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 pr_fmt(fmt) "kvm-s390: " fmt 14 15 #include <linux/compiler.h> 16 #include <linux/entry-virt.h> 17 #include <linux/export.h> 18 #include <linux/err.h> 19 #include <linux/fs.h> 20 #include <linux/hrtimer.h> 21 #include <linux/init.h> 22 #include <linux/kvm.h> 23 #include <linux/kvm_host.h> 24 #include <linux/mman.h> 25 #include <linux/module.h> 26 #include <linux/moduleparam.h> 27 #include <linux/cpufeature.h> 28 #include <linux/random.h> 29 #include <linux/slab.h> 30 #include <linux/timer.h> 31 #include <linux/vmalloc.h> 32 #include <linux/bitmap.h> 33 #include <linux/sched/signal.h> 34 #include <linux/string.h> 35 #include <linux/pgtable.h> 36 #include <linux/mmu_notifier.h> 37 38 #include <asm/access-regs.h> 39 #include <asm/asm-offsets.h> 40 #include <asm/lowcore.h> 41 #include <asm/machine.h> 42 #include <asm/stp.h> 43 #include <asm/gmap_helpers.h> 44 #include <asm/nmi.h> 45 #include <asm/isc.h> 46 #include <asm/sclp.h> 47 #include <asm/cpacf.h> 48 #include <asm/timex.h> 49 #include <asm/asm.h> 50 #include <asm/fpu.h> 51 #include <asm/ap.h> 52 #include <asm/uv.h> 53 #include "s390.h" 54 #include "gaccess.h" 55 #include "gmap.h" 56 #include "faultin.h" 57 #include "pci.h" 58 #include "kvm_mmu.h" 59 60 #define CREATE_TRACE_POINTS 61 #include "trace.h" 62 #include "trace-s390.h" 63 64 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */ 65 #define LOCAL_IRQS 32 66 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \ 67 (KVM_MAX_VCPUS + LOCAL_IRQS)) 68 69 const struct kvm_stats_desc kvm_vm_stats_desc[] = { 70 KVM_GENERIC_VM_STATS(), 71 STATS_DESC_COUNTER(VM, inject_io), 72 STATS_DESC_COUNTER(VM, io_390_adapter_map), 73 STATS_DESC_COUNTER(VM, io_390_adapter_unmap), 74 STATS_DESC_COUNTER(VM, io_390_inatomic), 75 STATS_DESC_COUNTER(VM, io_flic_inject_airq), 76 STATS_DESC_COUNTER(VM, io_set_adapter_int), 77 STATS_DESC_COUNTER(VM, io_390_inatomic_no_inject), 78 STATS_DESC_COUNTER(VM, inject_float_mchk), 79 STATS_DESC_COUNTER(VM, inject_pfault_done), 80 STATS_DESC_COUNTER(VM, inject_service_signal), 81 STATS_DESC_COUNTER(VM, inject_virtio), 82 STATS_DESC_COUNTER(VM, aen_forward), 83 STATS_DESC_COUNTER(VM, gmap_shadow_reuse), 84 STATS_DESC_COUNTER(VM, gmap_shadow_create), 85 STATS_DESC_COUNTER(VM, gmap_shadow_r1_entry), 86 STATS_DESC_COUNTER(VM, gmap_shadow_r2_entry), 87 STATS_DESC_COUNTER(VM, gmap_shadow_r3_entry), 88 STATS_DESC_COUNTER(VM, gmap_shadow_sg_entry), 89 STATS_DESC_COUNTER(VM, gmap_shadow_pg_entry), 90 }; 91 92 const struct kvm_stats_header kvm_vm_stats_header = { 93 .name_size = KVM_STATS_NAME_SIZE, 94 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc), 95 .id_offset = sizeof(struct kvm_stats_header), 96 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE, 97 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE + 98 sizeof(kvm_vm_stats_desc), 99 }; 100 101 const struct kvm_stats_desc kvm_vcpu_stats_desc[] = { 102 KVM_GENERIC_VCPU_STATS(), 103 STATS_DESC_COUNTER(VCPU, exit_userspace), 104 STATS_DESC_COUNTER(VCPU, exit_null), 105 STATS_DESC_COUNTER(VCPU, exit_external_request), 106 STATS_DESC_COUNTER(VCPU, exit_io_request), 107 STATS_DESC_COUNTER(VCPU, exit_external_interrupt), 108 STATS_DESC_COUNTER(VCPU, exit_stop_request), 109 STATS_DESC_COUNTER(VCPU, exit_validity), 110 STATS_DESC_COUNTER(VCPU, exit_instruction), 111 STATS_DESC_COUNTER(VCPU, exit_pei), 112 STATS_DESC_COUNTER(VCPU, halt_no_poll_steal), 113 STATS_DESC_COUNTER(VCPU, instruction_lctl), 114 STATS_DESC_COUNTER(VCPU, instruction_lctlg), 115 STATS_DESC_COUNTER(VCPU, instruction_stctl), 116 STATS_DESC_COUNTER(VCPU, instruction_stctg), 117 STATS_DESC_COUNTER(VCPU, exit_program_interruption), 118 STATS_DESC_COUNTER(VCPU, exit_instr_and_program), 119 STATS_DESC_COUNTER(VCPU, exit_operation_exception), 120 STATS_DESC_COUNTER(VCPU, deliver_ckc), 121 STATS_DESC_COUNTER(VCPU, deliver_cputm), 122 STATS_DESC_COUNTER(VCPU, deliver_external_call), 123 STATS_DESC_COUNTER(VCPU, deliver_emergency_signal), 124 STATS_DESC_COUNTER(VCPU, deliver_service_signal), 125 STATS_DESC_COUNTER(VCPU, deliver_virtio), 126 STATS_DESC_COUNTER(VCPU, deliver_stop_signal), 127 STATS_DESC_COUNTER(VCPU, deliver_prefix_signal), 128 STATS_DESC_COUNTER(VCPU, deliver_restart_signal), 129 STATS_DESC_COUNTER(VCPU, deliver_program), 130 STATS_DESC_COUNTER(VCPU, deliver_io), 131 STATS_DESC_COUNTER(VCPU, deliver_machine_check), 132 STATS_DESC_COUNTER(VCPU, exit_wait_state), 133 STATS_DESC_COUNTER(VCPU, inject_ckc), 134 STATS_DESC_COUNTER(VCPU, inject_cputm), 135 STATS_DESC_COUNTER(VCPU, inject_external_call), 136 STATS_DESC_COUNTER(VCPU, inject_emergency_signal), 137 STATS_DESC_COUNTER(VCPU, inject_mchk), 138 STATS_DESC_COUNTER(VCPU, inject_pfault_init), 139 STATS_DESC_COUNTER(VCPU, inject_program), 140 STATS_DESC_COUNTER(VCPU, inject_restart), 141 STATS_DESC_COUNTER(VCPU, inject_set_prefix), 142 STATS_DESC_COUNTER(VCPU, inject_stop_signal), 143 STATS_DESC_COUNTER(VCPU, instruction_epsw), 144 STATS_DESC_COUNTER(VCPU, instruction_gs), 145 STATS_DESC_COUNTER(VCPU, instruction_io_other), 146 STATS_DESC_COUNTER(VCPU, instruction_lpsw), 147 STATS_DESC_COUNTER(VCPU, instruction_lpswe), 148 STATS_DESC_COUNTER(VCPU, instruction_lpswey), 149 STATS_DESC_COUNTER(VCPU, instruction_pfmf), 150 STATS_DESC_COUNTER(VCPU, instruction_ptff), 151 STATS_DESC_COUNTER(VCPU, instruction_sck), 152 STATS_DESC_COUNTER(VCPU, instruction_sckpf), 153 STATS_DESC_COUNTER(VCPU, instruction_stidp), 154 STATS_DESC_COUNTER(VCPU, instruction_spx), 155 STATS_DESC_COUNTER(VCPU, instruction_stpx), 156 STATS_DESC_COUNTER(VCPU, instruction_stap), 157 STATS_DESC_COUNTER(VCPU, instruction_iske), 158 STATS_DESC_COUNTER(VCPU, instruction_ri), 159 STATS_DESC_COUNTER(VCPU, instruction_rrbe), 160 STATS_DESC_COUNTER(VCPU, instruction_sske), 161 STATS_DESC_COUNTER(VCPU, instruction_ipte_interlock), 162 STATS_DESC_COUNTER(VCPU, instruction_stsi), 163 STATS_DESC_COUNTER(VCPU, instruction_stfl), 164 STATS_DESC_COUNTER(VCPU, instruction_tb), 165 STATS_DESC_COUNTER(VCPU, instruction_tpi), 166 STATS_DESC_COUNTER(VCPU, instruction_tprot), 167 STATS_DESC_COUNTER(VCPU, instruction_tsch), 168 STATS_DESC_COUNTER(VCPU, instruction_sie), 169 STATS_DESC_COUNTER(VCPU, instruction_essa), 170 STATS_DESC_COUNTER(VCPU, instruction_sthyi), 171 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense), 172 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense_running), 173 STATS_DESC_COUNTER(VCPU, instruction_sigp_external_call), 174 STATS_DESC_COUNTER(VCPU, instruction_sigp_emergency), 175 STATS_DESC_COUNTER(VCPU, instruction_sigp_cond_emergency), 176 STATS_DESC_COUNTER(VCPU, instruction_sigp_start), 177 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop), 178 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop_store_status), 179 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_status), 180 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_adtl_status), 181 STATS_DESC_COUNTER(VCPU, instruction_sigp_arch), 182 STATS_DESC_COUNTER(VCPU, instruction_sigp_prefix), 183 STATS_DESC_COUNTER(VCPU, instruction_sigp_restart), 184 STATS_DESC_COUNTER(VCPU, instruction_sigp_init_cpu_reset), 185 STATS_DESC_COUNTER(VCPU, instruction_sigp_cpu_reset), 186 STATS_DESC_COUNTER(VCPU, instruction_sigp_unknown), 187 STATS_DESC_COUNTER(VCPU, instruction_diagnose_10), 188 STATS_DESC_COUNTER(VCPU, instruction_diagnose_44), 189 STATS_DESC_COUNTER(VCPU, instruction_diagnose_9c), 190 STATS_DESC_COUNTER(VCPU, diag_9c_ignored), 191 STATS_DESC_COUNTER(VCPU, diag_9c_forward), 192 STATS_DESC_COUNTER(VCPU, instruction_diagnose_258), 193 STATS_DESC_COUNTER(VCPU, instruction_diagnose_308), 194 STATS_DESC_COUNTER(VCPU, instruction_diagnose_500), 195 STATS_DESC_COUNTER(VCPU, instruction_diagnose_other), 196 STATS_DESC_COUNTER(VCPU, pfault_sync), 197 STATS_DESC_COUNTER(VCPU, signal_exits) 198 }; 199 200 const struct kvm_stats_header kvm_vcpu_stats_header = { 201 .name_size = KVM_STATS_NAME_SIZE, 202 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc), 203 .id_offset = sizeof(struct kvm_stats_header), 204 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE, 205 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE + 206 sizeof(kvm_vcpu_stats_desc), 207 }; 208 209 /* allow nested virtualization in KVM (if enabled by user space) */ 210 static int nested; 211 module_param(nested, int, S_IRUGO); 212 MODULE_PARM_DESC(nested, "Nested virtualization support"); 213 214 /* allow 1m huge page guest backing */ 215 static int hpage; 216 module_param(hpage, int, 0444); 217 MODULE_PARM_DESC(hpage, "1m huge page backing support"); 218 219 /* allow 2g huge page guest backing */ 220 static int hpage_2g; 221 module_param(hpage_2g, int, 0444); 222 MODULE_PARM_DESC(hpage_2g, "2g huge page backing support"); 223 224 /* maximum percentage of steal time for polling. >100 is treated like 100 */ 225 static u8 halt_poll_max_steal = 10; 226 module_param(halt_poll_max_steal, byte, 0644); 227 MODULE_PARM_DESC(halt_poll_max_steal, "Maximum percentage of steal time to allow polling"); 228 229 /* if set to true, the GISA will be initialized and used if available */ 230 static bool use_gisa = true; 231 module_param(use_gisa, bool, 0644); 232 MODULE_PARM_DESC(use_gisa, "Use the GISA if the host supports it."); 233 234 /* maximum diag9c forwarding per second */ 235 unsigned int diag9c_forwarding_hz; 236 module_param(diag9c_forwarding_hz, uint, 0644); 237 MODULE_PARM_DESC(diag9c_forwarding_hz, "Maximum diag9c forwarding per second, 0 to turn off"); 238 239 /* 240 * allow asynchronous deinit for protected guests; enable by default since 241 * the feature is opt-in anyway 242 */ 243 static int async_destroy = 1; 244 module_param(async_destroy, int, 0444); 245 MODULE_PARM_DESC(async_destroy, "Asynchronous destroy for protected guests"); 246 247 #define HMFAI_DWORDS 16 248 /* 249 * Base feature mask that defines default mask for facilities. Consists of the 250 * defines in FACILITIES_KVM and the non-hypervisor managed bits. 251 */ 252 static unsigned long kvm_s390_fac_base[HMFAI_DWORDS] = { FACILITIES_KVM }; 253 static_assert(ARRAY_SIZE(((long[]){ FACILITIES_KVM })) <= HMFAI_DWORDS); 254 static_assert(ARRAY_SIZE(kvm_s390_fac_base) <= S390_ARCH_FAC_MASK_SIZE_U64); 255 static_assert(ARRAY_SIZE(kvm_s390_fac_base) <= S390_ARCH_FAC_LIST_SIZE_U64); 256 static_assert(ARRAY_SIZE(kvm_s390_fac_base) <= ARRAY_SIZE(stfle_fac_list)); 257 258 /* 259 * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL 260 * and defines the facilities that can be enabled via a cpu model. 261 */ 262 static const unsigned long kvm_s390_fac_ext[] = { FACILITIES_KVM_CPUMODEL }; 263 static_assert(ARRAY_SIZE(kvm_s390_fac_ext) <= S390_ARCH_FAC_MASK_SIZE_U64); 264 static_assert(ARRAY_SIZE(kvm_s390_fac_ext) <= S390_ARCH_FAC_LIST_SIZE_U64); 265 static_assert(ARRAY_SIZE(kvm_s390_fac_ext) <= ARRAY_SIZE(stfle_fac_list)); 266 267 /* available cpu features supported by kvm */ 268 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS); 269 /* available subfunctions indicated via query / "test bit" */ 270 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc; 271 272 debug_info_t *kvm_s390_dbf; 273 debug_info_t *kvm_s390_dbf_uv; 274 275 /* Section: not file related */ 276 /* forward declarations */ 277 static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta) 278 { 279 u8 delta_idx = 0; 280 281 /* 282 * The TOD jumps by delta, we have to compensate this by adding 283 * -delta to the epoch. 284 */ 285 delta = -delta; 286 287 /* sign-extension - we're adding to signed values below */ 288 if ((s64)delta < 0) 289 delta_idx = -1; 290 291 scb->epoch += delta; 292 if (scb->ecd & ECD_MEF) { 293 scb->epdx += delta_idx; 294 if (scb->epoch < delta) 295 scb->epdx += 1; 296 } 297 } 298 299 /* 300 * This callback is executed during stop_machine(). All CPUs are therefore 301 * temporarily stopped. In order not to change guest behavior, we have to 302 * disable preemption whenever we touch the epoch of kvm and the VCPUs, 303 * so a CPU won't be stopped while calculating with the epoch. 304 */ 305 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val, 306 void *v) 307 { 308 struct kvm *kvm; 309 struct kvm_vcpu *vcpu; 310 unsigned long i; 311 unsigned long long *delta = v; 312 313 list_for_each_entry(kvm, &vm_list, vm_list) { 314 kvm_for_each_vcpu(i, vcpu, kvm) { 315 kvm_clock_sync_scb(vcpu->arch.sie_block, *delta); 316 if (i == 0) { 317 kvm->arch.epoch = vcpu->arch.sie_block->epoch; 318 kvm->arch.epdx = vcpu->arch.sie_block->epdx; 319 } 320 if (vcpu->arch.cputm_enabled) 321 vcpu->arch.cputm_start += *delta; 322 if (vcpu->arch.vsie_block) 323 kvm_clock_sync_scb(vcpu->arch.vsie_block, 324 *delta); 325 } 326 } 327 return NOTIFY_OK; 328 } 329 330 static struct notifier_block kvm_clock_notifier = { 331 .notifier_call = kvm_clock_sync, 332 }; 333 334 static void allow_cpu_feat(unsigned long nr) 335 { 336 set_bit_inv(nr, kvm_s390_available_cpu_feat); 337 } 338 339 static inline int plo_test_bit(unsigned char nr) 340 { 341 unsigned long function = (unsigned long)nr | 0x100; 342 int cc; 343 344 asm volatile( 345 " lgr 0,%[function]\n" 346 /* Parameter registers are ignored for "test bit" */ 347 " plo 0,0,0,0(0)\n" 348 CC_IPM(cc) 349 : CC_OUT(cc, cc) 350 : [function] "d" (function) 351 : CC_CLOBBER_LIST("0")); 352 return CC_TRANSFORM(cc) == 0; 353 } 354 355 static __always_inline void pfcr_query(u8 (*query)[16]) 356 { 357 asm volatile( 358 " lghi 0,0\n" 359 " .insn rsy,0xeb0000000016,0,0,%[query]" 360 : [query] "=QS" (*query) 361 : 362 : "cc", "0"); 363 } 364 365 static __always_inline void __sortl_query(u8 (*query)[32]) 366 { 367 asm volatile( 368 " lghi 0,0\n" 369 " la 1,%[query]\n" 370 /* Parameter registers are ignored */ 371 " .insn rre,0xb9380000,2,4" 372 : [query] "=R" (*query) 373 : 374 : "cc", "0", "1"); 375 } 376 377 static __always_inline void __dfltcc_query(u8 (*query)[32]) 378 { 379 asm volatile( 380 " lghi 0,0\n" 381 " la 1,%[query]\n" 382 /* Parameter registers are ignored */ 383 " .insn rrf,0xb9390000,2,4,6,0" 384 : [query] "=R" (*query) 385 : 386 : "cc", "0", "1"); 387 } 388 389 static void __init kvm_s390_cpu_feat_init(void) 390 { 391 int i; 392 393 for (i = 0; i < 256; ++i) { 394 if (plo_test_bit(i)) 395 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7); 396 } 397 398 if (test_facility(28)) /* TOD-clock steering */ 399 ptff(kvm_s390_available_subfunc.ptff, 400 sizeof(kvm_s390_available_subfunc.ptff), 401 PTFF_QAF); 402 403 if (test_facility(17)) { /* MSA */ 404 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *) 405 kvm_s390_available_subfunc.kmac); 406 __cpacf_query(CPACF_KMC, (cpacf_mask_t *) 407 kvm_s390_available_subfunc.kmc); 408 __cpacf_query(CPACF_KM, (cpacf_mask_t *) 409 kvm_s390_available_subfunc.km); 410 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *) 411 kvm_s390_available_subfunc.kimd); 412 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *) 413 kvm_s390_available_subfunc.klmd); 414 } 415 if (test_facility(76)) /* MSA3 */ 416 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *) 417 kvm_s390_available_subfunc.pckmo); 418 if (test_facility(77)) { /* MSA4 */ 419 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *) 420 kvm_s390_available_subfunc.kmctr); 421 __cpacf_query(CPACF_KMF, (cpacf_mask_t *) 422 kvm_s390_available_subfunc.kmf); 423 __cpacf_query(CPACF_KMO, (cpacf_mask_t *) 424 kvm_s390_available_subfunc.kmo); 425 __cpacf_query(CPACF_PCC, (cpacf_mask_t *) 426 kvm_s390_available_subfunc.pcc); 427 } 428 if (test_facility(57)) /* MSA5 */ 429 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *) 430 kvm_s390_available_subfunc.ppno); 431 432 if (test_facility(146)) /* MSA8 */ 433 __cpacf_query(CPACF_KMA, (cpacf_mask_t *) 434 kvm_s390_available_subfunc.kma); 435 436 if (test_facility(155)) /* MSA9 */ 437 __cpacf_query(CPACF_KDSA, (cpacf_mask_t *) 438 kvm_s390_available_subfunc.kdsa); 439 440 if (test_facility(150)) /* SORTL */ 441 __sortl_query(&kvm_s390_available_subfunc.sortl); 442 443 if (test_facility(151)) /* DFLTCC */ 444 __dfltcc_query(&kvm_s390_available_subfunc.dfltcc); 445 446 if (test_facility(201)) /* PFCR */ 447 pfcr_query(&kvm_s390_available_subfunc.pfcr); 448 449 if (machine_has_esop()) 450 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP); 451 /* 452 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow), 453 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing). 454 */ 455 if (!sclp.has_sief2 || !machine_has_esop() || !sclp.has_64bscao || 456 !test_facility(3) || !nested) 457 return; 458 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2); 459 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO); 460 if (sclp.has_siif) 461 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF); 462 if (sclp.has_gpere) 463 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE); 464 if (sclp.has_gsls) 465 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS); 466 if (sclp.has_ib) 467 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB); 468 if (sclp.has_cei) 469 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI); 470 if (sclp.has_ibs) 471 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS); 472 if (sclp.has_kss) 473 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS); 474 if (sclp.has_astfleie2) 475 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ASTFLEIE2); 476 /* 477 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make 478 * all skey handling functions read/set the skey from the PGSTE 479 * instead of the real storage key. 480 * 481 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make 482 * pages being detected as preserved although they are resident. 