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, 2020
   6  *
   7  *    Author(s): Carsten Otte <cotte@de.ibm.com>
   8  *               Christian Borntraeger <borntraeger@de.ibm.com>
   9  *               Christian Ehrhardt <ehrhardt@de.ibm.com>
  10  *               Jason J. Herne <jjherne@us.ibm.com>
  11  */
  12
  13 #define KMSG_COMPONENT "kvm-s390"
  14 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  15
  16 #include <linux/compiler.h>
  17 #include <linux/err.h>
  18 #include <linux/fs.h>
  19 #include <linux/hrtimer.h>
  20 #include <linux/init.h>
  21 #include <linux/kvm.h>
  22 #include <linux/kvm_host.h>
  23 #include <linux/mman.h>
  24 #include <linux/module.h>
  25 #include <linux/moduleparam.h>
  26 #include <linux/random.h>
  27 #include <linux/slab.h>
  28 #include <linux/timer.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/bitmap.h>
  31 #include <linux/sched/signal.h>
  32 #include <linux/string.h>
  33 #include <linux/pgtable.h>
  34 #include <linux/mmu_notifier.h>
  35
  36 #include <asm/asm-offsets.h>
  37 #include <asm/lowcore.h>
  38 #include <asm/stp.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 <asm/uv.h>
  48 #include <asm/fpu/api.h>
  49 #include "kvm-s390.h"
  50 #include "gaccess.h"
  51 #include "pci.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include "trace.h"
  55 #include "trace-s390.h"
  56
  57 #define MEM_OP_MAX_SIZE 65536   /* Maximum transfer size for KVM_S390_MEM_OP */
  58 #define LOCAL_IRQS 32
  59 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
  60                            (KVM_MAX_VCPUS + LOCAL_IRQS))
  61
  62 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
  63         KVM_GENERIC_VM_STATS(),
  64         STATS_DESC_COUNTER(VM, inject_io),
  65         STATS_DESC_COUNTER(VM, inject_float_mchk),
  66         STATS_DESC_COUNTER(VM, inject_pfault_done),
  67         STATS_DESC_COUNTER(VM, inject_service_signal),
  68         STATS_DESC_COUNTER(VM, inject_virtio),
  69         STATS_DESC_COUNTER(VM, aen_forward),
  70         STATS_DESC_COUNTER(VM, gmap_shadow_reuse),
  71         STATS_DESC_COUNTER(VM, gmap_shadow_create),
  72         STATS_DESC_COUNTER(VM, gmap_shadow_r1_entry),
  73         STATS_DESC_COUNTER(VM, gmap_shadow_r2_entry),
  74         STATS_DESC_COUNTER(VM, gmap_shadow_r3_entry),
  75         STATS_DESC_COUNTER(VM, gmap_shadow_sg_entry),
  76         STATS_DESC_COUNTER(VM, gmap_shadow_pg_entry),
  77 };
  78
  79 const struct kvm_stats_header kvm_vm_stats_header = {
  80         .name_size = KVM_STATS_NAME_SIZE,
  81         .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
  82         .id_offset = sizeof(struct kvm_stats_header),
  83         .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
  84         .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
  85                        sizeof(kvm_vm_stats_desc),
  86 };
  87
  88 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
  89         KVM_GENERIC_VCPU_STATS(),
  90         STATS_DESC_COUNTER(VCPU, exit_userspace),
  91         STATS_DESC_COUNTER(VCPU, exit_null),
  92         STATS_DESC_COUNTER(VCPU, exit_external_request),
  93         STATS_DESC_COUNTER(VCPU, exit_io_request),
  94         STATS_DESC_COUNTER(VCPU, exit_external_interrupt),
  95         STATS_DESC_COUNTER(VCPU, exit_stop_request),
  96         STATS_DESC_COUNTER(VCPU, exit_validity),
  97         STATS_DESC_COUNTER(VCPU, exit_instruction),
  98         STATS_DESC_COUNTER(VCPU, exit_pei),
  99         STATS_DESC_COUNTER(VCPU, halt_no_poll_steal),
 100         STATS_DESC_COUNTER(VCPU, instruction_lctl),
 101         STATS_DESC_COUNTER(VCPU, instruction_lctlg),
 102         STATS_DESC_COUNTER(VCPU, instruction_stctl),
 103         STATS_DESC_COUNTER(VCPU, instruction_stctg),
 104         STATS_DESC_COUNTER(VCPU, exit_program_interruption),
 105         STATS_DESC_COUNTER(VCPU, exit_instr_and_program),
 106         STATS_DESC_COUNTER(VCPU, exit_operation_exception),
 107         STATS_DESC_COUNTER(VCPU, deliver_ckc),
 108         STATS_DESC_COUNTER(VCPU, deliver_cputm),
 109         STATS_DESC_COUNTER(VCPU, deliver_external_call),
 110         STATS_DESC_COUNTER(VCPU, deliver_emergency_signal),
 111         STATS_DESC_COUNTER(VCPU, deliver_service_signal),
 112         STATS_DESC_COUNTER(VCPU, deliver_virtio),
 113         STATS_DESC_COUNTER(VCPU, deliver_stop_signal),
 114         STATS_DESC_COUNTER(VCPU, deliver_prefix_signal),
 115         STATS_DESC_COUNTER(VCPU, deliver_restart_signal),
 116         STATS_DESC_COUNTER(VCPU, deliver_program),
 117         STATS_DESC_COUNTER(VCPU, deliver_io),
 118         STATS_DESC_COUNTER(VCPU, deliver_machine_check),
 119         STATS_DESC_COUNTER(VCPU, exit_wait_state),
 120         STATS_DESC_COUNTER(VCPU, inject_ckc),
 121         STATS_DESC_COUNTER(VCPU, inject_cputm),
 122         STATS_DESC_COUNTER(VCPU, inject_external_call),
 123         STATS_DESC_COUNTER(VCPU, inject_emergency_signal),
 124         STATS_DESC_COUNTER(VCPU, inject_mchk),
 125         STATS_DESC_COUNTER(VCPU, inject_pfault_init),
 126         STATS_DESC_COUNTER(VCPU, inject_program),
 127         STATS_DESC_COUNTER(VCPU, inject_restart),
 128         STATS_DESC_COUNTER(VCPU, inject_set_prefix),
 129         STATS_DESC_COUNTER(VCPU, inject_stop_signal),
 130         STATS_DESC_COUNTER(VCPU, instruction_epsw),
 131         STATS_DESC_COUNTER(VCPU, instruction_gs),
 132         STATS_DESC_COUNTER(VCPU, instruction_io_other),
 133         STATS_DESC_COUNTER(VCPU, instruction_lpsw),
 134         STATS_DESC_COUNTER(VCPU, instruction_lpswe),
 135         STATS_DESC_COUNTER(VCPU, instruction_pfmf),
 136         STATS_DESC_COUNTER(VCPU, instruction_ptff),
 137         STATS_DESC_COUNTER(VCPU, instruction_sck),
 138         STATS_DESC_COUNTER(VCPU, instruction_sckpf),
 139         STATS_DESC_COUNTER(VCPU, instruction_stidp),
 140         STATS_DESC_COUNTER(VCPU, instruction_spx),
 141         STATS_DESC_COUNTER(VCPU, instruction_stpx),
 142         STATS_DESC_COUNTER(VCPU, instruction_stap),
 143         STATS_DESC_COUNTER(VCPU, instruction_iske),
 144         STATS_DESC_COUNTER(VCPU, instruction_ri),
 145         STATS_DESC_COUNTER(VCPU, instruction_rrbe),
 146         STATS_DESC_COUNTER(VCPU, instruction_sske),
 147         STATS_DESC_COUNTER(VCPU, instruction_ipte_interlock),
 148         STATS_DESC_COUNTER(VCPU, instruction_stsi),
 149         STATS_DESC_COUNTER(VCPU, instruction_stfl),
 150         STATS_DESC_COUNTER(VCPU, instruction_tb),
 151         STATS_DESC_COUNTER(VCPU, instruction_tpi),
 152         STATS_DESC_COUNTER(VCPU, instruction_tprot),
 153         STATS_DESC_COUNTER(VCPU, instruction_tsch),
 154         STATS_DESC_COUNTER(VCPU, instruction_sie),
 155         STATS_DESC_COUNTER(VCPU, instruction_essa),
 156         STATS_DESC_COUNTER(VCPU, instruction_sthyi),
 157         STATS_DESC_COUNTER(VCPU, instruction_sigp_sense),
 158         STATS_DESC_COUNTER(VCPU, instruction_sigp_sense_running),
 159         STATS_DESC_COUNTER(VCPU, instruction_sigp_external_call),
 160         STATS_DESC_COUNTER(VCPU, instruction_sigp_emergency),
 161         STATS_DESC_COUNTER(VCPU, instruction_sigp_cond_emergency),
 162         STATS_DESC_COUNTER(VCPU, instruction_sigp_start),
 163         STATS_DESC_COUNTER(VCPU, instruction_sigp_stop),
 164         STATS_DESC_COUNTER(VCPU, instruction_sigp_stop_store_status),
 165         STATS_DESC_COUNTER(VCPU, instruction_sigp_store_status),
 166         STATS_DESC_COUNTER(VCPU, instruction_sigp_store_adtl_status),
 167         STATS_DESC_COUNTER(VCPU, instruction_sigp_arch),
 168         STATS_DESC_COUNTER(VCPU, instruction_sigp_prefix),
 169         STATS_DESC_COUNTER(VCPU, instruction_sigp_restart),
 170         STATS_DESC_COUNTER(VCPU, instruction_sigp_init_cpu_reset),
 171         STATS_DESC_COUNTER(VCPU, instruction_sigp_cpu_reset),
 172         STATS_DESC_COUNTER(VCPU, instruction_sigp_unknown),
 173         STATS_DESC_COUNTER(VCPU, instruction_diagnose_10),
 174         STATS_DESC_COUNTER(VCPU, instruction_diagnose_44),
 175         STATS_DESC_COUNTER(VCPU, instruction_diagnose_9c),
 176         STATS_DESC_COUNTER(VCPU, diag_9c_ignored),
 177         STATS_DESC_COUNTER(VCPU, diag_9c_forward),
 178         STATS_DESC_COUNTER(VCPU, instruction_diagnose_258),
 179         STATS_DESC_COUNTER(VCPU, instruction_diagnose_308),
 180         STATS_DESC_COUNTER(VCPU, instruction_diagnose_500),
 181         STATS_DESC_COUNTER(VCPU, instruction_diagnose_other),
 182         STATS_DESC_COUNTER(VCPU, pfault_sync)
 183 };
 184
 185 const struct kvm_stats_header kvm_vcpu_stats_header = {
 186         .name_size = KVM_STATS_NAME_SIZE,
 187         .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
 188         .id_offset = sizeof(struct kvm_stats_header),
 189         .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
 190         .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
 191                        sizeof(kvm_vcpu_stats_desc),
 192 };
 193
 194 /* allow nested virtualization in KVM (if enabled by user space) */
 195 static int nested;
 196 module_param(nested, int, S_IRUGO);
 197 MODULE_PARM_DESC(nested, "Nested virtualization support");
 198
 199 /* allow 1m huge page guest backing, if !nested */
 200 static int hpage;
 201 module_param(hpage, int, 0444);
 202 MODULE_PARM_DESC(hpage, "1m huge page backing support");
 203
 204 /* maximum percentage of steal time for polling.  >100 is treated like 100 */
 205 static u8 halt_poll_max_steal = 10;
 206 module_param(halt_poll_max_steal, byte, 0644);
 207 MODULE_PARM_DESC(halt_poll_max_steal, "Maximum percentage of steal time to allow polling");
 208
 209 /* if set to true, the GISA will be initialized and used if available */
 210 static bool use_gisa  = true;
 211 module_param(use_gisa, bool, 0644);
 212 MODULE_PARM_DESC(use_gisa, "Use the GISA if the host supports it.");
 213
 214 /* maximum diag9c forwarding per second */
 215 unsigned int diag9c_forwarding_hz;
 216 module_param(diag9c_forwarding_hz, uint, 0644);
 217 MODULE_PARM_DESC(diag9c_forwarding_hz, "Maximum diag9c forwarding per second, 0 to turn off");
 218
 219 /*
 220  * allow asynchronous deinit for protected guests; enable by default since
 221  * the feature is opt-in anyway
 222  */
 223 static int async_destroy = 1;
 224 module_param(async_destroy, int, 0444);
 225 MODULE_PARM_DESC(async_destroy, "Asynchronous destroy for protected guests");
 226
 227 /*
 228  * For now we handle at most 16 double words as this is what the s390 base
 229  * kernel handles and stores in the prefix page. If we ever need to go beyond
 230  * this, this requires changes to code, but the external uapi can stay.
 231  */
 232 #define SIZE_INTERNAL 16
 233
 234 /*
 235  * Base feature mask that defines default mask for facilities. Consists of the
 236  * defines in FACILITIES_KVM and the non-hypervisor managed bits.
 237  */
 238 static unsigned long kvm_s390_fac_base[SIZE_INTERNAL] = { FACILITIES_KVM };
 239 /*
 240  * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL
 241  * and defines the facilities that can be enabled via a cpu model.
 242  */
 243 static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL] = { FACILITIES_KVM_CPUMODEL };
 244
 245 static unsigned long kvm_s390_fac_size(void)
 246 {
 247         BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_MASK_SIZE_U64);
 248         BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_LIST_SIZE_U64);
 249         BUILD_BUG_ON(SIZE_INTERNAL * sizeof(unsigned long) >
 250                 sizeof(stfle_fac_list));
 251
 252         return SIZE_INTERNAL;
 253 }
 254
 255 /* available cpu features supported by kvm */
 256 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
 257 /* available subfunctions indicated via query / "test bit" */
 258 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
 259
 260 static struct gmap_notifier gmap_notifier;
 261 static struct gmap_notifier vsie_gmap_notifier;
 262 debug_info_t *kvm_s390_dbf;
 263 debug_info_t *kvm_s390_dbf_uv;
 264
 265 /* Section: not file related */
 266 /* forward declarations */
 267 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
 268                               unsigned long end);
 269 static int sca_switch_to_extended(struct kvm *kvm);
 270
 271 static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta)
 272 {
 273         u8 delta_idx = 0;
 274
 275         /*
 276          * The TOD jumps by delta, we have to compensate this by adding
 277          * -delta to the epoch.
 278          */
 279         delta = -delta;
 280
 281         /* sign-extension - we're adding to signed values below */
 282         if ((s64)delta < 0)
 283                 delta_idx = -1;
 284
 285         scb->epoch += delta;
 286         if (scb->ecd & ECD_MEF) {
 287                 scb->epdx += delta_idx;
 288                 if (scb->epoch < delta)
 289                         scb->epdx += 1;
 290         }
 291 }
 292
 293 /*
 294  * This callback is executed during stop_machine(). All CPUs are therefore
 295  * temporarily stopped. In order not to change guest behavior, we have to
 296  * disable preemption whenever we touch the epoch of kvm and the VCPUs,
 297  * so a CPU won't be stopped while calculating with the epoch.
 298  */
 299 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
 300                           void *v)
 301 {
 302         struct kvm *kvm;
 303         struct kvm_vcpu *vcpu;
 304         unsigned long i;
 305         unsigned long long *delta = v;
 306
 307         list_for_each_entry(kvm, &vm_list, vm_list) {
 308                 kvm_for_each_vcpu(i, vcpu, kvm) {
 309                         kvm_clock_sync_scb(vcpu->arch.sie_block, *delta);
 310                         if (i == 0) {
 311                                 kvm->arch.epoch = vcpu->arch.sie_block->epoch;
 312                                 kvm->arch.epdx = vcpu->arch.sie_block->epdx;
 313                         }
 314                         if (vcpu->arch.cputm_enabled)
 315                                 vcpu->arch.cputm_start += *delta;
 316                         if (vcpu->arch.vsie_block)
 317                                 kvm_clock_sync_scb(vcpu->arch.vsie_block,
 318                                                    *delta);
 319                 }
 320         }
 321         return NOTIFY_OK;
 322 }
 323
 324 static struct notifier_block kvm_clock_notifier = {
 325         .notifier_call = kvm_clock_sync,
 326 };
 327
 328 static void allow_cpu_feat(unsigned long nr)
 329 {
 330         set_bit_inv(nr, kvm_s390_available_cpu_feat);
 331 }
 332
 333 static inline int plo_test_bit(unsigned char nr)
 334 {
 335         unsigned long function = (unsigned long)nr | 0x100;
 336         int cc;
 337
 338         asm volatile(
 339                 "       lgr     0,%[function]\n"
 340                 /* Parameter registers are ignored for "test bit" */
 341                 "       plo     0,0,0,0(0)\n"
 342                 "       ipm     %0\n"
 343                 "       srl     %0,28\n"
 344                 : "=d" (cc)
 345                 : [function] "d" (function)
 346                 : "cc", "0");
 347         return cc == 0;
 348 }
 349
 350 static __always_inline void __insn32_query(unsigned int opcode, u8 *query)
 351 {
 352         asm volatile(
 353                 "       lghi    0,0\n"
 354                 "       lgr     1,%[query]\n"
 355                 /* Parameter registers are ignored */
 356                 "       .insn   rrf,%[opc] << 16,2,4,6,0\n"
 357                 :
 358                 : [query] "d" ((unsigned long)query), [opc] "i" (opcode)
 359                 : "cc", "memory", "0", "1");
 360 }
 361
 362 #define INSN_SORTL 0xb938
 363 #define INSN_DFLTCC 0xb939
 364
 365 static void __init kvm_s390_cpu_feat_init(void)
 366 {
 367         int i;
 368
 369         for (i = 0; i < 256; ++i) {
 370                 if (plo_test_bit(i))
 371                         kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
 372         }
 373
 374         if (test_facility(28)) /* TOD-clock steering */
 375                 ptff(kvm_s390_available_subfunc.ptff,
 376                      sizeof(kvm_s390_available_subfunc.ptff),
 377                      PTFF_QAF);
 378
 379         if (test_facility(17)) { /* MSA */
 380                 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
 381                               kvm_s390_available_subfunc.kmac);
 382                 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
 383                               kvm_s390_available_subfunc.kmc);
 384                 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
 385                               kvm_s390_available_subfunc.km);
 386                 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
 387                               kvm_s390_available_subfunc.kimd);
 388                 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
 389                               kvm_s390_available_subfunc.klmd);
 390         }
 391         if (test_facility(76)) /* MSA3 */
 392                 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
 393                               kvm_s390_available_subfunc.pckmo);
 394         if (test_facility(77)) { /* MSA4 */
 395                 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
 396                               kvm_s390_available_subfunc.kmctr);
 397                 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
 398                               kvm_s390_available_subfunc.kmf);
 399                 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
 400                               kvm_s390_available_subfunc.kmo);
 401                 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
 402                               kvm_s390_available_subfunc.pcc);
 403         }
 404         if (test_facility(57)) /* MSA5 */
 405                 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
 406                               kvm_s390_available_subfunc.ppno);
 407
 408         if (test_facility(146)) /* MSA8 */
 409                 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
 410                               kvm_s390_available_subfunc.kma);
 411
 412         if (test_facility(155)) /* MSA9 */
 413                 __cpacf_query(CPACF_KDSA, (cpacf_mask_t *)
 414                               kvm_s390_available_subfunc.kdsa);
 415
 416         if (test_facility(150)) /* SORTL */
 417                 __insn32_query(INSN_SORTL, kvm_s390_available_subfunc.sortl);
 418
 419         if (test_facility(151)) /* DFLTCC */
 420                 __insn32_query(INSN_DFLTCC, kvm_s390_available_subfunc.dfltcc);
 421
 422         if (MACHINE_HAS_ESOP)
 423                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
 424         /*
 425          * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
 426          * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
 427          */
 428         if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
 429             !test_facility(3) || !nested)
 430                 return;
 431         allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
 432         if (sclp.has_64bscao)
 433                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
 434         if (sclp.has_siif)
 435                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
 436         if (sclp.has_gpere)
 437                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
 438         if (sclp.has_gsls)
 439                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
 440         if (sclp.has_ib)
 441                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
 442         if (sclp.has_cei)
 443                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
 444         if (sclp.has_ibs)
 445                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
 446         if (sclp.has_kss)
 447                 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
 448         /*
 449          * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
 450          * all skey handling functions read/set the skey from the PGSTE
 451          * instead of the real storage key.
 452          *
 453          * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
 454          * pages being detected as preserved although they are resident.
 455          *
 456          * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
 457          * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
 458          *
 459          * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
 460          * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
 461          * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
 462          *
 463          * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
 464          * cannot easily shadow the SCA because of the ipte lock.
 465          */
 466 }
 467
 468 static int __init __kvm_s390_init(void)
 469 {
 470         int rc = -ENOMEM;
 471
 472         kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
 473         if (!kvm_s390_dbf)
 474                 return -ENOMEM;
 475
 476         kvm_s390_dbf_uv = debug_register("kvm-uv", 32, 1, 7 * sizeof(long));
 477         if (!kvm_s390_dbf_uv)
 478                 goto err_kvm_uv;
 479
 480         if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view) ||
 481             debug_register_view(kvm_s390_dbf_uv, &debug_sprintf_view))
 482                 goto err_debug_view;
 483
 484         kvm_s390_cpu_feat_init();
 485
 486         /* Register floating interrupt controller interface. */
 487         rc = kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
 488         if (rc) {
 489                 pr_err("A FLIC registration call failed with rc=%d\n", rc);
 490                 goto err_flic;
 491         }
 492
 493         if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
 494                 rc = kvm_s390_pci_init();
 495                 if (rc) {
 496                         pr_err("Unable to allocate AIFT for PCI\n");
 497                         goto err_pci;
 498                 }
 499         }
 500
 501         rc = kvm_s390_gib_init(GAL_ISC);
 502         if (rc)
 503                 goto err_gib;
 504
 505         gmap_notifier.notifier_call = kvm_gmap_notifier;
 506         gmap_register_pte_notifier(&gmap_notifier);
 507         vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
 508         gmap_register_pte_notifier(&vsie_gmap_notifier);
 509         atomic_notifier_chain_register(&s390_epoch_delta_notifier,
 510                                        &kvm_clock_notifier);
 511
 512         return 0;
 513
 514 err_gib:
 515         if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
 516                 kvm_s390_pci_exit();
 517 err_pci:
 518 err_flic:
 519 err_debug_view:
 520         debug_unregister(kvm_s390_dbf_uv);
 521 err_kvm_uv:
 522         debug_unregister(kvm_s390_dbf);
 523         return rc;
 524 }
 525
 526 static void __kvm_s390_exit(void)
 527 {
 528         gmap_unregister_pte_notifier(&gmap_notifier);
 529         gmap_unregister_pte_notifier(&vsie_gmap_notifier);
 530         atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
 531                                          &kvm_clock_notifier);
 532
 533         kvm_s390_gib_destroy();
 534         if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
 535                 kvm_s390_pci_exit();
 536         debug_unregister(kvm_s390_dbf);
 537         debug_unregister(kvm_s390_dbf_uv);
 538 }
 539
 540 /* Section: device related */
 541 long kvm_arch_dev_ioctl(struct file *filp,
 542                         unsigned int ioctl, unsigned long arg)
 543 {
 544         if (ioctl == KVM_S390_ENABLE_SIE)
 545                 return s390_enable_sie();
 546         return -EINVAL;
 547 }
 548
 549 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 550 {
 551         int r;
 552
 553         switch (ext) {
 554         case KVM_CAP_S390_PSW:
 555         case KVM_CAP_S390_GMAP:
 556         case KVM_CAP_SYNC_MMU:
 557 #ifdef CONFIG_KVM_S390_UCONTROL
 558         case KVM_CAP_S390_UCONTROL:
 559 #endif
 560         case KVM_CAP_ASYNC_PF:
 561         case KVM_CAP_SYNC_REGS:
 562         case KVM_CAP_ONE_REG:
 563         case KVM_CAP_ENABLE_CAP:
 564         case KVM_CAP_S390_CSS_SUPPORT:
 565         case KVM_CAP_IOEVENTFD:
 566         case KVM_CAP_DEVICE_CTRL:
 567         case KVM_CAP_S390_IRQCHIP:
 568         case KVM_CAP_VM_ATTRIBUTES:
 569         case KVM_CAP_MP_STATE:
 570         case KVM_CAP_IMMEDIATE_EXIT:
 571         case KVM_CAP_S390_INJECT_IRQ:
 572         case KVM_CAP_S390_USER_SIGP:
 573         case KVM_CAP_S390_USER_STSI:
 574         case KVM_CAP_S390_SKEYS:
 575         case KVM_CAP_S390_IRQ_STATE:
 576         case KVM_CAP_S390_USER_INSTR0:
 577         case KVM_CAP_S390_CMMA_MIGRATION:
 578         case KVM_CAP_S390_AIS:
 579         case KVM_CAP_S390_AIS_MIGRATION:
 580         case KVM_CAP_S390_VCPU_RESETS:
 581         case KVM_CAP_SET_GUEST_DEBUG:
 582         case KVM_CAP_S390_DIAG318:
 583         case KVM_CAP_IRQFD_RESAMPLE:
 584                 r = 1;
 585                 break;
 586         case KVM_CAP_SET_GUEST_DEBUG2:
 587                 r = KVM_GUESTDBG_VALID_MASK;
 588                 break;
 589         case KVM_CAP_S390_HPAGE_1M:
 590                 r = 0;
 591                 if (hpage && !kvm_is_ucontrol(kvm))
 592                         r = 1;
 593                 break;
 594         case KVM_CAP_S390_MEM_OP:
 595                 r = MEM_OP_MAX_SIZE;
 596                 break;
 597         case KVM_CAP_S390_MEM_OP_EXTENSION:
 598                 /*
 599                  * Flag bits indicating which extensions are supported.
 600                  * If r > 0, the base extension must also be supported/indicated,
 601                  * in order to maintain backwards compatibility.
