACILITIES_KVM };
 199 /*
 200  * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL
 201  * and defines the facilities that can be enabled via a cpu model.
 202  */
 203 static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL] = { FACILITIES_KVM_CPUMODEL };
 204
 205 static unsigned long kvm_s390_fac_size(void)
 206 {
 207         BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_MASK_SIZE_U64);
 208         BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_LIST_SIZE_U64);
 209         BUILD_BUG_ON(SIZE_INTERNAL * sizeof(unsigned long) >
 210                 sizeof(S390_lowcore.stfle_fac_list));
 211
 212         return SIZE_INTERNAL;
 213 }
 214
 215 /* available cpu features supported by kvm */
 216 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
 217 /* available subfunctions indicated via query / "test bit" */
 218 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
 219
 220 static struct gmap_notifier gmap_notifier;
 221 static struct gmap_notifier vsie_gmap_notifier;
 222 debug_info_t *kvm_s390_dbf;
 223
 224 /* Section: not file related */
 225 int kvm_arch_hardware_enable(void)
 226 {
 227         /* every s390 is virtualization enabled ;-) */
 228         return 0;
 229 }
 230
 231 int kvm_arch_check_processor_compat(void)
 232 {
 233         return 0;
 234 }
 235
 236 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
 237                               unsigned long end);
 238
 239 static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta)
 240 {
 241         u8 delta_idx = 0;
 242
 243         /*
 244          * The TOD jumps by delta, we have to compensate this by adding
 245          * -delta to the epoch.
 246          */
 247         delta = -delta;
 248
 249         /* sign-extension - we're adding to signed values below */
 250         if ((s64)delta < 0)
 251                 delta_idx = -1;
 252
 253         scb->epoch += delta;
 254         if (scb->ecd & ECD_MEF) {
 255                 scb->epdx += delta_idx;
 256                 if (scb->epoch < delta)
 257                         scb->epdx += 1;
 258         }
 259 }
 260
 261 /*
 262  * This callback is executed during stop_machine(). All CPUs are therefore
 263  * temporarily stopped. In order not to change guest behavior, we have to
 264  * disable preemption whenever we touch the epoch of kvm and the VCPUs,
 265  * so a CPU won't be stopped while calculating with the epoch.
 266  */
 267 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
 268                           void *v)
 269 {
 270         struct kvm *kvm;
 271         struct kvm_vcpu *vcpu;
 272         int i;
 273         unsigned long long *delta = v;
 274
 275         list_for_each_entry(kvm, &vm_list, vm_list) {
 276                 kvm_for_each_vcpu(i, vcpu, kvm) {
 277                         kvm_clock_sync_scb(vcpu->arch.sie_block, *delta);
 278                         if (i == 0) {
 279                                 kvm->arch.epoch = vcpu->arch.sie_block->epoch;
 280                                 kvm->arch.epdx = vcpu->arch.sie_block->epdx;
 281                         }
 282                         if (vcpu->arch.cputm_enabled)
 283                                 vcpu->arch.cputm_start += *delta;
 284                         if (vcpu->arch.vsie_block)
 285                                 kvm_clock_sync_scb(vcpu->arch.vsie_block,
 286                                                    *delta);
 287                 }
 288         }
 289         return NOTIFY_OK;
 290 }
 291
 292 static struct notifier_block kvm_clock_notifier = {
 293         .notifier_call = kvm_clock_sync,
 294 };
 295
 296 int kvm_arch_hardware_setup(void)
 297 {
 298         gmap_notifier.notifier_call = kvm_gmap_notifier;
 299         gmap_register_pte_notifier(&gmap_notifier);
 300         vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
 301         gmap_register_pte_notifier(&vsie_gmap_notifier);
 302         atomic_notifier_chain_register(&s390_epoch_delta_notifier,
 303                                        &kvm_clock_notifier);
 304         return 0;
 305 }
 306
 307 void kvm_arch_hardware_unsetup(void)
 308 {
 309         gmap_unregister_pte_notifier(&gmap_notifier);
 310         gmap_unregister_pte_notifier(&vsie_gmap_notifier);
 311         atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
 312                                          &kvm_clock_notifier);
 313 }
 314
 315 static void allow_cpu_feat(unsigned long nr)
 316 {
 317         set_bit_inv(nr, kvm_s390_available_cpu_feat);
 318 }
 319
 320 static inline int plo_test_bit(unsigned char nr)
 321 {
 322         register unsigned long r0 asm("0") = (unsigned long) nr | 0x100;
 323         int cc;
 324
 325         asm volatile(
 326                 /* Parameter registers are ignored for "test bit" */
 327                 "       plo     0,0,0,0(0)\n"
 328                 "       ipm     %0\n"
 329                 "       srl     %0,28\n"
 330                 : "=d" (cc)
 331                 : "d" (r0)
 332                 : "cc");
 333         return cc == 0;
 334 }
 335
 336 static __always_inline void __insn32_query(unsigned int opcode, u8 *query)
 337 {
 338         register unsigned long r0 asm("0") = 0; /* query function */
 339         register unsigned long r1 asm("1") = (unsigned long) query;
 340
 341         asm volatile(
 342                 /* Parameter regs are ignored */
 343                 "       .insn   rrf,%[opc] << 16,2,4,6,0\n"
 344                 :
 345                 : "d" (r0), "a" (r1), [opc] "i" (opcode)
 346                 : "cc", "memory");
 347 }
 348
 349 #define INSN_SORTL 0xb938
 350 #define INSN_DFLTCC 0xb939
 351
 352 static void kvm_s390_cpu_feat_init(void)
 353 {
 354         int i;
 355
 356         for (i = 0; i < 256; ++i) {
 357                 if (plo_test_bit(i))
 358                         kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
 359         }
 360
 361         if (test_facility(28)) /* TOD-clock steering */
 362                 ptff(kvm_s390_available_subfunc.ptff,
 363                      sizeof(kvm_s390_available_subfunc.ptff),
 364                      PTFF_QAF);
 365
 366         if (test_facility(17)) { /* MSA */
 367                 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
 368                               kvm_s390_available_subfunc.kmac);
 369                 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
 370                               kvm_s390_available_subfunc.kmc);
 371                 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
 372                               kvm_s390_available_subfunc.km);
 373                 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
 374                               kvm_s390_available_subfunc.kimd);
 375                 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
 376                               kvm_s390_available_subfunc.klmd);
 377         }
 378         if (test_facility(76)) /* MSA3 */
 379                 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
 380                               kvm_s390_available_subfunc.pckmo);
 381         if (test_facility(77)) { /* MSA4 */
 382                 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
 383                               kvm_s390_available_subfunc.kmctr);
 384                 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
 385                               kvm_s390_available_subfunc.kmf);
 386                 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
 387                               kvm_s390_available_subfunc.kmo);
 388                 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
 389                               kvm_s390_available_subfunc.pcc);
 390         }
 391         if (test_facility(57)) /* MSA5 */
 392                 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
 393                               kvm_s390_available_subfunc.ppno);
 394
 395         if (test_facility(146)) /* MSA8 */
 396                 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
 397                               kvm_s390_available_subfunc.kma);
 398
 399         if (test_facility(155)) /* MSA9 */
 400                 __cpacf_query(CPACF_KDSA, (cpacf_mask_t *)
 401                               kvm_s390_available_subfunc.kdsa);
 402
 403         if (test_facility(150)) /* SORTL */
 404                 __insn32_query(INSN_SORTL, kvm_s390_available_subfunc.sortl);
 405
 406         if (test_facility(151)) /* DFLTCC */
 407                 __insn32_query(INSN_DFLTCC, kvm_s390_available_subfunc.dfltcc);
 408
 409         if (MACHINE_HAS_ESOP)
 410                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
 411         /*
 412          * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
 413          * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
 414          */
 415         if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
 416             !test_facility(3) || !nested)
 417                 return;
 418         allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
 419         if (sclp.has_64bscao)
 420                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
 421         if (sclp.has_siif)
 422                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
 423         if (sclp.has_gpere)
 424                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
 425         if (sclp.has_gsls)
 426                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
 427         if (sclp.has_ib)
 428                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
 429         if (sclp.has_cei)
 430                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
 431         if (sclp.has_ibs)
 432                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
 433         if (sclp.has_kss)
 434                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
 435         /*
 436          * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
 437          * all skey handling functions read/set the skey from the PGSTE
 438          * instead of the real storage key.
 439          *
 440          * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
 441          * pages being detected as preserved although they are resident.
 442          *
 443          * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
 444          * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
 445          *
 446          * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
 447          * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
 448          * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
 449          *
 450          * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
 451          * cannot easily shadow the SCA because of the ipte lock.
 452          */
 453 }
 454
 455 int kvm_arch_init(void *opaque)
 456 {
 457         int rc = -ENOMEM;
 458
 459         kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
 460         if (!kvm_s390_dbf)
 461                 return -ENOMEM;
 462
 463         if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view))
 464                 goto out;
 465
 466         kvm_s390_cpu_feat_init();
 467
 468         /* Register floating interrupt controller interface. */
 469         rc = kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
 470         if (rc) {
 471                 pr_err("A FLIC registration call failed with rc=%d\n", rc);
 472                 goto out;
 473         }
 474
 475         rc = kvm_s390_gib_init(GAL_ISC);
 476         if (rc)
 477                 goto out;
 478
 479         return 0;
 480
 481 out:
 482         kvm_arch_exit();
 483         return rc;
 484 }
 485
 486 void kvm_arch_exit(void)
 487 {
 488         kvm_s390_gib_destroy();
 489         debug_unregister(kvm_s390_dbf);
 490 }
 491
 492 /* Section: device related */
 493 long kvm_arch_dev_ioctl(struct file *filp,
 494                         unsigned int ioctl, unsigned long arg)
 495 {
 496         if (ioctl == KVM_S390_ENABLE_SIE)
 497                 return s390_enable_sie();
 498         return -EINVAL;
 499 }
 500
 501 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 502 {
 503         int r;
 504
 505         switch (ext) {
 506         case KVM_CAP_S390_PSW:
 507         case KVM_CAP_S390_GMAP:
 508         case KVM_CAP_SYNC_MMU:
 509 #ifdef CONFIG_KVM_S390_UCONTROL
 510         case KVM_CAP_S390_UCONTROL:
 511 #endif
 512         case KVM_CAP_ASYNC_PF:
 513         case KVM_CAP_SYNC_REGS:
 514         case KVM_CAP_ONE_REG:
 515         case KVM_CAP_ENABLE_CAP:
 516         case KVM_CAP_S390_CSS_SUPPORT:
 517         case KVM_CAP_IOEVENTFD:
 518         case KVM_CAP_DEVICE_CTRL:
 519         case KVM_CAP_S390_IRQCHIP:
 520         case KVM_CAP_VM_ATTRIBUTES:
 521         case KVM_CAP_MP_STATE:
 522         case KVM_CAP_IMMEDIATE_EXIT:
 523         case KVM_CAP_S390_INJECT_IRQ:
 524         case KVM_CAP_S390_USER_SIGP:
 525         case KVM_CAP_S390_USER_STSI:
 526         case KVM_CAP_S390_SKEYS:
 527         case KVM_CAP_S390_IRQ_STATE:
 528         case KVM_CAP_S390_USER_INSTR0:
 529         case KVM_CAP_S390_CMMA_MIGRATION:
 530         case KVM_CAP_S390_AIS:
 531         case KVM_CAP_S390_AIS_MIGRATION:
 532         case KVM_CAP_S390_VCPU_RESETS:
 533                 r = 1;
 534                 break;
 535         case KVM_CAP_S390_HPAGE_1M:
 536                 r = 0;
 537                 if (hpage && !kvm_is_ucontrol(kvm))
 538                         r = 1;
 539                 break;
 540         case KVM_CAP_S390_MEM_OP:
 541                 r = MEM_OP_MAX_SIZE;
 542                 break;
 543         case KVM_CAP_NR_VCPUS:
 544         case KVM_CAP_MAX_VCPUS:
 545         case KVM_CAP_MAX_VCPU_ID:
 546                 r = KVM_S390_BSCA_CPU_SLOTS;
 547                 if (!kvm_s390_use_sca_entries())
 548                         r = KVM_MAX_VCPUS;
 549                 else if (sclp.has_esca && sclp.has_64bscao)
 550                         r = KVM_S390_ESCA_CPU_SLOTS;
 551                 break;
 552         case KVM_CAP_S390_COW:
 553                 r = MACHINE_HAS_ESOP;
 554                 break;
 555         case KVM_CAP_S390_VECTOR_REGISTERS:
 556                 r = MACHINE_HAS_VX;
 557                 break;
 558         case KVM_CAP_S390_RI:
 559                 r = test_facility(64);
 560                 break;
 561         case KVM_CAP_S390_GS:
 562                 r = test_facility(133);
 563                 break;
 564         case KVM_CAP_S390_BPB:
 565                 r = test_facility(82);
 566                 break;
 567         default:
 568                 r = 0;
 569         }
 570         return r;
 571 }
 572
 573 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
 574                                     struct kvm_memory_slot *memslot)
 575 {
 576         int i;
 577         gfn_t cur_gfn, last_gfn;
 578         unsigned long gaddr, vmaddr;
 579         struct gmap *gmap = kvm->arch.gmap;
 580         DECLARE_BITMAP(bitmap, _PAGE_ENTRIES);
 581
 582         /* Loop over all guest segments */
 583         cur_gfn = memslot->base_gfn;
 584         last_gfn = memslot->base_gfn + memslot->npages;
 585         for (; cur_gfn <= last_gfn; cur_gfn += _PAGE_ENTRIES) {
 586                 gaddr = gfn_to_gpa(cur_gfn);
 587                 vmaddr = gfn_to_hva_memslot(memslot, cur_gfn);
 588                 if (kvm_is_error_hva(vmaddr))
 589                         continue;
 590
 591                 bitmap_zero(bitmap, _PAGE_ENTRIES);
 592                 gmap_sync_dirty_log_pmd(gmap, bitmap, gaddr, vmaddr);
 593                 for (i = 0; i < _PAGE_ENTRIES; i++) {
 594                         if (test_bit(i, bitmap))
 595                                 mark_page_dirty(kvm, cur_gfn + i);
 596                 }
 597
 598                 if (fatal_signal_pending(current))
 599                         return;
 600                 cond_resched();
 601         }
 602 }
 603
 604 /* Section: vm related */
 605 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
 606
 607 /*
 608  * Get (and clear) the dirty memory log for a memory slot.
 609  */
 610 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 611                                struct kvm_dirty_log *log)
 612 {
 613         int r;
 614         unsigned long n;
 615         struct kvm_memslots *slots;
 616         struct kvm_memory_slot *memslot;
 617         int is_dirty = 0;
 618
 619         if (kvm_is_ucontrol(kvm))
 620                 return -EINVAL;
 621
 622         mutex_lock(&kvm->slots_lock);
 623
 624         r = -EINVAL;
 625         if (log->slot >= KVM_USER_MEM_SLOTS)
 626                 goto out;
 627
 628         slots = kvm_memslots(kvm);
 629         memslot = id_to_memslot(slots, log->slot);
 630         r = -ENOENT;
 631         if (!memslot->dirty_bitmap)
 632                 goto out;
 633
 634         kvm_s390_sync_dirty_log(kvm, memslot);
 635         r = kvm_get_dirty_log(kvm, log, &is_dirty);
 636         if (r)
 637                 goto out;
 638
 639         /* Clear the dirty log */
 640         if (is_dirty) {
 641                 n = kvm_dirty_bitmap_bytes(memslot);
 642                 memset(memslot->dirty_bitmap, 0, n);
 643         }
 644         r = 0;
 645 out:
 646         mutex_unlock(&kvm->slots_lock);
 647         return r;
 648 }
 649
 650 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
 651 {
 652         unsigned int i;
 653         struct kvm_vcpu *vcpu;
 654
 655         kvm_for_each_vcpu(i, vcpu, kvm) {
 656                 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
 657         }
 658 }
 659
 660 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
 661 {
 662         int r;
 663
 664         if (cap->flags)
 665                 return -EINVAL;
 666
 667         switch (cap->cap) {
 668         case KVM_CAP_S390_IRQCHIP:
 669                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
 670                 kvm->arch.use_irqchip = 1;
 671                 r = 0;
 672                 break;
 673         case KVM_CAP_S390_USER_SIGP:
 674                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
 675                 kvm->arch.user_sigp = 1;
 676                 r = 0;
 677                 break;
 678         case KVM_CAP_S390_VECTOR_REGISTERS:
 679                 mutex_lock(&kvm->lock);
 680                 if (kvm->created_vcpus) {
 681                         r = -EBUSY;
 682                 } else if (MACHINE_HAS_VX) {
 683                         set_kvm_facility(kvm->arch.model.fac_mask, 129);
 684                         set_kvm_facility(kvm->arch.model.fac_list, 129);
 685                         if (test_facility(134)) {
 686                                 set_kvm_facility(kvm->arch.model.fac_mask, 134);
 687                                 set_kvm_facility(kvm->arch.model.fac_list, 134);
 688                         }
 689                         if (test_facility(135)) {
 690                                 set_kvm_facility(kvm->arch.model.fac_mask, 135);
 691                                 set_kvm_facility(kvm->arch.model.fac_list, 135);
 692                         }
 693                         if (test_facility(148)) {
 694                                 set_kvm_facility(kvm->arch.model.fac_mask, 148);
 695                                 set_kvm_facility(kvm->arch.model.fac_list, 148);
 696                         }
 697                         if (test_facility(152)) {
 698                                 set_kvm_facility(kvm->arch.model.fac_mask, 152);
 699                                 set_kvm_facility(kvm->arch.model.fac_list, 152);
 700                         }
 701                         r = 0;
 702                 } else
 703                         r = -EINVAL;
 704                 mutex_unlock(&kvm->lock);
 705                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
 706                          r ? "(not available)" : "(success)");
 707                 break;
 708         case KVM_CAP_S390_RI:
 709                 r = -EINVAL;
 710                 mutex_lock(&kvm->lock);
 711                 if (kvm->created_vcpus) {
 712                         r = -EBUSY;
 713                 } else if (test_facility(64)) {
 714                         set_kvm_facility(kvm->arch.model.fac_mask, 64);
 715                         set_kvm_facility(kvm->arch.model.fac_list, 64);
 716                         r = 0;
 717                 }
 718                 mutex_unlock(&kvm->lock);
 719                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
 720                          r ? "(not available)" : "(success)");
 721                 break;
 722         case KVM_CAP_S390_AIS:
 723                 mutex_lock(&kvm->lock);
 724                 if (kvm->created_vcpus) {
 725                         r = -EBUSY;
 726                 } else {
 727                         set_kvm_facility(kvm->arch.model.fac_mask, 72);
 728                         set_kvm_facility(kvm->arch.model.fac_list, 72);
 729                         r = 0;
 730                 }
 731                 mutex_unlock(&kvm->lock);
 732                 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
 733                          r ? "(not available)" : "(success)");
 734                 break;
 735         case KVM_CAP_S390_GS:
 736                 r = -EINVAL;
 737                 mutex_lock(&kvm->lock);
 738                 if (kvm->created_vcpus) {
 739                         r = -EBUSY;
 740                 } else if (test_facility(133)) {
 741                         set_kvm_facility(kvm->arch.model.fac_mask, 133);
 742                         set_kvm_facility(kvm->arch.model.fac_list, 133);
 743                         r = 0;
 744                 }
 745                 mutex_unlock(&kvm->lock);
 746                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
 747                          r ? "(not available)" : "(success)");
 748                 break;
 749         case KVM_CAP_S390_HPAGE_1M:
 750                 mutex_lock(&kvm->lock);
 751                 if (kvm->created_vcpus)
 752                         r = -EBUSY;
 753                 else if (!hpage || kvm->arch.use_cmma || kvm_is_ucontrol(kvm))
 754                         r = -EINVAL;
 755                 else {
 756                         r = 0;
 757                         down_write(&kvm->mm->mmap_sem);
 758                         kvm->mm->context.allow_gmap_hpage_1m = 1;
 759                         up_write(&kvm->mm->mmap_sem);
 760                         /*
 761                          * We might have to create fake 4k page
 762                          * tables. To avoid that the hardware works on
 763                          * stale PGSTEs, we emulate these instructions.