483 * 484 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will 485 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY. 486 * 487 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and 488 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be 489 * correctly shadowed. We can do that for the PGSTE but not for PTE.I. 490 * 491 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We 492 * cannot easily shadow the SCA because of the ipte lock. 493 */ 494 } 495 496 static int __init __kvm_s390_init(void) 497 { 498 int rc = -ENOMEM; 499 500 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long)); 501 if (!kvm_s390_dbf) 502 return -ENOMEM; 503 504 kvm_s390_dbf_uv = debug_register("kvm-uv", 32, 1, 7 * sizeof(long)); 505 if (!kvm_s390_dbf_uv) 506 goto err_kvm_uv; 507 508 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view) || 509 debug_register_view(kvm_s390_dbf_uv, &debug_sprintf_view)) 510 goto err_debug_view; 511 512 kvm_s390_cpu_feat_init(); 513 514 /* Register floating interrupt controller interface. */ 515 rc = kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC); 516 if (rc) { 517 pr_err("A FLIC registration call failed with rc=%d\n", rc); 518 goto err_flic; 519 } 520 521 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) { 522 rc = kvm_s390_pci_init(); 523 if (rc) { 524 pr_err("Unable to allocate AIFT for PCI\n"); 525 goto err_pci; 526 } 527 } 528 529 rc = kvm_s390_gib_init(GAL_ISC); 530 if (rc) 531 goto err_gib; 532 533 atomic_notifier_chain_register(&s390_epoch_delta_notifier, 534 &kvm_clock_notifier); 535 536 return 0; 537 538 err_gib: 539 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) 540 kvm_s390_pci_exit(); 541 err_pci: 542 err_flic: 543 err_debug_view: 544 debug_unregister(kvm_s390_dbf_uv); 545 err_kvm_uv: 546 debug_unregister(kvm_s390_dbf); 547 return rc; 548 } 549 550 static void __kvm_s390_exit(void) 551 { 552 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier, 553 &kvm_clock_notifier); 554 555 kvm_s390_gib_destroy(); 556 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) 557 kvm_s390_pci_exit(); 558 debug_unregister(kvm_s390_dbf); 559 debug_unregister(kvm_s390_dbf_uv); 560 } 561 562 static int kvm_s390_keyop(struct kvm_s390_mmu_cache *mc, struct kvm *kvm, int op, 563 unsigned long addr, union skey skey) 564 { 565 gfn_t gfn = gpa_to_gfn(addr); 566 union asce asce; 567 int r; 568 569 guard(read_lock)(&kvm->mmu_lock); 570 asce = kvm->arch.gmap->asce; 571 572 switch (op) { 573 case KVM_S390_KEYOP_SSKE: 574 r = dat_cond_set_storage_key(mc, asce, gfn, skey, &skey, 0, 0, 0); 575 if (r == 0 || r == 1) 576 return skey.skey; 577 break; 578 case KVM_S390_KEYOP_ISKE: 579 r = dat_get_storage_key(asce, gfn, &skey); 580 if (!r) 581 return skey.skey; 582 break; 583 case KVM_S390_KEYOP_RRBE: 584 r = dat_reset_reference_bit(asce, gfn, &skey); 585 if (!r) 586 return skey.skey; 587 break; 588 default: 589 return -EINVAL; 590 } 591 return r > 0 ? -EFAULT : r; 592 } 593 594 /* Section: device related */ 595 long kvm_arch_dev_ioctl(struct file *filp, 596 unsigned int ioctl, unsigned long arg) 597 { 598 if (ioctl == KVM_S390_ENABLE_SIE) 599 return 0; 600 return -EINVAL; 601 } 602 603 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 604 { 605 int r; 606 607 switch (ext) { 608 case KVM_CAP_S390_PSW: 609 case KVM_CAP_S390_GMAP: 610 #ifdef CONFIG_KVM_S390_UCONTROL 611 case KVM_CAP_S390_UCONTROL: 612 #endif 613 case KVM_CAP_ASYNC_PF: 614 case KVM_CAP_SYNC_REGS: 615 case KVM_CAP_ONE_REG: 616 case KVM_CAP_ENABLE_CAP: 617 case KVM_CAP_S390_CSS_SUPPORT: 618 case KVM_CAP_IOEVENTFD: 619 case KVM_CAP_S390_IRQCHIP: 620 case KVM_CAP_VM_ATTRIBUTES: 621 case KVM_CAP_MP_STATE: 622 case KVM_CAP_IMMEDIATE_EXIT: 623 case KVM_CAP_S390_INJECT_IRQ: 624 case KVM_CAP_S390_USER_SIGP: 625 case KVM_CAP_S390_USER_STSI: 626 case KVM_CAP_S390_SKEYS: 627 case KVM_CAP_S390_IRQ_STATE: 628 case KVM_CAP_S390_USER_INSTR0: 629 case KVM_CAP_S390_CMMA_MIGRATION: 630 case KVM_CAP_S390_AIS: 631 case KVM_CAP_S390_AIS_MIGRATION: 632 case KVM_CAP_S390_VCPU_RESETS: 633 case KVM_CAP_SET_GUEST_DEBUG: 634 case KVM_CAP_S390_DIAG318: 635 case KVM_CAP_IRQFD_RESAMPLE: 636 case KVM_CAP_S390_USER_OPEREXEC: 637 case KVM_CAP_S390_KEYOP: 638 case KVM_CAP_S390_VSIE_ESAMODE: 639 case KVM_CAP_PRE_FAULT_MEMORY: 640 r = 1; 641 break; 642 case KVM_CAP_SET_GUEST_DEBUG2: 643 r = KVM_GUESTDBG_VALID_MASK; 644 break; 645 case KVM_CAP_S390_HPAGE_1M: 646 r = 0; 647 if (hpage && !(kvm && kvm_is_ucontrol(kvm))) 648 r = 1; 649 break; 650 case KVM_CAP_S390_HPAGE_2G: 651 r = 0; 652 if (hpage_2g && !(kvm && kvm_is_ucontrol(kvm))) 653 r = 1; 654 break; 655 case KVM_CAP_S390_MEM_OP: 656 r = MEM_OP_MAX_SIZE; 657 break; 658 case KVM_CAP_S390_MEM_OP_EXTENSION: 659 /* 660 * Flag bits indicating which extensions are supported. 661 * If r > 0, the base extension must also be supported/indicated, 662 * in order to maintain backwards compatibility. 663 */ 664 r = KVM_S390_MEMOP_EXTENSION_CAP_BASE | 665 KVM_S390_MEMOP_EXTENSION_CAP_CMPXCHG; 666 break; 667 case KVM_CAP_NR_VCPUS: 668 case KVM_CAP_MAX_VCPUS: 669 case KVM_CAP_MAX_VCPU_ID: 670 /* 671 * Return the same value for KVM_CAP_MAX_VCPUS and 672 * KVM_CAP_MAX_VCPU_ID to conform with the KVM API. 673 */ 674 r = KVM_S390_ESCA_CPU_SLOTS; 675 if (!kvm_s390_use_sca_entries()) 676 r = KVM_MAX_VCPUS; 677 if (ext == KVM_CAP_NR_VCPUS) 678 r = min_t(unsigned int, num_online_cpus(), r); 679 break; 680 case KVM_CAP_S390_COW: 681 r = machine_has_esop(); 682 break; 683 case KVM_CAP_S390_VECTOR_REGISTERS: 684 r = test_facility(129); 685 break; 686 case KVM_CAP_S390_RI: 687 r = test_facility(64); 688 break; 689 case KVM_CAP_S390_GS: 690 r = test_facility(133); 691 break; 692 case KVM_CAP_S390_BPB: 693 r = test_facility(82); 694 break; 695 case KVM_CAP_S390_PROTECTED_ASYNC_DISABLE: 696 r = async_destroy && is_prot_virt_host(); 697 break; 698 case KVM_CAP_S390_PROTECTED: 699 r = is_prot_virt_host(); 700 break; 701 case KVM_CAP_S390_PROTECTED_DUMP: { 702 u64 pv_cmds_dump[] = { 703 BIT_UVC_CMD_DUMP_INIT, 704 BIT_UVC_CMD_DUMP_CONFIG_STOR_STATE, 705 BIT_UVC_CMD_DUMP_CPU, 706 BIT_UVC_CMD_DUMP_COMPLETE, 707 }; 708 int i; 709 710 r = is_prot_virt_host(); 711 712 for (i = 0; i < ARRAY_SIZE(pv_cmds_dump); i++) { 713 if (!test_bit_inv(pv_cmds_dump[i], 714 (unsigned long *)&uv_info.inst_calls_list)) { 715 r = 0; 716 break; 717 } 718 } 719 break; 720 } 721 case KVM_CAP_S390_ZPCI_OP: 722 r = kvm_s390_pci_interp_allowed(); 723 break; 724 case KVM_CAP_S390_CPU_TOPOLOGY: 725 r = test_facility(11); 726 break; 727 default: 728 r = 0; 729 } 730 return r; 731 } 732 733 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot) 734 { 735 gfn_t last_gfn = memslot->base_gfn + memslot->npages; 736 737 scoped_guard(read_lock, &kvm->mmu_lock) 738 gmap_sync_dirty_log(kvm->arch.gmap, memslot->base_gfn, last_gfn); 739 } 740 741 /* Section: vm related */ 742 static void sca_del_vcpu(struct kvm_vcpu *vcpu); 743 744 /* 745 * Get (and clear) the dirty memory log for a memory slot. 746 */ 747 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, 748 struct kvm_dirty_log *log) 749 { 750 return s390_kvm_mmu_get_dirty_log(kvm, log); 751 } 752 753 static void icpt_operexc_on_all_vcpus(struct kvm *kvm) 754 { 755 unsigned long i; 756 struct kvm_vcpu *vcpu; 757 758 kvm_for_each_vcpu(i, vcpu, kvm) { 759 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu); 760 } 761 } 762 763 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap) 764 { 765 int r; 766 767 if (cap->flags) 768 return -EINVAL; 769 770 switch (cap->cap) { 771 case KVM_CAP_S390_IRQCHIP: 772 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP"); 773 kvm->arch.use_irqchip = 1; 774 r = 0; 775 break; 776 case KVM_CAP_S390_USER_SIGP: 777 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP"); 778 kvm->arch.user_sigp = 1; 779 r = 0; 780 break; 781 case KVM_CAP_S390_VECTOR_REGISTERS: 782 mutex_lock(&kvm->lock); 783 if (kvm->created_vcpus) { 784 r = -EBUSY; 785 } else if (cpu_has_vx()) { 786 set_kvm_facility(kvm->arch.model.fac_mask, 129); 787 set_kvm_facility(kvm->arch.model.fac_list, 129); 788 if (test_facility(134)) { 789 set_kvm_facility(kvm->arch.model.fac_mask, 134); 790 set_kvm_facility(kvm->arch.model.fac_list, 134); 791 } 792 if (test_facility(135)) { 793 set_kvm_facility(kvm->arch.model.fac_mask, 135); 794 set_kvm_facility(kvm->arch.model.fac_list, 135); 795 } 796 if (test_facility(148)) { 797 set_kvm_facility(kvm->arch.model.fac_mask, 148); 798 set_kvm_facility(kvm->arch.model.fac_list, 148); 799 } 800 if (test_facility(152)) { 801 set_kvm_facility(kvm->arch.model.fac_mask, 152); 802 set_kvm_facility(kvm->arch.model.fac_list, 152); 803 } 804 if (test_facility(192)) { 805 set_kvm_facility(kvm->arch.model.fac_mask, 192); 806 set_kvm_facility(kvm->arch.model.fac_list, 192); 807 } 808 if (test_facility(198)) { 809 set_kvm_facility(kvm->arch.model.fac_mask, 198); 810 set_kvm_facility(kvm->arch.model.fac_list, 198); 811 } 812 if (test_facility(199)) { 813 set_kvm_facility(kvm->arch.model.fac_mask, 199); 814 set_kvm_facility(kvm->arch.model.fac_list, 199); 815 } 816 r = 0; 817 } else 818 r = -EINVAL; 819 mutex_unlock(&kvm->lock); 820 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s", 821 r ? "(not available)" : "(success)"); 822 break; 823 case KVM_CAP_S390_RI: 824 r = -EINVAL; 825 mutex_lock(&kvm->lock); 826 if (kvm->created_vcpus) { 827 r = -EBUSY; 828 } else if (test_facility(64)) { 829 set_kvm_facility(kvm->arch.model.fac_mask, 64); 830 set_kvm_facility(kvm->arch.model.fac_list, 64); 831 r = 0; 832 } 833 mutex_unlock(&kvm->lock); 834 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s", 835 r ? "(not available)" : "(success)"); 836 break; 837 case KVM_CAP_S390_AIS: 838 mutex_lock(&kvm->lock); 839 if (kvm->created_vcpus) { 840 r = -EBUSY; 841 } else { 842 set_kvm_facility(kvm->arch.model.fac_mask, 72); 843 set_kvm_facility(kvm->arch.model.fac_list, 72); 844 r = 0; 845 } 846 mutex_unlock(&kvm->lock); 847 VM_EVENT(kvm, 3, "ENABLE: AIS %s", 848 r ? "(not available)" : "(success)"); 849 break; 850 case KVM_CAP_S390_GS: 851 r = -EINVAL; 852 mutex_lock(&kvm->lock); 853 if (kvm->created_vcpus) { 854 r = -EBUSY; 855 } else if (test_facility(133)) { 856 set_kvm_facility(kvm->arch.model.fac_mask, 133); 857 set_kvm_facility(kvm->arch.model.fac_list, 133); 858 r = 0; 859 } 860 mutex_unlock(&kvm->lock); 861 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s", 862 r ? "(not available)" : "(success)"); 863 break; 864 case KVM_CAP_S390_HPAGE_1M: 865 mutex_lock(&kvm->lock); 866 if (kvm->created_vcpus) 867 r = -EBUSY; 868 else if (!hpage || kvm->arch.use_cmma || kvm_is_ucontrol(kvm)) 869 r = -EINVAL; 870 else { 871 r = 0; 872 set_bit(GMAP_FLAG_ALLOW_HPAGE_1M, &kvm->arch.gmap->flags); 873 /* 874 * We might have to create fake 4k page 875 * tables. To avoid that the hardware works on 876 * stale PGSTEs, we emulate these instructions. 877 */ 878 kvm->arch.use_skf = 0; 879 kvm->arch.use_pfmfi = 0; 880 } 881 mutex_unlock(&kvm->lock); 882 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_HPAGE %s", 883 r ? "(not available)" : "(success)"); 884 break; 885 case KVM_CAP_S390_HPAGE_2G: 886 mutex_lock(&kvm->lock); 887 if (kvm->created_vcpus) { 888 r = -EBUSY; 889 } else if (!hpage_2g || kvm->arch.use_cmma || kvm_is_ucontrol(kvm)) { 890 r = -EINVAL; 891 } else { 892 r = 0; 893 set_bit(GMAP_FLAG_ALLOW_HPAGE_2G, &kvm->arch.gmap->flags); 894 /* 895 * We might have to create fake 4k page 896 * tables. To avoid that the hardware works on 897 * stale PGSTEs, we emulate these instructions. 898 */ 899 kvm->arch.use_skf = 0; 900 kvm->arch.use_pfmfi = 0; 901 } 902 mutex_unlock(&kvm->lock); 903 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_HPAGE_2G %s", 904 r ? "(not available)" : "(success)"); 905 break; 906 case KVM_CAP_S390_USER_STSI: 907 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI"); 908 kvm->arch.user_stsi = 1; 909 r = 0; 910 break; 911 case KVM_CAP_S390_USER_INSTR0: 912 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0"); 913 kvm->arch.user_instr0 = 1; 914 icpt_operexc_on_all_vcpus(kvm); 915 r = 0; 916 break; 917 case KVM_CAP_S390_CPU_TOPOLOGY: 918 r = -EINVAL; 919 mutex_lock(&kvm->lock); 920 if (kvm->created_vcpus) { 921 r = -EBUSY; 922 } else if (test_facility(11)) { 923 set_kvm_facility(kvm->arch.model.fac_mask, 11); 924 set_kvm_facility(kvm->arch.model.fac_list, 11); 925 r = 0; 926 } 927 mutex_unlock(&kvm->lock); 928 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_CPU_TOPOLOGY %s", 929 r ? "(not available)" : "(success)"); 930 break; 931 case KVM_CAP_S390_USER_OPEREXEC: 932 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_OPEREXEC"); 933 kvm->arch.user_operexec = 1; 934 icpt_operexc_on_all_vcpus(kvm); 935 r = 0; 936 break; 937 case KVM_CAP_S390_VSIE_ESAMODE: 938 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_VSIE_ESAMODE"); 939 kvm->arch.allow_vsie_esamode = 1; 940 r = 0; 941 break; 942 default: 943 r = -EINVAL; 944 break; 945 } 946 return r; 947 } 948 949 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr) 950 { 951 int ret; 952 953 switch (attr->attr) { 954 case KVM_S390_VM_MEM_LIMIT_SIZE: 955 ret = 0; 956 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes", 957 kvm->arch.mem_limit); 958 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr)) 959 ret = -EFAULT; 960 break; 961 default: 962 ret = -ENXIO; 963 break; 964 } 965 return ret; 966 } 967 968 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr) 969 { 970 int ret; 971 972 switch (attr->attr) { 973 case KVM_S390_VM_MEM_ENABLE_CMMA: 974 ret = -ENXIO; 975 if (!sclp.has_cmma) 976 break; 977 978 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support"); 979 mutex_lock(&kvm->lock); 980 if (kvm->created_vcpus) 981 ret = -EBUSY; 982 else { 983 kvm->arch.use_cmma = 1; 984 /* Not compatible with cmma. */ 985 kvm->arch.use_pfmfi = 0; 986 ret = 0; 987 } 988 mutex_unlock(&kvm->lock); 989 break; 990 case KVM_S390_VM_MEM_CLR_CMMA: { 991 gfn_t start_gfn = 0; 992 993 ret = -ENXIO; 994 if (!sclp.has_cmma) 995 break; 996 ret = -EINVAL; 997 if (!kvm->arch.use_cmma) 998 break; 999 1000 guard(mutex)(&kvm->lock); 1001 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states"); 1002 do { 1003 scoped_guard(read_lock, &kvm->mmu_lock) 1004 start_gfn = dat_reset_cmma(kvm->arch.gmap->asce, start_gfn); 1005 } while (start_gfn); 1006 ret = 0; 1007 break; 1008 } 1009 case KVM_S390_VM_MEM_LIMIT_SIZE: { 1010 struct kvm_memslots *slots; 1011 struct kvm_memory_slot *ms; 1012 unsigned long new_limit; 1013 int bkt; 1014 1015 if (kvm_is_ucontrol(kvm)) 1016 return -EINVAL; 1017 1018 if (get_user(new_limit, (u64 __user *)attr->addr)) 1019 return -EFAULT; 1020 1021 guard(mutex)(&kvm->lock); 1022 1023 new_limit = ALIGN(new_limit, HPAGE_SIZE); 1024 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT && 1025 new_limit > kvm->arch.mem_limit) 1026 return -E2BIG; 1027 1028 if (!new_limit) 1029 return -EINVAL; 1030 1031 if (kvm->created_vcpus) 1032 return -EBUSY; 1033 1034 ret = 0; 1035 scoped_guard(mutex, &kvm->slots_lock) { 1036 slots = kvm_memslots(kvm); 1037 if (slots && !kvm_memslots_empty(slots)) { 1038 kvm_for_each_memslot(ms, bkt, slots) { 1039 if (gpa_to_gfn(new_limit) < ms->base_gfn + ms->npages) { 1040 ret = -EBUSY; 1041 break; 1042 } 1043 } 1044 } 1045 if (!ret) 1046 ret = gmap_set_limit(kvm->arch.gmap, gpa_to_gfn(new_limit)); 1047 } 1048 if (ret) 1049 break; 1050 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit); 1051 VM_EVENT(kvm, 3, "New guest asce: 0x%p", (void *)kvm->arch.gmap->asce.val); 1052 break; 1053 } 1054 default: 1055 ret = -ENXIO; 1056 break; 1057 } 1058 return ret; 1059 } 1060 1061 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu); 1062 1063 void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm) 1064 { 1065 struct kvm_vcpu *vcpu; 1066 unsigned long i; 1067 1068 kvm_s390_vcpu_block_all(kvm); 1069 1070 kvm_for_each_vcpu(i, vcpu, kvm) { 1071 kvm_s390_vcpu_crypto_setup(vcpu); 1072 /* recreate the shadow crycb by leaving the VSIE handler */ 1073 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu); 1074 } 1075 1076 kvm_s390_vcpu_unblock_all(kvm); 1077 } 1078 1079 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr) 1080 { 1081 mutex_lock(&kvm->lock); 1082 switch (attr->attr) { 1083 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW: 1084 if (!test_kvm_facility(kvm, 76)) { 1085 mutex_unlock(&kvm->lock); 1086 return -EINVAL; 1087 } 1088 get_random_bytes( 1089 kvm->arch.crypto.crycb->aes_wrapping_key_mask, 1090 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask)); 1091 kvm->arch.crypto.aes_kw = 1; 1092 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support"); 1093 break; 1094 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW: 1095 if (!test_kvm_facility(kvm, 76)) { 1096 mutex_unlock(&kvm->lock); 1097 return -EINVAL; 1098 } 1099 get_random_bytes( 1100 kvm->arch.crypto.crycb->dea_wrapping_key_mask, 1101 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask)); 1102 kvm->arch.crypto.dea_kw = 1; 1103 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support"); 1104 break; 1105 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW: 1106 if (!test_kvm_facility(kvm, 76)) { 1107 mutex_unlock(&kvm->lock); 1108 return -EINVAL; 1109 } 1110 kvm->arch.crypto.aes_kw = 0; 1111 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0, 1112 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask)); 1113 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support"); 1114 break; 1115 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW: 1116 if (!test_kvm_facility(kvm, 76)) { 1117 mutex_unlock(&kvm->lock); 1118 return -EINVAL; 1119 } 1120 kvm->arch.crypto.dea_kw = 0; 1121 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0, 1122 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask)); 1123 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support"); 1124 break; 1125 case KVM_S390_VM_CRYPTO_ENABLE_APIE: 1126 if (!ap_instructions_available()) { 1127 mutex_unlock(&kvm->lock); 1128 return -EOPNOTSUPP; 1129 } 1130 kvm->arch.crypto.apie = 1; 1131 break; 1132 case KVM_S390_VM_CRYPTO_DISABLE_APIE: 1133 if (!ap_instructions_available()) { 1134 mutex_unlock(&kvm->lock); 1135 return -EOPNOTSUPP; 1136 } 1137 kvm->arch.crypto.apie = 0; 1138 break; 1139 default: 1140 mutex_unlock(&kvm->lock); 1141 return -ENXIO; 1142 } 1143 1144 kvm_s390_vcpu_crypto_reset_all(kvm); 1145 mutex_unlock(&kvm->lock); 1146 return 0; 1147 } 1148 1149 static void kvm_s390_vcpu_pci_setup(struct kvm_vcpu *vcpu) 1150 { 1151 /* Only set the ECB bits after guest requests zPCI interpretation */ 1152 if (!vcpu->kvm->arch.use_zpci_interp) 1153 return; 1154 1155 vcpu->arch.sie_block->ecb2 |= ECB2_ZPCI_LSI; 1156 vcpu->arch.sie_block->ecb3 |= ECB3_AISII + ECB3_AISI; 1157 } 1158 1159 void kvm_s390_vcpu_pci_enable_interp(struct kvm *kvm) 1160 { 1161 struct kvm_vcpu *vcpu; 1162 unsigned long i; 1163 1164 lockdep_assert_held(&kvm->lock); 1165 1166 if (!kvm_s390_pci_interp_allowed()) 1167 return; 1168 1169 /* 1170 * If host is configured for PCI and the necessary facilities are 1171 * available, turn on interpretation for the life of this guest 1172 */ 1173 kvm->arch.use_zpci_interp = 1; 1174 1175 kvm_s390_vcpu_block_all(kvm); 1176 1177 kvm_for_each_vcpu(i, vcpu, kvm) { 1178 kvm_s390_vcpu_pci_setup(vcpu); 1179 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu); 1180 } 1181 1182 kvm_s390_vcpu_unblock_all(kvm); 1183 } 1184 1185 static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req) 1186 { 1187 unsigned long cx; 1188 struct kvm_vcpu *vcpu; 1189 1190 kvm_for_each_vcpu(cx, vcpu, kvm) 1191 kvm_s390_sync_request(req, vcpu); 1192 } 1193 1194 /* 1195 * Must be called with kvm->srcu held to avoid races on memslots, and with 1196 * kvm->slots_arch_lock to avoid races with ourselves, 1197 * kvm_s390_vm_stop_migration(), and kvm_s390_get_cmma_bits(). 