 602                  */
 603                 r = KVM_S390_MEMOP_EXTENSION_CAP_BASE |
 604                     KVM_S390_MEMOP_EXTENSION_CAP_CMPXCHG;
 605                 break;
 606         case KVM_CAP_NR_VCPUS:
 607         case KVM_CAP_MAX_VCPUS:
 608         case KVM_CAP_MAX_VCPU_ID:
 609                 r = KVM_S390_BSCA_CPU_SLOTS;
 610                 if (!kvm_s390_use_sca_entries())
 611                         r = KVM_MAX_VCPUS;
 612                 else if (sclp.has_esca && sclp.has_64bscao)
 613                         r = KVM_S390_ESCA_CPU_SLOTS;
 614                 if (ext == KVM_CAP_NR_VCPUS)
 615                         r = min_t(unsigned int, num_online_cpus(), r);
 616                 break;
 617         case KVM_CAP_S390_COW:
 618                 r = MACHINE_HAS_ESOP;
 619                 break;
 620         case KVM_CAP_S390_VECTOR_REGISTERS:
 621                 r = MACHINE_HAS_VX;
 622                 break;
 623         case KVM_CAP_S390_RI:
 624                 r = test_facility(64);
 625                 break;
 626         case KVM_CAP_S390_GS:
 627                 r = test_facility(133);
 628                 break;
 629         case KVM_CAP_S390_BPB:
 630                 r = test_facility(82);
 631                 break;
 632         case KVM_CAP_S390_PROTECTED_ASYNC_DISABLE:
 633                 r = async_destroy && is_prot_virt_host();
 634                 break;
 635         case KVM_CAP_S390_PROTECTED:
 636                 r = is_prot_virt_host();
 637                 break;
 638         case KVM_CAP_S390_PROTECTED_DUMP: {
 639                 u64 pv_cmds_dump[] = {
 640                         BIT_UVC_CMD_DUMP_INIT,
 641                         BIT_UVC_CMD_DUMP_CONFIG_STOR_STATE,
 642                         BIT_UVC_CMD_DUMP_CPU,
 643                         BIT_UVC_CMD_DUMP_COMPLETE,
 644                 };
 645                 int i;
 646
 647                 r = is_prot_virt_host();
 648
 649                 for (i = 0; i < ARRAY_SIZE(pv_cmds_dump); i++) {
 650                         if (!test_bit_inv(pv_cmds_dump[i],
 651                                           (unsigned long *)&uv_info.inst_calls_list)) {
 652                                 r = 0;
 653                                 break;
 654                         }
 655                 }
 656                 break;
 657         }
 658         case KVM_CAP_S390_ZPCI_OP:
 659                 r = kvm_s390_pci_interp_allowed();
 660                 break;
 661         case KVM_CAP_S390_CPU_TOPOLOGY:
 662                 r = test_facility(11);
 663                 break;
 664         default:
 665                 r = 0;
 666         }
 667         return r;
 668 }
 669
 670 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
 671 {
 672         int i;
 673         gfn_t cur_gfn, last_gfn;
 674         unsigned long gaddr, vmaddr;
 675         struct gmap *gmap = kvm->arch.gmap;
 676         DECLARE_BITMAP(bitmap, _PAGE_ENTRIES);
 677
 678         /* Loop over all guest segments */
 679         cur_gfn = memslot->base_gfn;
 680         last_gfn = memslot->base_gfn + memslot->npages;
 681         for (; cur_gfn <= last_gfn; cur_gfn += _PAGE_ENTRIES) {
 682                 gaddr = gfn_to_gpa(cur_gfn);
 683                 vmaddr = gfn_to_hva_memslot(memslot, cur_gfn);
 684                 if (kvm_is_error_hva(vmaddr))
 685                         continue;
 686
 687                 bitmap_zero(bitmap, _PAGE_ENTRIES);
 688                 gmap_sync_dirty_log_pmd(gmap, bitmap, gaddr, vmaddr);
 689                 for (i = 0; i < _PAGE_ENTRIES; i++) {
 690                         if (test_bit(i, bitmap))
 691                                 mark_page_dirty(kvm, cur_gfn + i);
 692                 }
 693
 694                 if (fatal_signal_pending(current))
 695                         return;
 696                 cond_resched();
 697         }
 698 }
 699
 700 /* Section: vm related */
 701 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
 702
 703 /*
 704  * Get (and clear) the dirty memory log for a memory slot.
 705  */
 706 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 707                                struct kvm_dirty_log *log)
 708 {
 709         int r;
 710         unsigned long n;
 711         struct kvm_memory_slot *memslot;
 712         int is_dirty;
 713
 714         if (kvm_is_ucontrol(kvm))
 715                 return -EINVAL;
 716
 717         mutex_lock(&kvm->slots_lock);
 718
 719         r = -EINVAL;
 720         if (log->slot >= KVM_USER_MEM_SLOTS)
 721                 goto out;
 722
 723         r = kvm_get_dirty_log(kvm, log, &is_dirty, &memslot);
 724         if (r)
 725                 goto out;
 726
 727         /* Clear the dirty log */
 728         if (is_dirty) {
 729                 n = kvm_dirty_bitmap_bytes(memslot);
 730                 memset(memslot->dirty_bitmap, 0, n);
 731         }
 732         r = 0;
 733 out:
 734         mutex_unlock(&kvm->slots_lock);
 735         return r;
 736 }
 737
 738 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
 739 {
 740         unsigned long i;
 741         struct kvm_vcpu *vcpu;
 742
 743         kvm_for_each_vcpu(i, vcpu, kvm) {
 744                 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
 745         }
 746 }
 747
 748 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
 749 {
 750         int r;
 751
 752         if (cap->flags)
 753                 return -EINVAL;
 754
 755         switch (cap->cap) {
 756         case KVM_CAP_S390_IRQCHIP:
 757                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
 758                 kvm->arch.use_irqchip = 1;
 759                 r = 0;
 760                 break;
 761         case KVM_CAP_S390_USER_SIGP:
 762                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
 763                 kvm->arch.user_sigp = 1;
 764                 r = 0;
 765                 break;
 766         case KVM_CAP_S390_VECTOR_REGISTERS:
 767                 mutex_lock(&kvm->lock);
 768                 if (kvm->created_vcpus) {
 769                         r = -EBUSY;
 770                 } else if (MACHINE_HAS_VX) {
 771                         set_kvm_facility(kvm->arch.model.fac_mask, 129);
 772                         set_kvm_facility(kvm->arch.model.fac_list, 129);
 773                         if (test_facility(134)) {
 774                                 set_kvm_facility(kvm->arch.model.fac_mask, 134);
 775                                 set_kvm_facility(kvm->arch.model.fac_list, 134);
 776                         }
 777                         if (test_facility(135)) {
 778                                 set_kvm_facility(kvm->arch.model.fac_mask, 135);
 779                                 set_kvm_facility(kvm->arch.model.fac_list, 135);
 780                         }
 781                         if (test_facility(148)) {
 782                                 set_kvm_facility(kvm->arch.model.fac_mask, 148);
 783                                 set_kvm_facility(kvm->arch.model.fac_list, 148);
 784                         }
 785                         if (test_facility(152)) {
 786                                 set_kvm_facility(kvm->arch.model.fac_mask, 152);
 787                                 set_kvm_facility(kvm->arch.model.fac_list, 152);
 788                         }
 789                         if (test_facility(192)) {
 790                                 set_kvm_facility(kvm->arch.model.fac_mask, 192);
 791                                 set_kvm_facility(kvm->arch.model.fac_list, 192);
 792                         }
 793                         r = 0;
 794                 } else
 795                         r = -EINVAL;
 796                 mutex_unlock(&kvm->lock);
 797                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
 798                          r ? "(not available)" : "(success)");
 799                 break;
 800         case KVM_CAP_S390_RI:
 801                 r = -EINVAL;
 802                 mutex_lock(&kvm->lock);
 803                 if (kvm->created_vcpus) {
 804                         r = -EBUSY;
 805                 } else if (test_facility(64)) {
 806                         set_kvm_facility(kvm->arch.model.fac_mask, 64);
 807                         set_kvm_facility(kvm->arch.model.fac_list, 64);
 808                         r = 0;
 809                 }
 810                 mutex_unlock(&kvm->lock);
 811                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
 812                          r ? "(not available)" : "(success)");
 813                 break;
 814         case KVM_CAP_S390_AIS:
 815                 mutex_lock(&kvm->lock);
 816                 if (kvm->created_vcpus) {
 817                         r = -EBUSY;
 818                 } else {
 819                         set_kvm_facility(kvm->arch.model.fac_mask, 72);
 820                         set_kvm_facility(kvm->arch.model.fac_list, 72);
 821                         r = 0;
 822                 }
 823                 mutex_unlock(&kvm->lock);
 824                 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
 825                          r ? "(not available)" : "(success)");
 826                 break;
 827         case KVM_CAP_S390_GS:
 828                 r = -EINVAL;
 829                 mutex_lock(&kvm->lock);
 830                 if (kvm->created_vcpus) {
 831                         r = -EBUSY;
 832                 } else if (test_facility(133)) {
 833                         set_kvm_facility(kvm->arch.model.fac_mask, 133);
 834                         set_kvm_facility(kvm->arch.model.fac_list, 133);
 835                         r = 0;
 836                 }
 837                 mutex_unlock(&kvm->lock);
 838                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
 839                          r ? "(not available)" : "(success)");
 840                 break;
 841         case KVM_CAP_S390_HPAGE_1M:
 842                 mutex_lock(&kvm->lock);
 843                 if (kvm->created_vcpus)
 844                         r = -EBUSY;
 845                 else if (!hpage || kvm->arch.use_cmma || kvm_is_ucontrol(kvm))
 846                         r = -EINVAL;
 847                 else {
 848                         r = 0;
 849                         mmap_write_lock(kvm->mm);
 850                         kvm->mm->context.allow_gmap_hpage_1m = 1;
 851                         mmap_write_unlock(kvm->mm);
 852                         /*
 853                          * We might have to create fake 4k page
 854                          * tables. To avoid that the hardware works on
 855                          * stale PGSTEs, we emulate these instructions.
 856                          */
 857                         kvm->arch.use_skf = 0;
 858                         kvm->arch.use_pfmfi = 0;
 859                 }
 860                 mutex_unlock(&kvm->lock);
 861                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_HPAGE %s",
 862                          r ? "(not available)" : "(success)");
 863                 break;
 864         case KVM_CAP_S390_USER_STSI:
 865                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
 866                 kvm->arch.user_stsi = 1;
 867                 r = 0;
 868                 break;
 869         case KVM_CAP_S390_USER_INSTR0:
 870                 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
 871                 kvm->arch.user_instr0 = 1;
 872                 icpt_operexc_on_all_vcpus(kvm);
 873                 r = 0;
 874                 break;
 875         case KVM_CAP_S390_CPU_TOPOLOGY:
 876                 r = -EINVAL;
 877                 mutex_lock(&kvm->lock);
 878                 if (kvm->created_vcpus) {
 879                         r = -EBUSY;
 880                 } else if (test_facility(11)) {
 881                         set_kvm_facility(kvm->arch.model.fac_mask, 11);
 882                         set_kvm_facility(kvm->arch.model.fac_list, 11);
 883                         r = 0;
 884                 }
 885                 mutex_unlock(&kvm->lock);
 886                 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_CPU_TOPOLOGY %s",
 887                          r ? "(not available)" : "(success)");
 888                 break;
 889         default:
 890                 r = -EINVAL;
 891                 break;
 892         }
 893         return r;
 894 }
 895
 896 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
 897 {
 898         int ret;
 899
 900         switch (attr->attr) {
 901         case KVM_S390_VM_MEM_LIMIT_SIZE:
 902                 ret = 0;
 903                 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
 904                          kvm->arch.mem_limit);
 905                 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
 906                         ret = -EFAULT;
 907                 break;
 908         default:
 909                 ret = -ENXIO;
 910                 break;
 911         }
 912         return ret;
 913 }
 914
 915 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
 916 {
 917         int ret;
 918         unsigned int idx;
 919         switch (attr->attr) {
 920         case KVM_S390_VM_MEM_ENABLE_CMMA:
 921                 ret = -ENXIO;
 922                 if (!sclp.has_cmma)
 923                         break;
 924
 925                 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
 926                 mutex_lock(&kvm->lock);
 927                 if (kvm->created_vcpus)
 928                         ret = -EBUSY;
 929                 else if (kvm->mm->context.allow_gmap_hpage_1m)
 930                         ret = -EINVAL;
 931                 else {
 932                         kvm->arch.use_cmma = 1;
 933                         /* Not compatible with cmma. */
 934                         kvm->arch.use_pfmfi = 0;
 935                         ret = 0;
 936                 }
 937                 mutex_unlock(&kvm->lock);
 938                 break;
 939         case KVM_S390_VM_MEM_CLR_CMMA:
 940                 ret = -ENXIO;
 941                 if (!sclp.has_cmma)
 942                         break;
 943                 ret = -EINVAL;
 944                 if (!kvm->arch.use_cmma)
 945                         break;
 946
 947                 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
 948                 mutex_lock(&kvm->lock);
 949                 idx = srcu_read_lock(&kvm->srcu);
 950                 s390_reset_cmma(kvm->arch.gmap->mm);
 951                 srcu_read_unlock(&kvm->srcu, idx);
 952                 mutex_unlock(&kvm->lock);
 953                 ret = 0;
 954                 break;
 955         case KVM_S390_VM_MEM_LIMIT_SIZE: {
 956                 unsigned long new_limit;
 957
 958                 if (kvm_is_ucontrol(kvm))
 959                         return -EINVAL;
 960
 961                 if (get_user(new_limit, (u64 __user *)attr->addr))
 962                         return -EFAULT;
 963
 964                 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
 965                     new_limit > kvm->arch.mem_limit)
 966                         return -E2BIG;
 967
 968                 if (!new_limit)
 969                         return -EINVAL;
 970
 971                 /* gmap_create takes last usable address */
 972                 if (new_limit != KVM_S390_NO_MEM_LIMIT)
 973                         new_limit -= 1;
 974
 975                 ret = -EBUSY;
 976                 mutex_lock(&kvm->lock);
 977                 if (!kvm->created_vcpus) {
 978                         /* gmap_create will round the limit up */
 979                         struct gmap *new = gmap_create(current->mm, new_limit);
 980
 981                         if (!new) {
 982                                 ret = -ENOMEM;
 983                         } else {
 984                                 gmap_remove(kvm->arch.gmap);
 985                                 new->private = kvm;
 986                                 kvm->arch.gmap = new;
 987                                 ret = 0;
 988                         }
 989                 }
 990                 mutex_unlock(&kvm->lock);
 991                 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
 992                 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
 993                          (void *) kvm->arch.gmap->asce);
 994                 break;
 995         }
 996         default:
 997                 ret = -ENXIO;
 998                 break;
 999         }
1000         return ret;
1001 }
1002
1003 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
1004
1005 void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm)
1006 {
1007         struct kvm_vcpu *vcpu;
1008         unsigned long i;
1009
1010         kvm_s390_vcpu_block_all(kvm);
1011
1012         kvm_for_each_vcpu(i, vcpu, kvm) {
1013                 kvm_s390_vcpu_crypto_setup(vcpu);
1014                 /* recreate the shadow crycb by leaving the VSIE handler */
1015                 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1016         }
1017
1018         kvm_s390_vcpu_unblock_all(kvm);
1019 }
1020
1021 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
1022 {
1023         mutex_lock(&kvm->lock);
1024         switch (attr->attr) {
1025         case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1026                 if (!test_kvm_facility(kvm, 76)) {
1027                         mutex_unlock(&kvm->lock);
1028                         return -EINVAL;
1029                 }
1030                 get_random_bytes(
1031                         kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1032                         sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1033                 kvm->arch.crypto.aes_kw = 1;
1034                 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
1035                 break;
1036         case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1037                 if (!test_kvm_facility(kvm, 76)) {
1038                         mutex_unlock(&kvm->lock);
1039                         return -EINVAL;
1040                 }
1041                 get_random_bytes(
1042                         kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1043                         sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1044                 kvm->arch.crypto.dea_kw = 1;
1045                 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
1046                 break;
1047         case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1048                 if (!test_kvm_facility(kvm, 76)) {
1049                         mutex_unlock(&kvm->lock);
1050                         return -EINVAL;
1051                 }
1052                 kvm->arch.crypto.aes_kw = 0;
1053                 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
1054                         sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1055                 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
1056                 break;
1057         case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1058                 if (!test_kvm_facility(kvm, 76)) {
1059                         mutex_unlock(&kvm->lock);
1060                         return -EINVAL;
1061                 }
1062                 kvm->arch.crypto.dea_kw = 0;
1063                 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
1064                         sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1065                 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
1066                 break;
1067         case KVM_S390_VM_CRYPTO_ENABLE_APIE:
1068                 if (!ap_instructions_available()) {
1069                         mutex_unlock(&kvm->lock);
1070                         return -EOPNOTSUPP;
1071                 }
1072                 kvm->arch.crypto.apie = 1;
1073                 break;
1074         case KVM_S390_VM_CRYPTO_DISABLE_APIE:
1075                 if (!ap_instructions_available()) {
1076                         mutex_unlock(&kvm->lock);
1077                         return -EOPNOTSUPP;
1078                 }
1079                 kvm->arch.crypto.apie = 0;
1080                 break;
1081         default:
1082                 mutex_unlock(&kvm->lock);
1083                 return -ENXIO;
1084         }
1085
1086         kvm_s390_vcpu_crypto_reset_all(kvm);
1087         mutex_unlock(&kvm->lock);
1088         return 0;
1089 }
1090
1091 static void kvm_s390_vcpu_pci_setup(struct kvm_vcpu *vcpu)
1092 {
1093         /* Only set the ECB bits after guest requests zPCI interpretation */
1094         if (!vcpu->kvm->arch.use_zpci_interp)
1095                 return;
1096
1097         vcpu->arch.sie_block->ecb2 |= ECB2_ZPCI_LSI;
1098         vcpu->arch.sie_block->ecb3 |= ECB3_AISII + ECB3_AISI;
1099 }
1100
1101 void kvm_s390_vcpu_pci_enable_interp(struct kvm *kvm)
1102 {
1103         struct kvm_vcpu *vcpu;
1104         unsigned long i;
1105
1106         lockdep_assert_held(&kvm->lock);
1107
1108         if (!kvm_s390_pci_interp_allowed())
1109                 return;
1110
1111         /*
1112          * If host is configured for PCI and the necessary facilities are
1113          * available, turn on interpretation for the life of this guest
1114          */
1115         kvm->arch.use_zpci_interp = 1;
1116
1117         kvm_s390_vcpu_block_all(kvm);
1118
1119         kvm_for_each_vcpu(i, vcpu, kvm) {
1120                 kvm_s390_vcpu_pci_setup(vcpu);
1121                 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1122         }
1123
1124         kvm_s390_vcpu_unblock_all(kvm);
1125 }
1126
1127 static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req)
1128 {
1129         unsigned long cx;
1130         struct kvm_vcpu *vcpu;
1131
1132         kvm_for_each_vcpu(cx, vcpu, kvm)
1133                 kvm_s390_sync_request(req, vcpu);
1134 }
1135
1136 /*
1137  * Must be called with kvm->srcu held to avoid races on memslots, and with
1138  * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration.
1139  */
1140 static int kvm_s390_vm_start_migration(struct kvm *kvm)
1141 {
1142         struct kvm_memory_slot *ms;
1143         struct kvm_memslots *slots;
1144         unsigned long ram_pages = 0;
1145         int bkt;
1146
1147         /* migration mode already enabled */
1148         if (kvm->arch.migration_mode)
1149                 return 0;
1150         slots = kvm_memslots(kvm);
1151         if (!slots || kvm_memslots_empty(slots))
1152                 return -EINVAL;
1153
1154         if (!kvm->arch.use_cmma) {
1155                 kvm->arch.migration_mode = 1;
1156                 return 0;
1157         }
1158         /* mark all the pages in active slots as dirty */
1159         kvm_for_each_memslot(ms, bkt, slots) {
1160                 if (!ms->dirty_bitmap)
1161                         return -EINVAL;
1162                 /*
1163                  * The second half of the bitmap is only used on x86,
1164                  * and would be wasted otherwise, so we put it to good
1165                  * use here to keep track of the state of the storage
1166                  * attributes.
1167                  */
1168                 memset(kvm_second_dirty_bitmap(ms), 0xff, kvm_dirty_bitmap_bytes(ms));
1169                 ram_pages += ms->npages;
1170         }
1171         atomic64_set(&kvm->arch.cmma_dirty_pages, ram_pages);
1172         kvm->arch.migration_mode = 1;
1173         kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION);
1174         return 0;
1175 }
1176
1177 /*
1178  * Must be called with kvm->slots_lock to avoid races with ourselves and
1179  * kvm_s390_vm_start_migration.
1180  */
1181 static int kvm_s390_vm_stop_migration(struct kvm *kvm)
1182 {
1183         /* migration mode already disabled */
1184         if (!kvm->arch.migration_mode)
1185                 return 0;
1186         kvm->arch.migration_mode = 0;
1187         if (kvm->arch.use_cmma)
1188                 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION);
1189         return 0;
1190 }
1191
1192 static int kvm_s390_vm_set_migration(struct kvm *kvm,
1193                                      struct kvm_device_attr *attr)
1194 {
1195         int res = -ENXIO;
1196
1197         mutex_lock(&kvm->slots_lock);
1198         switch (attr->attr) {
1199         case KVM_S390_VM_MIGRATION_START:
1200                 res = kvm_s390_vm_start_migration(kvm);
1201                 break;
1202         case KVM_S390_VM_MIGRATION_STOP:
1203                 res = kvm_s390_vm_stop_migration(kvm);
1204                 break;
1205         default:
1206                 break;
1207         }
1208         mutex_unlock(&kvm->slots_lock);
1209
1210         return res;
1211 }
1212
1213 static int kvm_s390_vm_get_migration(struct kvm *kvm,
1214                                      struct kvm_device_attr *attr)
1215 {
1216         u64 mig = kvm->arch.migration_mode;
1217
1218         if (attr->attr != KVM_S390_VM_MIGRATION_STATUS)
1219                 return -ENXIO;
1220
1221         if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig)))
1222                 return -EFAULT;
1223         return 0;
1224 }
1225
1226 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod);
1227
1228 static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1229 {
1230         struct kvm_s390_vm_tod_clock gtod;
1231
1232         if (copy_from_user(&gtod, (void __user *)attr->addr, sizeof(gtod)))
1233                 return -EFAULT;
1234
1235         if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx)
1236                 return -EINVAL;
1237         __kvm_s390_set_tod_clock(kvm, &gtod);
1238
1239         VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx",
1240                 gtod.epoch_idx, gtod.tod);
1241
1242         return 0;
1243 }
1244
1245 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1246 {
1247         u8 gtod_high;
1248
1249         if (copy_from_user(&gtod_high, (void __user *)attr->addr,
1250                                            sizeof(gtod_high)))
1251                 return -EFAULT;
1252
1253         if (gtod_high != 0)
1254                 return -EINVAL;
1255         VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
1256
1257         return 0;
1258 }
1259
1260 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1261 {
1262         struct kvm_s390_vm_tod_clock gtod = { 0 };
1263
1264         if (copy_from_user(&gtod.tod, (void __user *)attr->addr,
1265                            sizeof(gtod.tod)))
1266                 return -EFAULT;
1267
1268         __kvm_s390_set_tod_clock(kvm, &gtod);
1269         VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod);
1270         return 0;
1271 }
1272
1273 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1274 {
1275         int ret;
1276
1277         if (attr->flags)
1278                 return -EINVAL;
1279
1280         mutex_lock(&kvm->lock);
1281         /*
1282          * For protected guests, the TOD is managed by the ultravisor, so trying
1283          * to change it will never bring the expected results.
