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