 764                          */
 765                         kvm->arch.use_skf = 0;
 766                         kvm->arch.use_pfmfi = 0;
 767                 }
 768                 mutex_unlock(&kvm->lock);
 769                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_HPAGE %s",
 770                          r ? "(not available)" : "(success)");
 771                 break;
 772         case KVM_CAP_S390_USER_STSI:
 773                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
 774                 kvm->arch.user_stsi = 1;
 775                 r = 0;
 776                 break;
 777         case KVM_CAP_S390_USER_INSTR0:
 778                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
 779                 kvm->arch.user_instr0 = 1;
 780                 icpt_operexc_on_all_vcpus(kvm);
 781                 r = 0;
 782                 break;
 783         default:
 784                 r = -EINVAL;
 785                 break;
 786         }
 787         return r;
 788 }
 789
 790 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
 791 {
 792         int ret;
 793
 794         switch (attr->attr) {
 795         case KVM_S390_VM_MEM_LIMIT_SIZE:
 796                 ret = 0;
 797                 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
 798                          kvm->arch.mem_limit);
 799                 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
 800                         ret = -EFAULT;
 801                 break;
 802         default:
 803                 ret = -ENXIO;
 804                 break;
 805         }
 806         return ret;
 807 }
 808
 809 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
 810 {
 811         int ret;
 812         unsigned int idx;
 813         switch (attr->attr) {
 814         case KVM_S390_VM_MEM_ENABLE_CMMA:
 815                 ret = -ENXIO;
 816                 if (!sclp.has_cmma)
 817                         break;
 818
 819                 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
 820                 mutex_lock(&kvm->lock);
 821                 if (kvm->created_vcpus)
 822                         ret = -EBUSY;
 823                 else if (kvm->mm->context.allow_gmap_hpage_1m)
 824                         ret = -EINVAL;
 825                 else {
 826                         kvm->arch.use_cmma = 1;
 827                         /* Not compatible with cmma. */
 828                         kvm->arch.use_pfmfi = 0;
 829                         ret = 0;
 830                 }
 831                 mutex_unlock(&kvm->lock);
 832                 break;
 833         case KVM_S390_VM_MEM_CLR_CMMA:
 834                 ret = -ENXIO;
 835                 if (!sclp.has_cmma)
 836                         break;
 837                 ret = -EINVAL;
 838                 if (!kvm->arch.use_cmma)
 839                         break;
 840
 841                 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
 842                 mutex_lock(&kvm->lock);
 843                 idx = srcu_read_lock(&kvm->srcu);
 844                 s390_reset_cmma(kvm->arch.gmap->mm);
 845                 srcu_read_unlock(&kvm->srcu, idx);
 846                 mutex_unlock(&kvm->lock);
 847                 ret = 0;
 848                 break;
 849         case KVM_S390_VM_MEM_LIMIT_SIZE: {
 850                 unsigned long new_limit;
 851
 852                 if (kvm_is_ucontrol(kvm))
 853                         return -EINVAL;
 854
 855                 if (get_user(new_limit, (u64 __user *)attr->addr))
 856                         return -EFAULT;
 857
 858                 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
 859                     new_limit > kvm->arch.mem_limit)
 860                         return -E2BIG;
 861
 862                 if (!new_limit)
 863                         return -EINVAL;
 864
 865                 /* gmap_create takes last usable address */
 866                 if (new_limit != KVM_S390_NO_MEM_LIMIT)
 867                         new_limit -= 1;
 868
 869                 ret = -EBUSY;
 870                 mutex_lock(&kvm->lock);
 871                 if (!kvm->created_vcpus) {
 872                         /* gmap_create will round the limit up */
 873                         struct gmap *new = gmap_create(current->mm, new_limit);
 874
 875                         if (!new) {
 876                                 ret = -ENOMEM;
 877                         } else {
 878                                 gmap_remove(kvm->arch.gmap);
 879                                 new->private = kvm;
 880                                 kvm->arch.gmap = new;
 881                                 ret = 0;
 882                         }
 883                 }
 884                 mutex_unlock(&kvm->lock);
 885                 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
 886                 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
 887                          (void *) kvm->arch.gmap->asce);
 888                 break;
 889         }
 890         default:
 891                 ret = -ENXIO;
 892                 break;
 893         }
 894         return ret;
 895 }
 896
 897 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
 898
 899 void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm)
 900 {
 901         struct kvm_vcpu *vcpu;
 902         int i;
 903
 904         kvm_s390_vcpu_block_all(kvm);
 905
 906         kvm_for_each_vcpu(i, vcpu, kvm) {
 907                 kvm_s390_vcpu_crypto_setup(vcpu);
 908                 /* recreate the shadow crycb by leaving the VSIE handler */
 909                 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
 910         }
 911
 912         kvm_s390_vcpu_unblock_all(kvm);
 913 }
 914
 915 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
 916 {
 917         mutex_lock(&kvm->lock);
 918         switch (attr->attr) {
 919         case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
 920                 if (!test_kvm_facility(kvm, 76)) {
 921                         mutex_unlock(&kvm->lock);
 922                         return -EINVAL;
 923                 }
 924                 get_random_bytes(
 925                         kvm->arch.crypto.crycb->aes_wrapping_key_mask,
 926                         sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
 927                 kvm->arch.crypto.aes_kw = 1;
 928                 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
 929                 break;
 930         case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
 931                 if (!test_kvm_facility(kvm, 76)) {
 932                         mutex_unlock(&kvm->lock);
 933                         return -EINVAL;
 934                 }
 935                 get_random_bytes(
 936                         kvm->arch.crypto.crycb->dea_wrapping_key_mask,
 937                         sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
 938                 kvm->arch.crypto.dea_kw = 1;
 939                 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
 940                 break;
 941         case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
 942                 if (!test_kvm_facility(kvm, 76)) {
 943                         mutex_unlock(&kvm->lock);
 944                         return -EINVAL;
 945                 }
 946                 kvm->arch.crypto.aes_kw = 0;
 947                 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
 948                         sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
 949                 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
 950                 break;
 951         case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
 952                 if (!test_kvm_facility(kvm, 76)) {
 953                         mutex_unlock(&kvm->lock);
 954                         return -EINVAL;
 955                 }
 956                 kvm->arch.crypto.dea_kw = 0;
 957                 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
 958                         sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
 959                 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
 960                 break;
 961         case KVM_S390_VM_CRYPTO_ENABLE_APIE:
 962                 if (!ap_instructions_available()) {
 963                         mutex_unlock(&kvm->lock);
 964                         return -EOPNOTSUPP;
 965                 }
 966                 kvm->arch.crypto.apie = 1;
 967                 break;
 968         case KVM_S390_VM_CRYPTO_DISABLE_APIE:
 969                 if (!ap_instructions_available()) {
 970                         mutex_unlock(&kvm->lock);
 971                         return -EOPNOTSUPP;
 972                 }
 973                 kvm->arch.crypto.apie = 0;
 974                 break;
 975         default:
 976                 mutex_unlock(&kvm->lock);
 977                 return -ENXIO;
 978         }
 979
 980         kvm_s390_vcpu_crypto_reset_all(kvm);
 981         mutex_unlock(&kvm->lock);
 982         return 0;
 983 }
 984
 985 static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req)
 986 {
 987         int cx;
 988         struct kvm_vcpu *vcpu;
 989
 990         kvm_for_each_vcpu(cx, vcpu, kvm)
 991                 kvm_s390_sync_request(req, vcpu);
 992 }
 993
 994 /*
 995  * Must be called with kvm->srcu held to avoid races on memslots, and with
 996  * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration.
 997  */
 998 static int kvm_s390_vm_start_migration(struct kvm *kvm)
 999 {
1000         struct kvm_memory_slot *ms;
1001         struct kvm_memslots *slots;
1002         unsigned long ram_pages = 0;
1003         int slotnr;
1004
1005         /* migration mode already enabled */
1006         if (kvm->arch.migration_mode)
1007                 return 0;
1008         slots = kvm_memslots(kvm);
1009         if (!slots || !slots->used_slots)
1010                 return -EINVAL;
1011
1012         if (!kvm->arch.use_cmma) {
1013                 kvm->arch.migration_mode = 1;
1014                 return 0;
1015         }
1016         /* mark all the pages in active slots as dirty */
1017         for (slotnr = 0; slotnr < slots->used_slots; slotnr++) {
1018                 ms = slots->memslots + slotnr;
1019                 if (!ms->dirty_bitmap)
1020                         return -EINVAL;
1021                 /*
1022                  * The second half of the bitmap is only used on x86,
1023                  * and would be wasted otherwise, so we put it to good
1024                  * use here to keep track of the state of the storage
1025                  * attributes.
1026                  */
1027                 memset(kvm_second_dirty_bitmap(ms), 0xff, kvm_dirty_bitmap_bytes(ms));
1028                 ram_pages += ms->npages;
1029         }
1030         atomic64_set(&kvm->arch.cmma_dirty_pages, ram_pages);
1031         kvm->arch.migration_mode = 1;
1032         kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION);
1033         return 0;
1034 }
1035
1036 /*
1037  * Must be called with kvm->slots_lock to avoid races with ourselves and
1038  * kvm_s390_vm_start_migration.
1039  */
1040 static int kvm_s390_vm_stop_migration(struct kvm *kvm)
1041 {
1042         /* migration mode already disabled */
1043         if (!kvm->arch.migration_mode)
1044                 return 0;
1045         kvm->arch.migration_mode = 0;
1046         if (kvm->arch.use_cmma)
1047                 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION);
1048         return 0;
1049 }
1050
1051 static int kvm_s390_vm_set_migration(struct kvm *kvm,
1052                                      struct kvm_device_attr *attr)
1053 {
1054         int res = -ENXIO;
1055
1056         mutex_lock(&kvm->slots_lock);
1057         switch (attr->attr) {
1058         case KVM_S390_VM_MIGRATION_START:
1059                 res = kvm_s390_vm_start_migration(kvm);
1060                 break;
1061         case KVM_S390_VM_MIGRATION_STOP:
1062                 res = kvm_s390_vm_stop_migration(kvm);
1063                 break;
1064         default:
1065                 break;
1066         }
1067         mutex_unlock(&kvm->slots_lock);
1068
1069         return res;
1070 }
1071
1072 static int kvm_s390_vm_get_migration(struct kvm *kvm,
1073                                      struct kvm_device_attr *attr)
1074 {
1075         u64 mig = kvm->arch.migration_mode;
1076
1077         if (attr->attr != KVM_S390_VM_MIGRATION_STATUS)
1078                 return -ENXIO;
1079
1080         if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig)))
1081                 return -EFAULT;
1082         return 0;
1083 }
1084
1085 static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1086 {
1087         struct kvm_s390_vm_tod_clock gtod;
1088
1089         if (copy_from_user(&gtod, (void __user *)attr->addr, sizeof(gtod)))
1090                 return -EFAULT;
1091
1092         if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx)
1093                 return -EINVAL;
1094         kvm_s390_set_tod_clock(kvm, &gtod);
1095
1096         VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx",
1097                 gtod.epoch_idx, gtod.tod);
1098
1099         return 0;
1100 }
1101
1102 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1103 {
1104         u8 gtod_high;
1105
1106         if (copy_from_user(&gtod_high, (void __user *)attr->addr,
1107                                            sizeof(gtod_high)))
1108                 return -EFAULT;
1109
1110         if (gtod_high != 0)
1111                 return -EINVAL;
1112         VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
1113
1114         return 0;
1115 }
1116
1117 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1118 {
1119         struct kvm_s390_vm_tod_clock gtod = { 0 };
1120
1121         if (copy_from_user(&gtod.tod, (void __user *)attr->addr,
1122                            sizeof(gtod.tod)))
1123                 return -EFAULT;
1124
1125         kvm_s390_set_tod_clock(kvm, &gtod);
1126         VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod);
1127         return 0;
1128 }
1129
1130 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1131 {
1132         int ret;
1133
1134         if (attr->flags)
1135                 return -EINVAL;
1136
1137         switch (attr->attr) {
1138         case KVM_S390_VM_TOD_EXT:
1139                 ret = kvm_s390_set_tod_ext(kvm, attr);
1140                 break;
1141         case KVM_S390_VM_TOD_HIGH:
1142                 ret = kvm_s390_set_tod_high(kvm, attr);
1143                 break;
1144         case KVM_S390_VM_TOD_LOW:
1145                 ret = kvm_s390_set_tod_low(kvm, attr);
1146                 break;
1147         default:
1148                 ret = -ENXIO;
1149                 break;
1150         }
1151         return ret;
1152 }
1153
1154 static void kvm_s390_get_tod_clock(struct kvm *kvm,
1155                                    struct kvm_s390_vm_tod_clock *gtod)
1156 {
1157         struct kvm_s390_tod_clock_ext htod;
1158
1159         preempt_disable();
1160
1161         get_tod_clock_ext((char *)&htod);
1162
1163         gtod->tod = htod.tod + kvm->arch.epoch;
1164         gtod->epoch_idx = 0;
1165         if (test_kvm_facility(kvm, 139)) {
1166                 gtod->epoch_idx = htod.epoch_idx + kvm->arch.epdx;
1167                 if (gtod->tod < htod.tod)
1168                         gtod->epoch_idx += 1;
1169         }
1170
1171         preempt_enable();
1172 }
1173
1174 static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1175 {
1176         struct kvm_s390_vm_tod_clock gtod;
1177
1178         memset(&gtod, 0, sizeof(gtod));
1179         kvm_s390_get_tod_clock(kvm, &gtod);
1180         if (copy_to_user((void __user *)attr->addr, &gtod, sizeof(gtod)))
1181                 return -EFAULT;
1182
1183         VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx",
1184                 gtod.epoch_idx, gtod.tod);
1185         return 0;
1186 }
1187
1188 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1189 {
1190         u8 gtod_high = 0;
1191
1192         if (copy_to_user((void __user *)attr->addr, &gtod_high,
1193                                          sizeof(gtod_high)))
1194                 return -EFAULT;
1195         VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
1196
1197         return 0;
1198 }
1199
1200 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1201 {
1202         u64 gtod;
1203
1204         gtod = kvm_s390_get_tod_clock_fast(kvm);
1205         if (copy_to_user((void __user *)attr->addr, &gtod, sizeof(gtod)))
1206                 return -EFAULT;
1207         VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
1208
1209         return 0;
1210 }
1211
1212 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1213 {
1214         int ret;
1215
1216         if (attr->flags)
1217                 return -EINVAL;
1218
1219         switch (attr->attr) {
1220         case KVM_S390_VM_TOD_EXT:
1221                 ret = kvm_s390_get_tod_ext(kvm, attr);
1222                 break;
1223         case KVM_S390_VM_TOD_HIGH:
1224                 ret = kvm_s390_get_tod_high(kvm, attr);
1225                 break;
1226         case KVM_S390_VM_TOD_LOW:
1227                 ret = kvm_s390_get_tod_low(kvm, attr);
1228                 break;
1229         default:
1230                 ret = -ENXIO;
1231                 break;
1232         }
1233         return ret;
1234 }
1235
1236 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1237 {
1238         struct kvm_s390_vm_cpu_processor *proc;
1239         u16 lowest_ibc, unblocked_ibc;
1240         int ret = 0;
1241
1242         mutex_lock(&kvm->lock);
1243         if (kvm->created_vcpus) {
1244                 ret = -EBUSY;
1245                 goto out;
1246         }
1247         proc = kzalloc(sizeof(*proc), GFP_KERNEL);
1248         if (!proc) {
1249                 ret = -ENOMEM;
1250                 goto out;
1251         }
1252         if (!copy_from_user(proc, (void __user *)attr->addr,
1253                             sizeof(*proc))) {
1254                 kvm->arch.model.cpuid = proc->cpuid;
1255                 lowest_ibc = sclp.ibc >> 16 & 0xfff;
1256                 unblocked_ibc = sclp.ibc & 0xfff;
1257                 if (lowest_ibc && proc->ibc) {
1258                         if (proc->ibc > unblocked_ibc)
1259                                 kvm->arch.model.ibc = unblocked_ibc;
1260                         else if (proc->ibc < lowest_ibc)
1261                                 kvm->arch.model.ibc = lowest_ibc;
1262                         else
1263                                 kvm->arch.model.ibc = proc->ibc;
1264                 }
1265                 memcpy(kvm->arch.model.fac_list, proc->fac_list,
1266                        S390_ARCH_FAC_LIST_SIZE_BYTE);
1267                 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1268                          kvm->arch.model.ibc,
1269                          kvm->arch.model.cpuid);
1270                 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1271                          kvm->arch.model.fac_list[0],
1272                          kvm->arch.model.fac_list[1],
1273                          kvm->arch.model.fac_list[2]);
1274         } else
1275                 ret = -EFAULT;
1276         kfree(proc);
1277 out:
1278         mutex_unlock(&kvm->lock);
1279         return ret;
1280 }
1281
1282 static int kvm_s390_set_processor_feat(struct kvm *kvm,
1283                                        struct kvm_device_attr *attr)
1284 {
1285         struct kvm_s390_vm_cpu_feat data;
1286
1287         if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
1288                 return -EFAULT;
1289         if (!bitmap_subset((unsigned long *) data.feat,
1290                            kvm_s390_available_cpu_feat,
1291                            KVM_S390_VM_CPU_FEAT_NR_BITS))
1292                 return -EINVAL;
1293
1294         mutex_lock(&kvm->lock);
1295         if (kvm->created_vcpus) {
1296                 mutex_unlock(&kvm->lock);
1297                 return -EBUSY;
1298         }
1299         bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
1300                     KVM_S390_VM_CPU_FEAT_NR_BITS);
1301         mutex_unlock(&kvm->lock);
1302         VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1303                          data.feat[0],
1304                          data.feat[1],
1305                          data.feat[2]);
1306         return 0;
1307 }
1308
1309 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
1310                                           struct kvm_device_attr *attr)
1311 {
1312         mutex_lock(&kvm->lock);
1313         if (kvm->created_vcpus) {
1314                 mutex_unlock(&kvm->lock);
1315                 return -EBUSY;
1316         }
1317
1318         if (copy_from_user(&kvm->arch.model.subfuncs, (void __user *)attr->addr,
1319                            sizeof(struct kvm_s390_vm_cpu_subfunc))) {
1320                 mutex_unlock(&kvm->lock);
1321                 return -EFAULT;
1322         }
1323         mutex_unlock(&kvm->lock);
1324
1325         VM_EVENT(kvm, 3, "SET: guest PLO    subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1326                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1327                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1328                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1329                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1330         VM_EVENT(kvm, 3, "SET: guest PTFF   subfunc 0x%16.16lx.%16.16lx",
1331                  ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1332                  ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1333         VM_EVENT(kvm, 3, "SET: guest KMAC   subfunc 0x%16.16lx.%16.16lx",
1334                  ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1335                  ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1336         VM_EVENT(kvm, 3, "SET: guest KMC    subfunc 0x%16.16lx.%16.16lx",
1337                  ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1338                  ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1339         VM_EVENT(kvm, 3, "SET: guest KM     subfunc 0x%16.16lx.%16.16lx",
1340                  ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1341                  ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1342         VM_EVENT(kvm, 3, "SET: guest KIMD   subfunc 0x%16.16lx.%16.16lx",
1343                  ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1344                  ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1345         VM_EVENT(kvm, 3, "SET: guest KLMD   subfunc 0x%16.16lx.%16.16lx",
1346                  ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1347                  ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1348         VM_EVENT(kvm, 3, "SET: guest PCKMO  subfunc 0x%16.16lx.%16.16lx",
1349                  ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1350                  ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1351         VM_EVENT(kvm, 3, "SET: guest KMCTR  subfunc 0x%16.16lx.%16.16lx",
1352                  ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1353                  ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1354         VM_EVENT(kvm, 3, "SET: guest KMF    subfunc 0x%16.16lx.%16.16lx",
1355                  ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1356                  ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1357         VM_EVENT(kvm, 3, "SET: guest KMO    subfunc 0x%16.16lx.%16.16lx",
1358                  ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1359                  ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1360         VM_EVENT(kvm, 3, "SET: guest PCC    subfunc 0x%16.16lx.%16.16lx",
1361                  ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1362                  ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1363         VM_EVENT(kvm, 3, "SET: guest PPNO   subfunc 0x%16.16lx.%16.16lx",
1364                  ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1365                  ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1366         VM_EVENT(kvm, 3, "SET: guest KMA    subfunc 0x%16.16lx.%16.16lx",
1367                  ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1368                  ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1369         VM_EVENT(kvm, 3, "SET: guest KDSA   subfunc 0x%16.16lx.%16.16lx",