1284          */
1285         if (kvm_s390_pv_is_protected(kvm)) {
1286                 ret = -EOPNOTSUPP;
1287                 goto out_unlock;
1288         }
1289
1290         switch (attr->attr) {
1291         case KVM_S390_VM_TOD_EXT:
1292                 ret = kvm_s390_set_tod_ext(kvm, attr);
1293                 break;
1294         case KVM_S390_VM_TOD_HIGH:
1295                 ret = kvm_s390_set_tod_high(kvm, attr);
1296                 break;
1297         case KVM_S390_VM_TOD_LOW:
1298                 ret = kvm_s390_set_tod_low(kvm, attr);
1299                 break;
1300         default:
1301                 ret = -ENXIO;
1302                 break;
1303         }
1304
1305 out_unlock:
1306         mutex_unlock(&kvm->lock);
1307         return ret;
1308 }
1309
1310 static void kvm_s390_get_tod_clock(struct kvm *kvm,
1311                                    struct kvm_s390_vm_tod_clock *gtod)
1312 {
1313         union tod_clock clk;
1314
1315         preempt_disable();
1316
1317         store_tod_clock_ext(&clk);
1318
1319         gtod->tod = clk.tod + kvm->arch.epoch;
1320         gtod->epoch_idx = 0;
1321         if (test_kvm_facility(kvm, 139)) {
1322                 gtod->epoch_idx = clk.ei + kvm->arch.epdx;
1323                 if (gtod->tod < clk.tod)
1324                         gtod->epoch_idx += 1;
1325         }
1326
1327         preempt_enable();
1328 }
1329
1330 static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1331 {
1332         struct kvm_s390_vm_tod_clock gtod;
1333
1334         memset(&gtod, 0, sizeof(gtod));
1335         kvm_s390_get_tod_clock(kvm, &gtod);
1336         if (copy_to_user((void __user *)attr->addr, &gtod, sizeof(gtod)))
1337                 return -EFAULT;
1338
1339         VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx",
1340                 gtod.epoch_idx, gtod.tod);
1341         return 0;
1342 }
1343
1344 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1345 {
1346         u8 gtod_high = 0;
1347
1348         if (copy_to_user((void __user *)attr->addr, &gtod_high,
1349                                          sizeof(gtod_high)))
1350                 return -EFAULT;
1351         VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
1352
1353         return 0;
1354 }
1355
1356 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1357 {
1358         u64 gtod;
1359
1360         gtod = kvm_s390_get_tod_clock_fast(kvm);
1361         if (copy_to_user((void __user *)attr->addr, &gtod, sizeof(gtod)))
1362                 return -EFAULT;
1363         VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
1364
1365         return 0;
1366 }
1367
1368 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1369 {
1370         int ret;
1371
1372         if (attr->flags)
1373                 return -EINVAL;
1374
1375         switch (attr->attr) {
1376         case KVM_S390_VM_TOD_EXT:
1377                 ret = kvm_s390_get_tod_ext(kvm, attr);
1378                 break;
1379         case KVM_S390_VM_TOD_HIGH:
1380                 ret = kvm_s390_get_tod_high(kvm, attr);
1381                 break;
1382         case KVM_S390_VM_TOD_LOW:
1383                 ret = kvm_s390_get_tod_low(kvm, attr);
1384                 break;
1385         default:
1386                 ret = -ENXIO;
1387                 break;
1388         }
1389         return ret;
1390 }
1391
1392 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1393 {
1394         struct kvm_s390_vm_cpu_processor *proc;
1395         u16 lowest_ibc, unblocked_ibc;
1396         int ret = 0;
1397
1398         mutex_lock(&kvm->lock);
1399         if (kvm->created_vcpus) {
1400                 ret = -EBUSY;
1401                 goto out;
1402         }
1403         proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1404         if (!proc) {
1405                 ret = -ENOMEM;
1406                 goto out;
1407         }
1408         if (!copy_from_user(proc, (void __user *)attr->addr,
1409                             sizeof(*proc))) {
1410                 kvm->arch.model.cpuid = proc->cpuid;
1411                 lowest_ibc = sclp.ibc >> 16 & 0xfff;
1412                 unblocked_ibc = sclp.ibc & 0xfff;
1413                 if (lowest_ibc && proc->ibc) {
1414                         if (proc->ibc > unblocked_ibc)
1415                                 kvm->arch.model.ibc = unblocked_ibc;
1416                         else if (proc->ibc < lowest_ibc)
1417                                 kvm->arch.model.ibc = lowest_ibc;
1418                         else
1419                                 kvm->arch.model.ibc = proc->ibc;
1420                 }
1421                 memcpy(kvm->arch.model.fac_list, proc->fac_list,
1422                        S390_ARCH_FAC_LIST_SIZE_BYTE);
1423                 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1424                          kvm->arch.model.ibc,
1425                          kvm->arch.model.cpuid);
1426                 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1427                          kvm->arch.model.fac_list[0],
1428                          kvm->arch.model.fac_list[1],
1429                          kvm->arch.model.fac_list[2]);
1430         } else
1431                 ret = -EFAULT;
1432         kfree(proc);
1433 out:
1434         mutex_unlock(&kvm->lock);
1435         return ret;
1436 }
1437
1438 static int kvm_s390_set_processor_feat(struct kvm *kvm,
1439                                        struct kvm_device_attr *attr)
1440 {
1441         struct kvm_s390_vm_cpu_feat data;
1442
1443         if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
1444                 return -EFAULT;
1445         if (!bitmap_subset((unsigned long *) data.feat,
1446                            kvm_s390_available_cpu_feat,
1447                            KVM_S390_VM_CPU_FEAT_NR_BITS))
1448                 return -EINVAL;
1449
1450         mutex_lock(&kvm->lock);
1451         if (kvm->created_vcpus) {
1452                 mutex_unlock(&kvm->lock);
1453                 return -EBUSY;
1454         }
1455         bitmap_from_arr64(kvm->arch.cpu_feat, data.feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1456         mutex_unlock(&kvm->lock);
1457         VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1458                          data.feat[0],
1459                          data.feat[1],
1460                          data.feat[2]);
1461         return 0;
1462 }
1463
1464 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
1465                                           struct kvm_device_attr *attr)
1466 {
1467         mutex_lock(&kvm->lock);
1468         if (kvm->created_vcpus) {
1469                 mutex_unlock(&kvm->lock);
1470                 return -EBUSY;
1471         }
1472
1473         if (copy_from_user(&kvm->arch.model.subfuncs, (void __user *)attr->addr,
1474                            sizeof(struct kvm_s390_vm_cpu_subfunc))) {
1475                 mutex_unlock(&kvm->lock);
1476                 return -EFAULT;
1477         }
1478         mutex_unlock(&kvm->lock);
1479
1480         VM_EVENT(kvm, 3, "SET: guest PLO    subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1481                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1482                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1483                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1484                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1485         VM_EVENT(kvm, 3, "SET: guest PTFF   subfunc 0x%16.16lx.%16.16lx",
1486                  ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1487                  ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1488         VM_EVENT(kvm, 3, "SET: guest KMAC   subfunc 0x%16.16lx.%16.16lx",
1489                  ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1490                  ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1491         VM_EVENT(kvm, 3, "SET: guest KMC    subfunc 0x%16.16lx.%16.16lx",
1492                  ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1493                  ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1494         VM_EVENT(kvm, 3, "SET: guest KM     subfunc 0x%16.16lx.%16.16lx",
1495                  ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1496                  ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1497         VM_EVENT(kvm, 3, "SET: guest KIMD   subfunc 0x%16.16lx.%16.16lx",
1498                  ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1499                  ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1500         VM_EVENT(kvm, 3, "SET: guest KLMD   subfunc 0x%16.16lx.%16.16lx",
1501                  ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1502                  ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1503         VM_EVENT(kvm, 3, "SET: guest PCKMO  subfunc 0x%16.16lx.%16.16lx",
1504                  ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1505                  ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1506         VM_EVENT(kvm, 3, "SET: guest KMCTR  subfunc 0x%16.16lx.%16.16lx",
1507                  ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1508                  ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1509         VM_EVENT(kvm, 3, "SET: guest KMF    subfunc 0x%16.16lx.%16.16lx",
1510                  ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1511                  ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1512         VM_EVENT(kvm, 3, "SET: guest KMO    subfunc 0x%16.16lx.%16.16lx",
1513                  ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1514                  ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1515         VM_EVENT(kvm, 3, "SET: guest PCC    subfunc 0x%16.16lx.%16.16lx",
1516                  ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1517                  ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1518         VM_EVENT(kvm, 3, "SET: guest PPNO   subfunc 0x%16.16lx.%16.16lx",
1519                  ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1520                  ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1521         VM_EVENT(kvm, 3, "SET: guest KMA    subfunc 0x%16.16lx.%16.16lx",
1522                  ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1523                  ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1524         VM_EVENT(kvm, 3, "SET: guest KDSA   subfunc 0x%16.16lx.%16.16lx",
1525                  ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1526                  ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1527         VM_EVENT(kvm, 3, "SET: guest SORTL  subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1528                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1529                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1530                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1531                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1532         VM_EVENT(kvm, 3, "SET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1533                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1534                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1535                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1536                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1537
1538         return 0;
1539 }
1540
1541 #define KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK      \
1542 (                                               \
1543         ((struct kvm_s390_vm_cpu_uv_feat){      \
1544                 .ap = 1,                        \
1545                 .ap_intr = 1,                   \
1546         })                                      \
1547         .feat                                   \
1548 )
1549
1550 static int kvm_s390_set_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1551 {
1552         struct kvm_s390_vm_cpu_uv_feat __user *ptr = (void __user *)attr->addr;
1553         unsigned long data, filter;
1554
1555         filter = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK;
1556         if (get_user(data, &ptr->feat))
1557                 return -EFAULT;
1558         if (!bitmap_subset(&data, &filter, KVM_S390_VM_CPU_UV_FEAT_NR_BITS))
1559                 return -EINVAL;
1560
1561         mutex_lock(&kvm->lock);
1562         if (kvm->created_vcpus) {
1563                 mutex_unlock(&kvm->lock);
1564                 return -EBUSY;
1565         }
1566         kvm->arch.model.uv_feat_guest.feat = data;
1567         mutex_unlock(&kvm->lock);
1568
1569         VM_EVENT(kvm, 3, "SET: guest UV-feat: 0x%16.16lx", data);
1570
1571         return 0;
1572 }
1573
1574 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1575 {
1576         int ret = -ENXIO;
1577
1578         switch (attr->attr) {
1579         case KVM_S390_VM_CPU_PROCESSOR:
1580                 ret = kvm_s390_set_processor(kvm, attr);
1581                 break;
1582         case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1583                 ret = kvm_s390_set_processor_feat(kvm, attr);
1584                 break;
1585         case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1586                 ret = kvm_s390_set_processor_subfunc(kvm, attr);
1587                 break;
1588         case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
1589                 ret = kvm_s390_set_uv_feat(kvm, attr);
1590                 break;
1591         }
1592         return ret;
1593 }
1594
1595 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1596 {
1597         struct kvm_s390_vm_cpu_processor *proc;
1598         int ret = 0;
1599
1600         proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1601         if (!proc) {
1602                 ret = -ENOMEM;
1603                 goto out;
1604         }
1605         proc->cpuid = kvm->arch.model.cpuid;
1606         proc->ibc = kvm->arch.model.ibc;
1607         memcpy(&proc->fac_list, kvm->arch.model.fac_list,
1608                S390_ARCH_FAC_LIST_SIZE_BYTE);
1609         VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1610                  kvm->arch.model.ibc,
1611                  kvm->arch.model.cpuid);
1612         VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1613                  kvm->arch.model.fac_list[0],
1614                  kvm->arch.model.fac_list[1],
1615                  kvm->arch.model.fac_list[2]);
1616         if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
1617                 ret = -EFAULT;
1618         kfree(proc);
1619 out:
1620         return ret;
1621 }
1622
1623 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
1624 {
1625         struct kvm_s390_vm_cpu_machine *mach;
1626         int ret = 0;
1627
1628         mach = kzalloc(sizeof(*mach), GFP_KERNEL_ACCOUNT);
1629         if (!mach) {
1630                 ret = -ENOMEM;
1631                 goto out;
1632         }
1633         get_cpu_id((struct cpuid *) &mach->cpuid);
1634         mach->ibc = sclp.ibc;
1635         memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
1636                S390_ARCH_FAC_LIST_SIZE_BYTE);
1637         memcpy((unsigned long *)&mach->fac_list, stfle_fac_list,
1638                sizeof(stfle_fac_list));
1639         VM_EVENT(kvm, 3, "GET: host ibc:  0x%4.4x, host cpuid:  0x%16.16llx",
1640                  kvm->arch.model.ibc,
1641                  kvm->arch.model.cpuid);
1642         VM_EVENT(kvm, 3, "GET: host facmask:  0x%16.16llx.%16.16llx.%16.16llx",
1643                  mach->fac_mask[0],
1644                  mach->fac_mask[1],
1645                  mach->fac_mask[2]);
1646         VM_EVENT(kvm, 3, "GET: host faclist:  0x%16.16llx.%16.16llx.%16.16llx",
1647                  mach->fac_list[0],
1648                  mach->fac_list[1],
1649                  mach->fac_list[2]);
1650         if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1651                 ret = -EFAULT;
1652         kfree(mach);
1653 out:
1654         return ret;
1655 }
1656
1657 static int kvm_s390_get_processor_feat(struct kvm *kvm,
1658                                        struct kvm_device_attr *attr)
1659 {
1660         struct kvm_s390_vm_cpu_feat data;
1661
1662         bitmap_to_arr64(data.feat, kvm->arch.cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1663         if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1664                 return -EFAULT;
1665         VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1666                          data.feat[0],
1667                          data.feat[1],
1668                          data.feat[2]);
1669         return 0;
1670 }
1671
1672 static int kvm_s390_get_machine_feat(struct kvm *kvm,
1673                                      struct kvm_device_attr *attr)
1674 {
1675         struct kvm_s390_vm_cpu_feat data;
1676
1677         bitmap_to_arr64(data.feat, kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1678         if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1679                 return -EFAULT;
1680         VM_EVENT(kvm, 3, "GET: host feat:  0x%16.16llx.0x%16.16llx.0x%16.16llx",
1681                          data.feat[0],
1682                          data.feat[1],
1683                          data.feat[2]);
1684         return 0;
1685 }
1686
1687 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1688                                           struct kvm_device_attr *attr)
1689 {
1690         if (copy_to_user((void __user *)attr->addr, &kvm->arch.model.subfuncs,
1691             sizeof(struct kvm_s390_vm_cpu_subfunc)))
1692                 return -EFAULT;
1693
1694         VM_EVENT(kvm, 3, "GET: guest PLO    subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1695                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1696                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1697                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1698                  ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1699         VM_EVENT(kvm, 3, "GET: guest PTFF   subfunc 0x%16.16lx.%16.16lx",
1700                  ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1701                  ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1702         VM_EVENT(kvm, 3, "GET: guest KMAC   subfunc 0x%16.16lx.%16.16lx",
1703                  ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1704                  ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1705         VM_EVENT(kvm, 3, "GET: guest KMC    subfunc 0x%16.16lx.%16.16lx",
1706                  ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1707                  ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1708         VM_EVENT(kvm, 3, "GET: guest KM     subfunc 0x%16.16lx.%16.16lx",
1709                  ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1710                  ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1711         VM_EVENT(kvm, 3, "GET: guest KIMD   subfunc 0x%16.16lx.%16.16lx",
1712                  ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1713                  ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1714         VM_EVENT(kvm, 3, "GET: guest KLMD   subfunc 0x%16.16lx.%16.16lx",
1715                  ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1716                  ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1717         VM_EVENT(kvm, 3, "GET: guest PCKMO  subfunc 0x%16.16lx.%16.16lx",
1718                  ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1719                  ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1720         VM_EVENT(kvm, 3, "GET: guest KMCTR  subfunc 0x%16.16lx.%16.16lx",
1721                  ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1722                  ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1723         VM_EVENT(kvm, 3, "GET: guest KMF    subfunc 0x%16.16lx.%16.16lx",
1724                  ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1725                  ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1726         VM_EVENT(kvm, 3, "GET: guest KMO    subfunc 0x%16.16lx.%16.16lx",
1727                  ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1728                  ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1729         VM_EVENT(kvm, 3, "GET: guest PCC    subfunc 0x%16.16lx.%16.16lx",
1730                  ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1731                  ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1732         VM_EVENT(kvm, 3, "GET: guest PPNO   subfunc 0x%16.16lx.%16.16lx",
1733                  ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1734                  ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1735         VM_EVENT(kvm, 3, "GET: guest KMA    subfunc 0x%16.16lx.%16.16lx",
1736                  ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1737                  ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1738         VM_EVENT(kvm, 3, "GET: guest KDSA   subfunc 0x%16.16lx.%16.16lx",
1739                  ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1740                  ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1741         VM_EVENT(kvm, 3, "GET: guest SORTL  subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1742                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1743                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1744                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1745                  ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1746         VM_EVENT(kvm, 3, "GET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1747                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1748                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1749                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1750                  ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1751
1752         return 0;
1753 }
1754
1755 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1756                                         struct kvm_device_attr *attr)
1757 {
1758         if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1759             sizeof(struct kvm_s390_vm_cpu_subfunc)))
1760                 return -EFAULT;
1761
1762         VM_EVENT(kvm, 3, "GET: host  PLO    subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1763                  ((unsigned long *) &kvm_s390_available_subfunc.plo)[0],
1764                  ((unsigned long *) &kvm_s390_available_subfunc.plo)[1],
1765                  ((unsigned long *) &kvm_s390_available_subfunc.plo)[2],
1766                  ((unsigned long *) &kvm_s390_available_subfunc.plo)[3]);
1767         VM_EVENT(kvm, 3, "GET: host  PTFF   subfunc 0x%16.16lx.%16.16lx",
1768                  ((unsigned long *) &kvm_s390_available_subfunc.ptff)[0],
1769                  ((unsigned long *) &kvm_s390_available_subfunc.ptff)[1]);
1770         VM_EVENT(kvm, 3, "GET: host  KMAC   subfunc 0x%16.16lx.%16.16lx",
1771                  ((unsigned long *) &kvm_s390_available_subfunc.kmac)[0],
1772                  ((unsigned long *) &kvm_s390_available_subfunc.kmac)[1]);
1773         VM_EVENT(kvm, 3, "GET: host  KMC    subfunc 0x%16.16lx.%16.16lx",
1774                  ((unsigned long *) &kvm_s390_available_subfunc.kmc)[0],
1775                  ((unsigned long *) &kvm_s390_available_subfunc.kmc)[1]);
1776         VM_EVENT(kvm, 3, "GET: host  KM     subfunc 0x%16.16lx.%16.16lx",
1777                  ((unsigned long *) &kvm_s390_available_subfunc.km)[0],
1778                  ((unsigned long *) &kvm_s390_available_subfunc.km)[1]);
1779         VM_EVENT(kvm, 3, "GET: host  KIMD   subfunc 0x%16.16lx.%16.16lx",
1780                  ((unsigned long *) &kvm_s390_available_subfunc.kimd)[0],
1781                  ((unsigned long *) &kvm_s390_available_subfunc.kimd)[1]);
1782         VM_EVENT(kvm, 3, "GET: host  KLMD   subfunc 0x%16.16lx.%16.16lx",
1783                  ((unsigned long *) &kvm_s390_available_subfunc.klmd)[0],
1784                  ((unsigned long *) &kvm_s390_available_subfunc.klmd)[1]);
1785         VM_EVENT(kvm, 3, "GET: host  PCKMO  subfunc 0x%16.16lx.%16.16lx",
1786                  ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[0],
1787                  ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[1]);
1788         VM_EVENT(kvm, 3, "GET: host  KMCTR  subfunc 0x%16.16lx.%16.16lx",
1789                  ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[0],
1790                  ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[1]);
1791         VM_EVENT(kvm, 3, "GET: host  KMF    subfunc 0x%16.16lx.%16.16lx",
1792                  ((unsigned long *) &kvm_s390_available_subfunc.kmf)[0],
1793                  ((unsigned long *) &kvm_s390_available_subfunc.kmf)[1]);
1794         VM_EVENT(kvm, 3, "GET: host  KMO    subfunc 0x%16.16lx.%16.16lx",
1795                  ((unsigned long *) &kvm_s390_available_subfunc.kmo)[0],
1796                  ((unsigned long *) &kvm_s390_available_subfunc.kmo)[1]);
1797         VM_EVENT(kvm, 3, "GET: host  PCC    subfunc 0x%16.16lx.%16.16lx",
1798                  ((unsigned long *) &kvm_s390_available_subfunc.pcc)[0],
1799                  ((unsigned long *) &kvm_s390_available_subfunc.pcc)[1]);
1800         VM_EVENT(kvm, 3, "GET: host  PPNO   subfunc 0x%16.16lx.%16.16lx",
1801                  ((unsigned long *) &kvm_s390_available_subfunc.ppno)[0],
1802                  ((unsigned long *) &kvm_s390_available_subfunc.ppno)[1]);
1803         VM_EVENT(kvm, 3, "GET: host  KMA    subfunc 0x%16.16lx.%16.16lx",
1804                  ((unsigned long *) &kvm_s390_available_subfunc.kma)[0],
1805                  ((unsigned long *) &kvm_s390_available_subfunc.kma)[1]);
1806         VM_EVENT(kvm, 3, "GET: host  KDSA   subfunc 0x%16.16lx.%16.16lx",
1807                  ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[0],
1808                  ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[1]);
1809         VM_EVENT(kvm, 3, "GET: host  SORTL  subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1810                  ((unsigned long *) &kvm_s390_available_subfunc.sortl)[0],
1811                  ((unsigned long *) &kvm_s390_available_subfunc.sortl)[1],
1812                  ((unsigned long *) &kvm_s390_available_subfunc.sortl)[2],
1813                  ((unsigned long *) &kvm_s390_available_subfunc.sortl)[3]);
1814         VM_EVENT(kvm, 3, "GET: host  DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1815                  ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[0],
1816                  ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[1],
1817                  ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[2],
1818                  ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[3]);
1819
1820         return 0;
1821 }
1822
1823 static int kvm_s390_get_processor_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1824 {
1825         struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr;
1826         unsigned long feat = kvm->arch.model.uv_feat_guest.feat;
1827
1828         if (put_user(feat, &dst->feat))
1829                 return -EFAULT;
1830         VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat);
1831
1832         return 0;
1833 }
1834
1835 static int kvm_s390_get_machine_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1836 {
1837         struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr;
1838         unsigned long feat;
1839
1840         BUILD_BUG_ON(sizeof(*dst) != sizeof(uv_info.uv_feature_indications));
1841
1842         feat = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK;
1843         if (put_user(feat, &dst->feat))
1844                 return -EFAULT;
1845         VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat);
1846
1847         return 0;
1848 }
1849
1850 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1851 {
1852         int ret = -ENXIO;
1853
1854         switch (attr->attr) {
1855         case KVM_S390_VM_CPU_PROCESSOR:
1856                 ret = kvm_s390_get_processor(kvm, attr);
1857                 break;
1858         case KVM_S390_VM_CPU_MACHINE:
1859                 ret = kvm_s390_get_machine(kvm, attr);
1860                 break;
1861         case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1862                 ret = kvm_s390_get_processor_feat(kvm, attr);
1863                 break;
1864         case KVM_S390_VM_CPU_MACHINE_FEAT:
1865                 ret = kvm_s390_get_machine_feat(kvm, attr);
1866                 break;
1867         case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1868                 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1869                 break;
1870         case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1871                 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1872                 break;
1873         case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
1874                 ret = kvm_s390_get_processor_uv_feat(kvm, attr);
1875                 break;
1876         case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST:
1877                 ret = kvm_s390_get_machine_uv_feat(kvm, attr);
1878                 break;
1879         }
1880         return ret;
1881 }
1882
1883 /**
1884  * kvm_s390_update_topology_change_report - update CPU topology change report
1885  * @kvm: guest KVM description
1886  * @val: set or clear the MTCR bit
1887  *
1888  * Updates the Multiprocessor Topology-Change-Report bit to signal
1889  * the guest with a topology change.
1890  * This is only relevant if the topology facility is present.
1891  *
1892  * The SCA version, bsca or esca, doesn't matter as offset is the same.
1893  */
1894 static void kvm_s390_update_topology_change_report(struct kvm *kvm, bool val)
1895 {
1896         union sca_utility new, old;
1897         struct bsca_block *sca;
1898
1899         read_lock(&kvm->arch.sca_lock);
1900         sca = kvm->arch.sca;
1901         do {
1902                 old = READ_ONCE(sca->utility);
1903                 new = old;
1904                 new.mtcr = val;
1905         } while (cmpxchg(&sca->utility.val, old.val, new.val) != old.val);
1906         read_unlock(&kvm->arch.sca_lock);
1907 }
1908
1909 static int kvm_s390_set_topo_change_indication(struct kvm *kvm,
1910                                                struct kvm_device_attr *attr)
1911 {
1912         if (!test_kvm_facility(kvm, 11))
1913                 return -ENXIO;
1914
1915         kvm_s390_update_topology_change_report(kvm, !!attr->attr);
1916         return 0;
1917 }
1918
1919 static int kvm_s390_get_topo_change_indication(struct kvm *kvm,
1920                                                struct kvm_device_attr *attr)
1921 {
1922         u8 topo;
1923
1924         if (!test_kvm_facility(kvm, 11))
1925                 return -ENXIO;
1926
1927         read_lock(&kvm->arch.sca_lock);
1928         topo = ((struct bsca_block *)kvm->arch.sca)->utility.mtcr;
1929         read_unlock(&kvm->arch.sca_lock);
1930
1931         return put_user(topo, (u8 __user *)attr->addr);
1932 }
1933
1934 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1935 {
1936         int ret;
1937
1938         switch (attr->group) {
1939         case KVM_S390_VM_MEM_CTRL:
1940                 ret = kvm_s390_set_mem_control(kvm, attr);
1941                 break;
1942         case KVM_S390_VM_TOD:
1943                 ret = kvm_s390_set_tod(kvm, attr);
1944                 break;
1945         case KVM_S390_VM_CPU_MODEL:
1946                 ret = kvm_s390_set_cpu_model(kvm, attr);
1947                 break;
1948         case KVM_S390_VM_CRYPTO:
1949                 ret = kvm_s390_vm_set_crypto(kvm, attr);
1950                 break;
1951         case KVM_S390_VM_MIGRATION:
1952                 ret = kvm_s390_vm_set_migration(kvm, attr);
1953                 break;
1954         case KVM_S390_VM_CPU_TOPOLOGY:
1955                 ret = kvm_s390_set_topo_change_indication(kvm, attr);
1956                 break;
1957         default:
1958                 ret = -ENXIO;
1959                 break;
1960         }
1961
1962         return ret;
1963 }
1964
1965 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1966 {
1967         int ret;
1968
1969         switch (attr->group) {
1970         case KVM_S390_VM_MEM_CTRL:
1971                 ret = kvm_s390_get_mem_control(kvm, attr);
1972                 break;
1973         case KVM_S390_VM_TOD:
1974                 ret = kvm_s390_get_tod(kvm, attr);
1975                 break;
1976         case KVM_S390_VM_CPU_MODEL:
1977                 ret = kvm_s390_get_cpu_model(kvm, attr);
1978                 break;
1979         case KVM_S390_VM_MIGRATION:
1980                 ret = kvm_s390_vm_get_migration(kvm, attr);
1981                 break;
1982         case KVM_S390_VM_CPU_TOPOLOGY:
1983                 ret = kvm_s390_get_topo_change_indication(kvm, attr);
1984                 break;
1985         default:
1986                 ret = -ENXIO;
1987                 break;
1988         }
1989
1990         return ret;
1991 }
1992
1993 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1994 {
1995         int ret;
1996
1997         switch (attr->group) {
1998         case KVM_S390_VM_MEM_CTRL:
1999                 switch (attr->attr) {
2000                 case KVM_S390_VM_MEM_ENABLE_CMMA:
2001                 case KVM_S390_VM_MEM_CLR_CMMA:
2002                         ret = sclp.has_cmma ? 0 : -ENXIO;
2003                         break;
2004                 case KVM_S390_VM_MEM_LIMIT_SIZE:
2005                         ret = 0;
2006                         break;
2007                 default:
2008                         ret = -ENXIO;
2009                         break;
2010                 }
2011                 break;
2012         case KVM_S390_VM_TOD:
2013                 switch (attr->attr) {
2014                 case KVM_S390_VM_TOD_LOW:
2015                 case KVM_S390_VM_TOD_HIGH:
2016                         ret = 0;
2017                         break;
2018                 default:
2019                         ret = -ENXIO;
2020                         break;
2021                 }
2022                 break;
2023         case KVM_S390_VM_CPU_MODEL:
2024                 switch (attr->attr) {
2025                 case KVM_S390_VM_CPU_PROCESSOR:
2026                 case KVM_S390_VM_CPU_MACHINE:
2027                 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
2028                 case KVM_S390_VM_CPU_MACHINE_FEAT:
2029                 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
2030                 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
2031                 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST:
2032                 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
2033                         ret = 0;
2034                         break;
2035                 default:
2036                         ret = -ENXIO;
2037                         break;
2038                 }
2039                 break;
2040         case KVM_S390_VM_CRYPTO:
2041                 switch (attr->attr) {
2042                 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
2043                 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
2044                 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
2045                 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
2046                         ret = 0;
2047                         break;
2048                 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
2049                 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
2050                         ret = ap_instructions_available() ? 0 : -ENXIO;
2051                         break;
2052                 default:
2053                         ret = -ENXIO;
2054                         break;
2055                 }
2056                 break;
2057         case KVM_S390_VM_MIGRATION:
2058                 ret = 0;
2059                 break;
2060         case KVM_S390_VM_CPU_TOPOLOGY:
2061                 ret = test_kvm_facility(kvm, 11) ? 0 : -ENXIO;
2062                 break;
2063         default:
2064                 ret = -ENXIO;
2065                 break;
2066         }
2067
2068         return ret;
2069 }
2070
2071 static int kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2072 {
2073         uint8_t *keys;
2074         uint64_t hva;
2075         int srcu_idx, i, r = 0;
2076
2077         if (args->flags != 0)
2078                 return -EINVAL;
2079
2080         /* Is this guest using storage keys? */
2081         if (!mm_uses_skeys(current->mm))
2082                 return KVM_S390_GET_SKEYS_NONE;
2083
2084         /* Enforce sane limit on memory allocation */
2085         if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2086                 return -EINVAL;
2087
2088         keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2089         if (!keys)
2090                 return -ENOMEM;
2091
2092         mmap_read_lock(current->mm);
2093         srcu_idx = srcu_read_lock(&kvm->srcu);
2094         for (i = 0; i < args->count; i++) {
2095                 hva = gfn_to_hva(kvm, args->start_gfn + i);
2096                 if (kvm_is_error_hva(hva)) {
2097                         r = -EFAULT;
2098                         break;
2099                 }
2100
2101                 r = get_guest_storage_key(current->mm, hva, &keys[i]);
2102                 if (r)
2103                         break;
2104         }
2105         srcu_read_unlock(&kvm->srcu, srcu_idx);
2106         mmap_read_unlock(current->mm);
2107
2108         if (!r) {
2109                 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
2110                                  sizeof(uint8_t) * args->count);
2111                 if (r)
2112                         r = -EFAULT;
2113         }
2114
2115         kvfree(keys);
2116         return r;
2117 }
2118
2119 static int kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2120 {
2121         uint8_t *keys;
2122         uint64_t hva;
2123         int srcu_idx, i, r = 0;
2124         bool unlocked;
2125
2126         if (args->flags != 0)
2127                 return -EINVAL;
2128
2129         /* Enforce sane limit on memory allocation */
2130         if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2131                 return -EINVAL;
2132
2133         keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2134         if (!keys)
2135                 return -ENOMEM;
2136
2137         r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
2138                            sizeof(uint8_t) * args->count);
2139         if (r) {
2140                 r = -EFAULT;
2141                 goto out;
2142         }
2143
2144         /* Enable storage key handling for the guest */
2145         r = s390_enable_skey();
2146         if (r)
2147                 goto out;
2148
2149         i = 0;
2150         mmap_read_lock(current->mm);
2151         srcu_idx = srcu_read_lock(&kvm->srcu);
2152         while (i < args->count) {
2153                 unlocked = false;
2154                 hva = gfn_to_hva(kvm, args->start_gfn + i);
2155                 if (kvm_is_error_hva(hva)) {
2156                         r = -EFAULT;
2157                         break;
2158                 }
2159
2160                 /* Lowest order bit is reserved */
2161                 if (keys[i] & 0x01) {
2162                         r = -EINVAL;
2163                         break;
2164                 }
2165
2166                 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
2167                 if (r) {
2168                         r = fixup_user_fault(current->mm, hva,
2169                                              FAULT_FLAG_WRITE, &unlocked);
2170                         if (r)
2171                                 break;
2172                 }
2173                 if (!r)
2174                         i++;
2175         }
2176         srcu_read_unlock(&kvm->srcu, srcu_idx);
2177         mmap_read_unlock(current->mm);
2178 out:
2179         kvfree(keys);
2180         return r;
2181 }
2182
2183 /*
2184  * Base address and length must be sent at the start of each block, therefore
2185  * it's cheaper to send some clean data, as long as it's less than the size of
2186  * two longs.