1198 */ 1199 static int kvm_s390_vm_start_migration(struct kvm *kvm) 1200 { 1201 struct kvm_memory_slot *ms; 1202 struct kvm_memslots *slots; 1203 int bkt; 1204 1205 /* migration mode already enabled */ 1206 if (kvm->arch.migration_mode) 1207 return 0; 1208 slots = kvm_memslots(kvm); 1209 if (!slots || kvm_memslots_empty(slots)) 1210 return -EINVAL; 1211 1212 if (!kvm->arch.use_cmma) { 1213 kvm->arch.migration_mode = 1; 1214 return 0; 1215 } 1216 kvm_for_each_memslot(ms, bkt, slots) { 1217 if (!ms->dirty_bitmap) 1218 return -EINVAL; 1219 } 1220 /* 1221 * Set the flag and let KVM handle ESSA manually, potentially setting 1222 * the cmma_d bit in some PGSTEs and increasing cmma_dirty_pages. 1223 * At this point cmma_dirty_pages is still 0, and all existing PGSTEs 1224 * have their cmma_d bit set to 0. 1225 * Any newly allocated page table has its entries marked as cmma-clean, 1226 * which is fine because the CMMA values are not dirty. 1227 */ 1228 WRITE_ONCE(kvm->arch.migration_mode, 1); 1229 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION); 1230 /* 1231 * Mark all PGSTEs as cmma-dirty, increasing cmma_dirty_pages as needed, 1232 * but without double-counting pages that have become dirty on their own 1233 * in the meantime. 1234 * At this point some pages might have become dirty on their own already 1235 * and cmma_dirty_pages might therefore be non-zero. 1236 */ 1237 gmap_set_cmma_all_dirty(kvm->arch.gmap); 1238 return 0; 1239 } 1240 1241 /* 1242 * Must be called with kvm->slots_arch_lock to avoid races with ourselves, 1243 * kvm_s390_vm_start_migration() and kvm_s390_get_cmma_bits(). 1244 */ 1245 int kvm_s390_vm_stop_migration(struct kvm *kvm) 1246 { 1247 /* migration mode already disabled */ 1248 if (!kvm->arch.migration_mode) 1249 return 0; 1250 /* 1251 * Unset the flag and propagate to all vCPUs. From now on the cmma_d 1252 * bit will not be touched on any PGSTE. 1253 * At this point cmma_dirty_pages is possibly non-zero, and thus some 1254 * PGSTEs might have cmma_d set. 1255 */ 1256 WRITE_ONCE(kvm->arch.migration_mode, 0); 1257 if (!kvm->arch.use_cmma) 1258 return 0; 1259 1260 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION); 1261 /* Clear cmma_d on all existing PGSTEs and set cmma_dirty_pages to 0. */ 1262 gmap_set_cmma_all_clean(kvm->arch.gmap); 1263 atomic64_set(&kvm->arch.cmma_dirty_pages, 0); 1264 /* 1265 * At this point the system has the expected state: migration_mode is 0, 1266 * cmma_dirty_pages is 0, and all existing PGSTEs have their cmma_d bit 1267 * set to 0. 1268 */ 1269 return 0; 1270 } 1271 1272 static int kvm_s390_vm_set_migration(struct kvm *kvm, 1273 struct kvm_device_attr *attr) 1274 { 1275 int res = -ENXIO; 1276 1277 guard(srcu)(&kvm->srcu); 1278 guard(mutex)(&kvm->slots_arch_lock); 1279 1280 switch (attr->attr) { 1281 case KVM_S390_VM_MIGRATION_START: 1282 res = kvm_s390_vm_start_migration(kvm); 1283 break; 1284 case KVM_S390_VM_MIGRATION_STOP: 1285 res = kvm_s390_vm_stop_migration(kvm); 1286 break; 1287 default: 1288 break; 1289 } 1290 1291 return res; 1292 } 1293 1294 static int kvm_s390_vm_get_migration(struct kvm *kvm, 1295 struct kvm_device_attr *attr) 1296 { 1297 u64 mig = kvm->arch.migration_mode; 1298 1299 if (attr->attr != KVM_S390_VM_MIGRATION_STATUS) 1300 return -ENXIO; 1301 1302 if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig))) 1303 return -EFAULT; 1304 return 0; 1305 } 1306 1307 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod); 1308 1309 static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr) 1310 { 1311 struct kvm_s390_vm_tod_clock gtod; 1312 1313 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod))) 1314 return -EFAULT; 1315 1316 if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx) 1317 return -EINVAL; 1318 __kvm_s390_set_tod_clock(kvm, >od); 1319 1320 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx", 1321 gtod.epoch_idx, gtod.tod); 1322 1323 return 0; 1324 } 1325 1326 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr) 1327 { 1328 u8 gtod_high; 1329 1330 if (copy_from_user(>od_high, (void __user *)attr->addr, 1331 sizeof(gtod_high))) 1332 return -EFAULT; 1333 1334 if (gtod_high != 0) 1335 return -EINVAL; 1336 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high); 1337 1338 return 0; 1339 } 1340 1341 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr) 1342 { 1343 struct kvm_s390_vm_tod_clock gtod = { 0 }; 1344 1345 if (copy_from_user(>od.tod, (void __user *)attr->addr, 1346 sizeof(gtod.tod))) 1347 return -EFAULT; 1348 1349 __kvm_s390_set_tod_clock(kvm, >od); 1350 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod); 1351 return 0; 1352 } 1353 1354 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr) 1355 { 1356 int ret; 1357 1358 if (attr->flags) 1359 return -EINVAL; 1360 1361 mutex_lock(&kvm->lock); 1362 /* 1363 * For protected guests, the TOD is managed by the ultravisor, so trying 1364 * to change it will never bring the expected results. 1365 */ 1366 if (kvm_s390_pv_is_protected(kvm)) { 1367 ret = -EOPNOTSUPP; 1368 goto out_unlock; 1369 } 1370 1371 switch (attr->attr) { 1372 case KVM_S390_VM_TOD_EXT: 1373 ret = kvm_s390_set_tod_ext(kvm, attr); 1374 break; 1375 case KVM_S390_VM_TOD_HIGH: 1376 ret = kvm_s390_set_tod_high(kvm, attr); 1377 break; 1378 case KVM_S390_VM_TOD_LOW: 1379 ret = kvm_s390_set_tod_low(kvm, attr); 1380 break; 1381 default: 1382 ret = -ENXIO; 1383 break; 1384 } 1385 1386 out_unlock: 1387 mutex_unlock(&kvm->lock); 1388 return ret; 1389 } 1390 1391 static void kvm_s390_get_tod_clock(struct kvm *kvm, 1392 struct kvm_s390_vm_tod_clock *gtod) 1393 { 1394 union tod_clock clk; 1395 1396 preempt_disable(); 1397 1398 store_tod_clock_ext(&clk); 1399 1400 gtod->tod = clk.tod + kvm->arch.epoch; 1401 gtod->epoch_idx = 0; 1402 if (test_kvm_facility(kvm, 139)) { 1403 gtod->epoch_idx = clk.ei + kvm->arch.epdx; 1404 if (gtod->tod < clk.tod) 1405 gtod->epoch_idx += 1; 1406 } 1407 1408 preempt_enable(); 1409 } 1410 1411 static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr) 1412 { 1413 struct kvm_s390_vm_tod_clock gtod; 1414 1415 memset(>od, 0, sizeof(gtod)); 1416 kvm_s390_get_tod_clock(kvm, >od); 1417 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod))) 1418 return -EFAULT; 1419 1420 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx", 1421 gtod.epoch_idx, gtod.tod); 1422 return 0; 1423 } 1424 1425 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr) 1426 { 1427 u8 gtod_high = 0; 1428 1429 if (copy_to_user((void __user *)attr->addr, >od_high, 1430 sizeof(gtod_high))) 1431 return -EFAULT; 1432 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high); 1433 1434 return 0; 1435 } 1436 1437 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr) 1438 { 1439 u64 gtod; 1440 1441 gtod = kvm_s390_get_tod_clock_fast(kvm); 1442 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod))) 1443 return -EFAULT; 1444 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod); 1445 1446 return 0; 1447 } 1448 1449 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr) 1450 { 1451 int ret; 1452 1453 if (attr->flags) 1454 return -EINVAL; 1455 1456 switch (attr->attr) { 1457 case KVM_S390_VM_TOD_EXT: 1458 ret = kvm_s390_get_tod_ext(kvm, attr); 1459 break; 1460 case KVM_S390_VM_TOD_HIGH: 1461 ret = kvm_s390_get_tod_high(kvm, attr); 1462 break; 1463 case KVM_S390_VM_TOD_LOW: 1464 ret = kvm_s390_get_tod_low(kvm, attr); 1465 break; 1466 default: 1467 ret = -ENXIO; 1468 break; 1469 } 1470 return ret; 1471 } 1472 1473 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr) 1474 { 1475 struct kvm_s390_vm_cpu_processor *proc; 1476 u16 lowest_ibc, unblocked_ibc; 1477 int ret = 0; 1478 1479 mutex_lock(&kvm->lock); 1480 if (kvm->created_vcpus) { 1481 ret = -EBUSY; 1482 goto out; 1483 } 1484 proc = kzalloc_obj(*proc, GFP_KERNEL_ACCOUNT); 1485 if (!proc) { 1486 ret = -ENOMEM; 1487 goto out; 1488 } 1489 if (!copy_from_user(proc, (void __user *)attr->addr, 1490 sizeof(*proc))) { 1491 kvm->arch.model.cpuid = proc->cpuid; 1492 lowest_ibc = sclp.ibc >> 16 & 0xfff; 1493 unblocked_ibc = sclp.ibc & 0xfff; 1494 if (lowest_ibc && proc->ibc) { 1495 if (proc->ibc > unblocked_ibc) 1496 kvm->arch.model.ibc = unblocked_ibc; 1497 else if (proc->ibc < lowest_ibc) 1498 kvm->arch.model.ibc = lowest_ibc; 1499 else 1500 kvm->arch.model.ibc = proc->ibc; 1501 } 1502 memcpy(kvm->arch.model.fac_list, proc->fac_list, 1503 S390_ARCH_FAC_LIST_SIZE_BYTE); 1504 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx", 1505 kvm->arch.model.ibc, 1506 kvm->arch.model.cpuid); 1507 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx", 1508 kvm->arch.model.fac_list[0], 1509 kvm->arch.model.fac_list[1], 1510 kvm->arch.model.fac_list[2]); 1511 } else 1512 ret = -EFAULT; 1513 kfree(proc); 1514 out: 1515 mutex_unlock(&kvm->lock); 1516 return ret; 1517 } 1518 1519 static int kvm_s390_set_processor_feat(struct kvm *kvm, 1520 struct kvm_device_attr *attr) 1521 { 1522 struct kvm_s390_vm_cpu_feat data; 1523 1524 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data))) 1525 return -EFAULT; 1526 if (!bitmap_subset((unsigned long *) data.feat, 1527 kvm_s390_available_cpu_feat, 1528 KVM_S390_VM_CPU_FEAT_NR_BITS)) 1529 return -EINVAL; 1530 1531 mutex_lock(&kvm->lock); 1532 if (kvm->created_vcpus) { 1533 mutex_unlock(&kvm->lock); 1534 return -EBUSY; 1535 } 1536 bitmap_from_arr64(kvm->arch.cpu_feat, data.feat, KVM_S390_VM_CPU_FEAT_NR_BITS); 1537 mutex_unlock(&kvm->lock); 1538 VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx", 1539 data.feat[0], 1540 data.feat[1], 1541 data.feat[2]); 1542 return 0; 1543 } 1544 1545 static int kvm_s390_set_processor_subfunc(struct kvm *kvm, 1546 struct kvm_device_attr *attr) 1547 { 1548 mutex_lock(&kvm->lock); 1549 if (kvm->created_vcpus) { 1550 mutex_unlock(&kvm->lock); 1551 return -EBUSY; 1552 } 1553 1554 if (copy_from_user(&kvm->arch.model.subfuncs, (void __user *)attr->addr, 1555 sizeof(struct kvm_s390_vm_cpu_subfunc))) { 1556 mutex_unlock(&kvm->lock); 1557 return -EFAULT; 1558 } 1559 mutex_unlock(&kvm->lock); 1560 1561 VM_EVENT(kvm, 3, "SET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1562 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0], 1563 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1], 1564 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2], 1565 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]); 1566 VM_EVENT(kvm, 3, "SET: guest PTFF subfunc 0x%16.16lx.%16.16lx", 1567 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0], 1568 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]); 1569 VM_EVENT(kvm, 3, "SET: guest KMAC subfunc 0x%16.16lx.%16.16lx", 1570 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0], 1571 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]); 1572 VM_EVENT(kvm, 3, "SET: guest KMC subfunc 0x%16.16lx.%16.16lx", 1573 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0], 1574 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]); 1575 VM_EVENT(kvm, 3, "SET: guest KM subfunc 0x%16.16lx.%16.16lx", 1576 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0], 1577 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]); 1578 VM_EVENT(kvm, 3, "SET: guest KIMD subfunc 0x%16.16lx.%16.16lx", 1579 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0], 1580 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]); 1581 VM_EVENT(kvm, 3, "SET: guest KLMD subfunc 0x%16.16lx.%16.16lx", 1582 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0], 1583 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]); 1584 VM_EVENT(kvm, 3, "SET: guest PCKMO subfunc 0x%16.16lx.%16.16lx", 1585 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0], 1586 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]); 1587 VM_EVENT(kvm, 3, "SET: guest KMCTR subfunc 0x%16.16lx.%16.16lx", 1588 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0], 1589 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]); 1590 VM_EVENT(kvm, 3, "SET: guest KMF subfunc 0x%16.16lx.%16.16lx", 1591 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0], 1592 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]); 1593 VM_EVENT(kvm, 3, "SET: guest KMO subfunc 0x%16.16lx.%16.16lx", 1594 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0], 1595 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]); 1596 VM_EVENT(kvm, 3, "SET: guest PCC subfunc 0x%16.16lx.%16.16lx", 1597 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0], 1598 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]); 1599 VM_EVENT(kvm, 3, "SET: guest PPNO subfunc 0x%16.16lx.%16.16lx", 1600 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0], 1601 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]); 1602 VM_EVENT(kvm, 3, "SET: guest KMA subfunc 0x%16.16lx.%16.16lx", 1603 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0], 1604 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]); 1605 VM_EVENT(kvm, 3, "SET: guest KDSA subfunc 0x%16.16lx.%16.16lx", 1606 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0], 1607 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]); 1608 VM_EVENT(kvm, 3, "SET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1609 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0], 1610 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1], 1611 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2], 1612 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]); 1613 VM_EVENT(kvm, 3, "SET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1614 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0], 1615 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1], 1616 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2], 1617 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]); 1618 VM_EVENT(kvm, 3, "GET: guest PFCR subfunc 0x%16.16lx.