1370                  ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1371                  ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1372         VM_EVENT(kvm, 3, "SET: guest SORTL  subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1373                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1374                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1375                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1376                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1377         VM_EVENT(kvm, 3, "SET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1378                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1379                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1380                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1381                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1382
1383         return 0;
1384 }
1385
1386 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1387 {
1388         int ret = -ENXIO;
1389
1390         switch (attr->attr) {
1391         case KVM_S390_VM_CPU_PROCESSOR:
1392                 ret = kvm_s390_set_processor(kvm, attr);
1393                 break;
1394         case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1395                 ret = kvm_s390_set_processor_feat(kvm, attr);
1396                 break;
1397         case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1398                 ret = kvm_s390_set_processor_subfunc(kvm, attr);
1399                 break;
1400         }
1401         return ret;
1402 }
1403
1404 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1405 {
1406         struct kvm_s390_vm_cpu_processor *proc;
1407         int ret = 0;
1408
1409         proc = kzalloc(sizeof(*proc), GFP_KERNEL);
1410         if (!proc) {
1411                 ret = -ENOMEM;
1412                 goto out;
1413         }
1414         proc->cpuid = kvm->arch.model.cpuid;
1415         proc->ibc = kvm->arch.model.ibc;
1416         memcpy(&proc->fac_list, kvm->arch.model.fac_list,
1417                S390_ARCH_FAC_LIST_SIZE_BYTE);
1418         VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1419                  kvm->arch.model.ibc,
1420                  kvm->arch.model.cpuid);
1421         VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1422                  kvm->arch.model.fac_list[0],
1423                  kvm->arch.model.fac_list[1],
1424                  kvm->arch.model.fac_list[2]);
1425         if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
1426                 ret = -EFAULT;
1427         kfree(proc);
1428 out:
1429         return ret;
1430 }
1431
1432 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
1433 {
1434         struct kvm_s390_vm_cpu_machine *mach;
1435         int ret = 0;
1436
1437         mach = kzalloc(sizeof(*mach), GFP_KERNEL);
1438         if (!mach) {
1439                 ret = -ENOMEM;
1440                 goto out;
1441         }
1442         get_cpu_id((struct cpuid *) &mach->cpuid);
1443         mach->ibc = sclp.ibc;
1444         memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
1445                S390_ARCH_FAC_LIST_SIZE_BYTE);
1446         memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
1447                sizeof(S390_lowcore.stfle_fac_list));
1448         VM_EVENT(kvm, 3, "GET: host ibc:  0x%4.4x, host cpuid:  0x%16.16llx",
1449                  kvm->arch.model.ibc,
1450                  kvm->arch.model.cpuid);
1451         VM_EVENT(kvm, 3, "GET: host facmask:  0x%16.16llx.%16.16llx.%16.16llx",
1452                  mach->fac_mask[0],
1453                  mach->fac_mask[1],
1454                  mach->fac_mask[2]);
1455         VM_EVENT(kvm, 3, "GET: host faclist:  0x%16.16llx.%16.16llx.%16.16llx",
1456                  mach->fac_list[0],
1457                  mach->fac_list[1],
1458                  mach->fac_list[2]);
1459         if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1460                 ret = -EFAULT;
1461         kfree(mach);
1462 out:
1463         return ret;
1464 }
1465
1466 static int kvm_s390_get_processor_feat(struct kvm *kvm,
1467                                        struct kvm_device_attr *attr)
1468 {
1469         struct kvm_s390_vm_cpu_feat data;
1470
1471         bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
1472                     KVM_S390_VM_CPU_FEAT_NR_BITS);
1473         if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1474                 return -EFAULT;
1475         VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1476                          data.feat[0],
1477                          data.feat[1],
1478                          data.feat[2]);
1479         return 0;
1480 }
1481
1482 static int kvm_s390_get_machine_feat(struct kvm *kvm,
1483                                      struct kvm_device_attr *attr)
1484 {
1485         struct kvm_s390_vm_cpu_feat data;
1486
1487         bitmap_copy((unsigned long *) data.feat,
1488                     kvm_s390_available_cpu_feat,
1489                     KVM_S390_VM_CPU_FEAT_NR_BITS);
1490         if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1491                 return -EFAULT;
1492         VM_EVENT(kvm, 3, "GET: host feat:  0x%16.16llx.0x%16.16llx.0x%16.16llx",
1493                          data.feat[0],
1494                          data.feat[1],
1495                          data.feat[2]);
1496         return 0;
1497 }
1498
1499 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1500                                           struct kvm_device_attr *attr)
1501 {
1502         if (copy_to_user((void __user *)attr->addr, &kvm->arch.model.subfuncs,
1503             sizeof(struct kvm_s390_vm_cpu_subfunc)))
1504                 return -EFAULT;
1505
1506         VM_EVENT(kvm, 3, "GET: guest PLO    subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1507                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1508                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1509                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1510                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1511         VM_EVENT(kvm, 3, "GET: guest PTFF   subfunc 0x%16.16lx.%16.16lx",
1512                  ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1513                  ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1514         VM_EVENT(kvm, 3, "GET: guest KMAC   subfunc 0x%16.16lx.%16.16lx",
1515                  ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1516                  ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1517         VM_EVENT(kvm, 3, "GET: guest KMC    subfunc 0x%16.16lx.%16.16lx",
1518                  ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1519                  ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1520         VM_EVENT(kvm, 3, "GET: guest KM     subfunc 0x%16.16lx.%16.16lx",
1521                  ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1522                  ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1523         VM_EVENT(kvm, 3, "GET: guest KIMD   subfunc 0x%16.16lx.%16.16lx",
1524                  ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1525                  ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1526         VM_EVENT(kvm, 3, "GET: guest KLMD   subfunc 0x%16.16lx.%16.16lx",
1527                  ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1528                  ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1529         VM_EVENT(kvm, 3, "GET: guest PCKMO  subfunc 0x%16.16lx.%16.16lx",
1530                  ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1531                  ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1532         VM_EVENT(kvm, 3, "GET: guest KMCTR  subfunc 0x%16.16lx.%16.16lx",
1533                  ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1534                  ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1535         VM_EVENT(kvm, 3, "GET: guest KMF    subfunc 0x%16.16lx.%16.16lx",
1536                  ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1537                  ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1538         VM_EVENT(kvm, 3, "GET: guest KMO    subfunc 0x%16.16lx.%16.16lx",
1539                  ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1540                  ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1541         VM_EVENT(kvm, 3, "GET: guest PCC    subfunc 0x%16.16lx.%16.16lx",
1542                  ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1543                  ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1544         VM_EVENT(kvm, 3, "GET: guest PPNO   subfunc 0x%16.16lx.%16.16lx",
1545                  ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1546                  ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1547         VM_EVENT(kvm, 3, "GET: guest KMA    subfunc 0x%16.16lx.%16.16lx",
1548                  ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1549                  ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1550         VM_EVENT(kvm, 3, "GET: guest KDSA   subfunc 0x%16.16lx.%16.16lx",
1551                  ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1552                  ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1553         VM_EVENT(kvm, 3, "GET: guest SORTL  subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1554                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1555                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1556                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1557                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1558         VM_EVENT(kvm, 3, "GET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1559                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1560                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1561                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1562                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1563
1564         return 0;
1565 }
1566
1567 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1568                                         struct kvm_device_attr *attr)
1569 {
1570         if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1571             sizeof(struct kvm_s390_vm_cpu_subfunc)))
1572                 return -EFAULT;
1573
1574         VM_EVENT(kvm, 3, "GET: host  PLO    subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1575                  ((unsigned long *) &kvm_s390_available_subfunc.plo)[0],
1576                  ((unsigned long *) &kvm_s390_available_subfunc.plo)[1],
1577                  ((unsigned long *) &kvm_s390_available_subfunc.plo)[2],
1578                  ((unsigned long *) &kvm_s390_available_subfunc.plo)[3]);
1579         VM_EVENT(kvm, 3, "GET: host  PTFF   subfunc 0x%16.16lx.%16.16lx",
1580                  ((unsigned long *) &kvm_s390_available_subfunc.ptff)[0],
1581                  ((unsigned long *) &kvm_s390_available_subfunc.ptff)[1]);
1582         VM_EVENT(kvm, 3, "GET: host  KMAC   subfunc 0x%16.16lx.%16.16lx",
1583                  ((unsigned long *) &kvm_s390_available_subfunc.kmac)[0],
1584                  ((unsigned long *) &kvm_s390_available_subfunc.kmac)[1]);
1585         VM_EVENT(kvm, 3, "GET: host  KMC    subfunc 0x%16.16lx.%16.16lx",
1586                  ((unsigned long *) &kvm_s390_available_subfunc.kmc)[0],
1587                  ((unsigned long *) &kvm_s390_available_subfunc.kmc)[1]);
1588         VM_EVENT(kvm, 3, "GET: host  KM     subfunc 0x%16.16lx.%16.16lx",
1589                  ((unsigned long *) &kvm_s390_available_subfunc.km)[0],
1590                  ((unsigned long *) &kvm_s390_available_subfunc.km)[1]);
1591         VM_EVENT(kvm, 3, "GET: host  KIMD   subfunc 0x%16.16lx.%16.16lx",
1592                  ((unsigned long *) &kvm_s390_available_subfunc.kimd)[0],
1593                  ((unsigned long *) &kvm_s390_available_subfunc.kimd)[1]);
1594         VM_EVENT(kvm, 3, "GET: host  KLMD   subfunc 0x%16.16lx.%16.16lx",
1595                  ((unsigned long *) &kvm_s390_available_subfunc.klmd)[0],
1596                  ((unsigned long *) &kvm_s390_available_subfunc.klmd)[1]);
1597         VM_EVENT(kvm, 3, "GET: host  PCKMO  subfunc 0x%16.16lx.%16.16lx",
1598                  ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[0],
1599                  ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[1]);
1600         VM_EVENT(kvm, 3, "GET: host  KMCTR  subfunc 0x%16.16lx.%16.16lx",
1601                  ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[0],
1602                  ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[1]);
1603         VM_EVENT(kvm, 3, "GET: host  KMF    subfunc 0x%16.16lx.%16.16lx",
1604                  ((unsigned long *) &kvm_s390_available_subfunc.kmf)[0],
1605                  ((unsigned long *) &kvm_s390_available_subfunc.kmf)[1]);
1606         VM_EVENT(kvm, 3, "GET: host  KMO    subfunc 0x%16.16lx.%16.16lx",
1607                  ((unsigned long *) &kvm_s390_available_subfunc.kmo)[0],
1608                  ((unsigned long *) &kvm_s390_available_subfunc.kmo)[1]);
1609         VM_EVENT(kvm, 3, "GET: host  PCC    subfunc 0x%16.16lx.%16.16lx",
1610                  ((unsigned long *) &kvm_s390_available_subfunc.pcc)[0],
1611                  ((unsigned long *) &kvm_s390_available_subfunc.pcc)[1]);
1612         VM_EVENT(kvm, 3, "GET: host  PPNO   subfunc 0x%16.16lx.%16.16lx",
1613                  ((unsigned long *) &kvm_s390_available_subfunc.ppno)[0],
1614                  ((unsigned long *) &kvm_s390_available_subfunc.ppno)[1]);
1615         VM_EVENT(kvm, 3, "GET: host  KMA    subfunc 0x%16.16lx.%16.16lx",
1616                  ((unsigned long *) &kvm_s390_available_subfunc.kma)[0],
1617                  ((unsigned long *) &kvm_s390_available_subfunc.kma)[1]);
1618         VM_EVENT(kvm, 3, "GET: host  KDSA   subfunc 0x%16.16lx.%16.16lx",
1619                  ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[0],
1620                  ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[1]);
1621         VM_EVENT(kvm, 3, "GET: host  SORTL  subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1622                  ((unsigned long *) &kvm_s390_available_subfunc.sortl)[0],
1623                  ((unsigned long *) &kvm_s390_available_subfunc.sortl)[1],
1624                  ((unsigned long *) &kvm_s390_available_subfunc.sortl)[2],
1625                  ((unsigned long *) &kvm_s390_available_subfunc.sortl)[3]);
1626         VM_EVENT(kvm, 3, "GET: host  DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1627                  ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[0],
1628                  ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[1],
1629                  ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[2],
1630                  ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[3]);
1631
1632         return 0;
1633 }
1634
1635 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1636 {
1637         int ret = -ENXIO;
1638
1639         switch (attr->attr) {
1640         case KVM_S390_VM_CPU_PROCESSOR:
1641                 ret = kvm_s390_get_processor(kvm, attr);
1642                 break;
1643         case KVM_S390_VM_CPU_MACHINE:
1644                 ret = kvm_s390_get_machine(kvm, attr);
1645                 break;
1646         case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1647                 ret = kvm_s390_get_processor_feat(kvm, attr);
1648                 break;
1649         case KVM_S390_VM_CPU_MACHINE_FEAT:
1650                 ret = kvm_s390_get_machine_feat(kvm, attr);
1651                 break;
1652         case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1653                 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1654                 break;
1655         case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1656                 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1657                 break;
1658         }
1659         return ret;
1660 }
1661
1662 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1663 {
1664         int ret;
1665
1666         switch (attr->group) {
1667         case KVM_S390_VM_MEM_CTRL:
1668                 ret = kvm_s390_set_mem_control(kvm, attr);
1669                 break;
1670         case KVM_S390_VM_TOD:
1671                 ret = kvm_s390_set_tod(kvm, attr);
1672                 break;
1673         case KVM_S390_VM_CPU_MODEL:
1674                 ret = kvm_s390_set_cpu_model(kvm, attr);
1675                 break;
1676         case KVM_S390_VM_CRYPTO:
1677                 ret = kvm_s390_vm_set_crypto(kvm, attr);
1678                 break;
1679         case KVM_S390_VM_MIGRATION:
1680                 ret = kvm_s390_vm_set_migration(kvm, attr);
1681                 break;
1682         default:
1683                 ret = -ENXIO;
1684                 break;
1685         }
1686
1687         return ret;
1688 }
1689
1690 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1691 {
1692         int ret;
1693
1694         switch (attr->group) {
1695         case KVM_S390_VM_MEM_CTRL:
1696                 ret = kvm_s390_get_mem_control(kvm, attr);
1697                 break;
1698         case KVM_S390_VM_TOD:
1699                 ret = kvm_s390_get_tod(kvm, attr);
1700                 break;
1701         case KVM_S390_VM_CPU_MODEL:
1702                 ret = kvm_s390_get_cpu_model(kvm, attr);
1703                 break;
1704         case KVM_S390_VM_MIGRATION:
1705                 ret = kvm_s390_vm_get_migration(kvm, attr);
1706                 break;
1707         default:
1708                 ret = -ENXIO;
1709                 break;
1710         }
1711
1712         return ret;
1713 }
1714
1715 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1716 {
1717         int ret;
1718
1719         switch (attr->group) {
1720         case KVM_S390_VM_MEM_CTRL:
1721                 switch (attr->attr) {
1722                 case KVM_S390_VM_MEM_ENABLE_CMMA:
1723                 case KVM_S390_VM_MEM_CLR_CMMA:
1724                         ret = sclp.has_cmma ? 0 : -ENXIO;
1725                         break;
1726                 case KVM_S390_VM_MEM_LIMIT_SIZE:
1727                         ret = 0;
1728                         break;
1729                 default:
1730                         ret = -ENXIO;
1731                         break;
1732                 }
1733                 break;
1734         case KVM_S390_VM_TOD:
1735                 switch (attr->attr) {
1736                 case KVM_S390_VM_TOD_LOW:
1737                 case KVM_S390_VM_TOD_HIGH:
1738                         ret = 0;
1739                         break;
1740                 default:
1741                         ret = -ENXIO;
1742                         break;
1743                 }
1744                 break;
1745         case KVM_S390_VM_CPU_MODEL:
1746                 switch (attr->attr) {
1747                 case KVM_S390_VM_CPU_PROCESSOR:
1748                 case KVM_S390_VM_CPU_MACHINE:
1749                 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1750                 case KVM_S390_VM_CPU_MACHINE_FEAT:
1751                 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1752                 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1753                         ret = 0;
1754                         break;
1755                 default:
1756                         ret = -ENXIO;
1757                         break;
1758                 }
1759                 break;
1760         case KVM_S390_VM_CRYPTO:
1761                 switch (attr->attr) {
1762                 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1763                 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1764                 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1765                 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1766                         ret = 0;
1767                         break;
1768                 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
1769                 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
1770                         ret = ap_instructions_available() ? 0 : -ENXIO;
1771                         break;
1772                 default:
1773                         ret = -ENXIO;
1774                         break;
1775                 }
1776                 break;
1777         case KVM_S390_VM_MIGRATION:
1778                 ret = 0;
1779                 break;
1780         default:
1781                 ret = -ENXIO;
1782                 break;
1783         }
1784
1785         return ret;
1786 }
1787
1788 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1789 {
1790         uint8_t *keys;
1791         uint64_t hva;
1792         int srcu_idx, i, r = 0;
1793
1794         if (args->flags != 0)
1795                 return -EINVAL;
1796
1797         /* Is this guest using storage keys? */
1798         if (!mm_uses_skeys(current->mm))
1799                 return KVM_S390_GET_SKEYS_NONE;
1800
1801         /* Enforce sane limit on memory allocation */
1802         if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1803                 return -EINVAL;
1804
1805         keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL);
1806         if (!keys)
1807                 return -ENOMEM;
1808
1809         down_read(&current->mm->mmap_sem);
1810         srcu_idx = srcu_read_lock(&kvm->srcu);
1811         for (i = 0; i < args->count; i++) {
1812                 hva = gfn_to_hva(kvm, args->start_gfn + i);
1813                 if (kvm_is_error_hva(hva)) {
1814                         r = -EFAULT;
1815                         break;
1816                 }
1817
1818                 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1819                 if (r)
1820                         break;
1821         }
1822         srcu_read_unlock(&kvm->srcu, srcu_idx);
1823         up_read(&current->mm->mmap_sem);
1824
1825         if (!r) {
1826                 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1827                                  sizeof(uint8_t) * args->count);
1828                 if (r)
1829                         r = -EFAULT;
1830         }
1831
1832         kvfree(keys);
1833         return r;
1834 }
1835
1836 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1837 {
1838         uint8_t *keys;
1839         uint64_t hva;
1840         int srcu_idx, i, r = 0;
1841         bool unlocked;
1842
1843         if (args->flags != 0)
1844                 return -EINVAL;
1845
1846         /* Enforce sane limit on memory allocation */
1847         if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1848                 return -EINVAL;
1849
1850         keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL);
1851         if (!keys)
1852                 return -ENOMEM;
1853
1854         r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1855                            sizeof(uint8_t) * args->count);
1856         if (r) {
1857                 r = -EFAULT;
1858                 goto out;
1859         }
1860
1861         /* Enable storage key handling for the guest */
1862         r = s390_enable_skey();
1863         if (r)
1864                 goto out;
1865
1866         i = 0;
1867         down_read(&current->mm->mmap_sem);
1868         srcu_idx = srcu_read_lock(&kvm->srcu);
1869         while (i < args->count) {
1870                 unlocked = false;
1871                 hva = gfn_to_hva(kvm, args->start_gfn + i);
1872                 if (kvm_is_error_hva(hva)) {
1873                         r = -EFAULT;
1874                         break;
1875                 }
1876
1877                 /* Lowest order bit is reserved */
1878                 if (keys[i] & 0x01) {
1879                         r = -EINVAL;
1880                         break;
1881                 }
1882
1883                 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
1884                 if (r) {
1885                         r = fixup_user_fault(current, current->mm, hva,
1886                                              FAULT_FLAG_WRITE, &unlocked);
1887                         if (r)
1888                                 break;
1889                 }
1890                 if (!r)
1891                         i++;
1892         }
1893         srcu_read_unlock(&kvm->srcu, srcu_idx);
1894         up_read(&current->mm->mmap_sem);
1895 out:
1896         kvfree(keys);
1897         return r;
1898 }
1899
1900 /*
1901  * Base address and length must be sent at the start of each block, therefore
1902  * it's cheaper to send some clean data, as long as it's less than the size of
1903  * two longs.