2187  */
2188 #define KVM_S390_MAX_BIT_DISTANCE (2 * sizeof(void *))
2189 /* for consistency */
2190 #define KVM_S390_CMMA_SIZE_MAX ((u32)KVM_S390_SKEYS_MAX)
2191
2192 static int kvm_s390_peek_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2193                               u8 *res, unsigned long bufsize)
2194 {
2195         unsigned long pgstev, hva, cur_gfn = args->start_gfn;
2196
2197         args->count = 0;
2198         while (args->count < bufsize) {
2199                 hva = gfn_to_hva(kvm, cur_gfn);
2200                 /*
2201                  * We return an error if the first value was invalid, but we
2202                  * return successfully if at least one value was copied.
2203                  */
2204                 if (kvm_is_error_hva(hva))
2205                         return args->count ? 0 : -EFAULT;
2206                 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2207                         pgstev = 0;
2208                 res[args->count++] = (pgstev >> 24) & 0x43;
2209                 cur_gfn++;
2210         }
2211
2212         return 0;
2213 }
2214
2215 static struct kvm_memory_slot *gfn_to_memslot_approx(struct kvm_memslots *slots,
2216                                                      gfn_t gfn)
2217 {
2218         return ____gfn_to_memslot(slots, gfn, true);
2219 }
2220
2221 static unsigned long kvm_s390_next_dirty_cmma(struct kvm_memslots *slots,
2222                                               unsigned long cur_gfn)
2223 {
2224         struct kvm_memory_slot *ms = gfn_to_memslot_approx(slots, cur_gfn);
2225         unsigned long ofs = cur_gfn - ms->base_gfn;
2226         struct rb_node *mnode = &ms->gfn_node[slots->node_idx];
2227
2228         if (ms->base_gfn + ms->npages <= cur_gfn) {
2229                 mnode = rb_next(mnode);
2230                 /* If we are above the highest slot, wrap around */
2231                 if (!mnode)
2232                         mnode = rb_first(&slots->gfn_tree);
2233
2234                 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2235                 ofs = 0;
2236         }
2237
2238         if (cur_gfn < ms->base_gfn)
2239                 ofs = 0;
2240
2241         ofs = find_next_bit(kvm_second_dirty_bitmap(ms), ms->npages, ofs);
2242         while (ofs >= ms->npages && (mnode = rb_next(mnode))) {
2243                 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2244                 ofs = find_first_bit(kvm_second_dirty_bitmap(ms), ms->npages);
2245         }
2246         return ms->base_gfn + ofs;
2247 }
2248
2249 static int kvm_s390_get_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2250                              u8 *res, unsigned long bufsize)
2251 {
2252         unsigned long mem_end, cur_gfn, next_gfn, hva, pgstev;
2253         struct kvm_memslots *slots = kvm_memslots(kvm);
2254         struct kvm_memory_slot *ms;
2255
2256         if (unlikely(kvm_memslots_empty(slots)))
2257                 return 0;
2258
2259         cur_gfn = kvm_s390_next_dirty_cmma(slots, args->start_gfn);
2260         ms = gfn_to_memslot(kvm, cur_gfn);
2261         args->count = 0;
2262         args->start_gfn = cur_gfn;
2263         if (!ms)
2264                 return 0;
2265         next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2266         mem_end = kvm_s390_get_gfn_end(slots);
2267
2268         while (args->count < bufsize) {
2269                 hva = gfn_to_hva(kvm, cur_gfn);
2270                 if (kvm_is_error_hva(hva))
2271                         return 0;
2272                 /* Decrement only if we actually flipped the bit to 0 */
2273                 if (test_and_clear_bit(cur_gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
2274                         atomic64_dec(&kvm->arch.cmma_dirty_pages);
2275                 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2276                         pgstev = 0;
2277                 /* Save the value */
2278                 res[args->count++] = (pgstev >> 24) & 0x43;
2279                 /* If the next bit is too far away, stop. */
2280                 if (next_gfn > cur_gfn + KVM_S390_MAX_BIT_DISTANCE)
2281                         return 0;
2282                 /* If we reached the previous "next", find the next one */
2283                 if (cur_gfn == next_gfn)
2284                         next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2285                 /* Reached the end of memory or of the buffer, stop */
2286                 if ((next_gfn >= mem_end) ||
2287                     (next_gfn - args->start_gfn >= bufsize))
2288                         return 0;
2289                 cur_gfn++;
2290                 /* Reached the end of the current memslot, take the next one. */
2291                 if (cur_gfn - ms->base_gfn >= ms->npages) {
2292                         ms = gfn_to_memslot(kvm, cur_gfn);
2293                         if (!ms)
2294                                 return 0;
2295                 }
2296         }
2297         return 0;
2298 }
2299
2300 /*
2301  * This function searches for the next page with dirty CMMA attributes, and
2302  * saves the attributes in the buffer up to either the end of the buffer or
2303  * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found;
2304  * no trailing clean bytes are saved.
2305  * In case no dirty bits were found, or if CMMA was not enabled or used, the
2306  * output buffer will indicate 0 as length.
2307  */
2308 static int kvm_s390_get_cmma_bits(struct kvm *kvm,
2309                                   struct kvm_s390_cmma_log *args)
2310 {
2311         unsigned long bufsize;
2312         int srcu_idx, peek, ret;
2313         u8 *values;
2314
2315         if (!kvm->arch.use_cmma)
2316                 return -ENXIO;
2317         /* Invalid/unsupported flags were specified */
2318         if (args->flags & ~KVM_S390_CMMA_PEEK)
2319                 return -EINVAL;
2320         /* Migration mode query, and we are not doing a migration */
2321         peek = !!(args->flags & KVM_S390_CMMA_PEEK);
2322         if (!peek && !kvm->arch.migration_mode)
2323                 return -EINVAL;
2324         /* CMMA is disabled or was not used, or the buffer has length zero */
2325         bufsize = min(args->count, KVM_S390_CMMA_SIZE_MAX);
2326         if (!bufsize || !kvm->mm->context.uses_cmm) {
2327                 memset(args, 0, sizeof(*args));
2328                 return 0;
2329         }
2330         /* We are not peeking, and there are no dirty pages */
2331         if (!peek && !atomic64_read(&kvm->arch.cmma_dirty_pages)) {
2332                 memset(args, 0, sizeof(*args));
2333                 return 0;
2334         }
2335
2336         values = vmalloc(bufsize);
2337         if (!values)
2338                 return -ENOMEM;
2339
2340         mmap_read_lock(kvm->mm);
2341         srcu_idx = srcu_read_lock(&kvm->srcu);
2342         if (peek)
2343                 ret = kvm_s390_peek_cmma(kvm, args, values, bufsize);
2344         else
2345                 ret = kvm_s390_get_cmma(kvm, args, values, bufsize);
2346         srcu_read_unlock(&kvm->srcu, srcu_idx);
2347         mmap_read_unlock(kvm->mm);
2348
2349         if (kvm->arch.migration_mode)
2350                 args->remaining = atomic64_read(&kvm->arch.cmma_dirty_pages);
2351         else
2352                 args->remaining = 0;
2353
2354         if (copy_to_user((void __user *)args->values, values, args->count))
2355                 ret = -EFAULT;
2356
2357         vfree(values);
2358         return ret;
2359 }
2360
2361 /*
2362  * This function sets the CMMA attributes for the given pages. If the input
2363  * buffer has zero length, no action is taken, otherwise the attributes are
2364  * set and the mm->context.uses_cmm flag is set.
2365  */
2366 static int kvm_s390_set_cmma_bits(struct kvm *kvm,
2367                                   const struct kvm_s390_cmma_log *args)
2368 {
2369         unsigned long hva, mask, pgstev, i;
2370         uint8_t *bits;
2371         int srcu_idx, r = 0;
2372
2373         mask = args->mask;
2374
2375         if (!kvm->arch.use_cmma)
2376                 return -ENXIO;
2377         /* invalid/unsupported flags */
2378         if (args->flags != 0)
2379                 return -EINVAL;
2380         /* Enforce sane limit on memory allocation */
2381         if (args->count > KVM_S390_CMMA_SIZE_MAX)
2382                 return -EINVAL;
2383         /* Nothing to do */
2384         if (args->count == 0)
2385                 return 0;
2386
2387         bits = vmalloc(array_size(sizeof(*bits), args->count));
2388         if (!bits)
2389                 return -ENOMEM;
2390
2391         r = copy_from_user(bits, (void __user *)args->values, args->count);
2392         if (r) {
2393                 r = -EFAULT;
2394                 goto out;
2395         }
2396
2397         mmap_read_lock(kvm->mm);
2398         srcu_idx = srcu_read_lock(&kvm->srcu);
2399         for (i = 0; i < args->count; i++) {
2400                 hva = gfn_to_hva(kvm, args->start_gfn + i);
2401                 if (kvm_is_error_hva(hva)) {
2402                         r = -EFAULT;
2403                         break;
2404                 }
2405
2406                 pgstev = bits[i];
2407                 pgstev = pgstev << 24;
2408                 mask &= _PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT;
2409                 set_pgste_bits(kvm->mm, hva, mask, pgstev);
2410         }
2411         srcu_read_unlock(&kvm->srcu, srcu_idx);
2412         mmap_read_unlock(kvm->mm);
2413
2414         if (!kvm->mm->context.uses_cmm) {
2415                 mmap_write_lock(kvm->mm);
2416                 kvm->mm->context.uses_cmm = 1;
2417                 mmap_write_unlock(kvm->mm);
2418         }
2419 out:
2420         vfree(bits);
2421         return r;
2422 }
2423
2424 /**
2425  * kvm_s390_cpus_from_pv - Convert all protected vCPUs in a protected VM to
2426  * non protected.
2427  * @kvm: the VM whose protected vCPUs are to be converted
2428  * @rc: return value for the RC field of the UVC (in case of error)
2429  * @rrc: return value for the RRC field of the UVC (in case of error)
2430  *
2431  * Does not stop in case of error, tries to convert as many
2432  * CPUs as possible. In case of error, the RC and RRC of the last error are
2433  * returned.
2434  *
2435  * Return: 0 in case of success, otherwise -EIO
2436  */
2437 int kvm_s390_cpus_from_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2438 {
2439         struct kvm_vcpu *vcpu;
2440         unsigned long i;
2441         u16 _rc, _rrc;
2442         int ret = 0;
2443
2444         /*
2445          * We ignore failures and try to destroy as many CPUs as possible.
2446          * At the same time we must not free the assigned resources when
2447          * this fails, as the ultravisor has still access to that memory.
2448          * So kvm_s390_pv_destroy_cpu can leave a "wanted" memory leak
2449          * behind.
2450          * We want to return the first failure rc and rrc, though.
2451          */
2452         kvm_for_each_vcpu(i, vcpu, kvm) {
2453                 mutex_lock(&vcpu->mutex);
2454                 if (kvm_s390_pv_destroy_cpu(vcpu, &_rc, &_rrc) && !ret) {
2455                         *rc = _rc;
2456                         *rrc = _rrc;
2457                         ret = -EIO;
2458                 }
2459                 mutex_unlock(&vcpu->mutex);
2460         }
2461         /* Ensure that we re-enable gisa if the non-PV guest used it but the PV guest did not. */
2462         if (use_gisa)
2463                 kvm_s390_gisa_enable(kvm);
2464         return ret;
2465 }
2466
2467 /**
2468  * kvm_s390_cpus_to_pv - Convert all non-protected vCPUs in a protected VM
2469  * to protected.
2470  * @kvm: the VM whose protected vCPUs are to be converted
2471  * @rc: return value for the RC field of the UVC (in case of error)
2472  * @rrc: return value for the RRC field of the UVC (in case of error)
2473  *
2474  * Tries to undo the conversion in case of error.
2475  *
2476  * Return: 0 in case of success, otherwise -EIO
2477  */
2478 static int kvm_s390_cpus_to_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2479 {
2480         unsigned long i;
2481         int r = 0;
2482         u16 dummy;
2483
2484         struct kvm_vcpu *vcpu;
2485
2486         /* Disable the GISA if the ultravisor does not support AIV. */
2487         if (!uv_has_feature(BIT_UV_FEAT_AIV))
2488                 kvm_s390_gisa_disable(kvm);
2489
2490         kvm_for_each_vcpu(i, vcpu, kvm) {
2491                 mutex_lock(&vcpu->mutex);
2492                 r = kvm_s390_pv_create_cpu(vcpu, rc, rrc);
2493                 mutex_unlock(&vcpu->mutex);
2494                 if (r)
2495                         break;
2496         }
2497         if (r)
2498                 kvm_s390_cpus_from_pv(kvm, &dummy, &dummy);
2499         return r;
2500 }
2501
2502 /*
2503  * Here we provide user space with a direct interface to query UV
2504  * related data like UV maxima and available features as well as
2505  * feature specific data.
2506  *
2507  * To facilitate future extension of the data structures we'll try to
2508  * write data up to the maximum requested length.
2509  */
2510 static ssize_t kvm_s390_handle_pv_info(struct kvm_s390_pv_info *info)
2511 {
2512         ssize_t len_min;
2513
2514         switch (info->header.id) {
2515         case KVM_PV_INFO_VM: {
2516                 len_min =  sizeof(info->header) + sizeof(info->vm);
2517
2518                 if (info->header.len_max < len_min)
2519                         return -EINVAL;
2520
2521                 memcpy(info->vm.inst_calls_list,
2522                        uv_info.inst_calls_list,
2523                        sizeof(uv_info.inst_calls_list));
2524
2525                 /* It's max cpuid not max cpus, so it's off by one */
2526                 info->vm.max_cpus = uv_info.max_guest_cpu_id + 1;
2527                 info->vm.max_guests = uv_info.max_num_sec_conf;
2528                 info->vm.max_guest_addr = uv_info.max_sec_stor_addr;
2529                 info->vm.feature_indication = uv_info.uv_feature_indications;
2530
2531                 return len_min;
2532         }
2533         case KVM_PV_INFO_DUMP: {
2534                 len_min =  sizeof(info->header) + sizeof(info->dump);
2535
2536                 if (info->header.len_max < len_min)
2537                         return -EINVAL;
2538
2539                 info->dump.dump_cpu_buffer_len = uv_info.guest_cpu_stor_len;
2540                 info->dump.dump_config_mem_buffer_per_1m = uv_info.conf_dump_storage_state_len;
2541                 info->dump.dump_config_finalize_len = uv_info.conf_dump_finalize_len;
2542                 return len_min;
2543         }
2544         default:
2545                 return -EINVAL;
2546         }
2547 }
2548
2549 static int kvm_s390_pv_dmp(struct kvm *kvm, struct kvm_pv_cmd *cmd,
2550                            struct kvm_s390_pv_dmp dmp)
2551 {
2552         int r = -EINVAL;
2553         void __user *result_buff = (void __user *)dmp.buff_addr;
2554
2555         switch (dmp.subcmd) {
2556         case KVM_PV_DUMP_INIT: {
2557                 if (kvm->arch.pv.dumping)
2558                         break;
2559
2560                 /*
2561                  * Block SIE entry as concurrent dump UVCs could lead
2562                  * to validities.
2563                  */
2564                 kvm_s390_vcpu_block_all(kvm);
2565
2566                 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2567                                   UVC_CMD_DUMP_INIT, &cmd->rc, &cmd->rrc);
2568                 KVM_UV_EVENT(kvm, 3, "PROTVIRT DUMP INIT: rc %x rrc %x",
2569                              cmd->rc, cmd->rrc);
2570                 if (!r) {
2571                         kvm->arch.pv.dumping = true;
2572                 } else {
2573                         kvm_s390_vcpu_unblock_all(kvm);
2574                         r = -EINVAL;
2575                 }
2576                 break;
2577         }
2578         case KVM_PV_DUMP_CONFIG_STOR_STATE: {
2579                 if (!kvm->arch.pv.dumping)
2580                         break;
2581
2582                 /*
2583                  * gaddr is an output parameter since we might stop
2584                  * early. As dmp will be copied back in our caller, we
2585                  * don't need to do it ourselves.
2586                  */
2587                 r = kvm_s390_pv_dump_stor_state(kvm, result_buff, &dmp.gaddr, dmp.buff_len,
2588                                                 &cmd->rc, &cmd->rrc);
2589                 break;
2590         }
2591         case KVM_PV_DUMP_COMPLETE: {
2592                 if (!kvm->arch.pv.dumping)
2593                         break;
2594
2595                 r = -EINVAL;
2596                 if (dmp.buff_len < uv_info.conf_dump_finalize_len)
2597                         break;
2598
2599                 r = kvm_s390_pv_dump_complete(kvm, result_buff,
2600                                               &cmd->rc, &cmd->rrc);
2601                 break;
2602         }
2603         default:
2604                 r = -ENOTTY;
2605                 break;
2606         }
2607
2608         return r;
2609 }
2610
2611 static int kvm_s390_handle_pv(struct kvm *kvm, struct kvm_pv_cmd *cmd)
2612 {
2613         const bool need_lock = (cmd->cmd != KVM_PV_ASYNC_CLEANUP_PERFORM);
2614         void __user *argp = (void __user *)cmd->data;
2615         int r = 0;
2616         u16 dummy;
2617
2618         if (need_lock)
2619                 mutex_lock(&kvm->lock);
2620
2621         switch (cmd->cmd) {
2622         case KVM_PV_ENABLE: {
2623                 r = -EINVAL;
2624                 if (kvm_s390_pv_is_protected(kvm))
2625                         break;
2626
2627                 /*
2628                  *  FMT 4 SIE needs esca. As we never switch back to bsca from
2629                  *  esca, we need no cleanup in the error cases below
2630                  */
2631                 r = sca_switch_to_extended(kvm);
2632                 if (r)
2633                         break;
2634
2635                 mmap_write_lock(current->mm);
2636                 r = gmap_mark_unmergeable();
2637                 mmap_write_unlock(current->mm);
2638                 if (r)
2639                         break;
2640
2641                 r = kvm_s390_pv_init_vm(kvm, &cmd->rc, &cmd->rrc);
2642                 if (r)
2643                         break;
2644
2645                 r = kvm_s390_cpus_to_pv(kvm, &cmd->rc, &cmd->rrc);
2646                 if (r)
2647                         kvm_s390_pv_deinit_vm(kvm, &dummy, &dummy);
2648
2649                 /* we need to block service interrupts from now on */
2650                 set_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2651                 break;
2652         }
2653         case KVM_PV_ASYNC_CLEANUP_PREPARE:
2654                 r = -EINVAL;
2655                 if (!kvm_s390_pv_is_protected(kvm) || !async_destroy)
2656                         break;
2657
2658                 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2659                 /*
2660                  * If a CPU could not be destroyed, destroy VM will also fail.
2661                  * There is no point in trying to destroy it. Instead return
2662                  * the rc and rrc from the first CPU that failed destroying.
2663                  */
2664                 if (r)
2665                         break;
2666                 r = kvm_s390_pv_set_aside(kvm, &cmd->rc, &cmd->rrc);
2667
2668                 /* no need to block service interrupts any more */
2669                 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2670                 break;
2671         case KVM_PV_ASYNC_CLEANUP_PERFORM:
2672                 r = -EINVAL;
2673                 if (!async_destroy)
2674                         break;
2675                 /* kvm->lock must not be held; this is asserted inside the function. */
2676                 r = kvm_s390_pv_deinit_aside_vm(kvm, &cmd->rc, &cmd->rrc);
2677                 break;
2678         case KVM_PV_DISABLE: {
2679                 r = -EINVAL;
2680                 if (!kvm_s390_pv_is_protected(kvm))
2681                         break;
2682
2683                 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2684                 /*
2685                  * If a CPU could not be destroyed, destroy VM will also fail.
2686                  * There is no point in trying to destroy it. Instead return
2687                  * the rc and rrc from the first CPU that failed destroying.
2688                  */
2689                 if (r)
2690                         break;
2691                 r = kvm_s390_pv_deinit_cleanup_all(kvm, &cmd->rc, &cmd->rrc);
2692
2693                 /* no need to block service interrupts any more */
2694                 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2695                 break;
2696         }
2697         case KVM_PV_SET_SEC_PARMS: {
2698                 struct kvm_s390_pv_sec_parm parms = {};
2699                 void *hdr;
2700
2701                 r = -EINVAL;
2702                 if (!kvm_s390_pv_is_protected(kvm))
2703                         break;
2704
2705                 r = -EFAULT;
2706                 if (copy_from_user(&parms, argp, sizeof(parms)))
2707                         break;
2708
2709                 /* Currently restricted to 8KB */
2710                 r = -EINVAL;
2711                 if (parms.length > PAGE_SIZE * 2)
2712                         break;
2713
2714                 r = -ENOMEM;
2715                 hdr = vmalloc(parms.length);
2716                 if (!hdr)
2717                         break;
2718
2719                 r = -EFAULT;
2720                 if (!copy_from_user(hdr, (void __user *)parms.origin,
2721                                     parms.length))
2722                         r = kvm_s390_pv_set_sec_parms(kvm, hdr, parms.length,
2723                                                       &cmd->rc, &cmd->rrc);
2724
2725                 vfree(hdr);
2726                 break;
2727         }
2728         case KVM_PV_UNPACK: {
2729                 struct kvm_s390_pv_unp unp = {};
2730
2731                 r = -EINVAL;
2732                 if (!kvm_s390_pv_is_protected(kvm) || !mm_is_protected(kvm->mm))
2733                         break;
2734
2735                 r = -EFAULT;
2736                 if (copy_from_user(&unp, argp, sizeof(unp)))
2737                         break;
2738
2739                 r = kvm_s390_pv_unpack(kvm, unp.addr, unp.size, unp.tweak,
2740                                        &cmd->rc, &cmd->rrc);
2741                 break;
2742         }
2743         case KVM_PV_VERIFY: {
2744                 r = -EINVAL;
2745                 if (!kvm_s390_pv_is_protected(kvm))
2746                         break;
2747
2748                 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2749                                   UVC_CMD_VERIFY_IMG, &cmd->rc, &cmd->rrc);
2750                 KVM_UV_EVENT(kvm, 3, "PROTVIRT VERIFY: rc %x rrc %x", cmd->rc,
2751                              cmd->rrc);
2752                 break;
2753         }
2754         case KVM_PV_PREP_RESET: {
2755                 r = -EINVAL;
2756                 if (!kvm_s390_pv_is_protected(kvm))
2757                         break;
2758
2759                 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2760                                   UVC_CMD_PREPARE_RESET, &cmd->rc, &cmd->rrc);
2761                 KVM_UV_EVENT(kvm, 3, "PROTVIRT PREP RESET: rc %x rrc %x",
2762                              cmd->rc, cmd->rrc);
2763                 break;
2764         }
2765         case KVM_PV_UNSHARE_ALL: {
2766                 r = -EINVAL;
2767                 if (!kvm_s390_pv_is_protected(kvm))
2768                         break;
2769
2770                 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2771                                   UVC_CMD_SET_UNSHARE_ALL, &cmd->rc, &cmd->rrc);
2772                 KVM_UV_EVENT(kvm, 3, "PROTVIRT UNSHARE: rc %x rrc %x",
2773                              cmd->rc, cmd->rrc);
2774                 break;
2775         }
2776         case KVM_PV_INFO: {
2777                 struct kvm_s390_pv_info info = {};
2778                 ssize_t data_len;
2779
2780                 /*
2781                  * No need to check the VM protection here.