%16.16lx", 1619 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[0], 1620 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[1]); 1621 1622 return 0; 1623 } 1624 1625 #define KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK \ 1626 ( \ 1627 ((struct kvm_s390_vm_cpu_uv_feat){ \ 1628 .ap = 1, \ 1629 .ap_intr = 1, \ 1630 }) \ 1631 .feat \ 1632 ) 1633 1634 static int kvm_s390_set_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr) 1635 { 1636 struct kvm_s390_vm_cpu_uv_feat __user *ptr = (void __user *)attr->addr; 1637 unsigned long data, filter; 1638 1639 filter = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK; 1640 if (get_user(data, &ptr->feat)) 1641 return -EFAULT; 1642 if (!bitmap_subset(&data, &filter, KVM_S390_VM_CPU_UV_FEAT_NR_BITS)) 1643 return -EINVAL; 1644 1645 mutex_lock(&kvm->lock); 1646 if (kvm->created_vcpus) { 1647 mutex_unlock(&kvm->lock); 1648 return -EBUSY; 1649 } 1650 kvm->arch.model.uv_feat_guest.feat = data; 1651 mutex_unlock(&kvm->lock); 1652 1653 VM_EVENT(kvm, 3, "SET: guest UV-feat: 0x%16.16lx", data); 1654 1655 return 0; 1656 } 1657 1658 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr) 1659 { 1660 int ret = -ENXIO; 1661 1662 switch (attr->attr) { 1663 case KVM_S390_VM_CPU_PROCESSOR: 1664 ret = kvm_s390_set_processor(kvm, attr); 1665 break; 1666 case KVM_S390_VM_CPU_PROCESSOR_FEAT: 1667 ret = kvm_s390_set_processor_feat(kvm, attr); 1668 break; 1669 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC: 1670 ret = kvm_s390_set_processor_subfunc(kvm, attr); 1671 break; 1672 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST: 1673 ret = kvm_s390_set_uv_feat(kvm, attr); 1674 break; 1675 } 1676 return ret; 1677 } 1678 1679 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr) 1680 { 1681 struct kvm_s390_vm_cpu_processor *proc; 1682 int ret = 0; 1683 1684 proc = kzalloc_obj(*proc, GFP_KERNEL_ACCOUNT); 1685 if (!proc) { 1686 ret = -ENOMEM; 1687 goto out; 1688 } 1689 proc->cpuid = kvm->arch.model.cpuid; 1690 proc->ibc = kvm->arch.model.ibc; 1691 memcpy(&proc->fac_list, kvm->arch.model.fac_list, 1692 S390_ARCH_FAC_LIST_SIZE_BYTE); 1693 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx", 1694 kvm->arch.model.ibc, 1695 kvm->arch.model.cpuid); 1696 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx", 1697 kvm->arch.model.fac_list[0], 1698 kvm->arch.model.fac_list[1], 1699 kvm->arch.model.fac_list[2]); 1700 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc))) 1701 ret = -EFAULT; 1702 kfree(proc); 1703 out: 1704 return ret; 1705 } 1706 1707 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr) 1708 { 1709 struct kvm_s390_vm_cpu_machine *mach; 1710 int ret = 0; 1711 1712 mach = kzalloc_obj(*mach, GFP_KERNEL_ACCOUNT); 1713 if (!mach) { 1714 ret = -ENOMEM; 1715 goto out; 1716 } 1717 get_cpu_id((struct cpuid *) &mach->cpuid); 1718 mach->ibc = sclp.ibc; 1719 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask, 1720 S390_ARCH_FAC_LIST_SIZE_BYTE); 1721 memcpy((unsigned long *)&mach->fac_list, stfle_fac_list, 1722 sizeof(stfle_fac_list)); 1723 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx", 1724 kvm->arch.model.ibc, 1725 kvm->arch.model.cpuid); 1726 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx", 1727 mach->fac_mask[0], 1728 mach->fac_mask[1], 1729 mach->fac_mask[2]); 1730 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx", 1731 mach->fac_list[0], 1732 mach->fac_list[1], 1733 mach->fac_list[2]); 1734 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach))) 1735 ret = -EFAULT; 1736 kfree(mach); 1737 out: 1738 return ret; 1739 } 1740 1741 static int kvm_s390_get_processor_feat(struct kvm *kvm, 1742 struct kvm_device_attr *attr) 1743 { 1744 struct kvm_s390_vm_cpu_feat data; 1745 1746 bitmap_to_arr64(data.feat, kvm->arch.cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS); 1747 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data))) 1748 return -EFAULT; 1749 VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx", 1750 data.feat[0], 1751 data.feat[1], 1752 data.feat[2]); 1753 return 0; 1754 } 1755 1756 static int kvm_s390_get_machine_feat(struct kvm *kvm, 1757 struct kvm_device_attr *attr) 1758 { 1759 struct kvm_s390_vm_cpu_feat data; 1760 1761 bitmap_to_arr64(data.feat, kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS); 1762 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data))) 1763 return -EFAULT; 1764 VM_EVENT(kvm, 3, "GET: host feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx", 1765 data.feat[0], 1766 data.feat[1], 1767 data.feat[2]); 1768 return 0; 1769 } 1770 1771 static int kvm_s390_get_processor_subfunc(struct kvm *kvm, 1772 struct kvm_device_attr *attr) 1773 { 1774 if (copy_to_user((void __user *)attr->addr, &kvm->arch.model.subfuncs, 1775 sizeof(struct kvm_s390_vm_cpu_subfunc))) 1776 return -EFAULT; 1777 1778 VM_EVENT(kvm, 3, "GET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1779 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0], 1780 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1], 1781 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2], 1782 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]); 1783 VM_EVENT(kvm, 3, "GET: guest PTFF subfunc 0x%16.16lx.%16.16lx", 1784 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0], 1785 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]); 1786 VM_EVENT(kvm, 3, "GET: guest KMAC subfunc 0x%16.16lx.%16.16lx", 1787 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0], 1788 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]); 1789 VM_EVENT(kvm, 3, "GET: guest KMC subfunc 0x%16.16lx.%16.16lx", 1790 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0], 1791 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]); 1792 VM_EVENT(kvm, 3, "GET: guest KM subfunc 0x%16.16lx.%16.16lx", 1793 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0], 1794 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]); 1795 VM_EVENT(kvm, 3, "GET: guest KIMD subfunc 0x%16.16lx.%16.16lx", 1796 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0], 1797 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]); 1798 VM_EVENT(kvm, 3, "GET: guest KLMD subfunc 0x%16.16lx.%16.16lx", 1799 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0], 1800 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]); 1801 VM_EVENT(kvm, 3, "GET: guest PCKMO subfunc 0x%16.16lx.%16.16lx", 1802 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0], 1803 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]); 1804 VM_EVENT(kvm, 3, "GET: guest KMCTR subfunc 0x%16.16lx.%16.16lx", 1805 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0], 1806 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]); 1807 VM_EVENT(kvm, 3, "GET: guest KMF subfunc 0x%16.16lx.%16.16lx", 1808 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0], 1809 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]); 1810 VM_EVENT(kvm, 3, "GET: guest KMO subfunc 0x%16.16lx.%16.16lx", 1811 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0], 1812 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]); 1813 VM_EVENT(kvm, 3, "GET: guest PCC subfunc 0x%16.16lx.%16.16lx", 1814 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0], 1815 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]); 1816 VM_EVENT(kvm, 3, "GET: guest PPNO subfunc 0x%16.16lx.%16.16lx", 1817 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0], 1818 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]); 1819 VM_EVENT(kvm, 3, "GET: guest KMA subfunc 0x%16.16lx.%16.16lx", 1820 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0], 1821 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]); 1822 VM_EVENT(kvm, 3, "GET: guest KDSA subfunc 0x%16.16lx.%16.16lx", 1823 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0], 1824 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]); 1825 VM_EVENT(kvm, 3, "GET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1826 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0], 1827 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1], 1828 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2], 1829 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]); 1830 VM_EVENT(kvm, 3, "GET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1831 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0], 1832 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1], 1833 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2], 1834 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]); 1835 VM_EVENT(kvm, 3, "GET: guest PFCR subfunc 0x%16.16lx.%16.16lx", 1836 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[0], 1837 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[1]); 1838 1839 return 0; 1840 } 1841 1842 static int kvm_s390_get_machine_subfunc(struct kvm *kvm, 1843 struct kvm_device_attr *attr) 1844 { 1845 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc, 1846 sizeof(struct kvm_s390_vm_cpu_subfunc))) 1847 return -EFAULT; 1848 1849 VM_EVENT(kvm, 3, "GET: host PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1850 ((unsigned long *) &kvm_s390_available_subfunc.plo)[0], 1851 ((unsigned long *) &kvm_s390_available_subfunc.plo)[1], 1852 ((unsigned long *) &kvm_s390_available_subfunc.plo)[2], 1853 ((unsigned long *) &kvm_s390_available_subfunc.plo)[3]); 1854 VM_EVENT(kvm, 3, "GET: host PTFF subfunc 0x%16.16lx.%16.16lx", 1855 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[0], 1856 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[1]); 1857 VM_EVENT(kvm, 3, "GET: host KMAC subfunc 0x%16.16lx.%16.16lx", 1858 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[0], 1859 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[1]); 1860 VM_EVENT(kvm, 3, "GET: host KMC subfunc 0x%16.16lx.%16.16lx", 1861 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[0], 1862 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[1]); 1863 VM_EVENT(kvm, 3, "GET: host KM subfunc 0x%16.16lx.%16.16lx", 1864 ((unsigned long *) &kvm_s390_available_subfunc.km)[0], 1865 ((unsigned long *) &kvm_s390_available_subfunc.km)[1]); 1866 VM_EVENT(kvm, 3, "GET: host KIMD subfunc 0x%16.16lx.%16.16lx", 1867 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[0], 1868 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[1]); 1869 VM_EVENT(kvm, 3, "GET: host KLMD subfunc 0x%16.16lx.%16.16lx", 1870 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[0], 1871 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[1]); 1872 VM_EVENT(kvm, 3, "GET: host PCKMO subfunc 0x%16.16lx.%16.16lx", 1873 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[0], 1874 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[1]); 1875 VM_EVENT(kvm, 3, "GET: host KMCTR subfunc 0x%16.16lx.%16.16lx", 1876 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[0], 1877 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[1]); 1878 VM_EVENT(kvm, 3, "GET: host KMF subfunc 0x%16.16lx.%16.16lx", 1879 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[0], 1880 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[1]); 1881 VM_EVENT(kvm, 3, "GET: host KMO subfunc 0x%16.16lx.%16.16lx", 1882 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[0], 1883 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[1]); 1884 VM_EVENT(kvm, 3, "GET: host PCC subfunc 0x%16.16lx.%16.16lx", 1885 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[0], 1886 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[1]); 1887 VM_EVENT(kvm, 3, "GET: host PPNO subfunc 0x%16.16lx.%16.16lx", 1888 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[0], 1889 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[1]); 1890 VM_EVENT(kvm, 3, "GET: host KMA subfunc 0x%16.16lx.%16.16lx", 1891 ((unsigned long *) &kvm_s390_available_subfunc.kma)[0], 1892 ((unsigned long *) &kvm_s390_available_subfunc.kma)[1]); 1893 VM_EVENT(kvm, 3, "GET: host KDSA subfunc 0x%16.16lx.%16.16lx", 1894 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[0], 1895 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[1]); 1896 VM_EVENT(kvm, 3, "GET: host SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1897 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[0], 1898 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[1], 1899 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[2], 1900 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[3]); 1901 VM_EVENT(kvm, 3, "GET: host DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1902 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[0], 1903 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[1], 1904 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[2], 1905 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[3]); 1906 VM_EVENT(kvm, 3, "GET: host PFCR subfunc 0x%16.16lx.%16.16lx", 1907 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[0], 1908 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[1]); 1909 1910 return 0; 1911 } 1912 1913 static int kvm_s390_get_processor_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr) 1914 { 1915 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr; 1916 unsigned long feat = kvm->arch.model.uv_feat_guest.feat; 1917 1918 if (put_user(feat, &dst->feat)) 1919 return -EFAULT; 1920 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat); 1921 1922 return 0; 1923 } 1924 1925 static int kvm_s390_get_machine_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr) 1926 { 1927 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr; 1928 unsigned long feat; 1929 1930 BUILD_BUG_ON(sizeof(*dst) != sizeof(uv_info.uv_feature_indications)); 1931 1932 feat = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK; 1933 if (put_user(feat, &dst->feat)) 1934 return -EFAULT; 1935 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat); 1936 1937 return 0; 1938 } 1939 1940 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr) 1941 { 1942 int ret = -ENXIO; 1943 1944 switch (attr->attr) { 1945 case KVM_S390_VM_CPU_PROCESSOR: 1946 ret = kvm_s390_get_processor(kvm, attr); 1947 break; 1948 case KVM_S390_VM_CPU_MACHINE: 1949 ret = kvm_s390_get_machine(kvm, attr); 1950 break; 1951 case KVM_S390_VM_CPU_PROCESSOR_FEAT: 1952 ret = kvm_s390_get_processor_feat(kvm, attr); 1953 break; 1954 case KVM_S390_VM_CPU_MACHINE_FEAT: 1955 ret = kvm_s390_get_machine_feat(kvm, attr); 1956 break; 1957 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC: 1958 ret = kvm_s390_get_processor_subfunc(kvm, attr); 1959 break; 1960 case KVM_S390_VM_CPU_MACHINE_SUBFUNC: 1961 ret = kvm_s390_get_machine_subfunc(kvm, attr); 1962 break; 1963 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST: 1964 ret = kvm_s390_get_processor_uv_feat(kvm, attr); 1965 break; 1966 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST: 1967 ret = kvm_s390_get_machine_uv_feat(kvm, attr); 1968 break; 1969 } 1970 return ret; 1971 } 1972 1973 /** 1974 * kvm_s390_update_topology_change_report - update CPU topology change report 1975 * @kvm: guest KVM description 1976 * @val: set or clear the MTCR bit 1977 * 1978 * Updates the Multiprocessor Topology-Change-Report bit to signal 1979 * the guest with a topology change. 1980 * This is only relevant if the topology facility is present. 1981 */ 1982 static void kvm_s390_update_topology_change_report(struct kvm *kvm, bool val) 1983 { 1984 union sca_utility new, old; 1985 struct esca_block *sca; 1986 1987 sca = kvm->arch.sca; 1988 old = READ_ONCE(sca->utility); 1989 do { 1990 new = old; 1991 new.mtcr = val; 1992 } while (!try_cmpxchg(&sca->utility.val, &old.val, new.val)); 1993 } 1994 1995 static int kvm_s390_set_topo_change_indication(struct kvm *kvm, 1996 struct kvm_device_attr *attr) 1997 { 1998 if (!test_kvm_facility(kvm, 11)) 1999 return -ENXIO; 2000 2001 kvm_s390_update_topology_change_report(kvm, !!attr->attr); 2002 return 0; 2003 } 2004 2005 static int kvm_s390_get_topo_change_indication(struct kvm *kvm, 2006 struct kvm_device_attr *attr) 2007 { 2008 u8 topo; 2009 2010 if (!test_kvm_facility(kvm, 11)) 2011 return -ENXIO; 2012 2013 topo = kvm->arch.sca->utility.mtcr; 2014 2015 return put_user(topo, (u8 __user *)attr->addr); 2016 } 2017 2018 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr) 2019 { 2020 int ret; 2021 2022 switch (attr->group) { 2023 case KVM_S390_VM_MEM_CTRL: 2024 ret = kvm_s390_set_mem_control(kvm, attr); 2025 break; 2026 case KVM_S390_VM_TOD: 2027 ret = kvm_s390_set_tod(kvm, attr); 2028 break; 2029 case KVM_S390_VM_CPU_MODEL: 2030 ret = kvm_s390_set_cpu_model(kvm, attr); 2031 break; 2032 case KVM_S390_VM_CRYPTO: 2033 ret = kvm_s390_vm_set_crypto(kvm, attr); 2034 break; 2035 case KVM_S390_VM_MIGRATION: 2036 ret = kvm_s390_vm_set_migration(kvm, attr); 2037 break; 2038 case KVM_S390_VM_CPU_TOPOLOGY: 2039 ret = kvm_s390_set_topo_change_indication(kvm, attr); 2040 break; 2041 default: 2042 ret = -ENXIO; 2043 break; 2044 } 2045 2046 return ret; 2047 } 2048 2049 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr) 2050 { 2051 int ret; 2052 2053 switch (attr->group) { 2054 case KVM_S390_VM_MEM_CTRL: 2055 ret = kvm_s390_get_mem_control(kvm, attr); 2056 break; 2057 case KVM_S390_VM_TOD: 2058 ret = kvm_s390_get_tod(kvm, attr); 2059 break; 2060 case KVM_S390_VM_CPU_MODEL: 2061 ret = kvm_s390_get_cpu_model(kvm, attr); 2062 break; 2063 case KVM_S390_VM_MIGRATION: 2064 ret = kvm_s390_vm_get_migration(kvm, attr); 2065 break; 2066 case KVM_S390_VM_CPU_TOPOLOGY: 2067 ret = kvm_s390_get_topo_change_indication(kvm, attr); 2068 break; 2069 default: 2070 ret = -ENXIO; 2071 break; 2072 } 2073 2074 return ret; 2075 } 2076 2077 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr) 2078 { 2079 int ret; 2080 2081 switch (attr->group) { 2082 case KVM_S390_VM_MEM_CTRL: 2083 switch (attr->attr) { 2084 case KVM_S390_VM_MEM_ENABLE_CMMA: 2085 case KVM_S390_VM_MEM_CLR_CMMA: 2086 ret = sclp.has_cmma ? 0 : -ENXIO; 2087 break; 2088 case KVM_S390_VM_MEM_LIMIT_SIZE: 2089 ret = 0; 2090 break; 2091 default: 2092 ret = -ENXIO; 2093 break; 2094 } 2095 break; 2096 case KVM_S390_VM_TOD: 2097 switch (attr->attr) { 2098 case KVM_S390_VM_TOD_LOW: 2099 case KVM_S390_VM_TOD_HIGH: 2100 ret = 0; 2101 break; 2102 default: 2103 ret = -ENXIO; 2104 break; 2105 } 2106 break; 2107 case KVM_S390_VM_CPU_MODEL: 2108 switch (attr->attr) { 2109 case KVM_S390_VM_CPU_PROCESSOR: 2110 case KVM_S390_VM_CPU_MACHINE: 2111 case KVM_S390_VM_CPU_PROCESSOR_FEAT: 2112 case KVM_S390_VM_CPU_MACHINE_FEAT: 2113 case KVM_S390_VM_CPU_MACHINE_SUBFUNC: 2114 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC: 2115 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST: 2116 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST: 2117 ret = 0; 2118 break; 2119 default: 2120 ret = -ENXIO; 2121 break; 2122 } 2123 break; 2124 case KVM_S390_VM_CRYPTO: 2125 switch (attr->attr) { 2126 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW: 2127 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW: 2128 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW: 2129 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW: 2130 ret = 0; 2131 break; 2132 case KVM_S390_VM_CRYPTO_ENABLE_APIE: 2133 case KVM_S390_VM_CRYPTO_DISABLE_APIE: 2134 ret = ap_instructions_available() ? 0 : -ENXIO; 2135 break; 2136 default: 2137 ret = -ENXIO; 2138 break; 2139 } 2140 break; 2141 case KVM_S390_VM_MIGRATION: 2142 ret = 0; 2143 break; 2144 case KVM_S390_VM_CPU_TOPOLOGY: 2145 ret = test_kvm_facility(kvm, 11) ? 