1904  */
1905 #define KVM_S390_MAX_BIT_DISTANCE (2 * sizeof(void *))
1906 /* for consistency */
1907 #define KVM_S390_CMMA_SIZE_MAX ((u32)KVM_S390_SKEYS_MAX)
1908
1909 /*
1910  * Similar to gfn_to_memslot, but returns the index of a memslot also when the
1911  * address falls in a hole. In that case the index of one of the memslots
1912  * bordering the hole is returned.
1913  */
1914 static int gfn_to_memslot_approx(struct kvm_memslots *slots, gfn_t gfn)
1915 {
1916         int start = 0, end = slots->used_slots;
1917         int slot = atomic_read(&slots->lru_slot);
1918         struct kvm_memory_slot *memslots = slots->memslots;
1919
1920         if (gfn >= memslots[slot].base_gfn &&
1921             gfn < memslots[slot].base_gfn + memslots[slot].npages)
1922                 return slot;
1923
1924         while (start < end) {
1925                 slot = start + (end - start) / 2;
1926
1927                 if (gfn >= memslots[slot].base_gfn)
1928                         end = slot;
1929                 else
1930                         start = slot + 1;
1931         }
1932
1933         if (gfn >= memslots[start].base_gfn &&
1934             gfn < memslots[start].base_gfn + memslots[start].npages) {
1935                 atomic_set(&slots->lru_slot, start);
1936         }
1937
1938         return start;
1939 }
1940
1941 static int kvm_s390_peek_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
1942                               u8 *res, unsigned long bufsize)
1943 {
1944         unsigned long pgstev, hva, cur_gfn = args->start_gfn;
1945
1946         args->count = 0;
1947         while (args->count < bufsize) {
1948                 hva = gfn_to_hva(kvm, cur_gfn);
1949                 /*
1950                  * We return an error if the first value was invalid, but we
1951                  * return successfully if at least one value was copied.
1952                  */
1953                 if (kvm_is_error_hva(hva))
1954                         return args->count ? 0 : -EFAULT;
1955                 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
1956                         pgstev = 0;
1957                 res[args->count++] = (pgstev >> 24) & 0x43;
1958                 cur_gfn++;
1959         }
1960
1961         return 0;
1962 }
1963
1964 static unsigned long kvm_s390_next_dirty_cmma(struct kvm_memslots *slots,
1965                                               unsigned long cur_gfn)
1966 {
1967         int slotidx = gfn_to_memslot_approx(slots, cur_gfn);
1968         struct kvm_memory_slot *ms = slots->memslots + slotidx;
1969         unsigned long ofs = cur_gfn - ms->base_gfn;
1970
1971         if (ms->base_gfn + ms->npages <= cur_gfn) {
1972                 slotidx--;
1973                 /* If we are above the highest slot, wrap around */
1974                 if (slotidx < 0)
1975                         slotidx = slots->used_slots - 1;
1976
1977                 ms = slots->memslots + slotidx;
1978                 ofs = 0;
1979         }
1980         ofs = find_next_bit(kvm_second_dirty_bitmap(ms), ms->npages, ofs);
1981         while ((slotidx > 0) && (ofs >= ms->npages)) {
1982                 slotidx--;
1983                 ms = slots->memslots + slotidx;
1984                 ofs = find_next_bit(kvm_second_dirty_bitmap(ms), ms->npages, 0);
1985         }
1986         return ms->base_gfn + ofs;
1987 }
1988
1989 static int kvm_s390_get_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
1990                              u8 *res, unsigned long bufsize)
1991 {
1992         unsigned long mem_end, cur_gfn, next_gfn, hva, pgstev;
1993         struct kvm_memslots *slots = kvm_memslots(kvm);
1994         struct kvm_memory_slot *ms;
1995
1996         cur_gfn = kvm_s390_next_dirty_cmma(slots, args->start_gfn);
1997         ms = gfn_to_memslot(kvm, cur_gfn);
1998         args->count = 0;
1999         args->start_gfn = cur_gfn;
2000         if (!ms)
2001                 return 0;
2002         next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2003         mem_end = slots->memslots[0].base_gfn + slots->memslots[0].npages;
2004
2005         while (args->count < bufsize) {
2006                 hva = gfn_to_hva(kvm, cur_gfn);
2007                 if (kvm_is_error_hva(hva))
2008                         return 0;
2009                 /* Decrement only if we actually flipped the bit to 0 */
2010                 if (test_and_clear_bit(cur_gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
2011                         atomic64_dec(&kvm->arch.cmma_dirty_pages);
2012                 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2013                         pgstev = 0;
2014                 /* Save the value */
2015                 res[args->count++] = (pgstev >> 24) & 0x43;
2016                 /* If the next bit is too far away, stop. */
2017                 if (next_gfn > cur_gfn + KVM_S390_MAX_BIT_DISTANCE)
2018                         return 0;
2019                 /* If we reached the previous "next", find the next one */
2020                 if (cur_gfn == next_gfn)
2021                         next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2022                 /* Reached the end of memory or of the buffer, stop */
2023                 if ((next_gfn >= mem_end) ||
2024                     (next_gfn - args->start_gfn >= bufsize))
2025                         return 0;
2026                 cur_gfn++;
2027                 /* Reached the end of the current memslot, take the next one. */
2028                 if (cur_gfn - ms->base_gfn >= ms->npages) {
2029                         ms = gfn_to_memslot(kvm, cur_gfn);
2030                         if (!ms)
2031                                 return 0;
2032                 }
2033         }
2034         return 0;
2035 }
2036
2037 /*
2038  * This function searches for the next page with dirty CMMA attributes, and
2039  * saves the attributes in the buffer up to either the end of the buffer or
2040  * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found;
2041  * no trailing clean bytes are saved.
2042  * In case no dirty bits were found, or if CMMA was not enabled or used, the
2043  * output buffer will indicate 0 as length.
2044  */
2045 static int kvm_s390_get_cmma_bits(struct kvm *kvm,
2046                                   struct kvm_s390_cmma_log *args)
2047 {
2048         unsigned long bufsize;
2049         int srcu_idx, peek, ret;
2050         u8 *values;
2051
2052         if (!kvm->arch.use_cmma)
2053                 return -ENXIO;
2054         /* Invalid/unsupported flags were specified */
2055         if (args->flags & ~KVM_S390_CMMA_PEEK)
2056                 return -EINVAL;
2057         /* Migration mode query, and we are not doing a migration */
2058         peek = !!(args->flags & KVM_S390_CMMA_PEEK);
2059         if (!peek && !kvm->arch.migration_mode)
2060                 return -EINVAL;
2061         /* CMMA is disabled or was not used, or the buffer has length zero */
2062         bufsize = min(args->count, KVM_S390_CMMA_SIZE_MAX);
2063         if (!bufsize || !kvm->mm->context.uses_cmm) {
2064                 memset(args, 0, sizeof(*args));
2065                 return 0;
2066         }
2067         /* We are not peeking, and there are no dirty pages */
2068         if (!peek && !atomic64_read(&kvm->arch.cmma_dirty_pages)) {
2069                 memset(args, 0, sizeof(*args));
2070                 return 0;
2071         }
2072
2073         values = vmalloc(bufsize);
2074         if (!values)
2075                 return -ENOMEM;
2076
2077         down_read(&kvm->mm->mmap_sem);
2078         srcu_idx = srcu_read_lock(&kvm->srcu);
2079         if (peek)
2080                 ret = kvm_s390_peek_cmma(kvm, args, values, bufsize);
2081         else
2082                 ret = kvm_s390_get_cmma(kvm, args, values, bufsize);
2083         srcu_read_unlock(&kvm->srcu, srcu_idx);
2084         up_read(&kvm->mm->mmap_sem);
2085
2086         if (kvm->arch.migration_mode)
2087                 args->remaining = atomic64_read(&kvm->arch.cmma_dirty_pages);
2088         else
2089                 args->remaining = 0;
2090
2091         if (copy_to_user((void __user *)args->values, values, args->count))
2092                 ret = -EFAULT;
2093
2094         vfree(values);
2095         return ret;
2096 }
2097
2098 /*
2099  * This function sets the CMMA attributes for the given pages. If the input
2100  * buffer has zero length, no action is taken, otherwise the attributes are
2101  * set and the mm->context.uses_cmm flag is set.
2102  */
2103 static int kvm_s390_set_cmma_bits(struct kvm *kvm,
2104                                   const struct kvm_s390_cmma_log *args)
2105 {
2106         unsigned long hva, mask, pgstev, i;
2107         uint8_t *bits;
2108         int srcu_idx, r = 0;
2109
2110         mask = args->mask;
2111
2112         if (!kvm->arch.use_cmma)
2113                 return -ENXIO;
2114         /* invalid/unsupported flags */
2115         if (args->flags != 0)
2116                 return -EINVAL;
2117         /* Enforce sane limit on memory allocation */
2118         if (args->count > KVM_S390_CMMA_SIZE_MAX)
2119                 return -EINVAL;
2120         /* Nothing to do */
2121         if (args->count == 0)
2122                 return 0;
2123
2124         bits = vmalloc(array_size(sizeof(*bits), args->count));
2125         if (!bits)
2126                 return -ENOMEM;
2127
2128         r = copy_from_user(bits, (void __user *)args->values, args->count);
2129         if (r) {
2130                 r = -EFAULT;
2131                 goto out;
2132         }
2133
2134         down_read(&kvm->mm->mmap_sem);
2135         srcu_idx = srcu_read_lock(&kvm->srcu);
2136         for (i = 0; i < args->count; i++) {
2137                 hva = gfn_to_hva(kvm, args->start_gfn + i);
2138                 if (kvm_is_error_hva(hva)) {
2139                         r = -EFAULT;
2140                         break;
2141                 }
2142
2143                 pgstev = bits[i];
2144                 pgstev = pgstev << 24;
2145                 mask &= _PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT;
2146                 set_pgste_bits(kvm->mm, hva, mask, pgstev);
2147         }
2148         srcu_read_unlock(&kvm->srcu, srcu_idx);
2149         up_read(&kvm->mm->mmap_sem);
2150
2151         if (!kvm->mm->context.uses_cmm) {
2152                 down_write(&kvm->mm->mmap_sem);
2153                 kvm->mm->context.uses_cmm = 1;
2154                 up_write(&kvm->mm->mmap_sem);
2155         }
2156 out:
2157         vfree(bits);
2158         return r;
2159 }
2160
2161 long kvm_arch_vm_ioctl(struct file *filp,
2162                        unsigned int ioctl, unsigned long arg)
2163 {
2164         struct kvm *kvm = filp->private_data;
2165         void __user *argp = (void __user *)arg;
2166         struct kvm_device_attr attr;
2167         int r;
2168
2169         switch (ioctl) {
2170         case KVM_S390_INTERRUPT: {
2171                 struct kvm_s390_interrupt s390int;
2172
2173                 r = -EFAULT;
2174                 if (copy_from_user(&s390int, argp, sizeof(s390int)))
2175                         break;
2176                 r = kvm_s390_inject_vm(kvm, &s390int);
2177                 break;
2178         }
2179         case KVM_CREATE_IRQCHIP: {
2180                 struct kvm_irq_routing_entry routing;
2181
2182                 r = -EINVAL;
2183                 if (kvm->arch.use_irqchip) {
2184                         /* Set up dummy routing. */
2185                         memset(&routing, 0, sizeof(routing));
2186                         r = kvm_set_irq_routing(kvm, &routing, 0, 0);
2187                 }
2188                 break;
2189         }
2190         case KVM_SET_DEVICE_ATTR: {
2191                 r = -EFAULT;
2192                 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2193                         break;
2194                 r = kvm_s390_vm_set_attr(kvm, &attr);
2195                 break;
2196         }
2197         case KVM_GET_DEVICE_ATTR: {
2198                 r = -EFAULT;
2199                 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2200                         break;
2201                 r = kvm_s390_vm_get_attr(kvm, &attr);
2202                 break;
2203         }
2204         case KVM_HAS_DEVICE_ATTR: {
2205                 r = -EFAULT;
2206                 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2207                         break;
2208                 r = kvm_s390_vm_has_attr(kvm, &attr);
2209                 break;
2210         }
2211         case KVM_S390_GET_SKEYS: {
2212                 struct kvm_s390_skeys args;
2213
2214                 r = -EFAULT;
2215                 if (copy_from_user(&args, argp,
2216                                    sizeof(struct kvm_s390_skeys)))
2217                         break;
2218                 r = kvm_s390_get_skeys(kvm, &args);
2219                 break;
2220         }
2221         case KVM_S390_SET_SKEYS: {
2222                 struct kvm_s390_skeys args;
2223
2224                 r = -EFAULT;
2225                 if (copy_from_user(&args, argp,
2226                                    sizeof(struct kvm_s390_skeys)))
2227                         break;
2228                 r = kvm_s390_set_skeys(kvm, &args);
2229                 break;
2230         }
2231         case KVM_S390_GET_CMMA_BITS: {
2232                 struct kvm_s390_cmma_log args;
2233
2234                 r = -EFAULT;
2235                 if (copy_from_user(&args, argp, sizeof(args)))
2236                         break;
2237                 mutex_lock(&kvm->slots_lock);
2238                 r = kvm_s390_get_cmma_bits(kvm, &args);
2239                 mutex_unlock(&kvm->slots_lock);
2240                 if (!r) {
2241                         r = copy_to_user(argp, &args, sizeof(args));
2242                         if (r)
2243                                 r = -EFAULT;
2244                 }
2245                 break;
2246         }
2247         case KVM_S390_SET_CMMA_BITS: {
2248                 struct kvm_s390_cmma_log args;
2249
2250                 r = -EFAULT;
2251                 if (copy_from_user(&args, argp, sizeof(args)))
2252                         break;
2253                 mutex_lock(&kvm->slots_lock);
2254                 r = kvm_s390_set_cmma_bits(kvm, &args);
2255                 mutex_unlock(&kvm->slots_lock);
2256                 break;
2257         }
2258         default:
2259                 r = -ENOTTY;
2260         }
2261
2262         return r;
2263 }
2264
2265 static int kvm_s390_apxa_installed(void)
2266 {
2267         struct ap_config_info info;
2268
2269         if (ap_instructions_available()) {
2270                 if (ap_qci(&info) == 0)
2271                         return info.apxa;
2272         }
2273
2274         return 0;
2275 }
2276
2277 /*
2278  * The format of the crypto control block (CRYCB) is specified in the 3 low
2279  * order bits of the CRYCB designation (CRYCBD) field as follows:
2280  * Format 0: Neither the message security assist extension 3 (MSAX3) nor the
2281  *           AP extended addressing (APXA) facility are installed.
2282  * Format 1: The APXA facility is not installed but the MSAX3 facility is.