2782                  *
2783                  * Maybe user space wants to query some of the data
2784                  * when the VM is still unprotected. If we see the
2785                  * need to fence a new data command we can still
2786                  * return an error in the info handler.
2787                  */
2788
2789                 r = -EFAULT;
2790                 if (copy_from_user(&info, argp, sizeof(info.header)))
2791                         break;
2792
2793                 r = -EINVAL;
2794                 if (info.header.len_max < sizeof(info.header))
2795                         break;
2796
2797                 data_len = kvm_s390_handle_pv_info(&info);
2798                 if (data_len < 0) {
2799                         r = data_len;
2800                         break;
2801                 }
2802                 /*
2803                  * If a data command struct is extended (multiple
2804                  * times) this can be used to determine how much of it
2805                  * is valid.
2806                  */
2807                 info.header.len_written = data_len;
2808
2809                 r = -EFAULT;
2810                 if (copy_to_user(argp, &info, data_len))
2811                         break;
2812
2813                 r = 0;
2814                 break;
2815         }
2816         case KVM_PV_DUMP: {
2817                 struct kvm_s390_pv_dmp dmp;
2818
2819                 r = -EINVAL;
2820                 if (!kvm_s390_pv_is_protected(kvm))
2821                         break;
2822
2823                 r = -EFAULT;
2824                 if (copy_from_user(&dmp, argp, sizeof(dmp)))
2825                         break;
2826
2827                 r = kvm_s390_pv_dmp(kvm, cmd, dmp);
2828                 if (r)
2829                         break;
2830
2831                 if (copy_to_user(argp, &dmp, sizeof(dmp))) {
2832                         r = -EFAULT;
2833                         break;
2834                 }
2835
2836                 break;
2837         }
2838         default:
2839                 r = -ENOTTY;
2840         }
2841         if (need_lock)
2842                 mutex_unlock(&kvm->lock);
2843
2844         return r;
2845 }
2846
2847 static int mem_op_validate_common(struct kvm_s390_mem_op *mop, u64 supported_flags)
2848 {
2849         if (mop->flags & ~supported_flags || !mop->size)
2850                 return -EINVAL;
2851         if (mop->size > MEM_OP_MAX_SIZE)
2852                 return -E2BIG;
2853         if (mop->flags & KVM_S390_MEMOP_F_SKEY_PROTECTION) {
2854                 if (mop->key > 0xf)
2855                         return -EINVAL;
2856         } else {
2857                 mop->key = 0;
2858         }
2859         return 0;
2860 }
2861
2862 static int kvm_s390_vm_mem_op_abs(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2863 {
2864         void __user *uaddr = (void __user *)mop->buf;
2865         enum gacc_mode acc_mode;
2866         void *tmpbuf = NULL;
2867         int r, srcu_idx;
2868
2869         r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION |
2870                                         KVM_S390_MEMOP_F_CHECK_ONLY);
2871         if (r)
2872                 return r;
2873
2874         if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
2875                 tmpbuf = vmalloc(mop->size);
2876                 if (!tmpbuf)
2877                         return -ENOMEM;
2878         }
2879
2880         srcu_idx = srcu_read_lock(&kvm->srcu);
2881
2882         if (kvm_is_error_gpa(kvm, mop->gaddr)) {
2883                 r = PGM_ADDRESSING;
2884                 goto out_unlock;
2885         }
2886
2887         acc_mode = mop->op == KVM_S390_MEMOP_ABSOLUTE_READ ? GACC_FETCH : GACC_STORE;
2888         if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2889                 r = check_gpa_range(kvm, mop->gaddr, mop->size, acc_mode, mop->key);
2890                 goto out_unlock;
2891         }
2892         if (acc_mode == GACC_FETCH) {
2893                 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2894                                               mop->size, GACC_FETCH, mop->key);
2895                 if (r)
2896                         goto out_unlock;
2897                 if (copy_to_user(uaddr, tmpbuf, mop->size))
2898                         r = -EFAULT;
2899         } else {
2900                 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
2901                         r = -EFAULT;
2902                         goto out_unlock;
2903                 }
2904                 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2905                                               mop->size, GACC_STORE, mop->key);
2906         }
2907
2908 out_unlock:
2909         srcu_read_unlock(&kvm->srcu, srcu_idx);
2910
2911         vfree(tmpbuf);
2912         return r;
2913 }
2914
2915 static int kvm_s390_vm_mem_op_cmpxchg(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2916 {
2917         void __user *uaddr = (void __user *)mop->buf;
2918         void __user *old_addr = (void __user *)mop->old_addr;
2919         union {
2920                 __uint128_t quad;
2921                 char raw[sizeof(__uint128_t)];
2922         } old = { .quad = 0}, new = { .quad = 0 };
2923         unsigned int off_in_quad = sizeof(new) - mop->size;
2924         int r, srcu_idx;
2925         bool success;
2926
2927         r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION);
2928         if (r)
2929                 return r;
2930         /*
2931          * This validates off_in_quad. Checking that size is a power
2932          * of two is not necessary, as cmpxchg_guest_abs_with_key
2933          * takes care of that
2934          */
2935         if (mop->size > sizeof(new))
2936                 return -EINVAL;
2937         if (copy_from_user(&new.raw[off_in_quad], uaddr, mop->size))
2938                 return -EFAULT;
2939         if (copy_from_user(&old.raw[off_in_quad], old_addr, mop->size))
2940                 return -EFAULT;
2941
2942         srcu_idx = srcu_read_lock(&kvm->srcu);
2943
2944         if (kvm_is_error_gpa(kvm, mop->gaddr)) {
2945                 r = PGM_ADDRESSING;
2946                 goto out_unlock;
2947         }
2948
2949         r = cmpxchg_guest_abs_with_key(kvm, mop->gaddr, mop->size, &old.quad,
2950                                        new.quad, mop->key, &success);
2951         if (!success && copy_to_user(old_addr, &old.raw[off_in_quad], mop->size))
2952                 r = -EFAULT;
2953
2954 out_unlock:
2955         srcu_read_unlock(&kvm->srcu, srcu_idx);
2956         return r;
2957 }
2958
2959 static int kvm_s390_vm_mem_op(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2960 {
2961         /*
2962          * This is technically a heuristic only, if the kvm->lock is not
2963          * taken, it is not guaranteed that the vm is/remains non-protected.
2964          * This is ok from a kernel perspective, wrongdoing is detected
2965          * on the access, -EFAULT is returned and the vm may crash the
2966          * next time it accesses the memory in question.
2967          * There is no sane usecase to do switching and a memop on two
2968          * different CPUs at the same time.
2969          */
2970         if (kvm_s390_pv_get_handle(kvm))
2971                 return -EINVAL;
2972
2973         switch (mop->op) {
2974         case KVM_S390_MEMOP_ABSOLUTE_READ:
2975         case KVM_S390_MEMOP_ABSOLUTE_WRITE:
2976                 return kvm_s390_vm_mem_op_abs(kvm, mop);
2977         case KVM_S390_MEMOP_ABSOLUTE_CMPXCHG:
2978                 return kvm_s390_vm_mem_op_cmpxchg(kvm, mop);
2979         default:
2980                 return -EINVAL;
2981         }
2982 }
2983
2984 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
2985 {
2986         struct kvm *kvm = filp->private_data;
2987         void __user *argp = (void __user *)arg;
2988         struct kvm_device_attr attr;
2989         int r;
2990
2991         switch (ioctl) {
2992         case KVM_S390_INTERRUPT: {
2993                 struct kvm_s390_interrupt s390int;
2994
2995                 r = -EFAULT;
2996                 if (copy_from_user(&s390int, argp, sizeof(s390int)))
2997                         break;
2998                 r = kvm_s390_inject_vm(kvm, &s390int);
2999                 break;
3000         }
3001         case KVM_CREATE_IRQCHIP: {
3002                 struct kvm_irq_routing_entry routing;
3003
3004                 r = -EINVAL;
3005                 if (kvm->arch.use_irqchip) {
3006                         /* Set up dummy routing. */
3007                         memset(&routing, 0, sizeof(routing));
3008                         r = kvm_set_irq_routing(kvm, &routing, 0, 0);
3009                 }
3010                 break;
3011         }
3012         case KVM_SET_DEVICE_ATTR: {
3013                 r = -EFAULT;
3014                 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3015                         break;
3016                 r = kvm_s390_vm_set_attr(kvm, &attr);
3017                 break;
3018         }
3019         case KVM_GET_DEVICE_ATTR: {
3020                 r = -EFAULT;
3021                 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3022                         break;
3023                 r = kvm_s390_vm_get_attr(kvm, &attr);
3024                 break;
3025         }
3026         case KVM_HAS_DEVICE_ATTR: {
3027                 r = -EFAULT;
3028                 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3029                         break;
3030                 r = kvm_s390_vm_has_attr(kvm, &attr);
3031                 break;
3032         }
3033         case KVM_S390_GET_SKEYS: {
3034                 struct kvm_s390_skeys args;
3035
3036                 r = -EFAULT;
3037                 if (copy_from_user(&args, argp,
3038                                    sizeof(struct kvm_s390_skeys)))
3039                         break;
3040                 r = kvm_s390_get_skeys(kvm, &args);
3041                 break;
3042         }
3043         case KVM_S390_SET_SKEYS: {
3044                 struct kvm_s390_skeys args;
3045
3046                 r = -EFAULT;
3047                 if (copy_from_user(&args, argp,
3048                                    sizeof(struct kvm_s390_skeys)))
3049                         break;
3050                 r = kvm_s390_set_skeys(kvm, &args);
3051                 break;
3052         }
3053         case KVM_S390_GET_CMMA_BITS: {
3054                 struct kvm_s390_cmma_log args;
3055
3056                 r = -EFAULT;
3057                 if (copy_from_user(&args, argp, sizeof(args)))
3058                         break;
3059                 mutex_lock(&kvm->slots_lock);
3060                 r = kvm_s390_get_cmma_bits(kvm, &args);
3061                 mutex_unlock(&kvm->slots_lock);
3062                 if (!r) {
3063                         r = copy_to_user(argp, &args, sizeof(args));
3064                         if (r)
3065                                 r = -EFAULT;
3066                 }
3067                 break;
3068         }
3069         case KVM_S390_SET_CMMA_BITS: {
3070                 struct kvm_s390_cmma_log args;
3071
3072                 r = -EFAULT;
3073                 if (copy_from_user(&args, argp, sizeof(args)))
3074                         break;
3075                 mutex_lock(&kvm->slots_lock);
3076                 r = kvm_s390_set_cmma_bits(kvm, &args);
3077                 mutex_unlock(&kvm->slots_lock);
3078                 break;
3079         }
3080         case KVM_S390_PV_COMMAND: {
3081                 struct kvm_pv_cmd args;
3082
3083                 /* protvirt means user cpu state */
3084                 kvm_s390_set_user_cpu_state_ctrl(kvm);
3085                 r = 0;
3086                 if (!is_prot_virt_host()) {
3087                         r = -EINVAL;
3088                         break;
3089                 }
3090                 if (copy_from_user(&args, argp, sizeof(args))) {
3091                         r = -EFAULT;
3092                         break;
3093                 }
3094                 if (args.flags) {
3095                         r = -EINVAL;
3096                         break;
3097                 }
3098                 /* must be called without kvm->lock */
3099                 r = kvm_s390_handle_pv(kvm, &args);
3100                 if (copy_to_user(argp, &args, sizeof(args))) {
3101                         r = -EFAULT;
3102                         break;
3103                 }
3104                 break;
3105         }
3106         case KVM_S390_MEM_OP: {
3107                 struct kvm_s390_mem_op mem_op;
3108
3109                 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3110                         r = kvm_s390_vm_mem_op(kvm, &mem_op);
3111                 else
3112                         r = -EFAULT;
3113                 break;
3114         }
3115         case KVM_S390_ZPCI_OP: {
3116                 struct kvm_s390_zpci_op args;
3117
3118                 r = -EINVAL;
3119                 if (!IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3120                         break;
3121                 if (copy_from_user(&args, argp, sizeof(args))) {
3122                         r = -EFAULT;
3123                         break;
3124                 }
3125                 r = kvm_s390_pci_zpci_op(kvm, &args);
3126                 break;
3127         }
3128         default:
3129                 r = -ENOTTY;
3130         }
3131
3132         return r;
3133 }
3134
3135 static int kvm_s390_apxa_installed(void)
3136 {
3137         struct ap_config_info info;
3138
3139         if (ap_instructions_available()) {
3140                 if (ap_qci(&info) == 0)
3141                         return info.apxa;
3142         }
3143
3144         return 0;
3145 }
3146
3147 /*
3148  * The format of the crypto control block (CRYCB) is specified in the 3 low
3149  * order bits of the CRYCB designation (CRYCBD) field as follows:
3150  * Format 0: Neither the message security assist extension 3 (MSAX3) nor the
3151  *           AP extended addressing (APXA) facility are installed.
3152  * Format 1: The APXA facility is not installed but the MSAX3 facility is.
3153  * Format 2: Both the APXA and MSAX3 facilities are installed
3154  */
3155 static void kvm_s390_set_crycb_format(struct kvm *kvm)
3156 {
3157         kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
3158
3159         /* Clear the CRYCB format bits - i.e., set format 0 by default */
3160         kvm->arch.crypto.crycbd &= ~(CRYCB_FORMAT_MASK);
3161
3162         /* Check whether MSAX3 is installed */
3163         if (!test_kvm_facility(kvm, 76))
3164                 return;
3165
3166         if (kvm_s390_apxa_installed())
3167                 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
3168         else
3169                 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
3170 }
3171
3172 /*
3173  * kvm_arch_crypto_set_masks
3174  *
3175  * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3176  *       to be set.
3177  * @apm: the mask identifying the accessible AP adapters
3178  * @aqm: the mask identifying the accessible AP domains
3179  * @adm: the mask identifying the accessible AP control domains
3180  *
3181  * Set the masks that identify the adapters, domains and control domains to
3182  * which the KVM guest is granted access.
3183  *
3184  * Note: The kvm->lock mutex must be locked by the caller before invoking this
3185  *       function.
3186  */
3187 void kvm_arch_crypto_set_masks(struct kvm *kvm, unsigned long *apm,
3188                                unsigned long *aqm, unsigned long *adm)
3189 {
3190         struct kvm_s390_crypto_cb *crycb = kvm->arch.crypto.crycb;
3191
3192         kvm_s390_vcpu_block_all(kvm);
3193
3194         switch (kvm->arch.crypto.crycbd & CRYCB_FORMAT_MASK) {
3195         case CRYCB_FORMAT2: /* APCB1 use 256 bits */
3196                 memcpy(crycb->apcb1.apm, apm, 32);
3197                 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx %016lx %016lx %016lx",
3198                          apm[0], apm[1], apm[2], apm[3]);
3199                 memcpy(crycb->apcb1.aqm, aqm, 32);
3200                 VM_EVENT(kvm, 3, "SET CRYCB: aqm %016lx %016lx %016lx %016lx",
3201                          aqm[0], aqm[1], aqm[2], aqm[3]);
3202                 memcpy(crycb->apcb1.adm, adm, 32);
3203                 VM_EVENT(kvm, 3, "SET CRYCB: adm %016lx %016lx %016lx %016lx",
3204                          adm[0], adm[1], adm[2], adm[3]);
3205                 break;
3206         case CRYCB_FORMAT1:
3207         case CRYCB_FORMAT0: /* Fall through both use APCB0 */
3208                 memcpy(crycb->apcb0.apm, apm, 8);
3209                 memcpy(crycb->apcb0.aqm, aqm, 2);
3210                 memcpy(crycb->apcb0.adm, adm, 2);
3211                 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx aqm %04x adm %04x",
3212                          apm[0], *((unsigned short *)aqm),
3213                          *((unsigned short *)adm));
3214                 break;
3215         default:        /* Can not happen */
3216                 break;
3217         }
3218
3219         /* recreate the shadow crycb for each vcpu */
3220         kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3221         kvm_s390_vcpu_unblock_all(kvm);
3222 }
3223 EXPORT_SYMBOL_GPL(kvm_arch_crypto_set_masks);
3224
3225 /*
3226  * kvm_arch_crypto_clear_masks
3227  *
3228  * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3229  *       to be cleared.
3230  *
3231  * Clear the masks that identify the adapters, domains and control domains to
3232  * which the KVM guest is granted access.
3233  *
3234  * Note: The kvm->lock mutex must be locked by the caller before invoking this
3235  *       function.
3236  */
3237 void kvm_arch_crypto_clear_masks(struct kvm *kvm)
3238 {
3239         kvm_s390_vcpu_block_all(kvm);
3240
3241         memset(&kvm->arch.crypto.crycb->apcb0, 0,
3242                sizeof(kvm->arch.crypto.crycb->apcb0));
3243         memset(&kvm->arch.crypto.crycb->apcb1, 0,
3244                sizeof(kvm->arch.crypto.crycb->apcb1));
3245
3246         VM_EVENT(kvm, 3, "%s", "CLR CRYCB:");
3247         /* recreate the shadow crycb for each vcpu */
3248         kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3249         kvm_s390_vcpu_unblock_all(kvm);
3250 }
3251 EXPORT_SYMBOL_GPL(kvm_arch_crypto_clear_masks);
3252
3253 static u64 kvm_s390_get_initial_cpuid(void)
3254 {
3255         struct cpuid cpuid;
3256
3257         get_cpu_id(&cpuid);
3258         cpuid.version = 0xff;
3259         return *((u64 *) &cpuid);
3260 }
3261
3262 static void kvm_s390_crypto_init(struct kvm *kvm)
3263 {
3264         kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
3265         kvm_s390_set_crycb_format(kvm);
3266         init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem);
3267
3268         if (!test_kvm_facility(kvm, 76))
3269                 return;
3270
3271         /* Enable AES/DEA protected key functions by default */
3272         kvm->arch.crypto.aes_kw = 1;
3273         kvm->arch.crypto.dea_kw = 1;
3274         get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
3275                          sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
3276         get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
3277                          sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
3278 }
3279
3280 static void sca_dispose(struct kvm *kvm)
3281 {
3282         if (kvm->arch.use_esca)
3283                 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
3284         else
3285                 free_page((unsigned long)(kvm->arch.sca));
3286         kvm->arch.sca = NULL;
3287 }
3288
3289 void kvm_arch_free_vm(struct kvm *kvm)
3290 {
3291         if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3292                 kvm_s390_pci_clear_list(kvm);
3293
3294         __kvm_arch_free_vm(kvm);
3295 }
3296
3297 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
3298 {
3299         gfp_t alloc_flags = GFP_KERNEL_ACCOUNT;
3300         int i, rc;
3301         char debug_name[16];
3302         static unsigned long sca_offset;
3303
3304         rc = -EINVAL;
3305 #ifdef CONFIG_KVM_S390_UCONTROL
3306         if (type & ~KVM_VM_S390_UCONTROL)
3307                 goto out_err;
3308         if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
3309                 goto out_err;
3310 #else
3311         if (type)
3312                 goto out_err;
3313 #endif
3314
3315         rc = s390_enable_sie();
3316         if (rc)
3317                 goto out_err;
3318
3319         rc = -ENOMEM;
3320
3321         if (!sclp.has_64bscao)
3322                 alloc_flags |= GFP_DMA;
3323         rwlock_init(&kvm->arch.sca_lock);
3324         /* start with basic SCA */
3325         kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
3326         if (!kvm->arch.sca)
3327                 goto out_err;
3328         mutex_lock(&kvm_lock);
3329         sca_offset += 16;
3330         if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
3331                 sca_offset = 0;
3332         kvm->arch.sca = (struct bsca_block *)
3333                         ((char *) kvm->arch.sca + sca_offset);
3334         mutex_unlock(&kvm_lock);
3335
3336         sprintf(debug_name, "kvm-%u", current->pid);
3337
3338         kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
3339         if (!kvm->arch.dbf)
3340                 goto out_err;
3341
3342         BUILD_BUG_ON(sizeof(struct sie_page2) != 4096);
3343         kvm->arch.sie_page2 =
3344              (struct sie_page2 *) get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3345         if (!kvm->arch.sie_page2)
3346                 goto out_err;
3347
3348         kvm->arch.sie_page2->kvm = kvm;
3349         kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
3350
3351         for (i = 0; i < kvm_s390_fac_size(); i++) {
3352                 kvm->arch.model.fac_mask[i] = stfle_fac_list[i] &
3353                                               (kvm_s390_fac_base[i] |
3354                                                kvm_s390_fac_ext[i]);
3355                 kvm->arch.model.fac_list[i] = stfle_fac_list[i] &
3356                                               kvm_s390_fac_base[i];
3357         }
3358         kvm->arch.model.subfuncs = kvm_s390_available_subfunc;
3359
3360         /* we are always in czam mode - even on pre z14 machines */
3361         set_kvm_facility(kvm->arch.model.fac_mask, 138);
3362         set_kvm_facility(kvm->arch.model.fac_list, 138);
3363         /* we emulate STHYI in kvm */
3364         set_kvm_facility(kvm->arch.model.fac_mask, 74);
3365         set_kvm_facility(kvm->arch.model.fac_list, 74);
3366         if (MACHINE_HAS_TLB_GUEST) {
3367                 set_kvm_facility(kvm->arch.model.fac_mask, 147);
3368                 set_kvm_facility(kvm->arch.model.fac_list, 147);
3369         }
3370
3371         if (css_general_characteristics.aiv && test_facility(65))
3372                 set_kvm_facility(kvm->arch.model.fac_mask, 65);
3373
3374         kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
3375         kvm->arch.model.ibc = sclp.ibc & 0x0fff;
3376
3377         kvm->arch.model.uv_feat_guest.feat = 0;
3378
3379         kvm_s390_crypto_init(kvm);
3380
3381         if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3382                 mutex_lock(&kvm->lock);
3383                 kvm_s390_pci_init_list(kvm);
3384                 kvm_s390_vcpu_pci_enable_interp(kvm);
3385                 mutex_unlock(&kvm->lock);
3386         }
3387
3388         mutex_init(&kvm->arch.float_int.ais_lock);
3389         spin_lock_init(&kvm->arch.float_int.lock);
3390         for (i = 0; i < FIRQ_LIST_COUNT; i++)
3391                 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
3392         init_waitqueue_head(&kvm->arch.ipte_wq);
3393         mutex_init(&kvm->arch.ipte_mutex);
3394
3395         debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
3396         VM_EVENT(kvm, 3, "vm created with type %lu", type);
3397
3398         if (type & KVM_VM_S390_UCONTROL) {
3399                 kvm->arch.gmap = NULL;
3400                 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
3401         } else {
3402                 if (sclp.hamax == U64_MAX)
3403                         kvm->arch.mem_limit = TASK_SIZE_MAX;
3404                 else
3405                         kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
3406                                                     sclp.hamax + 1);
3407                 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
3408                 if (!kvm->arch.gmap)
3409                         goto out_err;
3410                 kvm->arch.gmap->private = kvm;
3411                 kvm->arch.gmap->pfault_enabled = 0;
3412         }
3413
3414         kvm->arch.use_pfmfi = sclp.has_pfmfi;
3415         kvm->arch.use_skf = sclp.has_skey;
3416         spin_lock_init(&kvm->arch.start_stop_lock);
3417         kvm_s390_vsie_init(kvm);
3418         if (use_gisa)
3419                 kvm_s390_gisa_init(kvm);
3420         INIT_LIST_HEAD(&kvm->arch.pv.need_cleanup);
3421         kvm->arch.pv.set_aside = NULL;
3422         KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
3423
3424         return 0;
3425 out_err:
3426         free_page((unsigned long)kvm->arch.sie_page2);
3427         debug_unregister(kvm->arch.dbf);
3428         sca_dispose(kvm);
3429         KVM_EVENT(3, "creation of vm failed: %d", rc);
3430         return rc;
3431 }
3432
3433 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3434 {
3435         u16 rc, rrc;
3436
3437         VCPU_EVENT(vcpu, 3, "%s", "free cpu");
3438         trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
3439         kvm_s390_clear_local_irqs(vcpu);
3440         kvm_clear_async_pf_completion_queue(vcpu);
3441         if (!kvm_is_ucontrol(vcpu->kvm))
3442                 sca_del_vcpu(vcpu);
3443         kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3444
3445         if (kvm_is_ucontrol(vcpu->kvm))
3446                 gmap_remove(vcpu->arch.gmap);
3447
3448         if (vcpu->kvm->arch.use_cmma)
3449                 kvm_s390_vcpu_unsetup_cmma(vcpu);
3450         /* We can not hold the vcpu mutex here, we are already dying */
3451         if (kvm_s390_pv_cpu_get_handle(vcpu))
3452                 kvm_s390_pv_destroy_cpu(vcpu, &rc, &rrc);
3453         free_page((unsigned long)(vcpu->arch.sie_block));
3454 }
3455
3456 void kvm_arch_destroy_vm(struct kvm *kvm)
3457 {
3458         u16 rc, rrc;
3459
3460         kvm_destroy_vcpus(kvm);
3461         sca_dispose(kvm);
3462         kvm_s390_gisa_destroy(kvm);
3463         /*
3464          * We are already at the end of life and kvm->lock is not taken.
3465          * This is ok as the file descriptor is closed by now and nobody
3466          * can mess with the pv state.
3467          */
3468         kvm_s390_pv_deinit_cleanup_all(kvm, &rc, &rrc);
3469         /*
3470          * Remove the mmu notifier only when the whole KVM VM is torn down,
3471          * and only if one was registered to begin with. If the VM is
3472          * currently not protected, but has been previously been protected,
3473          * then it's possible that the notifier is still registered.