0 : -ENXIO; 2146 break; 2147 default: 2148 ret = -ENXIO; 2149 break; 2150 } 2151 2152 return ret; 2153 } 2154 2155 static int kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args) 2156 { 2157 union skey *keys; 2158 int i, r = 0; 2159 2160 if (args->flags != 0) 2161 return -EINVAL; 2162 2163 /* Is this guest using storage keys? */ 2164 if (!uses_skeys(kvm->arch.gmap)) 2165 return KVM_S390_GET_SKEYS_NONE; 2166 2167 /* Enforce sane limit on memory allocation */ 2168 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX) 2169 return -EINVAL; 2170 2171 keys = kvmalloc_objs(*keys, args->count, GFP_KERNEL_ACCOUNT); 2172 if (!keys) 2173 return -ENOMEM; 2174 2175 scoped_guard(read_lock, &kvm->mmu_lock) { 2176 for (i = 0; i < args->count; i++) { 2177 r = dat_get_storage_key(kvm->arch.gmap->asce, 2178 args->start_gfn + i, keys + i); 2179 if (r) 2180 break; 2181 } 2182 } 2183 2184 if (!r) { 2185 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys, 2186 sizeof(uint8_t) * args->count); 2187 if (r) 2188 r = -EFAULT; 2189 } 2190 2191 kvfree(keys); 2192 return r <= 0 ? r : -EFAULT; 2193 } 2194 2195 static int kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args) 2196 { 2197 struct kvm_s390_mmu_cache *mc; 2198 union skey *keys; 2199 int i, r = 0; 2200 2201 if (args->flags != 0) 2202 return -EINVAL; 2203 2204 /* Enforce sane limit on memory allocation */ 2205 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX) 2206 return -EINVAL; 2207 2208 keys = kvmalloc_objs(*keys, args->count, GFP_KERNEL_ACCOUNT); 2209 if (!keys) 2210 return -ENOMEM; 2211 2212 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr, 2213 sizeof(uint8_t) * args->count); 2214 if (r) { 2215 r = -EFAULT; 2216 goto out; 2217 } 2218 2219 /* Enable storage key handling for the guest */ 2220 r = gmap_enable_skeys(kvm->arch.gmap); 2221 if (r) 2222 goto out; 2223 2224 r = -EINVAL; 2225 for (i = 0; i < args->count; i++) { 2226 /* Lowest order bit is reserved */ 2227 if (keys[i].zero) 2228 goto out; 2229 } 2230 2231 mc = kvm_s390_new_mmu_cache(); 2232 if (!mc) { 2233 r = -ENOMEM; 2234 goto out; 2235 } 2236 2237 r = 0; 2238 do { 2239 r = kvm_s390_mmu_cache_topup(mc); 2240 if (r == -ENOMEM) 2241 break; 2242 scoped_guard(read_lock, &kvm->mmu_lock) { 2243 for (i = 0 ; i < args->count; i++) { 2244 r = dat_set_storage_key(mc, kvm->arch.gmap->asce, 2245 args->start_gfn + i, keys[i], 0); 2246 if (r) 2247 break; 2248 } 2249 } 2250 } while (r == -ENOMEM); 2251 kvm_s390_free_mmu_cache(mc); 2252 out: 2253 kvfree(keys); 2254 return r <= 0 ? r : -EFAULT; 2255 } 2256 2257 /* 2258 * This function searches for the next page with dirty CMMA attributes, and 2259 * saves the attributes in the buffer up to either the end of the buffer or 2260 * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found; 2261 * no trailing clean bytes are saved. 2262 * In case no dirty bits were found, or if CMMA was not enabled or used, the 2263 * output buffer will indicate 0 as length. 2264 */ 2265 static int kvm_s390_get_cmma_bits(struct kvm *kvm, 2266 struct kvm_s390_cmma_log *args) 2267 { 2268 int peek, ret; 2269 u8 *values; 2270 2271 if (!kvm->arch.use_cmma) 2272 return -ENXIO; 2273 /* Invalid/unsupported flags were specified */ 2274 if (args->flags & ~KVM_S390_CMMA_PEEK) 2275 return -EINVAL; 2276 /* Migration mode query, and we are not doing a migration */ 2277 peek = !!(args->flags & KVM_S390_CMMA_PEEK); 2278 if (!peek && !kvm->arch.migration_mode) 2279 return -EINVAL; 2280 /* CMMA is disabled or was not used, or the buffer has length zero */ 2281 args->count = min(args->count, KVM_S390_CMMA_SIZE_MAX); 2282 if (!args->count || !uses_cmm(kvm->arch.gmap)) { 2283 memset(args, 0, sizeof(*args)); 2284 return 0; 2285 } 2286 /* We are not peeking, and there are no dirty pages */ 2287 if (!peek && !atomic64_read(&kvm->arch.cmma_dirty_pages)) { 2288 memset(args, 0, sizeof(*args)); 2289 return 0; 2290 } 2291 2292 values = vzalloc(args->count); 2293 if (!values) 2294 return -ENOMEM; 2295 2296 scoped_guard(read_lock, &kvm->mmu_lock) { 2297 if (peek) 2298 ret = dat_peek_cmma(args->start_gfn, kvm->arch.gmap->asce, &args->count, 2299 values); 2300 else 2301 ret = dat_get_cmma(kvm->arch.gmap->asce, &args->start_gfn, &args->count, 2302 values, &kvm->arch.cmma_dirty_pages); 2303 } 2304 2305 if (kvm->arch.migration_mode) 2306 args->remaining = atomic64_read(&kvm->arch.cmma_dirty_pages); 2307 else 2308 args->remaining = 0; 2309 2310 if (copy_to_user((void __user *)args->values, values, args->count)) 2311 ret = -EFAULT; 2312 2313 vfree(values); 2314 return ret; 2315 } 2316 2317 /* 2318 * This function sets the CMMA attributes for the given pages. If the input 2319 * buffer has zero length, no action is taken, otherwise the attributes are 2320 * set and the mm->context.uses_cmm flag is set. 2321 */ 2322 static int kvm_s390_set_cmma_bits(struct kvm *kvm, 2323 const struct kvm_s390_cmma_log *args) 2324 { 2325 struct kvm_s390_mmu_cache *mc __free(kvm_s390_mmu_cache) = NULL; 2326 u8 *bits __free(kvfree) = NULL; 2327 int r = 0; 2328 2329 if (!kvm->arch.use_cmma) 2330 return -ENXIO; 2331 /* invalid/unsupported flags */ 2332 if (args->flags != 0) 2333 return -EINVAL; 2334 /* Enforce sane limit on memory allocation */ 2335 if (args->count > KVM_S390_CMMA_SIZE_MAX) 2336 return -EINVAL; 2337 /* Nothing to do */ 2338 if (args->count == 0) 2339 return 0; 2340 2341 mc = kvm_s390_new_mmu_cache(); 2342 if (!mc) 2343 return -ENOMEM; 2344 bits = vmalloc(array_size(sizeof(*bits), args->count)); 2345 if (!bits) 2346 return -ENOMEM; 2347 2348 r = copy_from_user(bits, (void __user *)args->values, args->count); 2349 if (r) 2350 return -EFAULT; 2351 2352 do { 2353 r = kvm_s390_mmu_cache_topup(mc); 2354 if (r) 2355 return r; 2356 scoped_guard(read_lock, &kvm->mmu_lock) { 2357 r = dat_set_cmma_bits(mc, kvm->arch.gmap->asce, args->start_gfn, 2358 args->count, args->mask, bits); 2359 } 2360 } while (r == -ENOMEM); 2361 2362 set_bit(GMAP_FLAG_USES_CMM, &kvm->arch.gmap->flags); 2363 2364 return r <= 0 ? r : -EFAULT; 2365 } 2366 2367 /** 2368 * kvm_s390_cpus_from_pv - Convert all protected vCPUs in a protected VM to 2369 * non protected. 2370 * @kvm: the VM whose protected vCPUs are to be converted 2371 * @rc: return value for the RC field of the UVC (in case of error) 2372 * @rrc: return value for the RRC field of the UVC (in case of error) 2373 * 2374 * Does not stop in case of error, tries to convert as many 2375 * CPUs as possible. In case of error, the RC and RRC of the last error are 2376 * returned. 2377 * 2378 * Return: 0 in case of success, otherwise -EIO 2379 */ 2380 int kvm_s390_cpus_from_pv(struct kvm *kvm, u16 *rc, u16 *rrc) 2381 { 2382 struct kvm_vcpu *vcpu; 2383 unsigned long i; 2384 u16 _rc, _rrc; 2385 int ret = 0; 2386 2387 /* 2388 * We ignore failures and try to destroy as many CPUs as possible. 2389 * At the same time we must not free the assigned resources when 2390 * this fails, as the ultravisor has still access to that memory. 2391 * So kvm_s390_pv_destroy_cpu can leave a "wanted" memory leak 2392 * behind. 2393 * We want to return the first failure rc and rrc, though. 2394 */ 2395 kvm_for_each_vcpu(i, vcpu, kvm) { 2396 mutex_lock(&vcpu->mutex); 2397 if (kvm_s390_pv_destroy_cpu(vcpu, &_rc, &_rrc) && !ret) { 2398 *rc = _rc; 2399 *rrc = _rrc; 2400 ret = -EIO; 2401 } 2402 mutex_unlock(&vcpu->mutex); 2403 } 2404 /* Ensure that we re-enable gisa if the non-PV guest used it but the PV guest did not. */ 2405 if (use_gisa) 2406 kvm_s390_gisa_enable(kvm); 2407 return ret; 2408 } 2409 2410 /** 2411 * kvm_s390_cpus_to_pv - Convert all non-protected vCPUs in a protected VM 2412 * to protected. 2413 * @kvm: the VM whose protected vCPUs are to be converted 2414 * @rc: return value for the RC field of the UVC (in case of error) 2415 * @rrc: return value for the RRC field of the UVC (in case of error) 2416 * 2417 * Tries to undo the conversion in case of error. 2418 * 2419 * Return: 0 in case of success, otherwise -EIO 2420 */ 2421 static int kvm_s390_cpus_to_pv(struct kvm *kvm, u16 *rc, u16 *rrc) 2422 { 2423 unsigned long i; 2424 int r = 0; 2425 u16 dummy; 2426 2427 struct kvm_vcpu *vcpu; 2428 2429 /* Disable the GISA if the ultravisor does not support AIV. */ 2430 if (!uv_has_feature(BIT_UV_FEAT_AIV)) 2431 kvm_s390_gisa_disable(kvm); 2432 2433 kvm_for_each_vcpu(i, vcpu, kvm) { 2434 mutex_lock(&vcpu->mutex); 2435 r = kvm_s390_pv_create_cpu(vcpu, rc, rrc); 2436 mutex_unlock(&vcpu->mutex); 2437 if (r) 2438 break; 2439 } 2440 if (r) 2441 kvm_s390_cpus_from_pv(kvm, &dummy, &dummy); 2442 return r; 2443 } 2444 2445 /* 2446 * Here we provide user space with a direct interface to query UV 2447 * related data like UV maxima and available features as well as 2448 * feature specific data. 2449 * 2450 * To facilitate future extension of the data structures we'll try to 2451 * write data up to the maximum requested length. 2452 */ 2453 static ssize_t kvm_s390_handle_pv_info(struct kvm_s390_pv_info *info) 2454 { 2455 ssize_t len_min; 2456 2457 switch (info->header.id) { 2458 case KVM_PV_INFO_VM: { 2459 len_min = sizeof(info->header) + sizeof(info->vm); 2460 2461 if (info->header.len_max < len_min) 2462 return -EINVAL; 2463 2464 memcpy(info->vm.inst_calls_list, 2465 uv_info.inst_calls_list, 2466 sizeof(uv_info.inst_calls_list)); 2467 2468 /* It's max cpuid not max cpus, so it's off by one */ 2469 info->vm.max_cpus = uv_info.max_guest_cpu_id + 1; 2470 info->vm.max_guests = uv_info.max_num_sec_conf; 2471 info->vm.max_guest_addr = uv_info.max_sec_stor_addr; 2472 info->vm.feature_indication = uv_info.uv_feature_indications; 2473 2474 return len_min; 2475 } 2476 case KVM_PV_INFO_DUMP: { 2477 len_min = sizeof(info->header) + sizeof(info->dump); 2478 2479 if (info->header.len_max < len_min) 2480 return -EINVAL; 2481 2482 info->dump.dump_cpu_buffer_len = uv_info.guest_cpu_stor_len; 2483 info->dump.dump_config_mem_buffer_per_1m = uv_info.conf_dump_storage_state_len; 2484 info->dump.dump_config_finalize_len = uv_info.conf_dump_finalize_len; 2485 return len_min; 2486 } 2487 default: 2488 return -EINVAL; 2489 } 2490 } 2491 2492 static int kvm_s390_pv_dmp(struct kvm *kvm, struct kvm_pv_cmd *cmd, 2493 struct kvm_s390_pv_dmp dmp) 2494 { 2495 int r = -EINVAL; 2496 void __user *result_buff = (void __user *)dmp.buff_addr; 2497 2498 switch (dmp.subcmd) { 2499 case KVM_PV_DUMP_INIT: { 2500 if (kvm->arch.pv.dumping) 2501 break; 2502 2503 /* 2504 * Block SIE entry as concurrent dump UVCs could lead 2505 * to validities. 2506 */ 2507 kvm_s390_vcpu_block_all(kvm); 2508 2509 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm), 2510 UVC_CMD_DUMP_INIT, &cmd->rc, &cmd->rrc); 2511 KVM_UV_EVENT(kvm, 3, "PROTVIRT DUMP INIT: rc %x rrc %x", 2512 cmd->rc, cmd->rrc); 2513 if (!r) { 2514 kvm->arch.pv.dumping = true; 2515 } else { 2516 kvm_s390_vcpu_unblock_all(kvm); 2517 r = -EINVAL; 2518 } 2519 break; 2520 } 2521 case KVM_PV_DUMP_CONFIG_STOR_STATE: { 2522 if (!kvm->arch.pv.dumping) 2523 break; 2524 2525 /* 2526 * gaddr is an output parameter since we might stop 2527 * early. As dmp will be copied back in our caller, we 2528 * don't need to do it ourselves. 2529 */ 2530 r = kvm_s390_pv_dump_stor_state(kvm, result_buff, &dmp.gaddr, dmp.buff_len, 2531 &cmd->rc, &cmd->rrc); 2532 break; 2533 } 2534 case KVM_PV_DUMP_COMPLETE: { 2535 if (!kvm->arch.pv.dumping) 2536 break; 2537 2538 r = -EINVAL; 2539 if (dmp.buff_len < uv_info.conf_dump_finalize_len) 2540 break; 2541 2542 r = kvm_s390_pv_dump_complete(kvm, result_buff, 2543 &cmd->rc, &cmd->rrc); 2544 break; 2545 } 2546 default: 2547 r = -ENOTTY; 2548 break; 2549 } 2550 2551 return r; 2552 } 2553 2554 static int kvm_s390_handle_pv(struct kvm *kvm, struct kvm_pv_cmd *cmd) 2555 { 2556 const bool need_lock = (cmd->cmd != KVM_PV_ASYNC_CLEANUP_PERFORM); 2557 void __user *argp = (void __user *)cmd->data; 2558 int r = 0; 2559 u16 dummy; 2560 2561 if (need_lock) 2562 mutex_lock(&kvm->lock); 2563 2564 switch (cmd->cmd) { 2565 case KVM_PV_ENABLE: { 2566 r = -EINVAL; 2567 if (kvm_s390_pv_is_protected(kvm)) 2568 break; 2569 2570 kvm_s390_unmap_all_adapters(kvm); 2571 mmap_write_lock(kvm->mm); 2572 /* 2573 * Disable creation of new THPs. Existing THPs can stay, they 2574 * will be split when any part of them gets imported. 2575 */ 2576 mm_flags_clear(MMF_DISABLE_THP_EXCEPT_ADVISED, kvm->mm); 2577 mm_flags_set(MMF_DISABLE_THP_COMPLETELY, kvm->mm); 2578 set_bit(GMAP_FLAG_EXPORT_ON_UNMAP, &kvm->arch.gmap->flags); 2579 r = gmap_helper_disable_cow_sharing(); 2580 mmap_write_unlock(kvm->mm); 2581 if (r) 2582 break; 2583 2584 r = kvm_s390_pv_init_vm(kvm, &cmd->rc, &cmd->rrc); 2585 if (r) 2586 break; 2587 2588 r = kvm_s390_cpus_to_pv(kvm, &cmd->rc, &cmd->rrc); 2589 if (r) 2590 kvm_s390_pv_deinit_vm(kvm, &dummy, &dummy); 2591 2592 /* we need to block service interrupts from now on */ 2593 set_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs); 2594 break; 2595 } 2596 case KVM_PV_ASYNC_CLEANUP_PREPARE: 2597 r = -EINVAL; 2598 if (!kvm_s390_pv_is_protected(kvm) || !async_destroy) 2599 break; 2600 2601 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc); 2602 /* 2603 * If a CPU could not be destroyed, destroy VM will also fail. 2604 * There is no point in trying to destroy it. Instead return 2605 * the rc and rrc from the first CPU that failed destroying. 2606 */ 2607 if (r) 2608 break; 2609 r = kvm_s390_pv_set_aside(kvm, &cmd->rc, &cmd->rrc); 2610 2611 /* no need to block service interrupts any more */ 2612 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs); 2613 break; 2614 case KVM_PV_ASYNC_CLEANUP_PERFORM: 2615 r = -EINVAL; 2616 if (!async_destroy) 2617 break; 2618 /* kvm->lock must not be held; this is asserted inside the function. */ 2619 r = kvm_s390_pv_deinit_aside_vm(kvm, &cmd->rc, &cmd->rrc); 2620 break; 2621 case KVM_PV_DISABLE: { 2622 r = -EINVAL; 2623 if (!kvm_s390_pv_is_protected(kvm)) 2624 break; 2625 2626 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc); 2627 /* 2628 * If a CPU could not be destroyed, destroy VM will also fail. 2629 * There is no point in trying to destroy it. Instead return 2630 * the rc and rrc from the first CPU that failed destroying. 2631 */ 2632 if (r) 2633 break; 2634 r = kvm_s390_pv_deinit_cleanup_all(kvm, &cmd->rc, &cmd->rrc); 2635 2636 /* no need to block service interrupts any more */ 2637 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs); 2638 break; 2639 } 2640 case KVM_PV_SET_SEC_PARMS: { 2641 struct kvm_s390_pv_sec_parm parms = {}; 2642 void *hdr; 2643 2644 r = -EINVAL; 2645 if (!kvm_s390_pv_is_protected(kvm)) 2646 break; 2647 2648 r = -EFAULT; 2649 if (copy_from_user(&parms, argp, sizeof(parms))) 2650 break; 2651 2652 /* Currently restricted to 1MiB */ 2653 r = -EINVAL; 2654 if (parms.length > SZ_1M) 2655 break; 2656 2657 r = -ENOMEM; 2658 hdr = vmalloc(parms.length); 2659 if (!hdr) 2660 break; 2661 2662 r = -EFAULT; 2663 if (!copy_from_user(hdr, (void __user *)parms.origin, 2664 parms.length)) 2665 r = kvm_s390_pv_set_sec_parms(kvm, hdr, parms.length, 2666 &cmd->rc, &cmd->rrc); 2667 2668 vfree(hdr); 2669 break; 2670 } 2671 case KVM_PV_UNPACK: { 2672 struct kvm_s390_pv_unp unp = {}; 2673 2674 r = -EINVAL; 2675 if (!kvm_s390_pv_is_protected(kvm) || !mm_is_protected(kvm->mm)) 2676 break; 2677 2678 r = -EFAULT; 2679 if (copy_from_user(&unp, argp, sizeof(unp))) 2680 break; 2681 2682 r = kvm_s390_pv_unpack(kvm, unp.addr, unp.size, unp.tweak, 2683 &cmd->rc, &cmd->rrc); 2684 break; 2685 } 2686 case KVM_PV_VERIFY: { 2687 r = -EINVAL; 2688 if (!kvm_s390_pv_is_protected(kvm)) 2689 break; 2690 2691 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm), 2692 UVC_CMD_VERIFY_IMG, &cmd->rc, &cmd->rrc); 2693 KVM_UV_EVENT(kvm, 3, "PROTVIRT VERIFY: rc %x rrc %x", cmd->rc, 2694 cmd->rrc); 2695 break; 2696 } 2697 case KVM_PV_PREP_RESET: { 2698 r = -EINVAL; 2699 if (!kvm_s390_pv_is_protected(kvm)) 2700 break; 2701 2702 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm), 2703 UVC_CMD_PREPARE_RESET, &cmd->rc, &cmd->rrc); 2704 KVM_UV_EVENT(kvm, 3, "PROTVIRT PREP RESET: rc %x rrc %x", 2705 cmd->rc, cmd->rrc); 2706 break; 2707 } 2708 case KVM_PV_UNSHARE_ALL: { 2709 r = -EINVAL; 2710 if (!kvm_s390_pv_is_protected(kvm)) 2711 break; 2712 2713 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm), 2714 UVC_CMD_SET_UNSHARE_ALL, &cmd->rc, &cmd->rrc); 2715 KVM_UV_EVENT(kvm, 3, "PROTVIRT UNSHARE: rc %x rrc %x", 2716 cmd->rc, cmd->rrc); 2717 break; 2718 } 2719 case KVM_PV_INFO: { 2720 struct kvm_s390_pv_info info = {}; 2721 ssize_t data_len; 2722 2723 /* 2724 * No need to check the VM protection here. 2725 * 2726 * Maybe user space wants to query some of the data 2727 * when the VM is still unprotected. If we see the 2728 * need to fence a new data command we can still 2729 * return an error in the info handler. 2730 */ 2731 2732 r = -EFAULT; 2733 if (copy_from_user(&info, argp, sizeof(info.header))) 2734 break; 2735 2736 r = -EINVAL; 2737 if (info.header.len_max < sizeof(info.header)) 2738 break; 2739 2740 data_len = kvm_s390_handle_pv_info(&info); 2741 if (data_len < 0) { 2742 r = data_len; 2743 break; 2744 } 2745 /* 2746 * If a data command struct is extended (multiple 2747 * times) this can be used to determine how much of it 2748 * is valid. 