2283  * Format 2: Both the APXA and MSAX3 facilities are installed
2284  */
2285 static void kvm_s390_set_crycb_format(struct kvm *kvm)
2286 {
2287         kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
2288
2289         /* Clear the CRYCB format bits - i.e., set format 0 by default */
2290         kvm->arch.crypto.crycbd &= ~(CRYCB_FORMAT_MASK);
2291
2292         /* Check whether MSAX3 is installed */
2293         if (!test_kvm_facility(kvm, 76))
2294                 return;
2295
2296         if (kvm_s390_apxa_installed())
2297                 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
2298         else
2299                 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
2300 }
2301
2302 void kvm_arch_crypto_set_masks(struct kvm *kvm, unsigned long *apm,
2303                                unsigned long *aqm, unsigned long *adm)
2304 {
2305         struct kvm_s390_crypto_cb *crycb = kvm->arch.crypto.crycb;
2306
2307         mutex_lock(&kvm->lock);
2308         kvm_s390_vcpu_block_all(kvm);
2309
2310         switch (kvm->arch.crypto.crycbd & CRYCB_FORMAT_MASK) {
2311         case CRYCB_FORMAT2: /* APCB1 use 256 bits */
2312                 memcpy(crycb->apcb1.apm, apm, 32);
2313                 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx %016lx %016lx %016lx",
2314                          apm[0], apm[1], apm[2], apm[3]);
2315                 memcpy(crycb->apcb1.aqm, aqm, 32);
2316                 VM_EVENT(kvm, 3, "SET CRYCB: aqm %016lx %016lx %016lx %016lx",
2317                          aqm[0], aqm[1], aqm[2], aqm[3]);
2318                 memcpy(crycb->apcb1.adm, adm, 32);
2319                 VM_EVENT(kvm, 3, "SET CRYCB: adm %016lx %016lx %016lx %016lx",
2320                          adm[0], adm[1], adm[2], adm[3]);
2321                 break;
2322         case CRYCB_FORMAT1:
2323         case CRYCB_FORMAT0: /* Fall through both use APCB0 */
2324                 memcpy(crycb->apcb0.apm, apm, 8);
2325                 memcpy(crycb->apcb0.aqm, aqm, 2);
2326                 memcpy(crycb->apcb0.adm, adm, 2);
2327                 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx aqm %04x adm %04x",
2328                          apm[0], *((unsigned short *)aqm),
2329                          *((unsigned short *)adm));
2330                 break;
2331         default:        /* Can not happen */
2332                 break;
2333         }
2334
2335         /* recreate the shadow crycb for each vcpu */
2336         kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
2337         kvm_s390_vcpu_unblock_all(kvm);
2338         mutex_unlock(&kvm->lock);
2339 }
2340 EXPORT_SYMBOL_GPL(kvm_arch_crypto_set_masks);
2341
2342 void kvm_arch_crypto_clear_masks(struct kvm *kvm)
2343 {
2344         mutex_lock(&kvm->lock);
2345         kvm_s390_vcpu_block_all(kvm);
2346
2347         memset(&kvm->arch.crypto.crycb->apcb0, 0,
2348                sizeof(kvm->arch.crypto.crycb->apcb0));
2349         memset(&kvm->arch.crypto.crycb->apcb1, 0,
2350                sizeof(kvm->arch.crypto.crycb->apcb1));
2351
2352         VM_EVENT(kvm, 3, "%s", "CLR CRYCB:");
2353         /* recreate the shadow crycb for each vcpu */
2354         kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
2355         kvm_s390_vcpu_unblock_all(kvm);
2356         mutex_unlock(&kvm->lock);
2357 }
2358 EXPORT_SYMBOL_GPL(kvm_arch_crypto_clear_masks);
2359
2360 static u64 kvm_s390_get_initial_cpuid(void)
2361 {
2362         struct cpuid cpuid;
2363
2364         get_cpu_id(&cpuid);
2365         cpuid.version = 0xff;
2366         return *((u64 *) &cpuid);
2367 }
2368
2369 static void kvm_s390_crypto_init(struct kvm *kvm)
2370 {
2371         kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
2372         kvm_s390_set_crycb_format(kvm);
2373
2374         if (!test_kvm_facility(kvm, 76))
2375                 return;
2376
2377         /* Enable AES/DEA protected key functions by default */
2378         kvm->arch.crypto.aes_kw = 1;
2379         kvm->arch.crypto.dea_kw = 1;
2380         get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
2381                          sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
2382         get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
2383                          sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
2384 }
2385
2386 static void sca_dispose(struct kvm *kvm)
2387 {
2388         if (kvm->arch.use_esca)
2389                 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
2390         else
2391                 free_page((unsigned long)(kvm->arch.sca));
2392         kvm->arch.sca = NULL;
2393 }
2394
2395 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
2396 {
2397         gfp_t alloc_flags = GFP_KERNEL;
2398         int i, rc;
2399         char debug_name[16];
2400         static unsigned long sca_offset;
2401
2402         rc = -EINVAL;
2403 #ifdef CONFIG_KVM_S390_UCONTROL
2404         if (type & ~KVM_VM_S390_UCONTROL)
2405                 goto out_err;
2406         if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
2407                 goto out_err;
2408 #else
2409         if (type)
2410                 goto out_err;
2411 #endif
2412
2413         rc = s390_enable_sie();
2414         if (rc)
2415                 goto out_err;
2416
2417         rc = -ENOMEM;
2418
2419         if (!sclp.has_64bscao)
2420                 alloc_flags |= GFP_DMA;
2421         rwlock_init(&kvm->arch.sca_lock);
2422         /* start with basic SCA */
2423         kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
2424         if (!kvm->arch.sca)
2425                 goto out_err;
2426         mutex_lock(&kvm_lock);
2427         sca_offset += 16;
2428         if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
2429                 sca_offset = 0;
2430         kvm->arch.sca = (struct bsca_block *)
2431                         ((char *) kvm->arch.sca + sca_offset);
2432         mutex_unlock(&kvm_lock);
2433
2434         sprintf(debug_name, "kvm-%u", current->pid);
2435
2436         kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
2437         if (!kvm->arch.dbf)
2438                 goto out_err;
2439
2440         BUILD_BUG_ON(sizeof(struct sie_page2) != 4096);
2441         kvm->arch.sie_page2 =
2442              (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
2443         if (!kvm->arch.sie_page2)
2444                 goto out_err;
2445
2446         kvm->arch.sie_page2->kvm = kvm;
2447         kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
2448
2449         for (i = 0; i < kvm_s390_fac_size(); i++) {
2450                 kvm->arch.model.fac_mask[i] = S390_lowcore.stfle_fac_list[i] &
2451                                               (kvm_s390_fac_base[i] |
2452                                                kvm_s390_fac_ext[i]);
2453                 kvm->arch.model.fac_list[i] = S390_lowcore.stfle_fac_list[i] &
2454                                               kvm_s390_fac_base[i];
2455         }
2456         kvm->arch.model.subfuncs = kvm_s390_available_subfunc;
2457
2458         /* we are always in czam mode - even on pre z14 machines */
2459         set_kvm_facility(kvm->arch.model.fac_mask, 138);
2460         set_kvm_facility(kvm->arch.model.fac_list, 138);
2461         /* we emulate STHYI in kvm */
2462         set_kvm_facility(kvm->arch.model.fac_mask, 74);
2463         set_kvm_facility(kvm->arch.model.fac_list, 74);
2464         if (MACHINE_HAS_TLB_GUEST) {
2465                 set_kvm_facility(kvm->arch.model.fac_mask, 147);
2466                 set_kvm_facility(kvm->arch.model.fac_list, 147);
2467         }
2468
2469         if (css_general_characteristics.aiv && test_facility(65))
2470                 set_kvm_facility(kvm->arch.model.fac_mask, 65);
2471
2472         kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
2473         kvm->arch.model.ibc = sclp.ibc & 0x0fff;
2474
2475         kvm_s390_crypto_init(kvm);
2476
2477         mutex_init(&kvm->arch.float_int.ais_lock);
2478         spin_lock_init(&kvm->arch.float_int.lock);
2479         for (i = 0; i < FIRQ_LIST_COUNT; i++)
2480                 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
2481         init_waitqueue_head(&kvm->arch.ipte_wq);
2482         mutex_init(&kvm->arch.ipte_mutex);
2483
2484         debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
2485         VM_EVENT(kvm, 3, "vm created with type %lu", type);
2486
2487         if (type & KVM_VM_S390_UCONTROL) {
2488                 kvm->arch.gmap = NULL;
2489                 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
2490         } else {
2491                 if (sclp.hamax == U64_MAX)
2492                         kvm->arch.mem_limit = TASK_SIZE_MAX;
2493                 else
2494                         kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
2495                                                     sclp.hamax + 1);
2496                 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
2497                 if (!kvm->arch.gmap)
2498                         goto out_err;
2499                 kvm->arch.gmap->private = kvm;
2500                 kvm->arch.gmap->pfault_enabled = 0;
2501         }
2502
2503         kvm->arch.use_pfmfi = sclp.has_pfmfi;
2504         kvm->arch.use_skf = sclp.has_skey;
2505         spin_lock_init(&kvm->arch.start_stop_lock);
2506         kvm_s390_vsie_init(kvm);
2507         kvm_s390_gisa_init(kvm);
2508         KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
2509
2510         return 0;
2511 out_err:
2512         free_page((unsigned long)kvm->arch.sie_page2);
2513         debug_unregister(kvm->arch.dbf);
2514         sca_dispose(kvm);
2515         KVM_EVENT(3, "creation of vm failed: %d", rc);
2516         return rc;
2517 }
2518
2519 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
2520 {
2521         VCPU_EVENT(vcpu, 3, "%s", "free cpu");
2522         trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
2523         kvm_s390_clear_local_irqs(vcpu);
2524         kvm_clear_async_pf_completion_queue(vcpu);
2525         if (!kvm_is_ucontrol(vcpu->kvm))
2526                 sca_del_vcpu(vcpu);
2527
2528         if (kvm_is_ucontrol(vcpu->kvm))
2529                 gmap_remove(vcpu->arch.gmap);
2530
2531         if (vcpu->kvm->arch.use_cmma)
2532                 kvm_s390_vcpu_unsetup_cmma(vcpu);
2533         free_page((unsigned long)(vcpu->arch.sie_block));
2534
2535         kvm_vcpu_uninit(vcpu);
2536         kmem_cache_free(kvm_vcpu_cache, vcpu);
2537 }
2538
2539 static void kvm_free_vcpus(struct kvm *kvm)
2540 {
2541         unsigned int i;
2542         struct kvm_vcpu *vcpu;
2543
2544         kvm_for_each_vcpu(i, vcpu, kvm)
2545                 kvm_arch_vcpu_destroy(vcpu);
2546
2547         mutex_lock(&kvm->lock);
2548         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
2549                 kvm->vcpus[i] = NULL;
2550
2551         atomic_set(&kvm->online_vcpus, 0);
2552         mutex_unlock(&kvm->lock);
2553 }
2554
2555 void kvm_arch_destroy_vm(struct kvm *kvm)
2556 {
2557         kvm_free_vcpus(kvm);
2558         sca_dispose(kvm);
2559         debug_unregister(kvm->arch.dbf);
2560         kvm_s390_gisa_destroy(kvm);
2561         free_page((unsigned long)kvm->arch.sie_page2);
2562         if (!kvm_is_ucontrol(kvm))
2563                 gmap_remove(kvm->arch.gmap);
2564         kvm_s390_destroy_adapters(kvm);
2565         kvm_s390_clear_float_irqs(kvm);
2566         kvm_s390_vsie_destroy(kvm);
2567         KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
2568 }
2569
2570 /* Section: vcpu related */
2571 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
2572 {
2573         vcpu->arch.gmap = gmap_create(current->mm, -1UL);
2574         if (!vcpu->arch.gmap)
2575                 return -ENOMEM;
2576         vcpu->arch.gmap->private = vcpu->kvm;
2577
2578         return 0;
2579 }
2580
2581 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
2582 {
2583         if (!kvm_s390_use_sca_entries())
2584                 return;
2585         read_lock(&vcpu->kvm->arch.sca_lock);
2586         if (vcpu->kvm->arch.use_esca) {
2587                 struct esca_block *sca = vcpu->kvm->arch.sca;
2588
2589                 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
2590                 sca->cpu[vcpu->vcpu_id].sda = 0;
2591         } else {
2592                 struct bsca_block *sca = vcpu->kvm->arch.sca;
2593
2594                 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
2595                 sca->cpu[vcpu->vcpu_id].sda = 0;
2596         }
2597         read_unlock(&vcpu->kvm->arch.sca_lock);
2598 }
2599
2600 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
2601 {
2602         if (!kvm_s390_use_sca_entries()) {
2603                 struct bsca_block *sca = vcpu->kvm->arch.sca;
2604
2605                 /* we still need the basic sca for the ipte control */
2606                 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2607                 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
2608                 return;
2609         }
2610         read_lock(&vcpu->kvm->arch.sca_lock);
2611         if (vcpu->kvm->arch.use_esca) {
2612                 struct esca_block *sca = vcpu->kvm->arch.sca;
2613
2614                 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
2615                 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2616                 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
2617                 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
2618                 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
2619         } else {
2620                 struct bsca_block *sca = vcpu->kvm->arch.sca;
2621
2622                 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
2623                 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2624                 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
2625                 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
2626         }
2627         read_unlock(&vcpu->kvm->arch.sca_lock);
2628 }
2629
2630 /* Basic SCA to Extended SCA data copy routines */
2631 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
2632 {
2633         d->sda = s->sda;
2634         d->sigp_ctrl.c = s->sigp_ctrl.c;
2635         d->sigp_ctrl.scn = s->sigp_ctrl.scn;
2636 }
2637
2638 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
2639 {
2640         int i;
2641
2642         d->ipte_control = s->ipte_control;
2643         d->mcn[0] = s->mcn;
2644         for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
2645                 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
2646 }
2647
2648 static int sca_switch_to_extended(struct kvm *kvm)
2649 {
2650         struct bsca_block *old_sca = kvm->arch.sca;
2651         struct esca_block *new_sca;
2652         struct kvm_vcpu *vcpu;
2653         unsigned int vcpu_idx;
2654         u32 scaol, scaoh;
2655
2656         new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
2657         if (!new_sca)
2658                 return -ENOMEM;
2659
2660         scaoh = (u32)((u64)(new_sca) >> 32);
2661         scaol = (u32)(u64)(new_sca) & ~0x3fU;
2662
2663         kvm_s390_vcpu_block_all(kvm);
2664         write_lock(&kvm->arch.sca_lock);
2665
2666         sca_copy_b_to_e(new_sca, old_sca);
2667
2668         kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
2669                 vcpu->arch.sie_block->scaoh = scaoh;
2670                 vcpu->arch.sie_block->scaol = scaol;
2671                 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
2672         }
2673         kvm->arch.sca = new_sca;
2674         kvm->arch.use_esca = 1;
2675
2676         write_unlock(&kvm->arch.sca_lock);
2677         kvm_s390_vcpu_unblock_all(kvm);
2678
2679         free_page((unsigned long)old_sca);
2680
2681         VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
2682                  old_sca, kvm->arch.sca);
2683         return 0;
2684 }
2685
2686 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
2687 {
2688         int rc;
2689
2690         if (!kvm_s390_use_sca_entries()) {
2691                 if (id < KVM_MAX_VCPUS)
2692                         return true;
2693                 return false;
2694         }
2695         if (id < KVM_S390_BSCA_CPU_SLOTS)
2696                 return true;
2697         if (!sclp.has_esca || !sclp.has_64bscao)
2698                 return false;
2699
2700         mutex_lock(&kvm->lock);
2701         rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
2702         mutex_unlock(&kvm->lock);
2703
2704         return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
2705 }
2706
2707 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2708 {
2709         vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
2710         kvm_clear_async_pf_completion_queue(vcpu);
2711         vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
2712                                     KVM_SYNC_GPRS |
2713                                     KVM_SYNC_ACRS |
2714                                     KVM_SYNC_CRS |
2715                                     KVM_SYNC_ARCH0 |
2716                                     KVM_SYNC_PFAULT;
2717         kvm_s390_set_prefix(vcpu, 0);
2718         if (test_kvm_facility(vcpu->kvm, 64))
2719                 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
2720         if (test_kvm_facility(vcpu->kvm, 82))
2721                 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC;
2722         if (test_kvm_facility(vcpu->kvm, 133))
2723                 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
2724         if (test_kvm_facility(vcpu->kvm, 156))
2725                 vcpu->run->kvm_valid_regs |= KVM_SYNC_ETOKEN;
2726         /* fprs can be synchronized via vrs, even if the guest has no vx. With
2727          * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
2728          */
2729         if (MACHINE_HAS_VX)
2730                 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
2731         else
2732                 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
2733
2734         if (kvm_is_ucontrol(vcpu->kvm))
2735                 return __kvm_ucontrol_vcpu_init(vcpu);
2736
2737         return 0;
2738 }
2739
2740 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2741 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2742 {
2743         WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
2744         raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2745         vcpu->arch.cputm_start = get_tod_clock_fast();
2746         raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2747 }
2748
2749 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2750 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2751 {
2752         WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
2753         raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2754         vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
2755         vcpu->arch.cputm_start = 0;
2756         raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2757 }
2758
2759 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2760 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2761 {
2762         WARN_ON_ONCE(vcpu->arch.cputm_enabled);
2763         vcpu->arch.cputm_enabled = true;
2764         __start_cpu_timer_accounting(vcpu);
2765 }
2766
2767 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2768 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2769 {
2770         WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
2771         __stop_cpu_timer_accounting(vcpu);
2772         vcpu->arch.cputm_enabled = false;
2773 }
2774
2775 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2776 {
2777         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2778         __enable_cpu_timer_accounting(vcpu);
2779         preempt_enable();
2780 }
2781
2782 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2783 {
2784         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2785         __disable_cpu_timer_accounting(vcpu);
2786         preempt_enable();
2787 }
2788
2789 /* set the cpu timer - may only be called from the VCPU thread itself */
2790 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
2791 {
2792         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2793         raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2794         if (vcpu->arch.cputm_enabled)
2795                 vcpu->arch.cputm_start = get_tod_clock_fast();
2796         vcpu->arch.sie_block->cputm = cputm;
2797         raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2798         preempt_enable();
2799 }
2800
2801 /* update and get the cpu timer - can also be called from other VCPU threads */
2802 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
2803 {
2804         unsigned int seq;
2805         __u64 value;
2806
2807         if (unlikely(!vcpu->arch.cputm_enabled))
2808                 return vcpu->arch.sie_block->cputm;
2809
2810         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2811         do {
2812                 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
2813                 /*
2814                  * If the writer would ever execute a read in the critical
2815                  * section, e.g. in irq context, we have a deadlock.