3474          */
3475         if (kvm->arch.pv.mmu_notifier.ops)
3476                 mmu_notifier_unregister(&kvm->arch.pv.mmu_notifier, kvm->mm);
3477
3478         debug_unregister(kvm->arch.dbf);
3479         free_page((unsigned long)kvm->arch.sie_page2);
3480         if (!kvm_is_ucontrol(kvm))
3481                 gmap_remove(kvm->arch.gmap);
3482         kvm_s390_destroy_adapters(kvm);
3483         kvm_s390_clear_float_irqs(kvm);
3484         kvm_s390_vsie_destroy(kvm);
3485         KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
3486 }
3487
3488 /* Section: vcpu related */
3489 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
3490 {
3491         vcpu->arch.gmap = gmap_create(current->mm, -1UL);
3492         if (!vcpu->arch.gmap)
3493                 return -ENOMEM;
3494         vcpu->arch.gmap->private = vcpu->kvm;
3495
3496         return 0;
3497 }
3498
3499 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
3500 {
3501         if (!kvm_s390_use_sca_entries())
3502                 return;
3503         read_lock(&vcpu->kvm->arch.sca_lock);
3504         if (vcpu->kvm->arch.use_esca) {
3505                 struct esca_block *sca = vcpu->kvm->arch.sca;
3506
3507                 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3508                 sca->cpu[vcpu->vcpu_id].sda = 0;
3509         } else {
3510                 struct bsca_block *sca = vcpu->kvm->arch.sca;
3511
3512                 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3513                 sca->cpu[vcpu->vcpu_id].sda = 0;
3514         }
3515         read_unlock(&vcpu->kvm->arch.sca_lock);
3516 }
3517
3518 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
3519 {
3520         if (!kvm_s390_use_sca_entries()) {
3521                 phys_addr_t sca_phys = virt_to_phys(vcpu->kvm->arch.sca);
3522
3523                 /* we still need the basic sca for the ipte control */
3524                 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3525                 vcpu->arch.sie_block->scaol = sca_phys;
3526                 return;
3527         }
3528         read_lock(&vcpu->kvm->arch.sca_lock);
3529         if (vcpu->kvm->arch.use_esca) {
3530                 struct esca_block *sca = vcpu->kvm->arch.sca;
3531                 phys_addr_t sca_phys = virt_to_phys(sca);
3532
3533                 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3534                 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3535                 vcpu->arch.sie_block->scaol = sca_phys & ESCA_SCAOL_MASK;
3536                 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3537                 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3538         } else {
3539                 struct bsca_block *sca = vcpu->kvm->arch.sca;
3540                 phys_addr_t sca_phys = virt_to_phys(sca);
3541
3542                 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3543                 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3544                 vcpu->arch.sie_block->scaol = sca_phys;
3545                 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3546         }
3547         read_unlock(&vcpu->kvm->arch.sca_lock);
3548 }
3549
3550 /* Basic SCA to Extended SCA data copy routines */
3551 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
3552 {
3553         d->sda = s->sda;
3554         d->sigp_ctrl.c = s->sigp_ctrl.c;
3555         d->sigp_ctrl.scn = s->sigp_ctrl.scn;
3556 }
3557
3558 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
3559 {
3560         int i;
3561
3562         d->ipte_control = s->ipte_control;
3563         d->mcn[0] = s->mcn;
3564         for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
3565                 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
3566 }
3567
3568 static int sca_switch_to_extended(struct kvm *kvm)
3569 {
3570         struct bsca_block *old_sca = kvm->arch.sca;
3571         struct esca_block *new_sca;
3572         struct kvm_vcpu *vcpu;
3573         unsigned long vcpu_idx;
3574         u32 scaol, scaoh;
3575         phys_addr_t new_sca_phys;
3576
3577         if (kvm->arch.use_esca)
3578                 return 0;
3579
3580         new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3581         if (!new_sca)
3582                 return -ENOMEM;
3583
3584         new_sca_phys = virt_to_phys(new_sca);
3585         scaoh = new_sca_phys >> 32;
3586         scaol = new_sca_phys & ESCA_SCAOL_MASK;
3587
3588         kvm_s390_vcpu_block_all(kvm);
3589         write_lock(&kvm->arch.sca_lock);
3590
3591         sca_copy_b_to_e(new_sca, old_sca);
3592
3593         kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
3594                 vcpu->arch.sie_block->scaoh = scaoh;
3595                 vcpu->arch.sie_block->scaol = scaol;
3596                 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3597         }
3598         kvm->arch.sca = new_sca;
3599         kvm->arch.use_esca = 1;
3600
3601         write_unlock(&kvm->arch.sca_lock);
3602         kvm_s390_vcpu_unblock_all(kvm);
3603
3604         free_page((unsigned long)old_sca);
3605
3606         VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
3607                  old_sca, kvm->arch.sca);
3608         return 0;
3609 }
3610
3611 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
3612 {
3613         int rc;
3614
3615         if (!kvm_s390_use_sca_entries()) {
3616                 if (id < KVM_MAX_VCPUS)
3617                         return true;
3618                 return false;
3619         }
3620         if (id < KVM_S390_BSCA_CPU_SLOTS)
3621                 return true;
3622         if (!sclp.has_esca || !sclp.has_64bscao)
3623                 return false;
3624
3625         rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
3626
3627         return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
3628 }
3629
3630 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3631 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3632 {
3633         WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
3634         raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3635         vcpu->arch.cputm_start = get_tod_clock_fast();
3636         raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3637 }
3638
3639 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3640 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3641 {
3642         WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
3643         raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3644         vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3645         vcpu->arch.cputm_start = 0;
3646         raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3647 }
3648
3649 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3650 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3651 {
3652         WARN_ON_ONCE(vcpu->arch.cputm_enabled);
3653         vcpu->arch.cputm_enabled = true;
3654         __start_cpu_timer_accounting(vcpu);
3655 }
3656
3657 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3658 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3659 {
3660         WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
3661         __stop_cpu_timer_accounting(vcpu);
3662         vcpu->arch.cputm_enabled = false;
3663 }
3664
3665 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3666 {
3667         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3668         __enable_cpu_timer_accounting(vcpu);
3669         preempt_enable();
3670 }
3671
3672 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3673 {
3674         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3675         __disable_cpu_timer_accounting(vcpu);
3676         preempt_enable();
3677 }
3678
3679 /* set the cpu timer - may only be called from the VCPU thread itself */
3680 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
3681 {
3682         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3683         raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3684         if (vcpu->arch.cputm_enabled)
3685                 vcpu->arch.cputm_start = get_tod_clock_fast();
3686         vcpu->arch.sie_block->cputm = cputm;
3687         raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3688         preempt_enable();
3689 }
3690
3691 /* update and get the cpu timer - can also be called from other VCPU threads */
3692 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
3693 {
3694         unsigned int seq;
3695         __u64 value;
3696
3697         if (unlikely(!vcpu->arch.cputm_enabled))
3698                 return vcpu->arch.sie_block->cputm;
3699
3700         preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3701         do {
3702                 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
3703                 /*
3704                  * If the writer would ever execute a read in the critical
3705                  * section, e.g. in irq context, we have a deadlock.
3706                  */
3707                 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
3708                 value = vcpu->arch.sie_block->cputm;
3709                 /* if cputm_start is 0, accounting is being started/stopped */
3710                 if (likely(vcpu->arch.cputm_start))
3711                         value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3712         } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
3713         preempt_enable();
3714         return value;
3715 }
3716
3717 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3718 {
3719
3720         gmap_enable(vcpu->arch.enabled_gmap);
3721         kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING);
3722         if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3723                 __start_cpu_timer_accounting(vcpu);
3724         vcpu->cpu = cpu;
3725 }
3726
3727 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3728 {
3729         vcpu->cpu = -1;
3730         if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3731                 __stop_cpu_timer_accounting(vcpu);
3732         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING);
3733         vcpu->arch.enabled_gmap = gmap_get_enabled();
3734         gmap_disable(vcpu->arch.enabled_gmap);
3735
3736 }
3737
3738 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
3739 {
3740         mutex_lock(&vcpu->kvm->lock);
3741         preempt_disable();
3742         vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
3743         vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx;
3744         preempt_enable();
3745         mutex_unlock(&vcpu->kvm->lock);
3746         if (!kvm_is_ucontrol(vcpu->kvm)) {
3747                 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
3748                 sca_add_vcpu(vcpu);
3749         }
3750         if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
3751                 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
3752         /* make vcpu_load load the right gmap on the first trigger */
3753         vcpu->arch.enabled_gmap = vcpu->arch.gmap;
3754 }
3755
3756 static bool kvm_has_pckmo_subfunc(struct kvm *kvm, unsigned long nr)
3757 {
3758         if (test_bit_inv(nr, (unsigned long *)&kvm->arch.model.subfuncs.pckmo) &&
3759             test_bit_inv(nr, (unsigned long *)&kvm_s390_available_subfunc.pckmo))
3760                 return true;
3761         return false;
3762 }
3763
3764 static bool kvm_has_pckmo_ecc(struct kvm *kvm)
3765 {
3766         /* At least one ECC subfunction must be present */
3767         return kvm_has_pckmo_subfunc(kvm, 32) ||
3768                kvm_has_pckmo_subfunc(kvm, 33) ||
3769                kvm_has_pckmo_subfunc(kvm, 34) ||
3770                kvm_has_pckmo_subfunc(kvm, 40) ||
3771                kvm_has_pckmo_subfunc(kvm, 41);
3772
3773 }
3774
3775 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
3776 {
3777         /*
3778          * If the AP instructions are not being interpreted and the MSAX3
3779          * facility is not configured for the guest, there is nothing to set up.
3780          */
3781         if (!vcpu->kvm->arch.crypto.apie && !test_kvm_facility(vcpu->kvm, 76))
3782                 return;
3783
3784         vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
3785         vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
3786         vcpu->arch.sie_block->eca &= ~ECA_APIE;
3787         vcpu->arch.sie_block->ecd &= ~ECD_ECC;
3788
3789         if (vcpu->kvm->arch.crypto.apie)
3790                 vcpu->arch.sie_block->eca |= ECA_APIE;
3791
3792         /* Set up protected key support */
3793         if (vcpu->kvm->arch.crypto.aes_kw) {
3794                 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
3795                 /* ecc is also wrapped with AES key */
3796                 if (kvm_has_pckmo_ecc(vcpu->kvm))
3797                         vcpu->arch.sie_block->ecd |= ECD_ECC;
3798         }
3799
3800         if (vcpu->kvm->arch.crypto.dea_kw)
3801                 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
3802 }
3803
3804 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
3805 {
3806         free_page((unsigned long)phys_to_virt(vcpu->arch.sie_block->cbrlo));
3807         vcpu->arch.sie_block->cbrlo = 0;
3808 }
3809
3810 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
3811 {
3812         void *cbrlo_page = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
3813
3814         if (!cbrlo_page)
3815                 return -ENOMEM;
3816
3817         vcpu->arch.sie_block->cbrlo = virt_to_phys(cbrlo_page);
3818         return 0;
3819 }
3820
3821 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
3822 {
3823         struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
3824
3825         vcpu->arch.sie_block->ibc = model->ibc;
3826         if (test_kvm_facility(vcpu->kvm, 7))
3827                 vcpu->arch.sie_block->fac = virt_to_phys(model->fac_list);
3828 }
3829
3830 static int kvm_s390_vcpu_setup(struct kvm_vcpu *vcpu)
3831 {
3832         int rc = 0;
3833         u16 uvrc, uvrrc;
3834
3835         atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
3836                                                     CPUSTAT_SM |
3837                                                     CPUSTAT_STOPPED);
3838
3839         if (test_kvm_facility(vcpu->kvm, 78))
3840                 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2);
3841         else if (test_kvm_facility(vcpu->kvm, 8))
3842                 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED);
3843
3844         kvm_s390_vcpu_setup_model(vcpu);
3845
3846         /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
3847         if (MACHINE_HAS_ESOP)
3848                 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
3849         if (test_kvm_facility(vcpu->kvm, 9))
3850                 vcpu->arch.sie_block->ecb |= ECB_SRSI;
3851         if (test_kvm_facility(vcpu->kvm, 11))
3852                 vcpu->arch.sie_block->ecb |= ECB_PTF;
3853         if (test_kvm_facility(vcpu->kvm, 73))
3854                 vcpu->arch.sie_block->ecb |= ECB_TE;
3855         if (!kvm_is_ucontrol(vcpu->kvm))
3856                 vcpu->arch.sie_block->ecb |= ECB_SPECI;
3857
3858         if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi)
3859                 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
3860         if (test_kvm_facility(vcpu->kvm, 130))
3861                 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
3862         vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
3863         if (sclp.has_cei)
3864                 vcpu->arch.sie_block->eca |= ECA_CEI;
3865         if (sclp.has_ib)
3866                 vcpu->arch.sie_block->eca |= ECA_IB;
3867         if (sclp.has_siif)
3868                 vcpu->arch.sie_block->eca |= ECA_SII;
3869         if (sclp.has_sigpif)
3870                 vcpu->arch.sie_block->eca |= ECA_SIGPI;
3871         if (test_kvm_facility(vcpu->kvm, 129)) {
3872                 vcpu->arch.sie_block->eca |= ECA_VX;
3873                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
3874         }
3875         if (test_kvm_facility(vcpu->kvm, 139))
3876                 vcpu->arch.sie_block->ecd |= ECD_MEF;
3877         if (test_kvm_facility(vcpu->kvm, 156))
3878                 vcpu->arch.sie_block->ecd |= ECD_ETOKENF;
3879         if (vcpu->arch.sie_block->gd) {
3880                 vcpu->arch.sie_block->eca |= ECA_AIV;
3881                 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3882                            vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3883         }
3884         vcpu->arch.sie_block->sdnxo = virt_to_phys(&vcpu->run->s.regs.sdnx) | SDNXC;
3885         vcpu->arch.sie_block->riccbd = virt_to_phys(&vcpu->run->s.regs.riccb);
3886
3887         if (sclp.has_kss)
3888                 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS);
3889         else
3890                 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
3891
3892         if (vcpu->kvm->arch.use_cmma) {
3893                 rc = kvm_s390_vcpu_setup_cmma(vcpu);
3894                 if (rc)
3895                         return rc;
3896         }
3897         hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3898         vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
3899
3900         vcpu->arch.sie_block->hpid = HPID_KVM;
3901
3902         kvm_s390_vcpu_crypto_setup(vcpu);
3903
3904         kvm_s390_vcpu_pci_setup(vcpu);
3905
3906         mutex_lock(&vcpu->kvm->lock);
3907         if (kvm_s390_pv_is_protected(vcpu->kvm)) {
3908                 rc = kvm_s390_pv_create_cpu(vcpu, &uvrc, &uvrrc);
3909                 if (rc)
3910                         kvm_s390_vcpu_unsetup_cmma(vcpu);
3911         }
3912         mutex_unlock(&vcpu->kvm->lock);
3913
3914         return rc;
3915 }
3916
3917 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
3918 {
3919         if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
3920                 return -EINVAL;
3921         return 0;
3922 }
3923
3924 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
3925 {
3926         struct sie_page *sie_page;
3927         int rc;
3928
3929         BUILD_BUG_ON(sizeof(struct sie_page) != 4096);
3930         sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL_ACCOUNT);
3931         if (!sie_page)
3932                 return -ENOMEM;
3933
3934         vcpu->arch.sie_block = &sie_page->sie_block;
3935         vcpu->arch.sie_block->itdba = virt_to_phys(&sie_page->itdb);
3936
3937         /* the real guest size will always be smaller than msl */
3938         vcpu->arch.sie_block->mso = 0;
3939         vcpu->arch.sie_block->msl = sclp.hamax;
3940
3941         vcpu->arch.sie_block->icpua = vcpu->vcpu_id;
3942         spin_lock_init(&vcpu->arch.local_int.lock);
3943         vcpu->arch.sie_block->gd = kvm_s390_get_gisa_desc(vcpu->kvm);
3944         seqcount_init(&vcpu->arch.cputm_seqcount);
3945
3946         vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
3947         kvm_clear_async_pf_completion_queue(vcpu);
3948         vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
3949                                     KVM_SYNC_GPRS |
3950                                     KVM_SYNC_ACRS |
3951                                     KVM_SYNC_CRS |
3952                                     KVM_SYNC_ARCH0 |
3953                                     KVM_SYNC_PFAULT |
3954                                     KVM_SYNC_DIAG318;
3955         kvm_s390_set_prefix(vcpu, 0);
3956         if (test_kvm_facility(vcpu->kvm, 64))
3957                 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
3958         if (test_kvm_facility(vcpu->kvm, 82))
3959                 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC;
3960         if (test_kvm_facility(vcpu->kvm, 133))
3961                 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
3962         if (test_kvm_facility(vcpu->kvm, 156))
3963                 vcpu->run->kvm_valid_regs |= KVM_SYNC_ETOKEN;
3964         /* fprs can be synchronized via vrs, even if the guest has no vx. With
3965          * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
3966          */
3967         if (MACHINE_HAS_VX)
3968                 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
3969         else
3970                 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
3971
3972         if (kvm_is_ucontrol(vcpu->kvm)) {
3973                 rc = __kvm_ucontrol_vcpu_init(vcpu);
3974                 if (rc)
3975                         goto out_free_sie_block;
3976         }
3977
3978         VM_EVENT(vcpu->kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK",
3979                  vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
3980         trace_kvm_s390_create_vcpu(vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
3981
3982         rc = kvm_s390_vcpu_setup(vcpu);
3983         if (rc)
3984                 goto out_ucontrol_uninit;
3985
3986         kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3987         return 0;
3988
3989 out_ucontrol_uninit:
3990         if (kvm_is_ucontrol(vcpu->kvm))
3991                 gmap_remove(vcpu->arch.gmap);
3992 out_free_sie_block:
3993         free_page((unsigned long)(vcpu->arch.sie_block));
3994         return rc;
3995 }
3996
3997 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3998 {
3999         clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
4000         return kvm_s390_vcpu_has_irq(vcpu, 0);
4001 }
4002
4003 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
4004 {
4005         return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE);
4006 }
4007
4008 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
4009 {
4010         atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
4011         exit_sie(vcpu);
4012 }
4013
4014 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
4015 {
4016         atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
4017 }
4018
4019 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
4020 {
4021         atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
4022         exit_sie(vcpu);
4023 }
4024
4025 bool kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu *vcpu)
4026 {
4027         return atomic_read(&vcpu->arch.sie_block->prog20) &
4028                (PROG_BLOCK_SIE | PROG_REQUEST);
4029 }
4030
4031 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
4032 {
4033         atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
4034 }
4035
4036 /*
4037  * Kick a guest cpu out of (v)SIE and wait until (v)SIE is not running.
4038  * If the CPU is not running (e.g. waiting as idle) the function will
4039  * return immediately. */
4040 void exit_sie(struct kvm_vcpu *vcpu)
4041 {
4042         kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
4043         kvm_s390_vsie_kick(vcpu);
4044         while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
4045                 cpu_relax();
4046 }
4047
4048 /* Kick a guest cpu out of SIE to process a request synchronously */
4049 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
4050 {
4051         __kvm_make_request(req, vcpu);
4052         kvm_s390_vcpu_request(vcpu);
4053 }
4054
4055 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
4056                               unsigned long end)
4057 {
4058         struct kvm *kvm = gmap->private;
4059         struct kvm_vcpu *vcpu;
4060         unsigned long prefix;
4061         unsigned long i;
4062
4063         trace_kvm_s390_gmap_notifier(start, end, gmap_is_shadow(gmap));
4064
4065         if (gmap_is_shadow(gmap))
4066                 return;
4067         if (start >= 1UL << 31)
4068                 /* We are only interested in prefix pages */
4069                 return;
4070         kvm_for_each_vcpu(i, vcpu, kvm) {
4071                 /* match against both prefix pages */
4072                 prefix = kvm_s390_get_prefix(vcpu);
4073                 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
4074                         VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
4075                                    start, end);
4076                         kvm_s390_sync_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4077                 }
4078         }
4079 }
4080
4081 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
4082 {
4083         /* do not poll with more than halt_poll_max_steal percent of steal time */
4084         if (S390_lowcore.avg_steal_timer * 100 / (TICK_USEC << 12) >=
4085             READ_ONCE(halt_poll_max_steal)) {
4086                 vcpu->stat.halt_no_poll_steal++;
4087                 return true;
4088         }
4089         return false;
4090 }
4091
4092 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
4093 {
4094         /* kvm common code refers to this, but never calls it */
4095         BUG();
4096         return 0;
4097 }
4098
4099 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
4100                                            struct kvm_one_reg *reg)
4101 {
4102         int r = -EINVAL;
4103
4104         switch (reg->id) {
4105         case KVM_REG_S390_TODPR:
4106                 r = put_user(vcpu->arch.sie_block->todpr,
4107                              (u32 __user *)reg->addr);
4108                 break;
4109         case KVM_REG_S390_EPOCHDIFF:
4110                 r = put_user(vcpu->arch.sie_block->epoch,
4111                              (u64 __user *)reg->addr);
4112                 break;
4113         case KVM_REG_S390_CPU_TIMER:
4114                 r = put_user(kvm_s390_get_cpu_timer(vcpu),
4115                              (u64 __user *)reg->addr);
4116                 break;
4117         case KVM_REG_S390_CLOCK_COMP:
4118                 r = put_user(vcpu->arch.sie_block->ckc,
4119                              (u64 __user *)reg->addr);
4120                 break;
4121         case KVM_REG_S390_PFTOKEN:
4122                 r = put_user(vcpu->arch.pfault_token,
4123                              (u64 __user *)reg->addr);
4124                 break;
4125         case KVM_REG_S390_PFCOMPARE:
4126                 r = put_user(vcpu->arch.pfault_compare,
4127                              (u64 __user *)reg->addr);
4128                 break;
4129         case KVM_REG_S390_PFSELECT:
4130                 r = put_user(vcpu->arch.pfault_select,
4131                              (u64 __user *)reg->addr);
4132                 break;
4133         case KVM_REG_S390_PP:
4134                 r = put_user(vcpu->arch.sie_block->pp,
4135                              (u64 __user *)reg->addr);
4136                 break;
4137         case KVM_REG_S390_GBEA:
4138                 r = put_user(vcpu->arch.sie_block->gbea,
4139                              (u64 __user *)reg->addr);
4140                 break;
4141         default:
4142                 break;
4143         }
4144
4145         return r;
4146 }
4147
4148 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
4149                                            struct kvm_one_reg *reg)
4150 {
4151         int r = -EINVAL;
4152         __u64 val;
4153
4154         switch (reg->id) {
4155         case KVM_REG_S390_TODPR:
4156                 r = get_user(vcpu->arch.sie_block->todpr,
4157                              (u32 __user *)reg->addr);
4158                 break;
4159         case KVM_REG_S390_EPOCHDIFF:
4160                 r = get_user(vcpu->arch.sie_block->epoch,
4161                              (u64 __user *)reg->addr);
4162                 break;
4163         case KVM_REG_S390_CPU_TIMER:
4164                 r = get_user(val, (u64 __user *)reg->addr);
4165                 if (!r)
4166                         kvm_s390_set_cpu_timer(vcpu, val);
4167                 break;
4168         case KVM_REG_S390_CLOCK_COMP:
4169                 r = get_user(vcpu->arch.sie_block->ckc,
4170                              (u64 __user *)reg->addr);
4171                 break;
4172         case KVM_REG_S390_PFTOKEN:
4173                 r = get_user(vcpu->arch.pfault_token,
4174                              (u64 __user *)reg->addr);
4175                 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4176                         kvm_clear_async_pf_completion_queue(vcpu);
4177                 break;
4178         case KVM_REG_S390_PFCOMPARE:
4179                 r = get_user(vcpu->arch.pfault_compare,
4180                              (u64 __user *)reg->addr);
4181                 break;
4182         case KVM_REG_S390_PFSELECT:
4183                 r = get_user(vcpu->arch.pfault_select,
4184                              (u64 __user *)reg->addr);
4185                 break;
4186         case KVM_REG_S390_PP:
4187                 r = get_user(vcpu->arch.sie_block->pp,
4188                              (u64 __user *)reg->addr);
4189                 break;
4190         case KVM_REG_S390_GBEA:
4191                 r = get_user(vcpu->arch.sie_block->gbea,
4192                              (u64 __user *)reg->addr);
4193                 break;
4194         default:
4195                 break;
4196         }
4197
4198         return r;
4199 }
4200
4201 static void kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu *vcpu)
4202 {
4203         vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_RI;
4204         vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
4205         memset(vcpu->run->s.regs.riccb, 0, sizeof(vcpu->run->s.regs.riccb));
4206
4207         kvm_clear_async_pf_completion_queue(vcpu);
4208         if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
4209                 kvm_s390_vcpu_stop(vcpu);
4210         kvm_s390_clear_local_irqs(vcpu);
4211 }
4212
4213 static void kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
4214 {
4215         /* Initial reset is a superset of the normal reset */
4216         kvm_arch_vcpu_ioctl_normal_reset(vcpu);
4217
4218         /*
4219          * This equals initial cpu reset in pop, but we don't switch to ESA.
4220          * We do not only reset the internal data, but also ...
4221          */
4222         vcpu->arch.sie_block->gpsw.mask = 0;
4223         vcpu->arch.sie_block->gpsw.addr = 0;
4224         kvm_s390_set_prefix(vcpu, 0);
4225         kvm_s390_set_cpu_timer(vcpu, 0);
4226         vcpu->arch.sie_block->ckc = 0;
4227         memset(vcpu->arch.sie_block->gcr, 0, sizeof(vcpu->arch.sie_block->gcr));
4228         vcpu->arch.sie_block->gcr[0] = CR0_INITIAL_MASK;
4229         vcpu->arch.sie_block->gcr[14] = CR14_INITIAL_MASK;
4230
4231         /* ... the data in sync regs */
4232         memset(vcpu->run->s.regs.crs, 0, sizeof(vcpu->run->s.regs.crs));
4233         vcpu->run->s.regs.ckc = 0;
4234         vcpu->run->s.regs.crs[0] = CR0_INITIAL_MASK;
4235         vcpu->run->s.regs.crs[14] = CR14_INITIAL_MASK;
4236         vcpu->run->psw_addr = 0;
4237         vcpu->run->psw_mask = 0;
4238         vcpu->run->s.regs.todpr = 0;
4239         vcpu->run->s.regs.cputm = 0;
4240         vcpu->run->s.regs.ckc = 0;
4241         vcpu->run->s.regs.pp = 0;
4242         vcpu->run->s.regs.gbea = 1;
4243         vcpu->run->s.regs.fpc = 0;
4244         /*
4245          * Do not reset these registers in the protected case, as some of
4246          * them are overlaid and they are not accessible in this case
4247          * anyway.