2749 */ 2750 info.header.len_written = data_len; 2751 2752 r = -EFAULT; 2753 if (copy_to_user(argp, &info, data_len)) 2754 break; 2755 2756 r = 0; 2757 break; 2758 } 2759 case KVM_PV_DUMP: { 2760 struct kvm_s390_pv_dmp dmp; 2761 2762 r = -EINVAL; 2763 if (!kvm_s390_pv_is_protected(kvm)) 2764 break; 2765 2766 r = -EFAULT; 2767 if (copy_from_user(&dmp, argp, sizeof(dmp))) 2768 break; 2769 2770 r = kvm_s390_pv_dmp(kvm, cmd, dmp); 2771 if (r) 2772 break; 2773 2774 if (copy_to_user(argp, &dmp, sizeof(dmp))) { 2775 r = -EFAULT; 2776 break; 2777 } 2778 2779 break; 2780 } 2781 default: 2782 r = -ENOTTY; 2783 } 2784 if (need_lock) 2785 mutex_unlock(&kvm->lock); 2786 2787 return r; 2788 } 2789 2790 static int mem_op_validate_common(struct kvm_s390_mem_op *mop, u64 supported_flags) 2791 { 2792 if (mop->flags & ~supported_flags || !mop->size) 2793 return -EINVAL; 2794 if (mop->size > MEM_OP_MAX_SIZE) 2795 return -E2BIG; 2796 if (mop->flags & KVM_S390_MEMOP_F_SKEY_PROTECTION) { 2797 if (mop->key > 0xf) 2798 return -EINVAL; 2799 } else { 2800 mop->key = 0; 2801 } 2802 return 0; 2803 } 2804 2805 static int kvm_s390_vm_mem_op_abs(struct kvm *kvm, struct kvm_s390_mem_op *mop) 2806 { 2807 void __user *uaddr = (void __user *)mop->buf; 2808 void *tmpbuf __free(kvfree) = NULL; 2809 enum gacc_mode acc_mode; 2810 int r; 2811 2812 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION | 2813 KVM_S390_MEMOP_F_CHECK_ONLY); 2814 if (r) 2815 return r; 2816 2817 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) { 2818 tmpbuf = vmalloc(mop->size); 2819 if (!tmpbuf) 2820 return -ENOMEM; 2821 } 2822 2823 acc_mode = mop->op == KVM_S390_MEMOP_ABSOLUTE_READ ? GACC_FETCH : GACC_STORE; 2824 2825 scoped_guard(srcu, &kvm->srcu) { 2826 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) 2827 return check_gpa_range(kvm, mop->gaddr, mop->size, acc_mode, mop->key); 2828 2829 if (acc_mode == GACC_STORE && copy_from_user(tmpbuf, uaddr, mop->size)) 2830 return -EFAULT; 2831 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf, 2832 mop->size, acc_mode, mop->key); 2833 if (r) 2834 return r; 2835 if (acc_mode != GACC_STORE && copy_to_user(uaddr, tmpbuf, mop->size)) 2836 return -EFAULT; 2837 } 2838 return 0; 2839 } 2840 2841 static int kvm_s390_vm_mem_op_cmpxchg(struct kvm *kvm, struct kvm_s390_mem_op *mop) 2842 { 2843 void __user *uaddr = (void __user *)mop->buf; 2844 void __user *old_addr = (void __user *)mop->old_addr; 2845 union kvm_s390_quad old = { .sixteen = 0 }; 2846 union kvm_s390_quad new = { .sixteen = 0 }; 2847 bool success = false; 2848 int r; 2849 2850 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION); 2851 if (r) 2852 return r; 2853 /* 2854 * This validates off_in_quad. Checking that size is a power 2855 * of two is not necessary, as cmpxchg_guest_abs_with_key 2856 * takes care of that 2857 */ 2858 if (mop->size > sizeof(new)) 2859 return -EINVAL; 2860 if (copy_from_user(&new, uaddr, mop->size)) 2861 return -EFAULT; 2862 if (copy_from_user(&old, old_addr, mop->size)) 2863 return -EFAULT; 2864 2865 scoped_guard(srcu, &kvm->srcu) { 2866 r = cmpxchg_guest_abs_with_key(kvm, mop->gaddr, mop->size, &old, new, 2867 mop->key, &success); 2868 2869 if (!success && copy_to_user(old_addr, &old, mop->size)) 2870 return -EFAULT; 2871 } 2872 return r; 2873 } 2874 2875 static int kvm_s390_vm_mem_op(struct kvm *kvm, struct kvm_s390_mem_op *mop) 2876 { 2877 /* 2878 * This is technically a heuristic only, if the kvm->lock is not 2879 * taken, it is not guaranteed that the vm is/remains non-protected. 2880 * This is ok from a kernel perspective, wrongdoing is detected 2881 * on the access, -EFAULT is returned and the vm may crash the 2882 * next time it accesses the memory in question. 2883 * There is no sane usecase to do switching and a memop on two 2884 * different CPUs at the same time. 2885 */ 2886 if (kvm_s390_pv_get_handle(kvm)) 2887 return -EINVAL; 2888 2889 switch (mop->op) { 2890 case KVM_S390_MEMOP_ABSOLUTE_READ: 2891 case KVM_S390_MEMOP_ABSOLUTE_WRITE: 2892 return kvm_s390_vm_mem_op_abs(kvm, mop); 2893 case KVM_S390_MEMOP_ABSOLUTE_CMPXCHG: 2894 return kvm_s390_vm_mem_op_cmpxchg(kvm, mop); 2895 default: 2896 return -EINVAL; 2897 } 2898 } 2899 2900 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg) 2901 { 2902 struct kvm *kvm = filp->private_data; 2903 void __user *argp = (void __user *)arg; 2904 struct kvm_device_attr attr; 2905 int r; 2906 struct kvm_s390_interrupt_info *inti; 2907 2908 switch (ioctl) { 2909 case KVM_S390_INTERRUPT: { 2910 struct kvm_s390_interrupt s390int; 2911 2912 r = -EINVAL; 2913 if (kvm_is_ucontrol(kvm)) 2914 break; 2915 r = -EFAULT; 2916 if (copy_from_user(&s390int, argp, sizeof(s390int))) 2917 break; 2918 inti = kzalloc_obj(*inti, GFP_KERNEL_ACCOUNT); 2919 if (!inti) 2920 return -ENOMEM; 2921 r = kvm_s390_inject_vm(kvm, &s390int, inti); 2922 if (r) 2923 kfree(inti); 2924 break; 2925 } 2926 case KVM_CREATE_IRQCHIP: { 2927 r = -EINVAL; 2928 if (kvm->arch.use_irqchip) 2929 r = 0; 2930 break; 2931 } 2932 case KVM_SET_DEVICE_ATTR: { 2933 r = -EFAULT; 2934 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) 2935 break; 2936 r = kvm_s390_vm_set_attr(kvm, &attr); 2937 break; 2938 } 2939 case KVM_GET_DEVICE_ATTR: { 2940 r = -EFAULT; 2941 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) 2942 break; 2943 r = kvm_s390_vm_get_attr(kvm, &attr); 2944 break; 2945 } 2946 case KVM_HAS_DEVICE_ATTR: { 2947 r = -EFAULT; 2948 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) 2949 break; 2950 r = kvm_s390_vm_has_attr(kvm, &attr); 2951 break; 2952 } 2953 case KVM_S390_GET_SKEYS: { 2954 struct kvm_s390_skeys args; 2955 2956 r = -EFAULT; 2957 if (copy_from_user(&args, argp, 2958 sizeof(struct kvm_s390_skeys))) 2959 break; 2960 r = kvm_s390_get_skeys(kvm, &args); 2961 break; 2962 } 2963 case KVM_S390_SET_SKEYS: { 2964 struct kvm_s390_skeys args; 2965 2966 r = -EFAULT; 2967 if (copy_from_user(&args, argp, 2968 sizeof(struct kvm_s390_skeys))) 2969 break; 2970 r = kvm_s390_set_skeys(kvm, &args); 2971 break; 2972 } 2973 case KVM_S390_GET_CMMA_BITS: { 2974 struct kvm_s390_cmma_log args; 2975 2976 r = -EFAULT; 2977 if (copy_from_user(&args, argp, sizeof(args))) 2978 break; 2979 scoped_guard(mutex, &kvm->slots_arch_lock) 2980 r = kvm_s390_get_cmma_bits(kvm, &args); 2981 if (!r) { 2982 r = copy_to_user(argp, &args, sizeof(args)); 2983 if (r) 2984 r = -EFAULT; 2985 } 2986 break; 2987 } 2988 case KVM_S390_SET_CMMA_BITS: { 2989 struct kvm_s390_cmma_log args; 2990 2991 r = -EFAULT; 2992 if (copy_from_user(&args, argp, sizeof(args))) 2993 break; 2994 mutex_lock(&kvm->slots_arch_lock); 2995 r = kvm_s390_set_cmma_bits(kvm, &args); 2996 mutex_unlock(&kvm->slots_arch_lock); 2997 break; 2998 } 2999 case KVM_S390_PV_COMMAND: { 3000 struct kvm_pv_cmd args; 3001 3002 /* protvirt means user cpu state */ 3003 kvm_s390_set_user_cpu_state_ctrl(kvm); 3004 r = 0; 3005 if (!is_prot_virt_host()) { 3006 r = -EINVAL; 3007 break; 3008 } 3009 if (copy_from_user(&args, argp, sizeof(args))) { 3010 r = -EFAULT; 3011 break; 3012 } 3013 if (args.flags) { 3014 r = -EINVAL; 3015 break; 3016 } 3017 /* must be called without kvm->lock */ 3018 r = kvm_s390_handle_pv(kvm, &args); 3019 if (copy_to_user(argp, &args, sizeof(args))) { 3020 r = -EFAULT; 3021 break; 3022 } 3023 break; 3024 } 3025 case KVM_S390_MEM_OP: { 3026 struct kvm_s390_mem_op mem_op; 3027 3028 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0) 3029 r = kvm_s390_vm_mem_op(kvm, &mem_op); 3030 else 3031 r = -EFAULT; 3032 break; 3033 } 3034 case KVM_S390_KEYOP: { 3035 struct kvm_s390_mmu_cache *mc; 3036 struct kvm_s390_keyop kop; 3037 union skey skey; 3038 3039 if (copy_from_user(&kop, argp, sizeof(kop))) { 3040 r = -EFAULT; 3041 break; 3042 } 3043 skey.skey = kop.key; 3044 3045 mc = kvm_s390_new_mmu_cache(); 3046 if (!mc) 3047 return -ENOMEM; 3048 3049 r = kvm_s390_keyop(mc, kvm, kop.operation, kop.guest_addr, skey); 3050 kvm_s390_free_mmu_cache(mc); 3051 if (r < 0) 3052 break; 3053 3054 kop.key = r; 3055 r = 0; 3056 if (copy_to_user(argp, &kop, sizeof(kop))) 3057 r = -EFAULT; 3058 break; 3059 } 3060 case KVM_S390_ZPCI_OP: { 3061 struct kvm_s390_zpci_op args; 3062 3063 r = -EINVAL; 3064 if (!IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) 3065 break; 3066 if (copy_from_user(&args, argp, sizeof(args))) { 3067 r = -EFAULT; 3068 break; 3069 } 3070 r = kvm_s390_pci_zpci_op(kvm, &args); 3071 break; 3072 } 3073 default: 3074 r = -ENOTTY; 3075 } 3076 3077 return r; 3078 } 3079 3080 static int kvm_s390_apxa_installed(void) 3081 { 3082 struct ap_config_info info; 3083 3084 if (ap_instructions_available()) { 3085 if (ap_qci(&info) == 0) 3086 return info.apxa; 3087 } 3088 3089 return 0; 3090 } 3091 3092 /* 3093 * The format of the crypto control block (CRYCB) is specified in the 3 low 3094 * order bits of the CRYCB designation (CRYCBD) field as follows: 3095 * Format 0: Neither the message security assist extension 3 (MSAX3) nor the 3096 * AP extended addressing (APXA) facility are installed. 3097 * Format 1: The APXA facility is not installed but the MSAX3 facility is. 3098 * Format 2: Both the APXA and MSAX3 facilities are installed 3099 */ 3100 static void kvm_s390_set_crycb_format(struct kvm *kvm) 3101 { 3102 kvm->arch.crypto.crycbd = virt_to_phys(kvm->arch.crypto.crycb); 3103 3104 /* Clear the CRYCB format bits - i.e., set format 0 by default */ 3105 kvm->arch.crypto.crycbd &= ~(CRYCB_FORMAT_MASK); 3106 3107 /* Check whether MSAX3 is installed */ 3108 if (!test_kvm_facility(kvm, 76)) 3109 return; 3110 3111 if (kvm_s390_apxa_installed()) 3112 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2; 3113 else 3114 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1; 3115 } 3116 3117 /* 3118 * kvm_arch_crypto_set_masks 3119 * 3120 * @kvm: pointer to the target guest's KVM struct containing the crypto masks 3121 * to be set. 3122 * @apm: the mask identifying the accessible AP adapters 3123 * @aqm: the mask identifying the accessible AP domains 3124 * @adm: the mask identifying the accessible AP control domains 3125 * 3126 * Set the masks that identify the adapters, domains and control domains to 3127 * which the KVM guest is granted access. 3128 * 3129 * Note: The kvm->lock mutex must be locked by the caller before invoking this 3130 * function. 3131 */ 3132 void kvm_arch_crypto_set_masks(struct kvm *kvm, unsigned long *apm, 3133 unsigned long *aqm, unsigned long *adm) 3134 { 3135 struct kvm_s390_crypto_cb *crycb = kvm->arch.crypto.crycb; 3136 3137 kvm_s390_vcpu_block_all(kvm); 3138 3139 switch (kvm->arch.crypto.crycbd & CRYCB_FORMAT_MASK) { 3140 case CRYCB_FORMAT2: /* APCB1 use 256 bits */ 3141 memcpy(crycb->apcb1.apm, apm, 32); 3142 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx %016lx %016lx %016lx", 3143 apm[0], apm[1], apm[2], apm[3]); 3144 memcpy(crycb->apcb1.aqm, aqm, 32); 3145 VM_EVENT(kvm, 3, "SET CRYCB: aqm %016lx %016lx %016lx %016lx", 3146 aqm[0], aqm[1], aqm[2], aqm[3]); 3147 memcpy(crycb->apcb1.adm, adm, 32); 3148 VM_EVENT(kvm, 3, "SET CRYCB: adm %016lx %016lx %016lx %016lx", 3149 adm[0], adm[1], adm[2], adm[3]); 3150 break; 3151 case CRYCB_FORMAT1: 3152 case CRYCB_FORMAT0: /* Fall through both use APCB0 */ 3153 memcpy(crycb->apcb0.apm, apm, 8); 3154 memcpy(crycb->apcb0.aqm, aqm, 2); 3155 memcpy(crycb->apcb0.adm, adm, 2); 3156 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx aqm %04x adm %04x", 3157 apm[0], *((unsigned short *)aqm), 3158 *((unsigned short *)adm)); 3159 break; 3160 default: /* Can not happen */ 3161 break; 3162 } 3163 3164 /* recreate the shadow crycb for each vcpu */ 3165 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART); 3166 kvm_s390_vcpu_unblock_all(kvm); 3167 } 3168 EXPORT_SYMBOL_GPL(kvm_arch_crypto_set_masks); 3169 3170 /* 3171 * kvm_arch_crypto_clear_masks 3172 * 3173 * @kvm: pointer to the target guest's KVM struct containing the crypto masks 3174 * to be cleared. 3175 * 3176 * Clear the masks that identify the adapters, domains and control domains to 3177 * which the KVM guest is granted access. 3178 * 3179 * Note: The kvm->lock mutex must be locked by the caller before invoking this 3180 * function. 3181 */ 3182 void kvm_arch_crypto_clear_masks(struct kvm *kvm) 3183 { 3184 kvm_s390_vcpu_block_all(kvm); 3185 3186 memset(&kvm->arch.crypto.crycb->apcb0, 0, 3187 sizeof(kvm->arch.crypto.crycb->apcb0)); 3188 memset(&kvm->arch.crypto.crycb->apcb1, 0, 3189 sizeof(kvm->arch.crypto.crycb->apcb1)); 3190 3191 VM_EVENT(kvm, 3, "%s", "CLR CRYCB:"); 3192 /* recreate the shadow crycb for each vcpu */ 3193 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART); 3194 kvm_s390_vcpu_unblock_all(kvm); 3195 } 3196 EXPORT_SYMBOL_GPL(kvm_arch_crypto_clear_masks); 3197 3198 static u64 kvm_s390_get_initial_cpuid(void) 3199 { 3200 struct cpuid cpuid; 3201 3202 get_cpu_id(&cpuid); 3203 cpuid.version = 0xff; 3204 return *((u64 *) &cpuid); 3205 } 3206 3207 static void kvm_s390_crypto_init(struct kvm *kvm) 3208 { 3209 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb; 3210 kvm_s390_set_crycb_format(kvm); 3211 init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem); 3212 3213 if (!test_kvm_facility(kvm, 76)) 3214 return; 3215 3216 /* Enable AES/DEA protected key functions by default */ 3217 kvm->arch.crypto.aes_kw = 1; 3218 kvm->arch.crypto.dea_kw = 1; 3219 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 3220 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask)); 3221 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 3222 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask)); 3223 } 3224 3225 static void sca_dispose(struct kvm *kvm) 3226 { 3227 if (kvm->arch.sca) 3228 free_pages_exact(kvm->arch.sca, sizeof(*kvm->arch.sca)); 3229 kvm->arch.sca = NULL; 3230 } 3231 3232 void kvm_arch_free_vm(struct kvm *kvm) 3233 { 3234 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) 3235 kvm_s390_pci_clear_list(kvm); 3236 3237 __kvm_arch_free_vm(kvm); 3238 } 3239 3240 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 3241 { 3242 gfp_t alloc_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO; 3243 char debug_name[16]; 3244 int i, rc; 3245 3246 mutex_init(&kvm->arch.pv.import_lock); 3247 3248 rc = -EINVAL; 3249 #ifdef CONFIG_KVM_S390_UCONTROL 3250 if (type & ~KVM_VM_S390_UCONTROL) 3251 goto out_err; 3252 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN))) 3253 goto out_err; 3254 #else 3255 if (type) 3256 goto out_err; 3257 #endif 3258 rc = -ENOMEM; 3259 3260 if (!sclp.has_64bscao) 3261 alloc_flags |= GFP_DMA; 3262 mutex_lock(&kvm_lock); 3263 3264 kvm->arch.sca = alloc_pages_exact(sizeof(*kvm->arch.sca), alloc_flags); 3265 mutex_unlock(&kvm_lock); 3266 if (!kvm->arch.sca) 3267 goto out_err; 3268 3269 snprintf(debug_name, sizeof(debug_name), "kvm-%u", current->pid); 3270 3271 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long)); 3272 if (!kvm->arch.dbf) 3273 goto out_err; 3274 3275 BUILD_BUG_ON(sizeof(struct sie_page2) != 4096); 3276 kvm->arch.sie_page2 = 3277 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA); 3278 if (!kvm->arch.sie_page2) 3279 goto out_err; 3280 3281 kvm->arch.sie_page2->kvm = kvm; 3282 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list; 3283 3284 for (i = 0; i < ARRAY_SIZE(kvm_s390_fac_base); i++) { 3285 kvm->arch.model.fac_mask[i] = stfle_fac_list[i] & 3286 kvm_s390_fac_base[i]; 3287 kvm->arch.model.fac_list[i] = stfle_fac_list[i] & 3288 kvm_s390_fac_base[i]; 3289 } 3290 for (i = 0; i < ARRAY_SIZE(kvm_s390_fac_ext); i++) { 3291 kvm->arch.model.fac_mask[i] |= stfle_fac_list[i] & 3292 kvm_s390_fac_ext[i]; 3293 } 3294 kvm->arch.model.subfuncs = kvm_s390_available_subfunc; 3295 3296 /* we are always in czam mode - even on pre z14 machines */ 3297 set_kvm_facility(kvm->arch.model.fac_mask, 138); 3298 set_kvm_facility(kvm->arch.model.fac_list, 138); 3299 /* we emulate STHYI in kvm */ 3300 set_kvm_facility(kvm->arch.model.fac_mask, 74); 3301 set_kvm_facility(kvm->arch.model.fac_list, 74); 3302 if (machine_has_tlb_guest()) { 3303 set_kvm_facility(kvm->arch.model.fac_mask, 147); 3304 set_kvm_facility(kvm->arch.model.fac_list, 147); 3305 } 3306 3307 if (css_general_characteristics.aiv && test_facility(65)) 3308 set_kvm_facility(kvm->arch.model.fac_mask, 65); 3309 3310 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid(); 3311 kvm->arch.model.ibc = sclp.ibc & 0x0fff; 3312 3313 kvm->arch.model.uv_feat_guest.feat = 0; 3314 3315 kvm_s390_crypto_init(kvm); 3316 3317 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) { 3318 mutex_lock(&kvm->lock); 3319 kvm_s390_pci_init_list(kvm); 3320 kvm_s390_vcpu_pci_enable_interp(kvm); 3321 mutex_unlock(&kvm->lock); 3322 } 3323 3324 spin_lock_init(&kvm->arch.float_int.ais_lock); 3325 spin_lock_init(&kvm->arch.float_int.lock); 3326 for (i = 0; i < FIRQ_LIST_COUNT; i++) 3327 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]); 3328 init_waitqueue_head(&kvm->arch.ipte_wq); 3329 mutex_init(&kvm->arch.ipte_mutex); 3330 3331 debug_register_view(kvm->arch.dbf, &debug_sprintf_view); 3332 VM_EVENT(kvm, 3, "vm created with type %lu", type); 3333 3334 kvm->arch.mem_limit = type & KVM_VM_S390_UCONTROL ? KVM_S390_NO_MEM_LIMIT : sclp.hamax + 1; 3335 kvm->arch.gmap = gmap_new(kvm, gpa_to_gfn(kvm->arch.mem_limit)); 3336 if (!kvm->arch.gmap) 3337 goto out_err; 3338 clear_bit(GMAP_FLAG_PFAULT_ENABLED, &kvm->arch.gmap->flags); 3339 3340 if (type & KVM_VM_S390_UCONTROL) { 3341 struct kvm_userspace_memory_region2 fake_memslot = { 3342 .slot = KVM_S390_UCONTROL_MEMSLOT, 3343 .guest_phys_addr = 0, 3344 .userspace_addr = 0, 3345 .memory_size = ALIGN_DOWN(TASK_SIZE, _SEGMENT_SIZE), 3346 .flags = 0, 3347 }; 3348 3349 /* one flat fake memslot covering the whole address-space */ 3350 mutex_lock(&kvm->slots_lock); 3351 KVM_BUG_ON(kvm_set_internal_memslot(kvm, &fake_memslot), kvm); 3352 mutex_unlock(&kvm->slots_lock); 3353 set_bit(GMAP_FLAG_IS_UCONTROL, &kvm->arch.gmap->flags); 3354 } else { 3355 struct crst_table *table = dereference_asce(kvm->arch.gmap->asce); 3356 3357 crst_table_init((void *)table, _CRSTE_HOLE(table->crstes[0].h.tt).val); 3358 } 3359 3360 kvm->arch.use_pfmfi = sclp.has_pfmfi; 3361 kvm->arch.use_skf = sclp.has_skey; 3362 spin_lock_init(&kvm->arch.start_stop_lock); 3363 kvm_s390_vsie_init(kvm); 3364 if (use_gisa) 3365 kvm_s390_gisa_init(kvm); 3366 INIT_LIST_HEAD(&kvm->arch.pv.need_cleanup); 3367 kvm->arch.pv.set_aside = NULL; 3368 KVM_EVENT(3, "vm 0x%p created by pid %u", kvm, current->pid); 3369 3370 return 0; 3371 out_err: 3372 free_page((unsigned long)kvm->arch.sie_page2); 3373 debug_unregister(kvm->arch.dbf); 3374 sca_dispose(kvm); 3375 KVM_EVENT(3, "creation of vm failed: %d", rc); 3376 return rc; 3377 } 3378 3379 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 3380 { 3381 u16 rc, rrc; 3382 3383 VCPU_EVENT(vcpu, 3, "%s", "free cpu"); 3384 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id); 3385 kvm_s390_clear_local_irqs(vcpu); 3386 kvm_clear_async_pf_completion_queue(vcpu); 3387 kvm_s390_clear_bp_data(vcpu); 3388 if (!kvm_is_ucontrol(vcpu->kvm)) 3389 sca_del_vcpu(vcpu); 3390 kvm_s390_update_topology_change_report(vcpu->kvm, 1); 3391 3392 if (kvm_is_ucontrol(vcpu->kvm)) { 3393 scoped_guard(spinlock, &vcpu->kvm->arch.gmap->children_lock) 3394 gmap_remove_child(vcpu->arch.gmap); 3395 vcpu->arch.gmap = gmap_put(vcpu->arch.gmap); 3396 } 3397 3398 if (vcpu->kvm->arch.use_cmma) 3399 kvm_s390_vcpu_unsetup_cmma(vcpu); 3400 /* We can not hold the vcpu mutex here, we are already dying */ 3401 if (kvm_s390_pv_cpu_get_handle(vcpu)) 3402 kvm_s390_pv_destroy_cpu(vcpu, &rc, &rrc); 3403 free_page((unsigned long)(vcpu->arch.sie_block)); 3404 kvm_s390_free_mmu_cache(vcpu->arch.mc); 3405 } 3406 3407 void kvm_arch_destroy_vm(struct kvm *kvm) 3408 { 3409 u16 rc, rrc; 3410 3411 kvm_destroy_vcpus(kvm); 3412 sca_dispose(kvm); 3413 kvm_s390_gisa_destroy(kvm); 3414 /* 3415 * We are already at the end of life and kvm->lock is not taken. 