2816                  */
2817                 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
2818                 value = vcpu->arch.sie_block->cputm;
2819                 /* if cputm_start is 0, accounting is being started/stopped */
2820                 if (likely(vcpu->arch.cputm_start))
2821                         value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
2822         } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
2823         preempt_enable();
2824         return value;
2825 }
2826
2827 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2828 {
2829
2830         gmap_enable(vcpu->arch.enabled_gmap);
2831         kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING);
2832         if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
2833                 __start_cpu_timer_accounting(vcpu);
2834         vcpu->cpu = cpu;
2835 }
2836
2837 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2838 {
2839         vcpu->cpu = -1;
2840         if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
2841                 __stop_cpu_timer_accounting(vcpu);
2842         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING);
2843         vcpu->arch.enabled_gmap = gmap_get_enabled();
2844         gmap_disable(vcpu->arch.enabled_gmap);
2845
2846 }
2847
2848 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
2849 {
2850         mutex_lock(&vcpu->kvm->lock);
2851         preempt_disable();
2852         vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
2853         vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx;
2854         preempt_enable();
2855         mutex_unlock(&vcpu->kvm->lock);
2856         if (!kvm_is_ucontrol(vcpu->kvm)) {
2857                 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
2858                 sca_add_vcpu(vcpu);
2859         }
2860         if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
2861                 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2862         /* make vcpu_load load the right gmap on the first trigger */
2863         vcpu->arch.enabled_gmap = vcpu->arch.gmap;
2864 }
2865
2866 static bool kvm_has_pckmo_subfunc(struct kvm *kvm, unsigned long nr)
2867 {
2868         if (test_bit_inv(nr, (unsigned long *)&kvm->arch.model.subfuncs.pckmo) &&
2869             test_bit_inv(nr, (unsigned long *)&kvm_s390_available_subfunc.pckmo))
2870                 return true;
2871         return false;
2872 }
2873
2874 static bool kvm_has_pckmo_ecc(struct kvm *kvm)
2875 {
2876         /* At least one ECC subfunction must be present */
2877         return kvm_has_pckmo_subfunc(kvm, 32) ||
2878                kvm_has_pckmo_subfunc(kvm, 33) ||
2879                kvm_has_pckmo_subfunc(kvm, 34) ||
2880                kvm_has_pckmo_subfunc(kvm, 40) ||
2881                kvm_has_pckmo_subfunc(kvm, 41);
2882
2883 }
2884
2885 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
2886 {
2887         /*
2888          * If the AP instructions are not being interpreted and the MSAX3
2889          * facility is not configured for the guest, there is nothing to set up.
2890          */
2891         if (!vcpu->kvm->arch.crypto.apie && !test_kvm_facility(vcpu->kvm, 76))
2892                 return;
2893
2894         vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
2895         vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
2896         vcpu->arch.sie_block->eca &= ~ECA_APIE;
2897         vcpu->arch.sie_block->ecd &= ~ECD_ECC;
2898
2899         if (vcpu->kvm->arch.crypto.apie)
2900                 vcpu->arch.sie_block->eca |= ECA_APIE;
2901
2902         /* Set up protected key support */
2903         if (vcpu->kvm->arch.crypto.aes_kw) {
2904                 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
2905                 /* ecc is also wrapped with AES key */
2906                 if (kvm_has_pckmo_ecc(vcpu->kvm))
2907                         vcpu->arch.sie_block->ecd |= ECD_ECC;
2908         }
2909
2910         if (vcpu->kvm->arch.crypto.dea_kw)
2911                 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
2912 }
2913
2914 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
2915 {
2916         free_page(vcpu->arch.sie_block->cbrlo);
2917         vcpu->arch.sie_block->cbrlo = 0;
2918 }
2919
2920 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
2921 {
2922         vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
2923         if (!vcpu->arch.sie_block->cbrlo)
2924                 return -ENOMEM;
2925         return 0;
2926 }
2927
2928 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
2929 {
2930         struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
2931
2932         vcpu->arch.sie_block->ibc = model->ibc;
2933         if (test_kvm_facility(vcpu->kvm, 7))
2934                 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
2935 }
2936
2937 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
2938 {
2939         int rc = 0;
2940
2941         atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
2942                                                     CPUSTAT_SM |
2943                                                     CPUSTAT_STOPPED);
2944
2945         if (test_kvm_facility(vcpu->kvm, 78))
2946                 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2);
2947         else if (test_kvm_facility(vcpu->kvm, 8))
2948                 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED);
2949
2950         kvm_s390_vcpu_setup_model(vcpu);
2951
2952         /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
2953         if (MACHINE_HAS_ESOP)
2954                 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
2955         if (test_kvm_facility(vcpu->kvm, 9))
2956                 vcpu->arch.sie_block->ecb |= ECB_SRSI;
2957         if (test_kvm_facility(vcpu->kvm, 73))
2958                 vcpu->arch.sie_block->ecb |= ECB_TE;
2959
2960         if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi)
2961                 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
2962         if (test_kvm_facility(vcpu->kvm, 130))
2963                 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
2964         vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
2965         if (sclp.has_cei)
2966                 vcpu->arch.sie_block->eca |= ECA_CEI;
2967         if (sclp.has_ib)
2968                 vcpu->arch.sie_block->eca |= ECA_IB;
2969         if (sclp.has_siif)
2970                 vcpu->arch.sie_block->eca |= ECA_SII;
2971         if (sclp.has_sigpif)
2972                 vcpu->arch.sie_block->eca |= ECA_SIGPI;
2973         if (test_kvm_facility(vcpu->kvm, 129)) {
2974                 vcpu->arch.sie_block->eca |= ECA_VX;
2975                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
2976         }
2977         if (test_kvm_facility(vcpu->kvm, 139))
2978                 vcpu->arch.sie_block->ecd |= ECD_MEF;
2979         if (test_kvm_facility(vcpu->kvm, 156))
2980                 vcpu->arch.sie_block->ecd |= ECD_ETOKENF;
2981         if (vcpu->arch.sie_block->gd) {
2982                 vcpu->arch.sie_block->eca |= ECA_AIV;
2983                 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
2984                            vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
2985         }
2986         vcpu->arch.sie_block->sdnxo = ((unsigned long) &vcpu->run->s.regs.sdnx)
2987                                         | SDNXC;
2988         vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
2989
2990         if (sclp.has_kss)
2991                 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS);
2992         else
2993                 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
2994
2995         if (vcpu->kvm->arch.use_cmma) {
2996                 rc = kvm_s390_vcpu_setup_cmma(vcpu);
2997                 if (rc)
2998                         return rc;
2999         }
3000         hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3001         vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
3002
3003         vcpu->arch.sie_block->hpid = HPID_KVM;
3004
3005         kvm_s390_vcpu_crypto_setup(vcpu);
3006
3007         return rc;
3008 }
3009
3010 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3011                                       unsigned int id)
3012 {
3013         struct kvm_vcpu *vcpu;
3014         struct sie_page *sie_page;
3015         int rc = -EINVAL;
3016
3017         if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
3018                 goto out;
3019
3020         rc = -ENOMEM;
3021
3022         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3023         if (!vcpu)
3024                 goto out;
3025
3026         BUILD_BUG_ON(sizeof(struct sie_page) != 4096);
3027         sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
3028         if (!sie_page)
3029                 goto out_free_cpu;
3030
3031         vcpu->arch.sie_block = &sie_page->sie_block;
3032         vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
3033
3034         /* the real guest size will always be smaller than msl */
3035         vcpu->arch.sie_block->mso = 0;
3036         vcpu->arch.sie_block->msl = sclp.hamax;
3037
3038         vcpu->arch.sie_block->icpua = id;
3039         spin_lock_init(&vcpu->arch.local_int.lock);
3040         vcpu->arch.sie_block->gd = (u32)(u64)kvm->arch.gisa_int.origin;
3041         if (vcpu->arch.sie_block->gd && sclp.has_gisaf)
3042                 vcpu->arch.sie_block->gd |= GISA_FORMAT1;
3043         seqcount_init(&vcpu->arch.cputm_seqcount);
3044
3045         rc = kvm_vcpu_init(vcpu, kvm, id);
3046         if (rc)
3047                 goto out_free_sie_block;
3048         VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
3049                  vcpu->arch.sie_block);
3050         trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
3051
3052         return vcpu;
3053 out_free_sie_block:
3054         free_page((unsigned long)(vcpu->arch.sie_block));
3055 out_free_cpu:
3056         kmem_cache_free(kvm_vcpu_cache, vcpu);
3057 out:
3058         return ERR_PTR(rc);
3059 }
3060
3061 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3062 {
3063         return kvm_s390_vcpu_has_irq(vcpu, 0);
3064 }
3065
3066 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
3067 {
3068         return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE);
3069 }
3070
3071 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
3072 {
3073         atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
3074         exit_sie(vcpu);
3075 }
3076
3077 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
3078 {
3079         atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
3080 }
3081
3082 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
3083 {
3084         atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
3085         exit_sie(vcpu);
3086 }
3087
3088 bool kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu *vcpu)
3089 {
3090         return atomic_read(&vcpu->arch.sie_block->prog20) &
3091                (PROG_BLOCK_SIE | PROG_REQUEST);
3092 }
3093
3094 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
3095 {
3096         atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
3097 }
3098
3099 /*
3100  * Kick a guest cpu out of (v)SIE and wait until (v)SIE is not running.
3101  * If the CPU is not running (e.g. waiting as idle) the function will
3102  * return immediately. */
3103 void exit_sie(struct kvm_vcpu *vcpu)
3104 {
3105         kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
3106         kvm_s390_vsie_kick(vcpu);
3107         while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
3108                 cpu_relax();
3109 }
3110
3111 /* Kick a guest cpu out of SIE to process a request synchronously */
3112 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
3113 {
3114         kvm_make_request(req, vcpu);
3115         kvm_s390_vcpu_request(vcpu);
3116 }
3117
3118 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
3119                               unsigned long end)
3120 {
3121         struct kvm *kvm = gmap->private;
3122         struct kvm_vcpu *vcpu;
3123         unsigned long prefix;
3124         int i;
3125
3126         if (gmap_is_shadow(gmap))
3127                 return;
3128         if (start >= 1UL << 31)
3129                 /* We are only interested in prefix pages */
3130                 return;
3131         kvm_for_each_vcpu(i, vcpu, kvm) {
3132                 /* match against both prefix pages */
3133                 prefix = kvm_s390_get_prefix(vcpu);
3134                 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
3135                         VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
3136                                    start, end);
3137                         kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
3138                 }
3139         }
3140 }
3141
3142 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
3143 {
3144         /* do not poll with more than halt_poll_max_steal percent of steal time */
3145         if (S390_lowcore.avg_steal_timer * 100 / (TICK_USEC << 12) >=
3146             halt_poll_max_steal) {
3147                 vcpu->stat.halt_no_poll_steal++;
3148                 return true;
3149         }
3150         return false;
3151 }
3152
3153 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
3154 {
3155         /* kvm common code refers to this, but never calls it */
3156         BUG();
3157         return 0;
3158 }
3159
3160 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
3161                                            struct kvm_one_reg *reg)
3162 {
3163         int r = -EINVAL;
3164
3165         switch (reg->id) {
3166         case KVM_REG_S390_TODPR:
3167                 r = put_user(vcpu->arch.sie_block->todpr,
3168                              (u32 __user *)reg->addr);
3169                 break;
3170         case KVM_REG_S390_EPOCHDIFF:
3171                 r = put_user(vcpu->arch.sie_block->epoch,
3172                              (u64 __user *)reg->addr);
3173                 break;
3174         case KVM_REG_S390_CPU_TIMER:
3175                 r = put_user(kvm_s390_get_cpu_timer(vcpu),
3176                              (u64 __user *)reg->addr);
3177                 break;
3178         case KVM_REG_S390_CLOCK_COMP:
3179                 r = put_user(vcpu->arch.sie_block->ckc,
3180                              (u64 __user *)reg->addr);
3181                 break;
3182         case KVM_REG_S390_PFTOKEN:
3183                 r = put_user(vcpu->arch.pfault_token,
3184                              (u64 __user *)reg->addr);
3185                 break;
3186         case KVM_REG_S390_PFCOMPARE:
3187                 r = put_user(vcpu->arch.pfault_compare,
3188                              (u64 __user *)reg->addr);
3189                 break;
3190         case KVM_REG_S390_PFSELECT:
3191                 r = put_user(vcpu->arch.pfault_select,
3192                              (u64 __user *)reg->addr);
3193                 break;
3194         case KVM_REG_S390_PP:
3195                 r = put_user(vcpu->arch.sie_block->pp,
3196                              (u64 __user *)reg->addr);
3197                 break;
3198         case KVM_REG_S390_GBEA:
3199                 r = put_user(vcpu->arch.sie_block->gbea,
3200                              (u64 __user *)reg->addr);
3201                 break;
3202         default:
3203                 break;
3204         }
3205
3206         return r;
3207 }
3208
3209 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
3210                                            struct kvm_one_reg *reg)
3211 {
3212         int r = -EINVAL;
3213         __u64 val;
3214
3215         switch (reg->id) {
3216         case KVM_REG_S390_TODPR:
3217                 r = get_user(vcpu->arch.sie_block->todpr,
3218                              (u32 __user *)reg->addr);
3219                 break;
3220         case KVM_REG_S390_EPOCHDIFF:
3221                 r = get_user(vcpu->arch.sie_block->epoch,
3222                              (u64 __user *)reg->addr);
3223                 break;
3224         case KVM_REG_S390_CPU_TIMER:
3225                 r = get_user(val, (u64 __user *)reg->addr);
3226                 if (!r)
3227                         kvm_s390_set_cpu_timer(vcpu, val);
3228                 break;
3229         case KVM_REG_S390_CLOCK_COMP:
3230                 r = get_user(vcpu->arch.sie_block->ckc,
3231                              (u64 __user *)reg->addr);
3232                 break;
3233         case KVM_REG_S390_PFTOKEN:
3234                 r = get_user(vcpu->arch.pfault_token,
3235                              (u64 __user *)reg->addr);
3236                 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
3237                         kvm_clear_async_pf_completion_queue(vcpu);
3238                 break;
3239         case KVM_REG_S390_PFCOMPARE:
3240                 r = get_user(vcpu->arch.pfault_compare,
3241                              (u64 __user *)reg->addr);
3242                 break;
3243         case KVM_REG_S390_PFSELECT:
3244                 r = get_user(vcpu->arch.pfault_select,
3245                              (u64 __user *)reg->addr);
3246                 break;
3247         case KVM_REG_S390_PP:
3248                 r = get_user(vcpu->arch.sie_block->pp,
3249                              (u64 __user *)reg->addr);
3250                 break;
3251         case KVM_REG_S390_GBEA:
3252                 r = get_user(vcpu->arch.sie_block->gbea,
3253                              (u64 __user *)reg->addr);
3254                 break;
3255         default:
3256                 break;
3257         }
3258
3259         return r;
3260 }
3261
3262 static void kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu *vcpu)
3263 {
3264         vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_RI;
3265         vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
3266         memset(vcpu->run->s.regs.riccb, 0, sizeof(vcpu->run->s.regs.riccb));
3267
3268         kvm_clear_async_pf_completion_queue(vcpu);
3269         if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
3270                 kvm_s390_vcpu_stop(vcpu);
3271         kvm_s390_clear_local_irqs(vcpu);
3272 }
3273
3274 static void kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
3275 {
3276         /* Initial reset is a superset of the normal reset */
3277         kvm_arch_vcpu_ioctl_normal_reset(vcpu);
3278
3279         /* this equals initial cpu reset in pop, but we don't switch to ESA */
3280         vcpu->arch.sie_block->gpsw.mask = 0;
3281         vcpu->arch.sie_block->gpsw.addr = 0;
3282         kvm_s390_set_prefix(vcpu, 0);
3283         kvm_s390_set_cpu_timer(vcpu, 0);
3284         vcpu->arch.sie_block->ckc = 0;
3285         vcpu->arch.sie_block->todpr = 0;
3286         memset(vcpu->arch.sie_block->gcr, 0, sizeof(vcpu->arch.sie_block->gcr));
3287         vcpu->arch.sie_block->gcr[0] = CR0_INITIAL_MASK;
3288         vcpu->arch.sie_block->gcr[14] = CR14_INITIAL_MASK;
3289         vcpu->run->s.regs.fpc = 0;
3290         vcpu->arch.sie_block->gbea = 1;
3291         vcpu->arch.sie_block->pp = 0;
3292         vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
3293 }
3294
3295 static void kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu *vcpu)
3296 {
3297         struct kvm_sync_regs *regs = &vcpu->run->s.regs;
3298
3299         /* Clear reset is a superset of the initial reset */
3300         kvm_arch_vcpu_ioctl_initial_reset(vcpu);
3301
3302         memset(&regs->gprs, 0, sizeof(regs->gprs));
3303         memset(&regs->vrs, 0, sizeof(regs->vrs));
3304         memset(&regs->acrs, 0, sizeof(regs->acrs));
3305         memset(&regs->gscb, 0, sizeof(regs->gscb));
3306
3307         regs->etoken = 0;
3308         regs->etoken_extension = 0;
3309 }
3310
3311 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3312 {
3313         vcpu_load(vcpu);
3314         memcpy(&vcpu->run->s.regs.gprs, &regs->gprs, sizeof(regs->gprs));
3315         vcpu_put(vcpu);
3316         return 0;
3317 }
3318
3319 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3320 {
3321         vcpu_load(vcpu);
3322         memcpy(&regs->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
3323         vcpu_put(vcpu);
3324         return 0;
3325 }
3326
3327 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3328                                   struct kvm_sregs *sregs)
3329 {
3330         vcpu_load(vcpu);
3331
3332         memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
3333         memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
3334
3335         vcpu_put(vcpu);
3336         return 0;
3337 }
3338
3339 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3340                                   struct kvm_sregs *sregs)
3341 {
3342         vcpu_load(vcpu);
3343
3344         memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
3345         memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
3346
3347         vcpu_put(vcpu);
3348         return 0;
3349 }
3350
3351 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3352 {
3353         int ret = 0;
3354
3355         vcpu_load(vcpu);
3356
3357         if (test_fp_ctl(fpu->fpc)) {
3358                 ret = -EINVAL;
3359                 goto out;
3360         }
3361         vcpu->run->s.regs.fpc = fpu->fpc;
3362         if (MACHINE_HAS_VX)
3363                 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
3364                                  (freg_t *) fpu->fprs);
3365         else
3366                 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
3367
3368 out:
3369         vcpu_put(vcpu);
3370         return ret;
3371 }
3372
3373 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3374 {
3375         vcpu_load(vcpu);
3376
3377         /* make sure we have the latest values */
3378         save_fpu_regs();
3379         if (MACHINE_HAS_VX)
3380                 convert_vx_to_fp((freg_t *) fpu->fprs,
3381                                  (__vector128 *) vcpu->run->s.regs.vrs);
3382         else
3383                 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
3384         fpu->fpc = vcpu->run->s.regs.fpc;
3385
3386         vcpu_put(vcpu);
3387         return 0;
3388 }
3389
3390 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
3391 {
3392         int rc = 0;
3393
3394         if (!is_vcpu_stopped(vcpu))
3395                 rc = -EBUSY;
3396         else {
3397                 vcpu->run->psw_mask = psw.mask;
3398                 vcpu->run->psw_addr = psw.addr;
3399         }
3400         return rc;
3401 }
3402
3403 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3404                                   struct kvm_translation *tr)
3405 {
3406         return -EINVAL; /* not implemented yet */
3407 }
3408
3409 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
3410                               KVM_GUESTDBG_USE_HW_BP | \
3411                               KVM_GUESTDBG_ENABLE)
3412
3413 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
3414                                         struct kvm_guest_debug *dbg)
3415 {
3416         int rc = 0;
3417
3418         vcpu_load(vcpu);
3419
3420         vcpu->guest_debug = 0;
3421         kvm_s390_clear_bp_data(vcpu);
3422
3423         if (dbg->control & ~VALID_GUESTDBG_FLAGS) {
3424                 rc = -EINVAL;
3425                 goto out;
3426         }
3427         if (!sclp.has_gpere) {
3428                 rc = -EINVAL;
3429                 goto out;
3430         }
3431
3432         if (dbg->control & KVM_GUESTDBG_ENABLE) {
3433                 vcpu->guest_debug = dbg->control;
3434                 /* enforce guest PER */
3435                 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P);
3436
3437                 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
3438                         rc = kvm_s390_import_bp_data(vcpu, dbg);
3439         } else {
3440                 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
3441                 vcpu->arch.guestdbg.last_bp = 0;
3442         }
3443
3444         if (rc) {
3445                 vcpu->guest_debug = 0;
3446                 kvm_s390_clear_bp_data(vcpu);
3447                 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
3448         }
3449
3450 out:
3451         vcpu_put(vcpu);
3452         return rc;
3453 }
3454
3455 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3456                                     struct kvm_mp_state *mp_state)
3457 {
3458         int ret;
3459
3460         vcpu_load(vcpu);
3461
3462         /* CHECK_STOP and LOAD are not supported yet */
3463         ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
3464                                       KVM_MP_STATE_OPERATING;
3465
3466         vcpu_put(vcpu);
3467         return ret;
3468 }
3469
3470 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3471                                     struct kvm_mp_state *mp_state)
3472 {
3473         int rc = 0;
3474
3475         vcpu_load(vcpu);
3476
3477         /* user space knows about this interface - let it control the state */
3478         vcpu->kvm->arch.user_cpu_state_ctrl = 1;
3479
3480         switch (mp_state->mp_state) {
3481         case KVM_MP_STATE_STOPPED:
3482                 kvm_s390_vcpu_stop(vcpu);
3483                 break;
3484         case KVM_MP_STATE_OPERATING:
3485                 kvm_s390_vcpu_start(vcpu);
3486                 break;
3487         case KVM_MP_STATE_LOAD:
3488         case KVM_MP_STATE_CHECK_STOP:
3489                 /* fall through - CHECK_STOP and LOAD are not supported yet */
3490         default:
3491                 rc = -ENXIO;
3492         }
3493
3494         vcpu_put(vcpu);
3495         return rc;
3496 }
3497
3498 static bool ibs_enabled(struct kvm_vcpu *vcpu)
3499 {
3500         return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS);
3501 }
3502
3503 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
3504 {
3505 retry:
3506         kvm_s390_vcpu_request_handled(vcpu);
3507         if (!kvm_request_pending(vcpu))
3508                 return 0;
3509         /*
3510          * We use MMU_RELOAD just to re-arm the ipte notifier for the
3511          * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
3512          * This ensures that the ipte instruction for this request has
3513          * already finished. We might race against a second unmapper that
3514          * wants to set the blocking bit. Lets just retry the request loop.