4248          */
4249         if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4250                 vcpu->arch.sie_block->gbea = 1;
4251                 vcpu->arch.sie_block->pp = 0;
4252                 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4253                 vcpu->arch.sie_block->todpr = 0;
4254         }
4255 }
4256
4257 static void kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu *vcpu)
4258 {
4259         struct kvm_sync_regs *regs = &vcpu->run->s.regs;
4260
4261         /* Clear reset is a superset of the initial reset */
4262         kvm_arch_vcpu_ioctl_initial_reset(vcpu);
4263
4264         memset(&regs->gprs, 0, sizeof(regs->gprs));
4265         memset(&regs->vrs, 0, sizeof(regs->vrs));
4266         memset(&regs->acrs, 0, sizeof(regs->acrs));
4267         memset(&regs->gscb, 0, sizeof(regs->gscb));
4268
4269         regs->etoken = 0;
4270         regs->etoken_extension = 0;
4271 }
4272
4273 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4274 {
4275         vcpu_load(vcpu);
4276         memcpy(&vcpu->run->s.regs.gprs, &regs->gprs, sizeof(regs->gprs));
4277         vcpu_put(vcpu);
4278         return 0;
4279 }
4280
4281 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4282 {
4283         vcpu_load(vcpu);
4284         memcpy(&regs->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
4285         vcpu_put(vcpu);
4286         return 0;
4287 }
4288
4289 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4290                                   struct kvm_sregs *sregs)
4291 {
4292         vcpu_load(vcpu);
4293
4294         memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
4295         memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
4296
4297         vcpu_put(vcpu);
4298         return 0;
4299 }
4300
4301 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4302                                   struct kvm_sregs *sregs)
4303 {
4304         vcpu_load(vcpu);
4305
4306         memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
4307         memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
4308
4309         vcpu_put(vcpu);
4310         return 0;
4311 }
4312
4313 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4314 {
4315         int ret = 0;
4316
4317         vcpu_load(vcpu);
4318
4319         vcpu->run->s.regs.fpc = fpu->fpc;
4320         if (MACHINE_HAS_VX)
4321                 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
4322                                  (freg_t *) fpu->fprs);
4323         else
4324                 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
4325
4326         vcpu_put(vcpu);
4327         return ret;
4328 }
4329
4330 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4331 {
4332         vcpu_load(vcpu);
4333
4334         /* make sure we have the latest values */
4335         save_fpu_regs();
4336         if (MACHINE_HAS_VX)
4337                 convert_vx_to_fp((freg_t *) fpu->fprs,
4338                                  (__vector128 *) vcpu->run->s.regs.vrs);
4339         else
4340                 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
4341         fpu->fpc = vcpu->run->s.regs.fpc;
4342
4343         vcpu_put(vcpu);
4344         return 0;
4345 }
4346
4347 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
4348 {
4349         int rc = 0;
4350
4351         if (!is_vcpu_stopped(vcpu))
4352                 rc = -EBUSY;
4353         else {
4354                 vcpu->run->psw_mask = psw.mask;
4355                 vcpu->run->psw_addr = psw.addr;
4356         }
4357         return rc;
4358 }
4359
4360 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4361                                   struct kvm_translation *tr)
4362 {
4363         return -EINVAL; /* not implemented yet */
4364 }
4365
4366 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
4367                               KVM_GUESTDBG_USE_HW_BP | \
4368                               KVM_GUESTDBG_ENABLE)
4369
4370 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4371                                         struct kvm_guest_debug *dbg)
4372 {
4373         int rc = 0;
4374
4375         vcpu_load(vcpu);
4376
4377         vcpu->guest_debug = 0;
4378         kvm_s390_clear_bp_data(vcpu);
4379
4380         if (dbg->control & ~VALID_GUESTDBG_FLAGS) {
4381                 rc = -EINVAL;
4382                 goto out;
4383         }
4384         if (!sclp.has_gpere) {
4385                 rc = -EINVAL;
4386                 goto out;
4387         }
4388
4389         if (dbg->control & KVM_GUESTDBG_ENABLE) {
4390                 vcpu->guest_debug = dbg->control;
4391                 /* enforce guest PER */
4392                 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P);
4393
4394                 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
4395                         rc = kvm_s390_import_bp_data(vcpu, dbg);
4396         } else {
4397                 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4398                 vcpu->arch.guestdbg.last_bp = 0;
4399         }
4400
4401         if (rc) {
4402                 vcpu->guest_debug = 0;
4403                 kvm_s390_clear_bp_data(vcpu);
4404                 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4405         }
4406
4407 out:
4408         vcpu_put(vcpu);
4409         return rc;
4410 }
4411
4412 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4413                                     struct kvm_mp_state *mp_state)
4414 {
4415         int ret;
4416
4417         vcpu_load(vcpu);
4418
4419         /* CHECK_STOP and LOAD are not supported yet */
4420         ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
4421                                       KVM_MP_STATE_OPERATING;
4422
4423         vcpu_put(vcpu);
4424         return ret;
4425 }
4426
4427 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4428                                     struct kvm_mp_state *mp_state)
4429 {
4430         int rc = 0;
4431
4432         vcpu_load(vcpu);
4433
4434         /* user space knows about this interface - let it control the state */
4435         kvm_s390_set_user_cpu_state_ctrl(vcpu->kvm);
4436
4437         switch (mp_state->mp_state) {
4438         case KVM_MP_STATE_STOPPED:
4439                 rc = kvm_s390_vcpu_stop(vcpu);
4440                 break;
4441         case KVM_MP_STATE_OPERATING:
4442                 rc = kvm_s390_vcpu_start(vcpu);
4443                 break;
4444         case KVM_MP_STATE_LOAD:
4445                 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4446                         rc = -ENXIO;
4447                         break;
4448                 }
4449                 rc = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR_LOAD);
4450                 break;
4451         case KVM_MP_STATE_CHECK_STOP:
4452                 fallthrough;    /* CHECK_STOP and LOAD are not supported yet */
4453         default:
4454                 rc = -ENXIO;
4455         }
4456
4457         vcpu_put(vcpu);
4458         return rc;
4459 }
4460
4461 static bool ibs_enabled(struct kvm_vcpu *vcpu)
4462 {
4463         return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS);
4464 }
4465
4466 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
4467 {
4468 retry:
4469         kvm_s390_vcpu_request_handled(vcpu);
4470         if (!kvm_request_pending(vcpu))
4471                 return 0;
4472         /*
4473          * If the guest prefix changed, re-arm the ipte notifier for the
4474          * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
4475          * This ensures that the ipte instruction for this request has
4476          * already finished. We might race against a second unmapper that
4477          * wants to set the blocking bit. Lets just retry the request loop.
4478          */
4479         if (kvm_check_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu)) {
4480                 int rc;
4481                 rc = gmap_mprotect_notify(vcpu->arch.gmap,
4482                                           kvm_s390_get_prefix(vcpu),
4483                                           PAGE_SIZE * 2, PROT_WRITE);
4484                 if (rc) {
4485                         kvm_make_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4486                         return rc;
4487                 }
4488                 goto retry;
4489         }
4490
4491         if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
4492                 vcpu->arch.sie_block->ihcpu = 0xffff;
4493                 goto retry;
4494         }
4495
4496         if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
4497                 if (!ibs_enabled(vcpu)) {
4498                         trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
4499                         kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS);
4500                 }
4501                 goto retry;
4502         }
4503
4504         if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
4505                 if (ibs_enabled(vcpu)) {
4506                         trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
4507                         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS);
4508                 }
4509                 goto retry;
4510         }
4511
4512         if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
4513                 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
4514                 goto retry;
4515         }
4516
4517         if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) {
4518                 /*
4519                  * Disable CMM virtualization; we will emulate the ESSA
4520                  * instruction manually, in order to provide additional
4521                  * functionalities needed for live migration.
4522                  */
4523                 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA;
4524                 goto retry;
4525         }
4526
4527         if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) {
4528                 /*
4529                  * Re-enable CMM virtualization if CMMA is available and
4530                  * CMM has been used.
4531                  */
4532                 if ((vcpu->kvm->arch.use_cmma) &&
4533                     (vcpu->kvm->mm->context.uses_cmm))
4534                         vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
4535                 goto retry;
4536         }
4537
4538         /* we left the vsie handler, nothing to do, just clear the request */
4539         kvm_clear_request(KVM_REQ_VSIE_RESTART, vcpu);
4540
4541         return 0;
4542 }
4543
4544 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4545 {
4546         struct kvm_vcpu *vcpu;
4547         union tod_clock clk;
4548         unsigned long i;
4549
4550         preempt_disable();
4551
4552         store_tod_clock_ext(&clk);
4553
4554         kvm->arch.epoch = gtod->tod - clk.tod;
4555         kvm->arch.epdx = 0;
4556         if (test_kvm_facility(kvm, 139)) {
4557                 kvm->arch.epdx = gtod->epoch_idx - clk.ei;
4558                 if (kvm->arch.epoch > gtod->tod)
4559                         kvm->arch.epdx -= 1;
4560         }
4561
4562         kvm_s390_vcpu_block_all(kvm);
4563         kvm_for_each_vcpu(i, vcpu, kvm) {
4564                 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
4565                 vcpu->arch.sie_block->epdx  = kvm->arch.epdx;
4566         }
4567
4568         kvm_s390_vcpu_unblock_all(kvm);
4569         preempt_enable();
4570 }
4571
4572 int kvm_s390_try_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4573 {
4574         if (!mutex_trylock(&kvm->lock))
4575                 return 0;
4576         __kvm_s390_set_tod_clock(kvm, gtod);
4577         mutex_unlock(&kvm->lock);
4578         return 1;
4579 }
4580
4581 /**
4582  * kvm_arch_fault_in_page - fault-in guest page if necessary
4583  * @vcpu: The corresponding virtual cpu
4584  * @gpa: Guest physical address
4585  * @writable: Whether the page should be writable or not
4586  *
4587  * Make sure that a guest page has been faulted-in on the host.
4588  *
4589  * Return: Zero on success, negative error code otherwise.
4590  */
4591 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
4592 {
4593         return gmap_fault(vcpu->arch.gmap, gpa,
4594                           writable ? FAULT_FLAG_WRITE : 0);
4595 }
4596
4597 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
4598                                       unsigned long token)
4599 {
4600         struct kvm_s390_interrupt inti;
4601         struct kvm_s390_irq irq;
4602
4603         if (start_token) {
4604                 irq.u.ext.ext_params2 = token;
4605                 irq.type = KVM_S390_INT_PFAULT_INIT;
4606                 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
4607         } else {
4608                 inti.type = KVM_S390_INT_PFAULT_DONE;
4609                 inti.parm64 = token;
4610                 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
4611         }
4612 }
4613
4614 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
4615                                      struct kvm_async_pf *work)
4616 {
4617         trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
4618         __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
4619
4620         return true;
4621 }
4622
4623 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
4624                                  struct kvm_async_pf *work)
4625 {
4626         trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
4627         __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
4628 }
4629
4630 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
4631                                struct kvm_async_pf *work)
4632 {
4633         /* s390 will always inject the page directly */
4634 }
4635
4636 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
4637 {
4638         /*
4639          * s390 will always inject the page directly,
4640          * but we still want check_async_completion to cleanup
4641          */
4642         return true;
4643 }
4644
4645 static bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
4646 {
4647         hva_t hva;
4648         struct kvm_arch_async_pf arch;
4649
4650         if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4651                 return false;
4652         if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
4653             vcpu->arch.pfault_compare)
4654                 return false;
4655         if (psw_extint_disabled(vcpu))
4656                 return false;
4657         if (kvm_s390_vcpu_has_irq(vcpu, 0))
4658                 return false;
4659         if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
4660                 return false;
4661         if (!vcpu->arch.gmap->pfault_enabled)
4662                 return false;
4663
4664         hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
4665         hva += current->thread.gmap_addr & ~PAGE_MASK;
4666         if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
4667                 return false;
4668
4669         return kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
4670 }
4671
4672 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
4673 {
4674         int rc, cpuflags;
4675
4676         /*
4677          * On s390 notifications for arriving pages will be delivered directly
4678          * to the guest but the house keeping for completed pfaults is
4679          * handled outside the worker.
4680          */
4681         kvm_check_async_pf_completion(vcpu);
4682
4683         vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
4684         vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
4685
4686         if (need_resched())
4687                 schedule();
4688
4689         if (!kvm_is_ucontrol(vcpu->kvm)) {
4690                 rc = kvm_s390_deliver_pending_interrupts(vcpu);
4691                 if (rc || guestdbg_exit_pending(vcpu))
4692                         return rc;
4693         }
4694
4695         rc = kvm_s390_handle_requests(vcpu);
4696         if (rc)
4697                 return rc;
4698
4699         if (guestdbg_enabled(vcpu)) {
4700                 kvm_s390_backup_guest_per_regs(vcpu);
4701                 kvm_s390_patch_guest_per_regs(vcpu);
4702         }
4703
4704         clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
4705
4706         vcpu->arch.sie_block->icptcode = 0;
4707         cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
4708         VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
4709         trace_kvm_s390_sie_enter(vcpu, cpuflags);
4710
4711         return 0;
4712 }
4713
4714 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
4715 {
4716         struct kvm_s390_pgm_info pgm_info = {
4717                 .code = PGM_ADDRESSING,
4718         };
4719         u8 opcode, ilen;
4720         int rc;
4721
4722         VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
4723         trace_kvm_s390_sie_fault(vcpu);
4724
4725         /*
4726          * We want to inject an addressing exception, which is defined as a
4727          * suppressing or terminating exception. However, since we came here
4728          * by a DAT access exception, the PSW still points to the faulting
4729          * instruction since DAT exceptions are nullifying. So we've got
4730          * to look up the current opcode to get the length of the instruction
4731          * to be able to forward the PSW.
4732          */
4733         rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
4734         ilen = insn_length(opcode);
4735         if (rc < 0) {
4736                 return rc;
4737         } else if (rc) {
4738                 /* Instruction-Fetching Exceptions - we can't detect the ilen.
4739                  * Forward by arbitrary ilc, injection will take care of
4740                  * nullification if necessary.
4741                  */
4742                 pgm_info = vcpu->arch.pgm;
4743                 ilen = 4;
4744         }
4745         pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
4746         kvm_s390_forward_psw(vcpu, ilen);
4747         return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
4748 }
4749
4750 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
4751 {
4752         struct mcck_volatile_info *mcck_info;
4753         struct sie_page *sie_page;
4754
4755         VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
4756                    vcpu->arch.sie_block->icptcode);
4757         trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
4758
4759         if (guestdbg_enabled(vcpu))
4760                 kvm_s390_restore_guest_per_regs(vcpu);
4761
4762         vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
4763         vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
4764
4765         if (exit_reason == -EINTR) {
4766                 VCPU_EVENT(vcpu, 3, "%s", "machine check");
4767                 sie_page = container_of(vcpu->arch.sie_block,
4768                                         struct sie_page, sie_block);
4769                 mcck_info = &sie_page->mcck_info;
4770                 kvm_s390_reinject_machine_check(vcpu, mcck_info);
4771                 return 0;
4772         }
4773
4774         if (vcpu->arch.sie_block->icptcode > 0) {
4775                 int rc = kvm_handle_sie_intercept(vcpu);
4776
4777                 if (rc != -EOPNOTSUPP)
4778                         return rc;
4779                 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
4780                 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
4781                 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
4782                 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
4783                 return -EREMOTE;
4784         } else if (exit_reason != -EFAULT) {
4785                 vcpu->stat.exit_null++;
4786                 return 0;
4787         } else if (kvm_is_ucontrol(vcpu->kvm)) {
4788                 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
4789                 vcpu->run->s390_ucontrol.trans_exc_code =
4790                                                 current->thread.gmap_addr;
4791                 vcpu->run->s390_ucontrol.pgm_code = 0x10;
4792                 return -EREMOTE;
4793         } else if (current->thread.gmap_pfault) {
4794                 trace_kvm_s390_major_guest_pfault(vcpu);
4795                 current->thread.gmap_pfault = 0;
4796                 if (kvm_arch_setup_async_pf(vcpu))
4797                         return 0;
4798                 vcpu->stat.pfault_sync++;
4799                 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
4800         }
4801         return vcpu_post_run_fault_in_sie(vcpu);
4802 }
4803
4804 #define PSW_INT_MASK (PSW_MASK_EXT | PSW_MASK_IO | PSW_MASK_MCHECK)
4805 static int __vcpu_run(struct kvm_vcpu *vcpu)
4806 {
4807         int rc, exit_reason;
4808         struct sie_page *sie_page = (struct sie_page *)vcpu->arch.sie_block;
4809
4810         /*
4811          * We try to hold kvm->srcu during most of vcpu_run (except when run-
4812          * ning the guest), so that memslots (and other stuff) are protected
4813          */
4814         kvm_vcpu_srcu_read_lock(vcpu);
4815
4816         do {
4817                 rc = vcpu_pre_run(vcpu);
4818                 if (rc || guestdbg_exit_pending(vcpu))
4819                         break;
4820
4821                 kvm_vcpu_srcu_read_unlock(vcpu);
4822                 /*
4823                  * As PF_VCPU will be used in fault handler, between
4824                  * guest_enter and guest_exit should be no uaccess.
4825                  */
4826                 local_irq_disable();
4827                 guest_enter_irqoff();
4828                 __disable_cpu_timer_accounting(vcpu);
4829                 local_irq_enable();
4830                 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
4831                         memcpy(sie_page->pv_grregs,
4832                                vcpu->run->s.regs.gprs,
4833                                sizeof(sie_page->pv_grregs));
4834                 }
4835                 if (test_cpu_flag(CIF_FPU))
4836                         load_fpu_regs();
4837                 exit_reason = sie64a(vcpu->arch.sie_block,
4838                                      vcpu->run->s.regs.gprs);
4839                 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
4840                         memcpy(vcpu->run->s.regs.gprs,
4841                                sie_page->pv_grregs,
4842                                sizeof(sie_page->pv_grregs));
4843                         /*
4844                          * We're not allowed to inject interrupts on intercepts
4845                          * that leave the guest state in an "in-between" state
4846                          * where the next SIE entry will do a continuation.
4847                          * Fence interrupts in our "internal" PSW.
4848                          */
4849                         if (vcpu->arch.sie_block->icptcode == ICPT_PV_INSTR ||
4850                             vcpu->arch.sie_block->icptcode == ICPT_PV_PREF) {
4851                                 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
4852                         }
4853                 }
4854                 local_irq_disable();
4855                 __enable_cpu_timer_accounting(vcpu);
4856                 guest_exit_irqoff();
4857                 local_irq_enable();
4858                 kvm_vcpu_srcu_read_lock(vcpu);
4859
4860                 rc = vcpu_post_run(vcpu, exit_reason);
4861         } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
4862
4863         kvm_vcpu_srcu_read_unlock(vcpu);
4864         return rc;
4865 }
4866
4867 static void sync_regs_fmt2(struct kvm_vcpu *vcpu)
4868 {
4869         struct kvm_run *kvm_run = vcpu->run;
4870         struct runtime_instr_cb *riccb;
4871         struct gs_cb *gscb;
4872
4873         riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
4874         gscb = (struct gs_cb *) &kvm_run->s.regs.gscb;
4875         vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
4876         vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
4877         if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
4878                 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
4879                 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
4880                 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
4881         }
4882         if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
4883                 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
4884                 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
4885                 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
4886                 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4887                         kvm_clear_async_pf_completion_queue(vcpu);
4888         }
4889         if (kvm_run->kvm_dirty_regs & KVM_SYNC_DIAG318) {
4890                 vcpu->arch.diag318_info.val = kvm_run->s.regs.diag318;
4891                 vcpu->arch.sie_block->cpnc = vcpu->arch.diag318_info.cpnc;
4892                 VCPU_EVENT(vcpu, 3, "setting cpnc to %d", vcpu->arch.diag318_info.cpnc);
4893         }
4894         /*
4895          * If userspace sets the riccb (e.g. after migration) to a valid state,
4896          * we should enable RI here instead of doing the lazy enablement.
4897          */
4898         if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
4899             test_kvm_facility(vcpu->kvm, 64) &&
4900             riccb->v &&
4901             !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) {
4902                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)");
4903                 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
4904         }
4905         /*
4906          * If userspace sets the gscb (e.g. after migration) to non-zero,
4907          * we should enable GS here instead of doing the lazy enablement.
4908          */
4909         if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) &&
4910             test_kvm_facility(vcpu->kvm, 133) &&
4911             gscb->gssm &&
4912             !vcpu->arch.gs_enabled) {
4913                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)");
4914                 vcpu->arch.sie_block->ecb |= ECB_GS;
4915                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
4916                 vcpu->arch.gs_enabled = 1;
4917         }
4918         if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) &&
4919             test_kvm_facility(vcpu->kvm, 82)) {
4920                 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4921                 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0;
4922         }
4923         if (MACHINE_HAS_GS) {
4924                 preempt_disable();
4925                 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
4926                 if (current->thread.gs_cb) {
4927                         vcpu->arch.host_gscb = current->thread.gs_cb;
4928                         save_gs_cb(vcpu->arch.host_gscb);
4929                 }
4930                 if (vcpu->arch.gs_enabled) {
4931                         current->thread.gs_cb = (struct gs_cb *)
4932                                                 &vcpu->run->s.regs.gscb;
4933                         restore_gs_cb(current->thread.gs_cb);
4934                 }
4935                 preempt_enable();
4936         }
4937         /* SIE will load etoken directly from SDNX and therefore kvm_run */
4938 }
4939
4940 static void sync_regs(struct kvm_vcpu *vcpu)
4941 {
4942         struct kvm_run *kvm_run = vcpu->run;
4943
4944         if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
4945                 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
4946         if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
4947                 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
4948                 /* some control register changes require a tlb flush */
4949                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4950         }
4951         if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
4952                 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
4953                 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
4954         }
4955         save_access_regs(vcpu->arch.host_acrs);
4956         restore_access_regs(vcpu->run->s.regs.acrs);
4957         /* save host (userspace) fprs/vrs */
4958         save_fpu_regs();
4959         vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
4960         vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
4961         if (MACHINE_HAS_VX)
4962                 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
4963         else
4964                 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
4965         current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
4966         if (test_fp_ctl(current->thread.fpu.fpc))
4967                 /* User space provided an invalid FPC, let's clear it */
4968                 current->thread.fpu.fpc = 0;
4969
4970         /* Sync fmt2 only data */
4971         if (likely(!kvm_s390_pv_cpu_is_protected(vcpu))) {
4972                 sync_regs_fmt2(vcpu);
4973         } else {
4974                 /*
4975                  * In several places we have to modify our internal view to
4976                  * not do things that are disallowed by the ultravisor. For
4977                  * example we must not inject interrupts after specific exits
4978                  * (e.g. 112 prefix page not secure). We do this by turning
4979                  * off the machine check, external and I/O interrupt bits
4980                  * of our PSW copy. To avoid getting validity intercepts, we
4981                  * do only accept the condition code from userspace.
4982                  */
4983                 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_CC;
4984                 vcpu->arch.sie_block->gpsw.mask |= kvm_run->psw_mask &
4985                                                    PSW_MASK_CC;
4986         }
4987
4988         kvm_run->kvm_dirty_regs = 0;
4989 }
4990
4991 static void store_regs_fmt2(struct kvm_vcpu *vcpu)
4992 {
4993         struct kvm_run *kvm_run = vcpu->run;
4994
4995         kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
4996         kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
4997         kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
4998         kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC;
4999         kvm_run->s.regs.diag318 = vcpu->arch.diag318_info.val;
5000         if (MACHINE_HAS_GS) {
5001                 preempt_disable();
5002                 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
5003                 if (vcpu->arch.gs_enabled)
5004                         save_gs_cb(current->thread.gs_cb);
5005                 current->thread.gs_cb = vcpu->arch.host_gscb;
5006                 restore_gs_cb(vcpu->arch.host_gscb);
5007                 if (!vcpu->arch.host_gscb)
5008                         local_ctl_clear_bit(2, CR2_GUARDED_STORAGE_BIT);
5009                 vcpu->arch.host_gscb = NULL;
5010                 preempt_enable();
5011         }
5012         /* SIE will save etoken directly into SDNX and therefore kvm_run */
5013 }
5014
5015 static void store_regs(struct kvm_vcpu *vcpu)
5016 {
5017         struct kvm_run *kvm_run = vcpu->run;
5018
5019         kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
5020         kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
5021         kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
5022         memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
5023         kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
5024         kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
5025         kvm_run->s.regs.pft = vcpu->arch.pfault_token;
5026         kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
5027         kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
5028         save_access_regs(vcpu->run->s.regs.acrs);
5029         restore_access_regs(vcpu->arch.host_acrs);
5030         /* Save guest register state */
5031         save_fpu_regs();
5032         vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
5033         /* Restore will be done lazily at return */
5034         current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
5035         current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
5036         if (likely(!kvm_s390_pv_cpu_is_protected(vcpu)))
5037                 store_regs_fmt2(vcpu);
5038 }
5039
5040 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
5041 {
5042         struct kvm_run *kvm_run = vcpu->run;
5043         int rc;
5044
5045         /*
5046          * Running a VM while dumping always has the potential to
5047          * produce inconsistent dump data. But for PV vcpus a SIE
5048          * entry while dumping could also lead to a fatal validity
5049          * intercept which we absolutely want to avoid.