3416 * This is ok as the file descriptor is closed by now and nobody 3417 * can mess with the pv state. 3418 */ 3419 kvm_s390_pv_deinit_cleanup_all(kvm, &rc, &rrc); 3420 /* 3421 * Remove the mmu notifier only when the whole KVM VM is torn down, 3422 * and only if one was registered to begin with. If the VM is 3423 * currently not protected, but has been previously been protected, 3424 * then it's possible that the notifier is still registered. 3425 */ 3426 if (kvm->arch.pv.mmu_notifier.ops) 3427 mmu_notifier_unregister(&kvm->arch.pv.mmu_notifier, kvm->mm); 3428 3429 debug_unregister(kvm->arch.dbf); 3430 free_page((unsigned long)kvm->arch.sie_page2); 3431 kvm_s390_destroy_adapters(kvm); 3432 kvm_s390_clear_float_irqs(kvm); 3433 kvm_s390_vsie_destroy(kvm); 3434 kvm->arch.gmap = gmap_put(kvm->arch.gmap); 3435 KVM_EVENT(3, "vm 0x%p destroyed", kvm); 3436 } 3437 3438 /* Section: vcpu related */ 3439 static void sca_del_vcpu(struct kvm_vcpu *vcpu) 3440 { 3441 struct esca_block *sca = vcpu->kvm->arch.sca; 3442 3443 if (!kvm_s390_use_sca_entries() || !vcpu->arch.initialized) 3444 return; 3445 3446 clear_bit_inv(vcpu->vcpu_id, (unsigned long *)sca->mcn); 3447 sca->cpu[vcpu->vcpu_id].sda = 0; 3448 } 3449 3450 static void sca_add_vcpu(struct kvm_vcpu *vcpu) 3451 { 3452 struct esca_block *sca = vcpu->kvm->arch.sca; 3453 phys_addr_t sca_phys = virt_to_phys(sca); 3454 3455 /* we still need the sca header for the ipte control */ 3456 vcpu->arch.sie_block->scaoh = sca_phys >> 32; 3457 vcpu->arch.sie_block->scaol = sca_phys & ESCA_SCAOL_MASK; 3458 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA; 3459 3460 if (!kvm_s390_use_sca_entries()) 3461 return; 3462 3463 WRITE_ONCE(sca->cpu[vcpu->vcpu_id].sda, virt_to_phys(vcpu->arch.sie_block)); 3464 set_bit_inv(vcpu->vcpu_id, (unsigned long *)sca->mcn); 3465 } 3466 3467 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id) 3468 { 3469 if (!kvm_s390_use_sca_entries()) 3470 return id < KVM_MAX_VCPUS; 3471 3472 return id < KVM_S390_ESCA_CPU_SLOTS; 3473 } 3474 3475 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */ 3476 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3477 { 3478 WARN_ON_ONCE(vcpu->arch.cputm_start != 0); 3479 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount); 3480 vcpu->arch.cputm_start = get_tod_clock_fast(); 3481 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount); 3482 } 3483 3484 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */ 3485 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3486 { 3487 WARN_ON_ONCE(vcpu->arch.cputm_start == 0); 3488 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount); 3489 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start; 3490 vcpu->arch.cputm_start = 0; 3491 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount); 3492 } 3493 3494 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */ 3495 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3496 { 3497 WARN_ON_ONCE(vcpu->arch.cputm_enabled); 3498 vcpu->arch.cputm_enabled = true; 3499 __start_cpu_timer_accounting(vcpu); 3500 } 3501 3502 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */ 3503 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3504 { 3505 WARN_ON_ONCE(!vcpu->arch.cputm_enabled); 3506 __stop_cpu_timer_accounting(vcpu); 3507 vcpu->arch.cputm_enabled = false; 3508 } 3509 3510 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3511 { 3512 preempt_disable(); /* protect from TOD sync and vcpu_load/put */ 3513 __enable_cpu_timer_accounting(vcpu); 3514 preempt_enable(); 3515 } 3516 3517 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3518 { 3519 preempt_disable(); /* protect from TOD sync and vcpu_load/put */ 3520 __disable_cpu_timer_accounting(vcpu); 3521 preempt_enable(); 3522 } 3523 3524 /* set the cpu timer - may only be called from the VCPU thread itself */ 3525 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm) 3526 { 3527 preempt_disable(); /* protect from TOD sync and vcpu_load/put */ 3528 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount); 3529 if (vcpu->arch.cputm_enabled) 3530 vcpu->arch.cputm_start = get_tod_clock_fast(); 3531 vcpu->arch.sie_block->cputm = cputm; 3532 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount); 3533 preempt_enable(); 3534 } 3535 3536 /* update and get the cpu timer - can also be called from other VCPU threads */ 3537 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu) 3538 { 3539 unsigned int seq; 3540 __u64 value; 3541 3542 if (unlikely(!vcpu->arch.cputm_enabled)) 3543 return vcpu->arch.sie_block->cputm; 3544 3545 preempt_disable(); /* protect from TOD sync and vcpu_load/put */ 3546 do { 3547 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount); 3548 /* 3549 * If the writer would ever execute a read in the critical 3550 * section, e.g. in irq context, we have a deadlock. 3551 */ 3552 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu); 3553 value = vcpu->arch.sie_block->cputm; 3554 /* if cputm_start is 0, accounting is being started/stopped */ 3555 if (likely(vcpu->arch.cputm_start)) 3556 value -= get_tod_clock_fast() - vcpu->arch.cputm_start; 3557 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1)); 3558 preempt_enable(); 3559 return value; 3560 } 3561 3562 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 3563 { 3564 3565 kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING); 3566 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu)) 3567 __start_cpu_timer_accounting(vcpu); 3568 vcpu->cpu = cpu; 3569 } 3570 3571 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 3572 { 3573 vcpu->cpu = -1; 3574 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu)) 3575 __stop_cpu_timer_accounting(vcpu); 3576 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING); 3577 3578 } 3579 3580 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 3581 { 3582 preempt_disable(); 3583 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch; 3584 vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx; 3585 preempt_enable(); 3586 3587 if (!kvm_is_ucontrol(vcpu->kvm)) { 3588 vcpu->arch.gmap = vcpu->kvm->arch.gmap; 3589 sca_add_vcpu(vcpu); 3590 } 3591 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0 || 3592 vcpu->kvm->arch.user_operexec) 3593 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC; 3594 3595 /* Pairs with smp_load_acquire() in kvm_arch_vcpu_ioctl_run() and kvm_arch_vcpu_ioctl() */ 3596 smp_store_release(&vcpu->arch.initialized, true); 3597 } 3598 3599 static bool kvm_has_pckmo_subfunc(struct kvm *kvm, unsigned long nr) 3600 { 3601 if (test_bit_inv(nr, (unsigned long *)&kvm->arch.model.subfuncs.pckmo) && 3602 test_bit_inv(nr, (unsigned long *)&kvm_s390_available_subfunc.pckmo)) 3603 return true; 3604 return false; 3605 } 3606 3607 static bool kvm_has_pckmo_ecc(struct kvm *kvm) 3608 { 3609 /* At least one ECC subfunction must be present */ 3610 return kvm_has_pckmo_subfunc(kvm, 32) || 3611 kvm_has_pckmo_subfunc(kvm, 33) || 3612 kvm_has_pckmo_subfunc(kvm, 34) || 3613 kvm_has_pckmo_subfunc(kvm, 40) || 3614 kvm_has_pckmo_subfunc(kvm, 41); 3615 3616 } 3617 3618 static bool kvm_has_pckmo_hmac(struct kvm *kvm) 3619 { 3620 /* At least one HMAC subfunction must be present */ 3621 return kvm_has_pckmo_subfunc(kvm, 118) || 3622 kvm_has_pckmo_subfunc(kvm, 122); 3623 } 3624 3625 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu) 3626 { 3627 /* 3628 * If the AP instructions are not being interpreted and the MSAX3 3629 * facility is not configured for the guest, there is nothing to set up. 3630 */ 3631 if (!vcpu->kvm->arch.crypto.apie && !test_kvm_facility(vcpu->kvm, 76)) 3632 return; 3633 3634 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd; 3635 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA); 3636 vcpu->arch.sie_block->eca &= ~ECA_APIE; 3637 vcpu->arch.sie_block->ecd &= ~(ECD_ECC | ECD_HMAC); 3638 3639 if (vcpu->kvm->arch.crypto.apie) 3640 vcpu->arch.sie_block->eca |= ECA_APIE; 3641 3642 /* Set up protected key support */ 3643 if (vcpu->kvm->arch.crypto.aes_kw) { 3644 vcpu->arch.sie_block->ecb3 |= ECB3_AES; 3645 /* ecc/hmac is also wrapped with AES key */ 3646 if (kvm_has_pckmo_ecc(vcpu->kvm)) 3647 vcpu->arch.sie_block->ecd |= ECD_ECC; 3648 if (kvm_has_pckmo_hmac(vcpu->kvm)) 3649 vcpu->arch.sie_block->ecd |= ECD_HMAC; 3650 } 3651 3652 if (vcpu->kvm->arch.crypto.dea_kw) 3653 vcpu->arch.sie_block->ecb3 |= ECB3_DEA; 3654 } 3655 3656 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu) 3657 { 3658 if (vcpu->arch.sie_block->cbrlo) 3659 free_page((unsigned long)phys_to_virt(vcpu->arch.sie_block->cbrlo)); 3660 vcpu->arch.sie_block->cbrlo = 0; 3661 } 3662 3663 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu) 3664 { 3665 void *cbrlo_page = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT); 3666 3667 if (!cbrlo_page) 3668 return -ENOMEM; 3669 3670 vcpu->arch.sie_block->cbrlo = virt_to_phys(cbrlo_page); 3671 return 0; 3672 } 3673 3674 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu) 3675 { 3676 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model; 3677 3678 vcpu->arch.sie_block->ibc = model->ibc; 3679 if (test_kvm_facility(vcpu->kvm, 7)) 3680 vcpu->arch.sie_block->fac = virt_to_phys(model->fac_list); 3681 } 3682 3683 static int kvm_s390_vcpu_setup(struct kvm_vcpu *vcpu) 3684 { 3685 int rc = 0; 3686 u16 uvrc, uvrrc; 3687 3688 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH | 3689 CPUSTAT_SM | 3690 CPUSTAT_STOPPED); 3691 3692 if (test_kvm_facility(vcpu->kvm, 78)) 3693 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2); 3694 else if (test_kvm_facility(vcpu->kvm, 8)) 3695 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED); 3696 3697 kvm_s390_vcpu_setup_model(vcpu); 3698 3699 /* pgste_set_pte has special handling for !machine_has_esop() */ 3700 if (machine_has_esop()) 3701 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT; 3702 if (test_kvm_facility(vcpu->kvm, 9)) 3703 vcpu->arch.sie_block->ecb |= ECB_SRSI; 3704 if (test_kvm_facility(vcpu->kvm, 11)) 3705 vcpu->arch.sie_block->ecb |= ECB_PTF; 3706 if (test_kvm_facility(vcpu->kvm, 73)) 3707 vcpu->arch.sie_block->ecb |= ECB_TE; 3708 if (!kvm_is_ucontrol(vcpu->kvm)) 3709 vcpu->arch.sie_block->ecb |= ECB_SPECI; 3710 3711 if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi) 3712 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI; 3713 if (test_kvm_facility(vcpu->kvm, 130)) 3714 vcpu->arch.sie_block->ecb2 |= ECB2_IEP; 3715 vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI; 3716 if (sclp.has_cei) 3717 vcpu->arch.sie_block->eca |= ECA_CEI; 3718 if (sclp.has_ib) 3719 vcpu->arch.sie_block->eca |= ECA_IB; 3720 if (sclp.has_siif) 3721 vcpu->arch.sie_block->eca |= ECA_SII; 3722 if (kvm_s390_use_sca_entries()) 3723 vcpu->arch.sie_block->eca |= ECA_SIGPI; 3724 if (test_kvm_facility(vcpu->kvm, 129)) { 3725 vcpu->arch.sie_block->eca |= ECA_VX; 3726 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT; 3727 } 3728 if (test_kvm_facility(vcpu->kvm, 139)) 3729 vcpu->arch.sie_block->ecd |= ECD_MEF; 3730 if (test_kvm_facility(vcpu->kvm, 156)) 3731 vcpu->arch.sie_block->ecd |= ECD_ETOKENF; 3732 if (vcpu->arch.sie_block->gd) { 3733 vcpu->arch.sie_block->eca |= ECA_AIV; 3734 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u", 3735 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id); 3736 } 3737 vcpu->arch.sie_block->sdnxo = virt_to_phys(&vcpu->run->s.regs.sdnx) | SDNXC; 3738 vcpu->arch.sie_block->riccbd = virt_to_phys(&vcpu->run->s.regs.riccb); 3739 3740 if (sclp.has_kss) 3741 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS); 3742 else 3743 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE; 3744 3745 if (vcpu->kvm->arch.use_cmma) { 3746 rc = kvm_s390_vcpu_setup_cmma(vcpu); 3747 if (rc) 3748 return rc; 3749 } 3750 hrtimer_setup(&vcpu->arch.ckc_timer, kvm_s390_idle_wakeup, CLOCK_MONOTONIC, 3751 HRTIMER_MODE_REL); 3752 3753 vcpu->arch.sie_block->hpid = HPID_KVM; 3754 3755 kvm_s390_vcpu_crypto_setup(vcpu); 3756 3757 kvm_s390_vcpu_pci_setup(vcpu); 3758 3759 if (kvm_s390_pv_is_protected(vcpu->kvm)) { 3760 rc = kvm_s390_pv_create_cpu(vcpu, &uvrc, &uvrrc); 3761 if (rc) 3762 kvm_s390_vcpu_unsetup_cmma(vcpu); 3763 } 3764 3765 return rc; 3766 } 3767 3768 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) 3769 { 3770 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id)) 3771 return -EINVAL; 3772 return 0; 3773 } 3774 3775 DEFINE_FREE(sie_page, struct sie_page *, if (_T) free_page((unsigned long)(_T))) 3776 3777 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) 3778 { 3779 struct kvm_s390_mmu_cache *mc __free(kvm_s390_mmu_cache) = NULL; 3780 struct sie_page *sie_page __free(sie_page) = NULL; 3781 int rc; 3782 3783 BUILD_BUG_ON(sizeof(struct sie_page) != 4096); 3784 mc = kvm_s390_new_mmu_cache(); 3785 if (!mc) 3786 return -ENOMEM; 3787 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL_ACCOUNT); 3788 if (!sie_page) 3789 return -ENOMEM; 3790 3791 vcpu->arch.sie_block = &sie_page->sie_block; 3792 vcpu->arch.sie_block->itdba = virt_to_phys(&sie_page->itdb); 3793 3794 /* the real guest size will always be smaller than msl */ 3795 vcpu->arch.sie_block->mso = 0; 3796 vcpu->arch.sie_block->msl = sclp.hamax; 3797 3798 vcpu->arch.sie_block->icpua = vcpu->vcpu_id; 3799 spin_lock_init(&vcpu->arch.local_int.lock); 3800 vcpu->arch.sie_block->gd = kvm_s390_get_gisa_desc(vcpu->kvm); 3801 seqcount_init(&vcpu->arch.cputm_seqcount); 3802 3803 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID; 3804 kvm_clear_async_pf_completion_queue(vcpu); 3805 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX | 3806 KVM_SYNC_GPRS | 3807 KVM_SYNC_ACRS | 3808 KVM_SYNC_CRS | 3809 KVM_SYNC_ARCH0 | 3810 KVM_SYNC_PFAULT | 3811 KVM_SYNC_DIAG318; 3812 vcpu->arch.acrs_loaded = false; 3813 kvm_s390_set_prefix(vcpu, 0); 3814 if (test_kvm_facility(vcpu->kvm, 64)) 3815 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB; 3816 if (test_kvm_facility(vcpu->kvm, 82)) 3817 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC; 3818 if (test_kvm_facility(vcpu->kvm, 133)) 3819 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB; 3820 if (test_kvm_facility(vcpu->kvm, 156)) 3821 vcpu->run->kvm_valid_regs |= KVM_SYNC_ETOKEN; 3822 /* fprs can be synchronized via vrs, even if the guest has no vx. With 3823 * cpu_has_vx(), (load|store)_fpu_regs() will work with vrs format. 3824 */ 3825 if (cpu_has_vx()) 3826 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS; 3827 else 3828 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS; 3829 3830 if (kvm_is_ucontrol(vcpu->kvm)) { 3831 vcpu->arch.gmap = gmap_new_child(vcpu->kvm->arch.gmap, -1UL); 3832 if (!vcpu->arch.gmap) 3833 return -ENOMEM; 3834 } 3835 3836 VM_EVENT(vcpu->kvm, 3, "create cpu %d at 0x%p, sie block at 0x%p", 3837 vcpu->vcpu_id, vcpu, vcpu->arch.sie_block); 3838 trace_kvm_s390_create_vcpu(vcpu->vcpu_id, vcpu, vcpu->arch.sie_block); 3839 3840 rc = kvm_s390_vcpu_setup(vcpu); 3841 if (rc) { 3842 if (kvm_is_ucontrol(vcpu->kvm)) { 3843 scoped_guard(spinlock, &vcpu->kvm->arch.gmap->children_lock) 3844 gmap_remove_child(vcpu->arch.gmap); 3845 vcpu->arch.gmap = gmap_put(vcpu->arch.gmap); 3846 } 3847 return rc; 3848 } 3849 3850 vcpu->arch.mc = no_free_ptr(mc); 3851 sie_page = NULL; 3852 kvm_s390_update_topology_change_report(vcpu->kvm, 1); 3853 return 0; 3854 } 3855 3856 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) 3857 { 3858 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask); 3859 return kvm_s390_vcpu_has_irq(vcpu, 0); 3860 } 3861 3862 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 3863 { 3864 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE); 3865 } 3866 3867 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu) 3868 { 3869 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20); 3870 exit_sie(vcpu); 3871 } 3872 3873 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu) 3874 { 3875 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20); 3876 } 3877 3878 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu) 3879 { 3880 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20); 3881 exit_sie(vcpu); 3882 } 3883 3884 bool kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu *vcpu) 3885 { 3886 return atomic_read(&vcpu->arch.sie_block->prog20) & 3887 (PROG_BLOCK_SIE | PROG_REQUEST); 3888 } 3889 3890 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu) 3891 { 3892 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20); 3893 } 3894 3895 /* 3896 * Kick a guest cpu out of (v)SIE and wait until (v)SIE is not running. 