3515          */
3516         if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
3517                 int rc;
3518                 rc = gmap_mprotect_notify(vcpu->arch.gmap,
3519                                           kvm_s390_get_prefix(vcpu),
3520                                           PAGE_SIZE * 2, PROT_WRITE);
3521                 if (rc) {
3522                         kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
3523                         return rc;
3524                 }
3525                 goto retry;
3526         }
3527
3528         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
3529                 vcpu->arch.sie_block->ihcpu = 0xffff;
3530                 goto retry;
3531         }
3532
3533         if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
3534                 if (!ibs_enabled(vcpu)) {
3535                         trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
3536                         kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS);
3537                 }
3538                 goto retry;
3539         }
3540
3541         if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
3542                 if (ibs_enabled(vcpu)) {
3543                         trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
3544                         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS);
3545                 }
3546                 goto retry;
3547         }
3548
3549         if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
3550                 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
3551                 goto retry;
3552         }
3553
3554         if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) {
3555                 /*
3556                  * Disable CMM virtualization; we will emulate the ESSA
3557                  * instruction manually, in order to provide additional
3558                  * functionalities needed for live migration.
3559                  */
3560                 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA;
3561                 goto retry;
3562         }
3563
3564         if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) {
3565                 /*
3566                  * Re-enable CMM virtualization if CMMA is available and
3567                  * CMM has been used.
3568                  */
3569                 if ((vcpu->kvm->arch.use_cmma) &&
3570                     (vcpu->kvm->mm->context.uses_cmm))
3571                         vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
3572                 goto retry;
3573         }
3574
3575         /* nothing to do, just clear the request */
3576         kvm_clear_request(KVM_REQ_UNHALT, vcpu);
3577         /* we left the vsie handler, nothing to do, just clear the request */
3578         kvm_clear_request(KVM_REQ_VSIE_RESTART, vcpu);
3579
3580         return 0;
3581 }
3582
3583 void kvm_s390_set_tod_clock(struct kvm *kvm,
3584                             const struct kvm_s390_vm_tod_clock *gtod)
3585 {
3586         struct kvm_vcpu *vcpu;
3587         struct kvm_s390_tod_clock_ext htod;
3588         int i;
3589
3590         mutex_lock(&kvm->lock);
3591         preempt_disable();
3592
3593         get_tod_clock_ext((char *)&htod);
3594
3595         kvm->arch.epoch = gtod->tod - htod.tod;
3596         kvm->arch.epdx = 0;
3597         if (test_kvm_facility(kvm, 139)) {
3598                 kvm->arch.epdx = gtod->epoch_idx - htod.epoch_idx;
3599                 if (kvm->arch.epoch > gtod->tod)
3600                         kvm->arch.epdx -= 1;
3601         }
3602
3603         kvm_s390_vcpu_block_all(kvm);
3604         kvm_for_each_vcpu(i, vcpu, kvm) {
3605                 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
3606                 vcpu->arch.sie_block->epdx  = kvm->arch.epdx;
3607         }
3608
3609         kvm_s390_vcpu_unblock_all(kvm);
3610         preempt_enable();
3611         mutex_unlock(&kvm->lock);
3612 }
3613
3614 /**
3615  * kvm_arch_fault_in_page - fault-in guest page if necessary
3616  * @vcpu: The corresponding virtual cpu
3617  * @gpa: Guest physical address
3618  * @writable: Whether the page should be writable or not
3619  *
3620  * Make sure that a guest page has been faulted-in on the host.
3621  *
3622  * Return: Zero on success, negative error code otherwise.
3623  */
3624 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
3625 {
3626         return gmap_fault(vcpu->arch.gmap, gpa,
3627                           writable ? FAULT_FLAG_WRITE : 0);
3628 }
3629
3630 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
3631                                       unsigned long token)
3632 {
3633         struct kvm_s390_interrupt inti;
3634         struct kvm_s390_irq irq;
3635
3636         if (start_token) {
3637                 irq.u.ext.ext_params2 = token;
3638                 irq.type = KVM_S390_INT_PFAULT_INIT;
3639                 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
3640         } else {
3641                 inti.type = KVM_S390_INT_PFAULT_DONE;
3642                 inti.parm64 = token;
3643                 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
3644         }
3645 }
3646
3647 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
3648                                      struct kvm_async_pf *work)
3649 {
3650         trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
3651         __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
3652 }
3653
3654 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
3655                                  struct kvm_async_pf *work)
3656 {
3657         trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
3658         __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
3659 }
3660
3661 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
3662                                struct kvm_async_pf *work)
3663 {
3664         /* s390 will always inject the page directly */
3665 }
3666
3667 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
3668 {
3669         /*
3670          * s390 will always inject the page directly,
3671          * but we still want check_async_completion to cleanup
3672          */
3673         return true;
3674 }
3675
3676 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
3677 {
3678         hva_t hva;
3679         struct kvm_arch_async_pf arch;
3680         int rc;
3681
3682         if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
3683                 return 0;
3684         if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
3685             vcpu->arch.pfault_compare)
3686                 return 0;
3687         if (psw_extint_disabled(vcpu))
3688                 return 0;
3689         if (kvm_s390_vcpu_has_irq(vcpu, 0))
3690                 return 0;
3691         if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
3692                 return 0;
3693         if (!vcpu->arch.gmap->pfault_enabled)
3694                 return 0;
3695
3696         hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
3697         hva += current->thread.gmap_addr & ~PAGE_MASK;
3698         if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
3699                 return 0;
3700
3701         rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
3702         return rc;
3703 }
3704
3705 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
3706 {
3707         int rc, cpuflags;
3708
3709         /*
3710          * On s390 notifications for arriving pages will be delivered directly
3711          * to the guest but the house keeping for completed pfaults is
3712          * handled outside the worker.
3713          */
3714         kvm_check_async_pf_completion(vcpu);
3715
3716         vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
3717         vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
3718
3719         if (need_resched())
3720                 schedule();
3721
3722         if (test_cpu_flag(CIF_MCCK_PENDING))
3723                 s390_handle_mcck();
3724
3725         if (!kvm_is_ucontrol(vcpu->kvm)) {
3726                 rc = kvm_s390_deliver_pending_interrupts(vcpu);
3727                 if (rc)
3728                         return rc;
3729         }
3730
3731         rc = kvm_s390_handle_requests(vcpu);
3732         if (rc)
3733                 return rc;
3734
3735         if (guestdbg_enabled(vcpu)) {
3736                 kvm_s390_backup_guest_per_regs(vcpu);
3737                 kvm_s390_patch_guest_per_regs(vcpu);
3738         }
3739
3740         clear_bit(vcpu->vcpu_id, vcpu->kvm->arch.gisa_int.kicked_mask);
3741
3742         vcpu->arch.sie_block->icptcode = 0;
3743         cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
3744         VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
3745         trace_kvm_s390_sie_enter(vcpu, cpuflags);
3746
3747         return 0;
3748 }
3749
3750 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
3751 {
3752         struct kvm_s390_pgm_info pgm_info = {
3753                 .code = PGM_ADDRESSING,
3754         };
3755         u8 opcode, ilen;
3756         int rc;
3757
3758         VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
3759         trace_kvm_s390_sie_fault(vcpu);
3760
3761         /*
3762          * We want to inject an addressing exception, which is defined as a
3763          * suppressing or terminating exception. However, since we came here
3764          * by a DAT access exception, the PSW still points to the faulting
3765          * instruction since DAT exceptions are nullifying. So we've got
3766          * to look up the current opcode to get the length of the instruction
3767          * to be able to forward the PSW.
3768          */
3769         rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
3770         ilen = insn_length(opcode);
3771         if (rc < 0) {
3772                 return rc;
3773         } else if (rc) {
3774                 /* Instruction-Fetching Exceptions - we can't detect the ilen.
3775                  * Forward by arbitrary ilc, injection will take care of
3776                  * nullification if necessary.
3777                  */
3778                 pgm_info = vcpu->arch.pgm;
3779                 ilen = 4;
3780         }
3781         pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
3782         kvm_s390_forward_psw(vcpu, ilen);
3783         return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
3784 }
3785
3786 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
3787 {
3788         struct mcck_volatile_info *mcck_info;
3789         struct sie_page *sie_page;
3790
3791         VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
3792                    vcpu->arch.sie_block->icptcode);
3793         trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
3794
3795         if (guestdbg_enabled(vcpu))
3796                 kvm_s390_restore_guest_per_regs(vcpu);
3797
3798         vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
3799         vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
3800
3801         if (exit_reason == -EINTR) {
3802                 VCPU_EVENT(vcpu, 3, "%s", "machine check");
3803                 sie_page = container_of(vcpu->arch.sie_block,
3804                                         struct sie_page, sie_block);
3805                 mcck_info = &sie_page->mcck_info;
3806                 kvm_s390_reinject_machine_check(vcpu, mcck_info);
3807                 return 0;
3808         }
3809
3810         if (vcpu->arch.sie_block->icptcode > 0) {
3811                 int rc = kvm_handle_sie_intercept(vcpu);
3812
3813                 if (rc != -EOPNOTSUPP)
3814                         return rc;
3815                 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
3816                 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
3817                 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
3818                 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
3819                 return -EREMOTE;
3820         } else if (exit_reason != -EFAULT) {
3821                 vcpu->stat.exit_null++;
3822                 return 0;
3823         } else if (kvm_is_ucontrol(vcpu->kvm)) {
3824                 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
3825                 vcpu->run->s390_ucontrol.trans_exc_code =
3826                                                 current->thread.gmap_addr;
3827                 vcpu->run->s390_ucontrol.pgm_code = 0x10;
3828                 return -EREMOTE;
3829         } else if (current->thread.gmap_pfault) {
3830                 trace_kvm_s390_major_guest_pfault(vcpu);
3831                 current->thread.gmap_pfault = 0;
3832                 if (kvm_arch_setup_async_pf(vcpu))
3833                         return 0;
3834                 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
3835         }
3836         return vcpu_post_run_fault_in_sie(vcpu);
3837 }
3838
3839 static int __vcpu_run(struct kvm_vcpu *vcpu)
3840 {
3841         int rc, exit_reason;
3842
3843         /*
3844          * We try to hold kvm->srcu during most of vcpu_run (except when run-
3845          * ning the guest), so that memslots (and other stuff) are protected
3846          */
3847         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3848
3849         do {
3850                 rc = vcpu_pre_run(vcpu);
3851                 if (rc)
3852                         break;
3853
3854                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3855                 /*
3856                  * As PF_VCPU will be used in fault handler, between
3857                  * guest_enter and guest_exit should be no uaccess.
3858                  */
3859                 local_irq_disable();
3860                 guest_enter_irqoff();
3861                 __disable_cpu_timer_accounting(vcpu);
3862                 local_irq_enable();
3863                 exit_reason = sie64a(vcpu->arch.sie_block,
3864                                      vcpu->run->s.regs.gprs);
3865                 local_irq_disable();
3866                 __enable_cpu_timer_accounting(vcpu);
3867                 guest_exit_irqoff();
3868                 local_irq_enable();
3869                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3870
3871                 rc = vcpu_post_run(vcpu, exit_reason);
3872         } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
3873
3874         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3875         return rc;
3876 }
3877
3878 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3879 {
3880         struct runtime_instr_cb *riccb;
3881         struct gs_cb *gscb;
3882
3883         riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
3884         gscb = (struct gs_cb *) &kvm_run->s.regs.gscb;
3885         vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
3886         vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
3887         if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
3888                 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
3889         if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
3890                 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
3891                 /* some control register changes require a tlb flush */
3892                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3893         }
3894         if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
3895                 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
3896                 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
3897                 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
3898                 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
3899                 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
3900         }
3901         if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
3902                 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
3903                 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
3904                 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
3905                 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
3906                         kvm_clear_async_pf_completion_queue(vcpu);
3907         }
3908         /*
3909          * If userspace sets the riccb (e.g. after migration) to a valid state,
3910          * we should enable RI here instead of doing the lazy enablement.
3911          */
3912         if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
3913             test_kvm_facility(vcpu->kvm, 64) &&
3914             riccb->v &&
3915             !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) {
3916                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)");
3917                 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
3918         }
3919         /*
3920          * If userspace sets the gscb (e.g. after migration) to non-zero,
3921          * we should enable GS here instead of doing the lazy enablement.