5050          */
5051         if (vcpu->kvm->arch.pv.dumping)
5052                 return -EINVAL;
5053
5054         if (kvm_run->immediate_exit)
5055                 return -EINTR;
5056
5057         if (kvm_run->kvm_valid_regs & ~KVM_SYNC_S390_VALID_FIELDS ||
5058             kvm_run->kvm_dirty_regs & ~KVM_SYNC_S390_VALID_FIELDS)
5059                 return -EINVAL;
5060
5061         vcpu_load(vcpu);
5062
5063         if (guestdbg_exit_pending(vcpu)) {
5064                 kvm_s390_prepare_debug_exit(vcpu);
5065                 rc = 0;
5066                 goto out;
5067         }
5068
5069         kvm_sigset_activate(vcpu);
5070
5071         /*
5072          * no need to check the return value of vcpu_start as it can only have
5073          * an error for protvirt, but protvirt means user cpu state
5074          */
5075         if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
5076                 kvm_s390_vcpu_start(vcpu);
5077         } else if (is_vcpu_stopped(vcpu)) {
5078                 pr_err_ratelimited("can't run stopped vcpu %d\n",
5079                                    vcpu->vcpu_id);
5080                 rc = -EINVAL;
5081                 goto out;
5082         }
5083
5084         sync_regs(vcpu);
5085         enable_cpu_timer_accounting(vcpu);
5086
5087         might_fault();
5088         rc = __vcpu_run(vcpu);
5089
5090         if (signal_pending(current) && !rc) {
5091                 kvm_run->exit_reason = KVM_EXIT_INTR;
5092                 rc = -EINTR;
5093         }
5094
5095         if (guestdbg_exit_pending(vcpu) && !rc)  {
5096                 kvm_s390_prepare_debug_exit(vcpu);
5097                 rc = 0;
5098         }
5099
5100         if (rc == -EREMOTE) {
5101                 /* userspace support is needed, kvm_run has been prepared */
5102                 rc = 0;
5103         }
5104
5105         disable_cpu_timer_accounting(vcpu);
5106         store_regs(vcpu);
5107
5108         kvm_sigset_deactivate(vcpu);
5109
5110         vcpu->stat.exit_userspace++;
5111 out:
5112         vcpu_put(vcpu);
5113         return rc;
5114 }
5115
5116 /*
5117  * store status at address
5118  * we use have two special cases:
5119  * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
5120  * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
5121  */
5122 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
5123 {
5124         unsigned char archmode = 1;
5125         freg_t fprs[NUM_FPRS];
5126         unsigned int px;
5127         u64 clkcomp, cputm;
5128         int rc;
5129
5130         px = kvm_s390_get_prefix(vcpu);
5131         if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
5132                 if (write_guest_abs(vcpu, 163, &archmode, 1))
5133                         return -EFAULT;
5134                 gpa = 0;
5135         } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
5136                 if (write_guest_real(vcpu, 163, &archmode, 1))
5137                         return -EFAULT;
5138                 gpa = px;
5139         } else
5140                 gpa -= __LC_FPREGS_SAVE_AREA;
5141
5142         /* manually convert vector registers if necessary */
5143         if (MACHINE_HAS_VX) {
5144                 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
5145                 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5146                                      fprs, 128);
5147         } else {
5148                 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5149                                      vcpu->run->s.regs.fprs, 128);
5150         }
5151         rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
5152                               vcpu->run->s.regs.gprs, 128);
5153         rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
5154                               &vcpu->arch.sie_block->gpsw, 16);
5155         rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
5156                               &px, 4);
5157         rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
5158                               &vcpu->run->s.regs.fpc, 4);
5159         rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
5160                               &vcpu->arch.sie_block->todpr, 4);
5161         cputm = kvm_s390_get_cpu_timer(vcpu);
5162         rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
5163                               &cputm, 8);
5164         clkcomp = vcpu->arch.sie_block->ckc >> 8;
5165         rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
5166                               &clkcomp, 8);
5167         rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
5168                               &vcpu->run->s.regs.acrs, 64);
5169         rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
5170                               &vcpu->arch.sie_block->gcr, 128);
5171         return rc ? -EFAULT : 0;
5172 }
5173
5174 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
5175 {
5176         /*
5177          * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
5178          * switch in the run ioctl. Let's update our copies before we save
5179          * it into the save area
5180          */
5181         save_fpu_regs();
5182         vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
5183         save_access_regs(vcpu->run->s.regs.acrs);
5184
5185         return kvm_s390_store_status_unloaded(vcpu, addr);
5186 }
5187
5188 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5189 {
5190         kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
5191         kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
5192 }
5193
5194 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
5195 {
5196         unsigned long i;
5197         struct kvm_vcpu *vcpu;
5198
5199         kvm_for_each_vcpu(i, vcpu, kvm) {
5200                 __disable_ibs_on_vcpu(vcpu);
5201         }
5202 }
5203
5204 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5205 {
5206         if (!sclp.has_ibs)
5207                 return;
5208         kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
5209         kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
5210 }
5211
5212 int kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
5213 {
5214         int i, online_vcpus, r = 0, started_vcpus = 0;
5215
5216         if (!is_vcpu_stopped(vcpu))
5217                 return 0;
5218
5219         trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
5220         /* Only one cpu at a time may enter/leave the STOPPED state. */
5221         spin_lock(&vcpu->kvm->arch.start_stop_lock);
5222         online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5223
5224         /* Let's tell the UV that we want to change into the operating state */
5225         if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5226                 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR);
5227                 if (r) {
5228                         spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5229                         return r;
5230                 }
5231         }
5232
5233         for (i = 0; i < online_vcpus; i++) {
5234                 if (!is_vcpu_stopped(kvm_get_vcpu(vcpu->kvm, i)))
5235                         started_vcpus++;
5236         }
5237
5238         if (started_vcpus == 0) {
5239                 /* we're the only active VCPU -> speed it up */
5240                 __enable_ibs_on_vcpu(vcpu);
5241         } else if (started_vcpus == 1) {
5242                 /*
5243                  * As we are starting a second VCPU, we have to disable
5244                  * the IBS facility on all VCPUs to remove potentially
5245                  * outstanding ENABLE requests.
5246                  */
5247                 __disable_ibs_on_all_vcpus(vcpu->kvm);
5248         }
5249
5250         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED);
5251         /*
5252          * The real PSW might have changed due to a RESTART interpreted by the
5253          * ultravisor. We block all interrupts and let the next sie exit
5254          * refresh our view.
5255          */
5256         if (kvm_s390_pv_cpu_is_protected(vcpu))
5257                 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
5258         /*
5259          * Another VCPU might have used IBS while we were offline.
5260          * Let's play safe and flush the VCPU at startup.
5261          */
5262         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5263         spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5264         return 0;
5265 }
5266
5267 int kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
5268 {
5269         int i, online_vcpus, r = 0, started_vcpus = 0;
5270         struct kvm_vcpu *started_vcpu = NULL;
5271
5272         if (is_vcpu_stopped(vcpu))
5273                 return 0;
5274
5275         trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
5276         /* Only one cpu at a time may enter/leave the STOPPED state. */
5277         spin_lock(&vcpu->kvm->arch.start_stop_lock);
5278         online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5279
5280         /* Let's tell the UV that we want to change into the stopped state */
5281         if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5282                 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_STP);
5283                 if (r) {
5284                         spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5285                         return r;
5286                 }
5287         }
5288
5289         /*
5290          * Set the VCPU to STOPPED and THEN clear the interrupt flag,
5291          * now that the SIGP STOP and SIGP STOP AND STORE STATUS orders
5292          * have been fully processed. This will ensure that the VCPU
5293          * is kept BUSY if another VCPU is inquiring with SIGP SENSE.
5294          */
5295         kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED);
5296         kvm_s390_clear_stop_irq(vcpu);
5297
5298         __disable_ibs_on_vcpu(vcpu);
5299
5300         for (i = 0; i < online_vcpus; i++) {
5301                 struct kvm_vcpu *tmp = kvm_get_vcpu(vcpu->kvm, i);
5302
5303                 if (!is_vcpu_stopped(tmp)) {
5304                         started_vcpus++;
5305                         started_vcpu = tmp;
5306                 }
5307         }
5308
5309         if (started_vcpus == 1) {
5310                 /*
5311                  * As we only have one VCPU left, we want to enable the
5312                  * IBS facility for that VCPU to speed it up.
5313                  */
5314                 __enable_ibs_on_vcpu(started_vcpu);
5315         }
5316
5317         spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5318         return 0;
5319 }
5320
5321 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5322                                      struct kvm_enable_cap *cap)
5323 {
5324         int r;
5325
5326         if (cap->flags)
5327                 return -EINVAL;
5328
5329         switch (cap->cap) {
5330         case KVM_CAP_S390_CSS_SUPPORT:
5331                 if (!vcpu->kvm->arch.css_support) {
5332                         vcpu->kvm->arch.css_support = 1;
5333                         VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
5334                         trace_kvm_s390_enable_css(vcpu->kvm);
5335                 }
5336                 r = 0;
5337                 break;
5338         default:
5339                 r = -EINVAL;
5340                 break;
5341         }
5342         return r;
5343 }
5344
5345 static long kvm_s390_vcpu_sida_op(struct kvm_vcpu *vcpu,
5346                                   struct kvm_s390_mem_op *mop)
5347 {
5348         void __user *uaddr = (void __user *)mop->buf;
5349         void *sida_addr;
5350         int r = 0;
5351
5352         if (mop->flags || !mop->size)
5353                 return -EINVAL;
5354         if (mop->size + mop->sida_offset < mop->size)
5355                 return -EINVAL;
5356         if (mop->size + mop->sida_offset > sida_size(vcpu->arch.sie_block))
5357                 return -E2BIG;
5358         if (!kvm_s390_pv_cpu_is_protected(vcpu))
5359                 return -EINVAL;
5360
5361         sida_addr = (char *)sida_addr(vcpu->arch.sie_block) + mop->sida_offset;
5362
5363         switch (mop->op) {
5364         case KVM_S390_MEMOP_SIDA_READ:
5365                 if (copy_to_user(uaddr, sida_addr, mop->size))
5366                         r = -EFAULT;
5367
5368                 break;
5369         case KVM_S390_MEMOP_SIDA_WRITE:
5370                 if (copy_from_user(sida_addr, uaddr, mop->size))
5371                         r = -EFAULT;
5372                 break;
5373         }
5374         return r;
5375 }
5376
5377 static long kvm_s390_vcpu_mem_op(struct kvm_vcpu *vcpu,
5378                                  struct kvm_s390_mem_op *mop)
5379 {
5380         void __user *uaddr = (void __user *)mop->buf;
5381         enum gacc_mode acc_mode;
5382         void *tmpbuf = NULL;
5383         int r;
5384
5385         r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_INJECT_EXCEPTION |
5386                                         KVM_S390_MEMOP_F_CHECK_ONLY |
5387                                         KVM_S390_MEMOP_F_SKEY_PROTECTION);
5388         if (r)
5389                 return r;
5390         if (mop->ar >= NUM_ACRS)
5391                 return -EINVAL;
5392         if (kvm_s390_pv_cpu_is_protected(vcpu))
5393                 return -EINVAL;
5394         if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
5395                 tmpbuf = vmalloc(mop->size);
5396                 if (!tmpbuf)
5397                         return -ENOMEM;
5398         }
5399
5400         acc_mode = mop->op == KVM_S390_MEMOP_LOGICAL_READ ? GACC_FETCH : GACC_STORE;
5401         if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
5402                 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size,
5403                                     acc_mode, mop->key);
5404                 goto out_inject;
5405         }
5406         if (acc_mode == GACC_FETCH) {
5407                 r = read_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5408                                         mop->size, mop->key);
5409                 if (r)
5410                         goto out_inject;
5411                 if (copy_to_user(uaddr, tmpbuf, mop->size)) {
5412                         r = -EFAULT;
5413                         goto out_free;
5414                 }
5415         } else {
5416                 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
5417                         r = -EFAULT;
5418                         goto out_free;
5419                 }
5420                 r = write_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5421                                          mop->size, mop->key);
5422         }
5423
5424 out_inject:
5425         if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
5426                 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
5427
5428 out_free:
5429         vfree(tmpbuf);
5430         return r;
5431 }
5432
5433 static long kvm_s390_vcpu_memsida_op(struct kvm_vcpu *vcpu,
5434                                      struct kvm_s390_mem_op *mop)
5435 {
5436         int r, srcu_idx;
5437
5438         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5439
5440         switch (mop->op) {
5441         case KVM_S390_MEMOP_LOGICAL_READ:
5442         case KVM_S390_MEMOP_LOGICAL_WRITE:
5443                 r = kvm_s390_vcpu_mem_op(vcpu, mop);
5444                 break;
5445         case KVM_S390_MEMOP_SIDA_READ:
5446         case KVM_S390_MEMOP_SIDA_WRITE:
5447                 /* we are locked against sida going away by the vcpu->mutex */
5448                 r = kvm_s390_vcpu_sida_op(vcpu, mop);
5449                 break;
5450         default:
5451                 r = -EINVAL;
5452         }
5453
5454         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
5455         return r;
5456 }
5457
5458 long kvm_arch_vcpu_async_ioctl(struct file *filp,
5459                                unsigned int ioctl, unsigned long arg)
5460 {
5461         struct kvm_vcpu *vcpu = filp->private_data;
5462         void __user *argp = (void __user *)arg;
5463         int rc;
5464
5465         switch (ioctl) {
5466         case KVM_S390_IRQ: {
5467                 struct kvm_s390_irq s390irq;
5468
5469                 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
5470                         return -EFAULT;
5471                 rc = kvm_s390_inject_vcpu(vcpu, &s390irq);
5472                 break;
5473         }
5474         case KVM_S390_INTERRUPT: {
5475                 struct kvm_s390_interrupt s390int;
5476                 struct kvm_s390_irq s390irq = {};
5477
5478                 if (copy_from_user(&s390int, argp, sizeof(s390int)))
5479                         return -EFAULT;
5480                 if (s390int_to_s390irq(&s390int, &s390irq))
5481                         return -EINVAL;
5482                 rc = kvm_s390_inject_vcpu(vcpu, &s390irq);
5483                 break;
5484         }
5485         default:
5486                 rc = -ENOIOCTLCMD;
5487                 break;
5488         }
5489
5490         /*
5491          * To simplify single stepping of userspace-emulated instructions,
5492          * KVM_EXIT_S390_SIEIC exit sets KVM_GUESTDBG_EXIT_PENDING (see
5493          * should_handle_per_ifetch()). However, if userspace emulation injects
5494          * an interrupt, it needs to be cleared, so that KVM_EXIT_DEBUG happens
5495          * after (and not before) the interrupt delivery.
5496          */
5497         if (!rc)
5498                 vcpu->guest_debug &= ~KVM_GUESTDBG_EXIT_PENDING;
5499
5500         return rc;
5501 }
5502
5503 static int kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu *vcpu,
5504                                         struct kvm_pv_cmd *cmd)
5505 {
5506         struct kvm_s390_pv_dmp dmp;
5507         void *data;
5508         int ret;
5509
5510         /* Dump initialization is a prerequisite */
5511         if (!vcpu->kvm->arch.pv.dumping)
5512                 return -EINVAL;
5513
5514         if (copy_from_user(&dmp, (__u8 __user *)cmd->data, sizeof(dmp)))
5515                 return -EFAULT;
5516
5517         /* We only handle this subcmd right now */
5518         if (dmp.subcmd != KVM_PV_DUMP_CPU)
5519                 return -EINVAL;
5520
5521         /* CPU dump length is the same as create cpu storage donation. */
5522         if (dmp.buff_len != uv_info.guest_cpu_stor_len)
5523                 return -EINVAL;
5524
5525         data = kvzalloc(uv_info.guest_cpu_stor_len, GFP_KERNEL);
5526         if (!data)
5527                 return -ENOMEM;
5528
5529         ret = kvm_s390_pv_dump_cpu(vcpu, data, &cmd->rc, &cmd->rrc);
5530
5531         VCPU_EVENT(vcpu, 3, "PROTVIRT DUMP CPU %d rc %x rrc %x",
5532                    vcpu->vcpu_id, cmd->rc, cmd->rrc);
5533
5534         if (ret)
5535                 ret = -EINVAL;
5536
5537         /* On success copy over the dump data */
5538         if (!ret && copy_to_user((__u8 __user *)dmp.buff_addr, data, uv_info.guest_cpu_stor_len))
5539                 ret = -EFAULT;
5540
5541         kvfree(data);
5542         return ret;
5543 }
5544
5545 long kvm_arch_vcpu_ioctl(struct file *filp,
5546                          unsigned int ioctl, unsigned long arg)
5547 {
5548         struct kvm_vcpu *vcpu = filp->private_data;
5549         void __user *argp = (void __user *)arg;
5550         int idx;
5551         long r;
5552         u16 rc, rrc;
5553
5554         vcpu_load(vcpu);
5555
5556         switch (ioctl) {
5557         case KVM_S390_STORE_STATUS:
5558                 idx = srcu_read_lock(&vcpu->kvm->srcu);
5559                 r = kvm_s390_store_status_unloaded(vcpu, arg);
5560                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5561                 break;
5562         case KVM_S390_SET_INITIAL_PSW: {
5563                 psw_t psw;
5564
5565                 r = -EFAULT;
5566                 if (copy_from_user(&psw, argp, sizeof(psw)))
5567                         break;
5568                 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
5569                 break;
5570         }
5571         case KVM_S390_CLEAR_RESET:
5572                 r = 0;
5573                 kvm_arch_vcpu_ioctl_clear_reset(vcpu);
5574                 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5575                         r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5576                                           UVC_CMD_CPU_RESET_CLEAR, &rc, &rrc);
5577                         VCPU_EVENT(vcpu, 3, "PROTVIRT RESET CLEAR VCPU: rc %x rrc %x",
5578                                    rc, rrc);
5579                 }
5580                 break;
5581         case KVM_S390_INITIAL_RESET:
5582                 r = 0;
5583                 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
5584                 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5585                         r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5586                                           UVC_CMD_CPU_RESET_INITIAL,
5587                                           &rc, &rrc);
5588                         VCPU_EVENT(vcpu, 3, "PROTVIRT RESET INITIAL VCPU: rc %x rrc %x",
5589                                    rc, rrc);
5590                 }
5591                 break;
5592         case KVM_S390_NORMAL_RESET:
5593                 r = 0;
5594                 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
5595                 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5596                         r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5597                                           UVC_CMD_CPU_RESET, &rc, &rrc);
5598                         VCPU_EVENT(vcpu, 3, "PROTVIRT RESET NORMAL VCPU: rc %x rrc %x",
5599                                    rc, rrc);
5600                 }
5601                 break;
5602         case KVM_SET_ONE_REG:
5603         case KVM_GET_ONE_REG: {
5604                 struct kvm_one_reg reg;
5605                 r = -EINVAL;
5606                 if (kvm_s390_pv_cpu_is_protected(vcpu))
5607                         break;
5608                 r = -EFAULT;
5609                 if (copy_from_user(&reg, argp, sizeof(reg)))
5610                         break;
5611                 if (ioctl == KVM_SET_ONE_REG)
5612                         r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, &reg);
5613                 else
5614                         r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, &reg);
5615                 break;
5616         }
5617 #ifdef CONFIG_KVM_S390_UCONTROL
5618         case KVM_S390_UCAS_MAP: {
5619                 struct kvm_s390_ucas_mapping ucasmap;
5620
5621                 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5622                         r = -EFAULT;
5623                         break;
5624                 }
5625
5626                 if (!kvm_is_ucontrol(vcpu->kvm)) {
5627                         r = -EINVAL;
5628                         break;
5629                 }
5630
5631                 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
5632                                      ucasmap.vcpu_addr, ucasmap.length);
5633                 break;
5634         }
5635         case KVM_S390_UCAS_UNMAP: {
5636                 struct kvm_s390_ucas_mapping ucasmap;
5637
5638                 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5639                         r = -EFAULT;
5640                         break;
5641                 }
5642
5643                 if (!kvm_is_ucontrol(vcpu->kvm)) {
5644                         r = -EINVAL;
5645                         break;
5646                 }
5647
5648                 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
5649                         ucasmap.length);
5650                 break;
5651         }
5652 #endif
5653         case KVM_S390_VCPU_FAULT: {
5654                 r = gmap_fault(vcpu->arch.gmap, arg, 0);
5655                 break;
5656         }
5657         case KVM_ENABLE_CAP:
5658         {
5659                 struct kvm_enable_cap cap;
5660                 r = -EFAULT;
5661                 if (copy_from_user(&cap, argp, sizeof(cap)))
5662                         break;
5663                 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5664                 break;
5665         }
5666         case KVM_S390_MEM_OP: {
5667                 struct kvm_s390_mem_op mem_op;
5668
5669                 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
5670                         r = kvm_s390_vcpu_memsida_op(vcpu, &mem_op);
5671                 else
5672                         r = -EFAULT;
5673                 break;
5674         }
5675         case KVM_S390_SET_IRQ_STATE: {
5676                 struct kvm_s390_irq_state irq_state;
5677
5678                 r = -EFAULT;
5679                 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5680                         break;
5681                 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
5682                     irq_state.len == 0 ||
5683                     irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
5684                         r = -EINVAL;
5685                         break;
5686                 }
5687                 /* do not use irq_state.flags, it will break old QEMUs */
5688                 r = kvm_s390_set_irq_state(vcpu,
5689                                            (void __user *) irq_state.buf,
5690                                            irq_state.len);
5691                 break;
5692         }
5693         case KVM_S390_GET_IRQ_STATE: {
5694                 struct kvm_s390_irq_state irq_state;
5695
5696                 r = -EFAULT;
5697                 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5698                         break;
5699                 if (irq_state.len == 0) {
5700                         r = -EINVAL;
5701                         break;
5702                 }
5703                 /* do not use irq_state.flags, it will break old QEMUs */
5704                 r = kvm_s390_get_irq_state(vcpu,
5705                                            (__u8 __user *)  irq_state.buf,
5706                                            irq_state.len);
5707                 break;
5708         }
5709         case KVM_S390_PV_CPU_COMMAND: {
5710                 struct kvm_pv_cmd cmd;
5711
5712                 r = -EINVAL;
5713                 if (!is_prot_virt_host())
5714                         break;
5715
5716                 r = -EFAULT;
5717                 if (copy_from_user(&cmd, argp, sizeof(cmd)))
5718                         break;
5719
5720                 r = -EINVAL;
5721                 if (cmd.flags)
5722                         break;
5723
5724                 /* We only handle this cmd right now */
5725                 if (cmd.cmd != KVM_PV_DUMP)
5726                         break;
5727
5728                 r = kvm_s390_handle_pv_vcpu_dump(vcpu, &cmd);
5729
5730                 /* Always copy over UV rc / rrc data */
5731                 if (copy_to_user((__u8 __user *)argp, &cmd.rc,
5732                                  sizeof(cmd.rc) + sizeof(cmd.rrc)))
5733                         r = -EFAULT;
5734                 break;
5735         }
5736         default:
5737                 r = -ENOTTY;
5738         }
5739
5740         vcpu_put(vcpu);
5741         return r;
5742 }
5743
5744 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5745 {
5746 #ifdef CONFIG_KVM_S390_UCONTROL
5747         if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
5748                  && (kvm_is_ucontrol(vcpu->kvm))) {
5749                 vmf->page = virt_to_page(vcpu->arch.sie_block);
5750                 get_page(vmf->page);
5751                 return 0;
5752         }
5753 #endif
5754         return VM_FAULT_SIGBUS;
5755 }
5756
5757 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
5758 {
5759         return true;
5760 }
5761
5762 /* Section: memory related */
5763 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5764                                    const struct kvm_memory_slot *old,
5765                                    struct kvm_memory_slot *new,
5766                                    enum kvm_mr_change change)
5767 {
5768         gpa_t size;
5769
5770         /* When we are protected, we should not change the memory slots */
5771         if (kvm_s390_pv_get_handle(kvm))
5772                 return -EINVAL;
5773
5774         if (change != KVM_MR_DELETE && change != KVM_MR_FLAGS_ONLY) {
5775                 /*
5776                  * A few sanity checks. We can have memory slots which have to be
5777                  * located/ended at a segment boundary (1MB). The memory in userland is
5778                  * ok to be fragmented into various different vmas. It is okay to mmap()
5779                  * and munmap() stuff in this slot after doing this call at any time
5780                  */
5781
5782                 if (new->userspace_addr & 0xffffful)
5783                         return -EINVAL;
5784
5785                 size = new->npages * PAGE_SIZE;
5786                 if (size & 0xffffful)
5787                         return -EINVAL;
5788
5789                 if ((new->base_gfn * PAGE_SIZE) + size > kvm->arch.mem_limit)
5790                         return -EINVAL;
5791         }
5792
5793         if (!kvm->arch.migration_mode)
5794                 return 0;
5795
5796         /*
5797          * Turn off migration mode when:
5798          * - userspace creates a new memslot with dirty logging off,
5799          * - userspace modifies an existing memslot (MOVE or FLAGS_ONLY) and
5800          *   dirty logging is turned off.
5801          * Migration mode expects dirty page logging being enabled to store
5802          * its dirty bitmap.
5803          */
5804         if (change != KVM_MR_DELETE &&
5805             !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
5806                 WARN(kvm_s390_vm_stop_migration(kvm),
5807                      "Failed to stop migration mode");
5808
5809         return 0;
5810 }
5811
5812 void kvm_arch_commit_memory_region(struct kvm *kvm,
5813                                 struct kvm_memory_slot *old,
5814                                 const struct kvm_memory_slot *new,
5815                                 enum kvm_mr_change change)
5816 {
5817         int rc = 0;
5818
5819         switch (change) {
5820         case KVM_MR_DELETE:
5821                 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
5822                                         old->npages * PAGE_SIZE);
5823                 break;
5824         case KVM_MR_MOVE:
5825                 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
5826                                         old->npages * PAGE_SIZE);
5827                 if (rc)
5828                         break;
5829                 fallthrough;
5830         case KVM_MR_CREATE:
5831                 rc = gmap_map_segment(kvm->arch.gmap, new->userspace_addr,
5832                                       new->base_gfn * PAGE_SIZE,
5833                                       new->npages * PAGE_SIZE);
5834                 break;
5835         case KVM_MR_FLAGS_ONLY:
5836                 break;
5837         default:
5838                 WARN(1, "Unknown KVM MR CHANGE: %d\n", change);
5839         }
5840         if (rc)
5841                 pr_warn("failed to commit memory region\n");
5842         return;
5843 }
5844
5845 static inline unsigned long nonhyp_mask(int i)
5846 {
5847         unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
5848
5849         return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
5850 }
5851
5852 static int __init kvm_s390_init(void)
5853 {
5854         int i, r;
5855
5856         if (!sclp.has_sief2) {
5857                 pr_info("SIE is not available\n");
5858                 return -ENODEV;
5859         }
5860
5861         if (nested && hpage) {
5862                 pr_info("A KVM host that supports nesting cannot back its KVM guests with huge pages\n");
5863                 return -EINVAL;
5864         }
5865
5866         for (i = 0; i < 16; i++)
5867                 kvm_s390_fac_base[i] |=
5868                         stfle_fac_list[i] & nonhyp_mask(i);
5869
5870         r = __kvm_s390_init();
5871         if (r)
5872                 return r;
5873
5874         r = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
5875         if (r) {
5876                 __kvm_s390_exit();
5877                 return r;
5878         }
5879         return 0;
5880 }
5881
5882 static void __exit kvm_s390_exit(void)
5883 {
5884         kvm_exit();
5885
5886         __kvm_s390_exit();
5887 }
5888
5889 module_init(kvm_s390_init);
5890 module_exit(kvm_s390_exit);
5891
5892 /*
5893  * Enable autoloading of the kvm module.
5894  * Note that we add the module alias here instead of virt/kvm/kvm_main.c
5895  * since x86 takes a different approach.
5896  */
5897 #include <linux/miscdevice.h>
5898 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5899 MODULE_ALIAS("devname:kvm");