3922          */
3923         if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) &&
3924             test_kvm_facility(vcpu->kvm, 133) &&
3925             gscb->gssm &&
3926             !vcpu->arch.gs_enabled) {
3927                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)");
3928                 vcpu->arch.sie_block->ecb |= ECB_GS;
3929                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
3930                 vcpu->arch.gs_enabled = 1;
3931         }
3932         if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) &&
3933             test_kvm_facility(vcpu->kvm, 82)) {
3934                 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
3935                 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0;
3936         }
3937         save_access_regs(vcpu->arch.host_acrs);
3938         restore_access_regs(vcpu->run->s.regs.acrs);
3939         /* save host (userspace) fprs/vrs */
3940         save_fpu_regs();
3941         vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
3942         vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
3943         if (MACHINE_HAS_VX)
3944                 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
3945         else
3946                 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
3947         current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
3948         if (test_fp_ctl(current->thread.fpu.fpc))
3949                 /* User space provided an invalid FPC, let's clear it */
3950                 current->thread.fpu.fpc = 0;
3951         if (MACHINE_HAS_GS) {
3952                 preempt_disable();
3953                 __ctl_set_bit(2, 4);
3954                 if (current->thread.gs_cb) {
3955                         vcpu->arch.host_gscb = current->thread.gs_cb;
3956                         save_gs_cb(vcpu->arch.host_gscb);
3957                 }
3958                 if (vcpu->arch.gs_enabled) {
3959                         current->thread.gs_cb = (struct gs_cb *)
3960                                                 &vcpu->run->s.regs.gscb;
3961                         restore_gs_cb(current->thread.gs_cb);
3962                 }
3963                 preempt_enable();
3964         }
3965         /* SIE will load etoken directly from SDNX and therefore kvm_run */
3966
3967         kvm_run->kvm_dirty_regs = 0;
3968 }
3969
3970 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3971 {
3972         kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
3973         kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
3974         kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
3975         memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
3976         kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
3977         kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
3978         kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
3979         kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
3980         kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
3981         kvm_run->s.regs.pft = vcpu->arch.pfault_token;
3982         kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
3983         kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
3984         kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC;
3985         save_access_regs(vcpu->run->s.regs.acrs);
3986         restore_access_regs(vcpu->arch.host_acrs);
3987         /* Save guest register state */
3988         save_fpu_regs();
3989         vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
3990         /* Restore will be done lazily at return */
3991         current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
3992         current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
3993         if (MACHINE_HAS_GS) {
3994                 __ctl_set_bit(2, 4);
3995                 if (vcpu->arch.gs_enabled)
3996                         save_gs_cb(current->thread.gs_cb);
3997                 preempt_disable();
3998                 current->thread.gs_cb = vcpu->arch.host_gscb;
3999                 restore_gs_cb(vcpu->arch.host_gscb);
4000                 preempt_enable();
4001                 if (!vcpu->arch.host_gscb)
4002                         __ctl_clear_bit(2, 4);
4003                 vcpu->arch.host_gscb = NULL;
4004         }
4005         /* SIE will save etoken directly into SDNX and therefore kvm_run */
4006 }
4007
4008 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4009 {
4010         int rc;
4011
4012         if (kvm_run->immediate_exit)
4013                 return -EINTR;
4014
4015         if (kvm_run->kvm_valid_regs & ~KVM_SYNC_S390_VALID_FIELDS ||
4016             kvm_run->kvm_dirty_regs & ~KVM_SYNC_S390_VALID_FIELDS)
4017                 return -EINVAL;
4018
4019         vcpu_load(vcpu);
4020
4021         if (guestdbg_exit_pending(vcpu)) {
4022                 kvm_s390_prepare_debug_exit(vcpu);
4023                 rc = 0;
4024                 goto out;
4025         }
4026
4027         kvm_sigset_activate(vcpu);
4028
4029         if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
4030                 kvm_s390_vcpu_start(vcpu);
4031         } else if (is_vcpu_stopped(vcpu)) {
4032                 pr_err_ratelimited("can't run stopped vcpu %d\n",
4033                                    vcpu->vcpu_id);
4034                 rc = -EINVAL;
4035                 goto out;
4036         }
4037
4038         sync_regs(vcpu, kvm_run);
4039         enable_cpu_timer_accounting(vcpu);
4040
4041         might_fault();
4042         rc = __vcpu_run(vcpu);
4043
4044         if (signal_pending(current) && !rc) {
4045                 kvm_run->exit_reason = KVM_EXIT_INTR;
4046                 rc = -EINTR;
4047         }
4048
4049         if (guestdbg_exit_pending(vcpu) && !rc)  {
4050                 kvm_s390_prepare_debug_exit(vcpu);
4051                 rc = 0;
4052         }
4053
4054         if (rc == -EREMOTE) {
4055                 /* userspace support is needed, kvm_run has been prepared */
4056                 rc = 0;
4057         }
4058
4059         disable_cpu_timer_accounting(vcpu);
4060         store_regs(vcpu, kvm_run);
4061
4062         kvm_sigset_deactivate(vcpu);
4063
4064         vcpu->stat.exit_userspace++;
4065 out:
4066         vcpu_put(vcpu);
4067         return rc;
4068 }
4069
4070 /*
4071  * store status at address
4072  * we use have two special cases:
4073  * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
4074  * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
4075  */
4076 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
4077 {
4078         unsigned char archmode = 1;
4079         freg_t fprs[NUM_FPRS];
4080         unsigned int px;
4081         u64 clkcomp, cputm;
4082         int rc;
4083
4084         px = kvm_s390_get_prefix(vcpu);
4085         if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
4086                 if (write_guest_abs(vcpu, 163, &archmode, 1))
4087                         return -EFAULT;
4088                 gpa = 0;
4089         } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
4090                 if (write_guest_real(vcpu, 163, &archmode, 1))
4091                         return -EFAULT;
4092                 gpa = px;
4093         } else
4094                 gpa -= __LC_FPREGS_SAVE_AREA;
4095
4096         /* manually convert vector registers if necessary */
4097         if (MACHINE_HAS_VX) {
4098                 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
4099                 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
4100                                      fprs, 128);
4101         } else {
4102                 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
4103                                      vcpu->run->s.regs.fprs, 128);
4104         }
4105         rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
4106                               vcpu->run->s.regs.gprs, 128);
4107         rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
4108                               &vcpu->arch.sie_block->gpsw, 16);
4109         rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
4110                               &px, 4);
4111         rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
4112                               &vcpu->run->s.regs.fpc, 4);
4113         rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
4114                               &vcpu->arch.sie_block->todpr, 4);
4115         cputm = kvm_s390_get_cpu_timer(vcpu);
4116         rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
4117                               &cputm, 8);
4118         clkcomp = vcpu->arch.sie_block->ckc >> 8;
4119         rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
4120                               &clkcomp, 8);
4121         rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
4122                               &vcpu->run->s.regs.acrs, 64);
4123         rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
4124                               &vcpu->arch.sie_block->gcr, 128);
4125         return rc ? -EFAULT : 0;
4126 }
4127
4128 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
4129 {
4130         /*
4131          * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
4132          * switch in the run ioctl. Let's update our copies before we save
4133          * it into the save area
4134          */
4135         save_fpu_regs();
4136         vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
4137         save_access_regs(vcpu->run->s.regs.acrs);
4138
4139         return kvm_s390_store_status_unloaded(vcpu, addr);
4140 }
4141
4142 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
4143 {
4144         kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
4145         kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
4146 }
4147
4148 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
4149 {
4150         unsigned int i;
4151         struct kvm_vcpu *vcpu;
4152
4153         kvm_for_each_vcpu(i, vcpu, kvm) {
4154                 __disable_ibs_on_vcpu(vcpu);
4155         }
4156 }
4157
4158 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
4159 {
4160         if (!sclp.has_ibs)
4161                 return;
4162         kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
4163         kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
4164 }
4165
4166 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
4167 {
4168         int i, online_vcpus, started_vcpus = 0;
4169
4170         if (!is_vcpu_stopped(vcpu))
4171                 return;
4172
4173         trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
4174         /* Only one cpu at a time may enter/leave the STOPPED state. */
4175         spin_lock(&vcpu->kvm->arch.start_stop_lock);
4176         online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
4177
4178         for (i = 0; i < online_vcpus; i++) {
4179                 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
4180                         started_vcpus++;
4181         }
4182
4183         if (started_vcpus == 0) {
4184                 /* we're the only active VCPU -> speed it up */
4185                 __enable_ibs_on_vcpu(vcpu);
4186         } else if (started_vcpus == 1) {
4187                 /*
4188                  * As we are starting a second VCPU, we have to disable
4189                  * the IBS facility on all VCPUs to remove potentially
4190                  * oustanding ENABLE requests.
4191                  */
4192                 __disable_ibs_on_all_vcpus(vcpu->kvm);
4193         }
4194
4195         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED);
4196         /*
4197          * Another VCPU might have used IBS while we were offline.
4198          * Let's play safe and flush the VCPU at startup.
4199          */
4200         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4201         spin_unlock(&vcpu->kvm->arch.start_stop_lock);
4202         return;
4203 }
4204
4205 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
4206 {
4207         int i, online_vcpus, started_vcpus = 0;
4208         struct kvm_vcpu *started_vcpu = NULL;
4209
4210         if (is_vcpu_stopped(vcpu))
4211                 return;
4212
4213         trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
4214         /* Only one cpu at a time may enter/leave the STOPPED state. */
4215         spin_lock(&vcpu->kvm->arch.start_stop_lock);
4216         online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
4217
4218         /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
4219         kvm_s390_clear_stop_irq(vcpu);
4220
4221         kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED);
4222         __disable_ibs_on_vcpu(vcpu);
4223
4224         for (i = 0; i < online_vcpus; i++) {
4225                 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
4226                         started_vcpus++;
4227                         started_vcpu = vcpu->kvm->vcpus[i];
4228                 }
4229         }
4230
4231         if (started_vcpus == 1) {
4232                 /*
4233                  * As we only have one VCPU left, we want to enable the
4234                  * IBS facility for that VCPU to speed it up.
4235                  */
4236                 __enable_ibs_on_vcpu(started_vcpu);
4237         }
4238
4239         spin_unlock(&vcpu->kvm->arch.start_stop_lock);
4240         return;
4241 }
4242
4243 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4244                                      struct kvm_enable_cap *cap)
4245 {
4246         int r;
4247
4248         if (cap->flags)
4249                 return -EINVAL;
4250
4251         switch (cap->cap) {
4252         case KVM_CAP_S390_CSS_SUPPORT:
4253                 if (!vcpu->kvm->arch.css_support) {
4254                         vcpu->kvm->arch.css_support = 1;
4255                         VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
4256                         trace_kvm_s390_enable_css(vcpu->kvm);
4257                 }
4258                 r = 0;
4259                 break;
4260         default:
4261                 r = -EINVAL;
4262                 break;
4263         }
4264         return r;
4265 }
4266
4267 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
4268                                   struct kvm_s390_mem_op *mop)
4269 {
4270         void __user *uaddr = (void __user *)mop->buf;
4271         void *tmpbuf = NULL;
4272         int r, srcu_idx;
4273         const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
4274                                     | KVM_S390_MEMOP_F_CHECK_ONLY;
4275
4276         if (mop->flags & ~supported_flags || mop->ar >= NUM_ACRS || !mop->size)
4277                 return -EINVAL;
4278
4279         if (mop->size > MEM_OP_MAX_SIZE)
4280                 return -E2BIG;
4281
4282         if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
4283                 tmpbuf = vmalloc(mop->size);
4284                 if (!tmpbuf)
4285                         return -ENOMEM;
4286         }
4287
4288         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4289
4290         switch (mop->op) {
4291         case KVM_S390_MEMOP_LOGICAL_READ:
4292                 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
4293                         r = check_gva_range(vcpu, mop->gaddr, mop->ar,
4294                                             mop->size, GACC_FETCH);
4295                         break;
4296                 }
4297                 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
4298                 if (r == 0) {
4299                         if (copy_to_user(uaddr, tmpbuf, mop->size))
4300                                 r = -EFAULT;
4301                 }
4302                 break;
4303         case KVM_S390_MEMOP_LOGICAL_WRITE:
4304                 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
4305                         r = check_gva_range(vcpu, mop->gaddr, mop->ar,
4306                                             mop->size, GACC_STORE);
4307                         break;
4308                 }
4309                 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
4310                         r = -EFAULT;
4311                         break;
4312                 }
4313                 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
4314                 break;
4315         default:
4316                 r = -EINVAL;
4317         }
4318
4319         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
4320
4321         if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
4322                 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
4323
4324         vfree(tmpbuf);
4325         return r;
4326 }
4327
4328 long kvm_arch_vcpu_async_ioctl(struct file *filp,
4329                                unsigned int ioctl, unsigned long arg)
4330 {
4331         struct kvm_vcpu *vcpu = filp->private_data;
4332         void __user *argp = (void __user *)arg;
4333
4334         switch (ioctl) {
4335         case KVM_S390_IRQ: {
4336                 struct kvm_s390_irq s390irq;
4337
4338                 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
4339                         return -EFAULT;
4340                 return kvm_s390_inject_vcpu(vcpu, &s390irq);
4341         }
4342         case KVM_S390_INTERRUPT: {
4343                 struct kvm_s390_interrupt s390int;
4344                 struct kvm_s390_irq s390irq = {};
4345
4346                 if (copy_from_user(&s390int, argp, sizeof(s390int)))
4347                         return -EFAULT;
4348                 if (s390int_to_s390irq(&s390int, &s390irq))
4349                         return -EINVAL;
4350                 return kvm_s390_inject_vcpu(vcpu, &s390irq);
4351         }
4352         }
4353         return -ENOIOCTLCMD;
4354 }
4355
4356 long kvm_arch_vcpu_ioctl(struct file *filp,
4357                          unsigned int ioctl, unsigned long arg)
4358 {
4359         struct kvm_vcpu *vcpu = filp->private_data;
4360         void __user *argp = (void __user *)arg;
4361         int idx;
4362         long r;
4363
4364         vcpu_load(vcpu);
4365
4366         switch (ioctl) {
4367         case KVM_S390_STORE_STATUS:
4368                 idx = srcu_read_lock(&vcpu->kvm->srcu);
4369                 r = kvm_s390_store_status_unloaded(vcpu, arg);
4370                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4371                 break;
4372         case KVM_S390_SET_INITIAL_PSW: {
4373                 psw_t psw;
4374
4375                 r = -EFAULT;
4376                 if (copy_from_user(&psw, argp, sizeof(psw)))
4377                         break;
4378                 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
4379                 break;
4380         }
4381         case KVM_S390_CLEAR_RESET:
4382                 r = 0;
4383                 kvm_arch_vcpu_ioctl_clear_reset(vcpu);
4384                 break;
4385         case KVM_S390_INITIAL_RESET:
4386                 r = 0;
4387                 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
4388                 break;
4389         case KVM_S390_NORMAL_RESET:
4390                 r = 0;
4391                 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
4392                 break;
4393         case KVM_SET_ONE_REG:
4394         case KVM_GET_ONE_REG: {
4395                 struct kvm_one_reg reg;
4396                 r = -EFAULT;
4397                 if (copy_from_user(&reg, argp, sizeof(reg)))
4398                         break;
4399                 if (ioctl == KVM_SET_ONE_REG)
4400                         r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, &reg);
4401                 else
4402                         r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, &reg);
4403                 break;
4404         }
4405 #ifdef CONFIG_KVM_S390_UCONTROL
4406         case KVM_S390_UCAS_MAP: {
4407                 struct kvm_s390_ucas_mapping ucasmap;
4408
4409                 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
4410                         r = -EFAULT;
4411                         break;
4412                 }
4413
4414                 if (!kvm_is_ucontrol(vcpu->kvm)) {
4415                         r = -EINVAL;
4416                         break;
4417                 }
4418
4419                 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
4420                                      ucasmap.vcpu_addr, ucasmap.length);
4421                 break;
4422         }
4423         case KVM_S390_UCAS_UNMAP: {
4424                 struct kvm_s390_ucas_mapping ucasmap;
4425
4426                 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
4427                         r = -EFAULT;
4428                         break;
4429                 }
4430
4431                 if (!kvm_is_ucontrol(vcpu->kvm)) {
4432                         r = -EINVAL;
4433                         break;
4434                 }
4435
4436                 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
4437                         ucasmap.length);
4438                 break;
4439         }
4440 #endif
4441         case KVM_S390_VCPU_FAULT: {
4442                 r = gmap_fault(vcpu->arch.gmap, arg, 0);
4443                 break;
4444         }
4445         case KVM_ENABLE_CAP:
4446         {
4447                 struct kvm_enable_cap cap;
4448                 r = -EFAULT;
4449                 if (copy_from_user(&cap, argp, sizeof(cap)))
4450                         break;
4451                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4452                 break;
4453         }
4454         case KVM_S390_MEM_OP: {
4455                 struct kvm_s390_mem_op mem_op;
4456
4457                 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
4458                         r = kvm_s390_guest_mem_op(vcpu, &mem_op);
4459                 else
4460                         r = -EFAULT;
4461                 break;
4462         }
4463         case KVM_S390_SET_IRQ_STATE: {
4464                 struct kvm_s390_irq_state irq_state;
4465
4466                 r = -EFAULT;
4467                 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
4468                         break;
4469                 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
4470                     irq_state.len == 0 ||
4471                     irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
4472                         r = -EINVAL;
4473                         break;
4474                 }
4475                 /* do not use irq_state.flags, it will break old QEMUs */
4476                 r = kvm_s390_set_irq_state(vcpu,
4477                                            (void __user *) irq_state.buf,
4478                                            irq_state.len);
4479                 break;
4480         }
4481         case KVM_S390_GET_IRQ_STATE: {
4482                 struct kvm_s390_irq_state irq_state;
4483
4484                 r = -EFAULT;
4485                 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
4486                         break;
4487                 if (irq_state.len == 0) {
4488                         r = -EINVAL;
4489                         break;
4490                 }
4491                 /* do not use irq_state.flags, it will break old QEMUs */
4492                 r = kvm_s390_get_irq_state(vcpu,
4493                                            (__u8 __user *)  irq_state.buf,
4494                                            irq_state.len);
4495                 break;
4496         }
4497         default:
4498                 r = -ENOTTY;
4499         }
4500
4501         vcpu_put(vcpu);
4502         return r;
4503 }
4504
4505 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4506 {
4507 #ifdef CONFIG_KVM_S390_UCONTROL
4508         if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
4509                  && (kvm_is_ucontrol(vcpu->kvm))) {
4510                 vmf->page = virt_to_page(vcpu->arch.sie_block);
4511                 get_page(vmf->page);
4512                 return 0;
4513         }
4514 #endif
4515         return VM_FAULT_SIGBUS;
4516 }
4517
4518 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
4519                             unsigned long npages)
4520 {
4521         return 0;
4522 }
4523
4524 /* Section: memory related */
4525 int kvm_arch_prepare_memory_region(struct kvm *kvm,
4526                                    struct kvm_memory_slot *memslot,
4527                                    const struct kvm_userspace_memory_region *mem,
4528                                    enum kvm_mr_change change)
4529 {
4530         /* A few sanity checks. We can have memory slots which have to be
4531            located/ended at a segment boundary (1MB). The memory in userland is
4532            ok to be fragmented into various different vmas. It is okay to mmap()
4533            and munmap() stuff in this slot after doing this call at any time */
4534
4535         if (mem->userspace_addr & 0xffffful)
4536                 return -EINVAL;
4537
4538         if (mem->memory_size & 0xffffful)
4539                 return -EINVAL;
4540
4541         if (mem->guest_phys_addr + mem->memory_size > kvm->arch.mem_limit)
4542                 return -EINVAL;
4543
4544         return 0;
4545 }
4546
4547 void kvm_arch_commit_memory_region(struct kvm *kvm,
4548                                 const struct kvm_userspace_memory_region *mem,
4549                                 const struct kvm_memory_slot *old,
4550                                 const struct kvm_memory_slot *new,
4551                                 enum kvm_mr_change change)
4552 {
4553         int rc = 0;
4554
4555         switch (change) {
4556         case KVM_MR_DELETE:
4557                 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
4558                                         old->npages * PAGE_SIZE);
4559                 break;
4560         case KVM_MR_MOVE:
4561                 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
4562                                         old->npages * PAGE_SIZE);
4563                 if (rc)
4564                         break;
4565                 /* FALLTHROUGH */
4566         case KVM_MR_CREATE:
4567                 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
4568                                       mem->guest_phys_addr, mem->memory_size);
4569                 break;
4570         case KVM_MR_FLAGS_ONLY:
4571                 break;
4572         default:
4573                 WARN(1, "Unknown KVM MR CHANGE: %d\n", change);
4574         }
4575         if (rc)
4576                 pr_warn("failed to commit memory region\n");
4577         return;
4578 }
4579
4580 static inline unsigned long nonhyp_mask(int i)
4581 {
4582         unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
4583
4584         return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
4585 }
4586
4587 void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
4588 {
4589         vcpu->valid_wakeup = false;
4590 }
4591
4592 static int __init kvm_s390_init(void)
4593 {
4594         int i;
4595
4596         if (!sclp.has_sief2) {
4597                 pr_info("SIE is not available\n");
4598                 return -ENODEV;
4599         }
4600
4601         if (nested && hpage) {
4602                 pr_info("A KVM host that supports nesting cannot back its KVM guests with huge pages\n");
4603                 return -EINVAL;
4604         }
4605
4606         for (i = 0; i < 16; i++)
4607                 kvm_s390_fac_base[i] |=
4608                         S390_lowcore.stfle_fac_list[i] & nonhyp_mask(i);
4609
4610         return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
4611 }
4612
4613 static void __exit kvm_s390_exit(void)
4614 {
4615         kvm_exit();
4616 }
4617
4618 module_init(kvm_s390_init);
4619 module_exit(kvm_s390_exit);
4620
4621 /*
4622  * Enable autoloading of the kvm module.
4623  * Note that we add the module alias here instead of virt/kvm/kvm_main.c
4624  * since x86 takes a different approach.
4625  */
4626 #include <linux/miscdevice.h>
4627 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4628 MODULE_ALIAS("devname:kvm");