3897 * If the CPU is not running (e.g. waiting as idle) the function will 3898 * return immediately. */ 3899 void exit_sie(struct kvm_vcpu *vcpu) 3900 { 3901 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT); 3902 kvm_s390_vsie_kick(vcpu); 3903 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE) 3904 cpu_relax(); 3905 } 3906 3907 /* Kick a guest cpu out of SIE to process a request synchronously */ 3908 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu) 3909 { 3910 __kvm_make_request(req, vcpu); 3911 kvm_s390_vcpu_request(vcpu); 3912 } 3913 3914 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu) 3915 { 3916 /* do not poll with more than halt_poll_max_steal percent of steal time */ 3917 if (get_lowcore()->avg_steal_timer * 100 / (TICK_USEC << 12) >= 3918 READ_ONCE(halt_poll_max_steal)) { 3919 vcpu->stat.halt_no_poll_steal++; 3920 return true; 3921 } 3922 return false; 3923 } 3924 3925 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 3926 { 3927 /* kvm common code refers to this, but never calls it */ 3928 BUG(); 3929 return 0; 3930 } 3931 3932 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, 3933 struct kvm_one_reg *reg) 3934 { 3935 int r = -EINVAL; 3936 3937 switch (reg->id) { 3938 case KVM_REG_S390_TODPR: 3939 r = put_user(vcpu->arch.sie_block->todpr, 3940 (u32 __user *)reg->addr); 3941 break; 3942 case KVM_REG_S390_EPOCHDIFF: 3943 r = put_user(vcpu->arch.sie_block->epoch, 3944 (u64 __user *)reg->addr); 3945 break; 3946 case KVM_REG_S390_CPU_TIMER: 3947 r = put_user(kvm_s390_get_cpu_timer(vcpu), 3948 (u64 __user *)reg->addr); 3949 break; 3950 case KVM_REG_S390_CLOCK_COMP: 3951 r = put_user(vcpu->arch.sie_block->ckc, 3952 (u64 __user *)reg->addr); 3953 break; 3954 case KVM_REG_S390_PFTOKEN: 3955 r = put_user(vcpu->arch.pfault_token, 3956 (u64 __user *)reg->addr); 3957 break; 3958 case KVM_REG_S390_PFCOMPARE: 3959 r = put_user(vcpu->arch.pfault_compare, 3960 (u64 __user *)reg->addr); 3961 break; 3962 case KVM_REG_S390_PFSELECT: 3963 r = put_user(vcpu->arch.pfault_select, 3964 (u64 __user *)reg->addr); 3965 break; 3966 case KVM_REG_S390_PP: 3967 r = put_user(vcpu->arch.sie_block->pp, 3968 (u64 __user *)reg->addr); 3969 break; 3970 case KVM_REG_S390_GBEA: 3971 r = put_user(vcpu->arch.sie_block->gbea, 3972 (u64 __user *)reg->addr); 3973 break; 3974 default: 3975 break; 3976 } 3977 3978 return r; 3979 } 3980 3981 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, 3982 struct kvm_one_reg *reg) 3983 { 3984 int r = -EINVAL; 3985 __u64 val; 3986 3987 switch (reg->id) { 3988 case KVM_REG_S390_TODPR: 3989 r = get_user(vcpu->arch.sie_block->todpr, 3990 (u32 __user *)reg->addr); 3991 break; 3992 case KVM_REG_S390_EPOCHDIFF: 3993 r = get_user(vcpu->arch.sie_block->epoch, 3994 (u64 __user *)reg->addr); 3995 break; 3996 case KVM_REG_S390_CPU_TIMER: 3997 r = get_user(val, (u64 __user *)reg->addr); 3998 if (!r) 3999 kvm_s390_set_cpu_timer(vcpu, val); 4000 break; 4001 case KVM_REG_S390_CLOCK_COMP: 4002 r = get_user(vcpu->arch.sie_block->ckc, 4003 (u64 __user *)reg->addr); 4004 break; 4005 case KVM_REG_S390_PFTOKEN: 4006 r = get_user(vcpu->arch.pfault_token, 4007 (u64 __user *)reg->addr); 4008 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) 4009 kvm_clear_async_pf_completion_queue(vcpu); 4010 break; 4011 case KVM_REG_S390_PFCOMPARE: 4012 r = get_user(vcpu->arch.pfault_compare, 4013 (u64 __user *)reg->addr); 4014 break; 4015 case KVM_REG_S390_PFSELECT: 4016 r = get_user(vcpu->arch.pfault_select, 4017 (u64 __user *)reg->addr); 4018 break; 4019 case KVM_REG_S390_PP: 4020 r = get_user(vcpu->arch.sie_block->pp, 4021 (u64 __user *)reg->addr); 4022 break; 4023 case KVM_REG_S390_GBEA: 4024 r = get_user(vcpu->arch.sie_block->gbea, 4025 (u64 __user *)reg->addr); 4026 break; 4027 default: 4028 break; 4029 } 4030 4031 return r; 4032 } 4033 4034 static void kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu *vcpu) 4035 { 4036 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_RI; 4037 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID; 4038 memset(vcpu->run->s.regs.riccb, 0, sizeof(vcpu->run->s.regs.riccb)); 4039 4040 kvm_clear_async_pf_completion_queue(vcpu); 4041 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) 4042 kvm_s390_vcpu_stop(vcpu); 4043 kvm_s390_clear_local_irqs(vcpu); 4044 } 4045 4046 static void kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu) 4047 { 4048 /* Initial reset is a superset of the normal reset */ 4049 kvm_arch_vcpu_ioctl_normal_reset(vcpu); 4050 4051 /* 4052 * This equals initial cpu reset in pop, but we don't switch to ESA. 4053 * We do not only reset the internal data, but also ... 4054 */ 4055 vcpu->arch.sie_block->gpsw.mask = 0; 4056 vcpu->arch.sie_block->gpsw.addr = 0; 4057 kvm_s390_set_prefix(vcpu, 0); 4058 kvm_s390_set_cpu_timer(vcpu, 0); 4059 vcpu->arch.sie_block->ckc = 0; 4060 memset(vcpu->arch.sie_block->gcr, 0, sizeof(vcpu->arch.sie_block->gcr)); 4061 vcpu->arch.sie_block->gcr[0] = CR0_INITIAL_MASK; 4062 vcpu->arch.sie_block->gcr[14] = CR14_INITIAL_MASK; 4063 4064 /* ... the data in sync regs */ 4065 memset(vcpu->run->s.regs.crs, 0, sizeof(vcpu->run->s.regs.crs)); 4066 vcpu->run->s.regs.ckc = 0; 4067 vcpu->run->s.regs.crs[0] = CR0_INITIAL_MASK; 4068 vcpu->run->s.regs.crs[14] = CR14_INITIAL_MASK; 4069 vcpu->run->psw_addr = 0; 4070 vcpu->run->psw_mask = 0; 4071 vcpu->run->s.regs.todpr = 0; 4072 vcpu->run->s.regs.cputm = 0; 4073 vcpu->run->s.regs.ckc = 0; 4074 vcpu->run->s.regs.pp = 0; 4075 vcpu->run->s.regs.gbea = 1; 4076 vcpu->run->s.regs.fpc = 0; 4077 /* 4078 * Do not reset these registers in the protected case, as some of 4079 * them are overlaid and they are not accessible in this case 4080 * anyway. 4081 */ 4082 if (!kvm_s390_pv_cpu_is_protected(vcpu)) { 4083 vcpu->arch.sie_block->gbea = 1; 4084 vcpu->arch.sie_block->pp = 0; 4085 vcpu->arch.sie_block->fpf &= ~FPF_BPBC; 4086 vcpu->arch.sie_block->todpr = 0; 4087 } 4088 } 4089 4090 static void kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu *vcpu) 4091 { 4092 struct kvm_sync_regs *regs = &vcpu->run->s.regs; 4093 4094 /* Clear reset is a superset of the initial reset */ 4095 kvm_arch_vcpu_ioctl_initial_reset(vcpu); 4096 4097 memset(®s->gprs, 0, sizeof(regs->gprs)); 4098 memset(®s->vrs, 0, sizeof(regs->vrs)); 4099 memset(®s->acrs, 0, sizeof(regs->acrs)); 4100 memset(®s->gscb, 0, sizeof(regs->gscb)); 4101 4102 regs->etoken = 0; 4103 regs->etoken_extension = 0; 4104 } 4105 4106 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) 4107 { 4108 vcpu_load(vcpu); 4109 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs)); 4110 vcpu_put(vcpu); 4111 return 0; 4112 } 4113 4114 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) 4115 { 4116 vcpu_load(vcpu); 4117 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs)); 4118 vcpu_put(vcpu); 4119 return 0; 4120 } 4121 4122 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, 4123 struct kvm_sregs *sregs) 4124 { 4125 vcpu_load(vcpu); 4126 4127 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs)); 4128 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs)); 4129 4130 vcpu_put(vcpu); 4131 return 0; 4132 } 4133 4134 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, 4135 struct kvm_sregs *sregs) 4136 { 4137 vcpu_load(vcpu); 4138 4139 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs)); 4140 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs)); 4141 4142 vcpu_put(vcpu); 4143 return 0; 4144 } 4145 4146 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) 4147 { 4148 vcpu_load(vcpu); 4149 4150 vcpu->run->s.regs.fpc = fpu->fpc; 4151 if (cpu_has_vx()) 4152 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs, 4153 (freg_t *) fpu->fprs); 4154 else 4155 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs)); 4156 4157 vcpu_put(vcpu); 4158 return 0; 4159 } 4160 4161 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) 4162 { 4163 vcpu_load(vcpu); 4164 4165 if (cpu_has_vx()) 4166 convert_vx_to_fp((freg_t *) fpu->fprs, 4167 (__vector128 *) vcpu->run->s.regs.vrs); 4168 else 4169 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs)); 4170 fpu->fpc = vcpu->run->s.regs.fpc; 4171 4172 vcpu_put(vcpu); 4173 return 0; 4174 } 4175 4176 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw) 4177 { 4178 int rc = 0; 4179 4180 if (!is_vcpu_stopped(vcpu)) 4181 rc = -EBUSY; 4182 else { 4183 vcpu->run->psw_mask = psw.mask; 4184 vcpu->run->psw_addr = psw.addr; 4185 } 4186 return rc; 4187 } 4188 4189 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, 4190 struct kvm_translation *tr) 4191 { 4192 return -EINVAL; /* not implemented yet */ 4193 } 4194 4195 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \ 4196 KVM_GUESTDBG_USE_HW_BP | \ 4197 KVM_GUESTDBG_ENABLE) 4198 4199 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, 4200 struct kvm_guest_debug *dbg) 4201 { 4202 int rc = 0; 4203 4204 vcpu_load(vcpu); 4205 4206 vcpu->guest_debug = 0; 4207 kvm_s390_clear_bp_data(vcpu); 4208 4209 if (dbg->control & ~VALID_GUESTDBG_FLAGS) { 4210 rc = -EINVAL; 4211 goto out; 4212 } 4213 if (!sclp.has_gpere) { 4214 rc = -EINVAL; 4215 goto out; 4216 } 4217 4218 if (dbg->control & KVM_GUESTDBG_ENABLE) { 4219 vcpu->guest_debug = dbg->control; 4220 /* enforce guest PER */ 4221 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P); 4222 4223 if (dbg->control & KVM_GUESTDBG_USE_HW_BP) { 4224 scoped_guard(srcu, &vcpu->kvm->srcu) 4225 rc = kvm_s390_import_bp_data(vcpu, dbg); 4226 } 4227 } else { 4228 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P); 4229 vcpu->arch.guestdbg.last_bp = 0; 4230 } 4231 4232 if (rc) { 4233 vcpu->guest_debug = 0; 4234 kvm_s390_clear_bp_data(vcpu); 4235 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P); 4236 } 4237 4238 out: 4239 vcpu_put(vcpu); 4240 return rc; 4241 } 4242 4243 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 4244 struct kvm_mp_state *mp_state) 4245 { 4246 int ret; 4247 4248 vcpu_load(vcpu); 4249 4250 /* CHECK_STOP and LOAD are not supported yet */ 4251 ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED : 4252 KVM_MP_STATE_OPERATING; 4253 4254 vcpu_put(vcpu); 4255 return ret; 4256 } 4257 4258 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 4259 struct kvm_mp_state *mp_state) 4260 { 4261 int rc = 0; 4262 4263 vcpu_load(vcpu); 4264 4265 /* user space knows about this interface - let it control the state */ 4266 kvm_s390_set_user_cpu_state_ctrl(vcpu->kvm); 4267 4268 switch (mp_state->mp_state) { 4269 case KVM_MP_STATE_STOPPED: 4270 rc = kvm_s390_vcpu_stop(vcpu); 4271 break; 4272 case KVM_MP_STATE_OPERATING: 4273 rc = kvm_s390_vcpu_start(vcpu); 4274 break; 4275 case KVM_MP_STATE_LOAD: 4276 if (!kvm_s390_pv_cpu_is_protected(vcpu)) { 4277 rc = -ENXIO; 4278 break; 4279 } 4280 rc = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR_LOAD); 4281 break; 4282 case KVM_MP_STATE_CHECK_STOP: 4283 fallthrough; /* CHECK_STOP and LOAD are not supported yet */ 4284 default: 4285 rc = -ENXIO; 4286 } 4287 4288 vcpu_put(vcpu); 4289 return rc; 4290 } 4291 4292 static bool ibs_enabled(struct kvm_vcpu *vcpu) 4293 { 4294 return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS); 4295 } 4296 4297 static int vcpu_ucontrol_translate(struct kvm_vcpu *vcpu, gpa_t *gaddr) 4298 { 4299 int rc; 4300 4301 if (kvm_is_ucontrol(vcpu->kvm)) { 4302 rc = gmap_ucas_translate(vcpu->arch.mc, vcpu->arch.gmap, gaddr); 4303 if (rc == -EREMOTE) { 4304 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL; 4305 vcpu->run->s390_ucontrol.trans_exc_code = *gaddr; 4306 vcpu->run->s390_ucontrol.pgm_code = PGM_SEGMENT_TRANSLATION; 4307 } 4308 return rc; 4309 } 4310 return 0; 4311 } 4312 4313 static int kvm_s390_fixup_prefix(struct kvm_vcpu *vcpu) 4314 { 4315 gpa_t gaddr = kvm_s390_get_prefix(vcpu); 4316 gfn_t gfn; 4317 int rc; 4318 4319 if (vcpu_ucontrol_translate(vcpu, &gaddr)) 4320 return -EREMOTE; 4321 gfn = gpa_to_gfn(gaddr); 4322 4323 rc = kvm_s390_faultin_gfn_simple(vcpu, NULL, gfn, true); 4324 if (rc) 4325 return rc; 4326 rc = kvm_s390_faultin_gfn_simple(vcpu, NULL, gfn + 1, true); 4327 if (rc) 4328 return rc; 4329 4330 scoped_guard(write_lock, &vcpu->kvm->mmu_lock) 4331 rc = dat_set_prefix_notif_bit(vcpu->kvm->arch.gmap->asce, gfn); 4332 return rc; 4333 } 4334 4335 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu) 4336 { 4337 retry: 4338 kvm_s390_vcpu_request_handled(vcpu); 4339 if (!kvm_request_pending(vcpu)) 4340 return 0; 4341 /* 4342 * If the guest prefix changed, re-arm the ipte notifier for the 4343 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock. 4344 * This ensures that the ipte instruction for this request has 4345 * already finished. We might race against a second unmapper that 4346 * wants to set the blocking bit. Lets just retry the request loop. 4347 */ 4348 if (kvm_check_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu)) { 4349 int rc; 4350 4351 rc = kvm_s390_fixup_prefix(vcpu); 4352 if (rc) { 4353 kvm_make_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu); 4354 return rc; 4355 } 4356 goto retry; 4357 } 4358 4359 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) { 4360 vcpu->arch.sie_block->ihcpu = 0xffff; 4361 goto retry; 4362 } 4363 4364 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) { 4365 if (!ibs_enabled(vcpu)) { 4366 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1); 4367 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS); 4368 } 4369 goto retry; 4370 } 4371 4372 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) { 4373 if (ibs_enabled(vcpu)) { 4374 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0); 4375 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS); 4376 } 4377 goto retry; 4378 } 4379 4380 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) { 4381 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC; 4382 goto retry; 4383 } 4384 4385 if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) { 4386 /* 4387 * Disable CMM virtualization; we will emulate the ESSA 4388 * instruction manually, in order to provide additional 4389 * functionalities needed for live migration. 4390 */ 4391 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA; 4392 goto retry; 4393 } 4394 4395 if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) { 4396 /* 4397 * Re-enable CMM virtualization if CMMA is available and 4398 * CMM has been used. 4399 */ 4400 if (vcpu->kvm->arch.use_cmma && uses_cmm(vcpu->arch.gmap)) 4401 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA; 4402 goto retry; 4403 } 4404 4405 /* we left the vsie handler, nothing to do, just clear the request */ 4406 kvm_clear_request(KVM_REQ_VSIE_RESTART, vcpu); 4407 4408 return 0; 4409 } 4410 4411 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod) 4412 { 4413 struct kvm_vcpu *vcpu; 4414 union tod_clock clk; 4415 unsigned long i; 4416 4417 preempt_disable(); 4418 4419 store_tod_clock_ext(&clk); 4420 4421 kvm->arch.epoch = gtod->tod - clk.tod; 4422 kvm->arch.epdx = 0; 4423 if (test_kvm_facility(kvm, 139)) { 4424 kvm->arch.epdx = gtod->epoch_idx - clk.ei; 4425 if (kvm->arch.epoch > gtod->tod) 4426 kvm->arch.epdx -= 1; 4427 } 4428 4429 kvm_s390_vcpu_block_all(kvm); 4430 kvm_for_each_vcpu(i, vcpu, kvm) { 4431 vcpu->arch.sie_block->epoch = kvm->arch.epoch; 4432 vcpu->arch.sie_block->epdx = kvm->arch.epdx; 4433 } 4434 4435 kvm_s390_vcpu_unblock_all(kvm); 4436 preempt_enable(); 4437 } 4438 4439 int kvm_s390_try_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod) 4440 { 4441 if (!mutex_trylock(&kvm->lock)) 4442 return 0; 4443 __kvm_s390_set_tod_clock(kvm, gtod); 4444 mutex_unlock(&kvm->lock); 4445 return 1; 4446 } 4447 4448 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token, 4449 unsigned long token) 4450 { 4451 struct kvm_s390_interrupt inti = {}; 4452 struct kvm_s390_irq irq = {}; 4453 struct kvm_s390_interrupt_info *inti_mem = NULL; 4454 int ret = 0; 4455 4456 if (start_token) { 4457 irq.u.ext.ext_params2 = token; 4458 irq.type = KVM_S390_INT_PFAULT_INIT; 4459 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq)); 4460 } else { 4461 inti_mem = kzalloc_obj(*inti_mem, GFP_KERNEL_ACCOUNT); 4462 if (WARN_ON_ONCE(!inti_mem)) 4463 return; 4464 4465 inti.type = KVM_S390_INT_PFAULT_DONE; 4466 inti.parm64 = token; 4467 ret = kvm_s390_inject_vm(vcpu->kvm, &inti, inti_mem); 4468 if (ret) 4469 kfree(inti_mem); 4470 WARN_ON_ONCE(ret); 4471 } 4472 } 4473 4474 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, 4475 struct kvm_async_pf *work) 4476 { 4477 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token); 4478 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token); 4479 4480 return true; 4481 } 4482 4483 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, 4484 struct kvm_async_pf *work) 4485 { 4486 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token); 4487 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token); 4488 } 4489 4490 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, 4491 struct kvm_async_pf *work) 4492 { 4493 /* s390 will always inject the page directly */ 4494 } 4495 4496 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu) 4497 { 4498 /* 4499 * s390 will always inject the page directly, 4500 * but we still want check_async_completion to cleanup 4501 */ 4502 return true; 4503 } 4504 4505 bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu) 4506 { 4507 hva_t hva; 4508 struct kvm_arch_async_pf arch; 4509 4510 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) 4511 return false; 4512 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) != 4513 vcpu->arch.pfault_compare) 4514 return false; 4515 if (psw_extint_disabled(vcpu)) 4516 return false; 4517 if (kvm_s390_vcpu_has_irq(vcpu, 0)) 4518 return false; 4519 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) 4520 return false; 4521 if (!pfault_enabled(vcpu->arch.gmap)) 4522 return false; 4523 4524 hva = gfn_to_hva(vcpu->kvm, current->thread.gmap_teid.addr); 4525 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8)) 4526 return false; 4527 4528 return kvm_setup_async_pf(vcpu, current->thread.gmap_teid.addr * PAGE_SIZE, hva, &arch); 4529 } 4530 4531 static int vcpu_pre_run(struct kvm_vcpu *vcpu) 4532 { 4533 int rc, cpuflags; 4534 4535 /* 4536 * On s390 notifications for arriving pages will be delivered directly 4537 * to the guest but the house keeping for completed pfaults is 4538 * handled outside the worker. 4539 */ 4540 kvm_check_async_pf_completion(vcpu); 4541 4542 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14]; 4543 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15]; 4544 4545 if (!kvm_is_ucontrol(vcpu->kvm)) { 4546 rc = kvm_s390_deliver_pending_interrupts(vcpu); 4547 if (rc || guestdbg_exit_pending(vcpu)) 4548 return rc; 4549 } 4550 4551 rc = kvm_s390_handle_requests(vcpu); 4552 if (rc) 4553 return rc; 4554 4555 if (guestdbg_enabled(vcpu)) { 4556 kvm_s390_backup_guest_per_regs(vcpu); 4557 kvm_s390_patch_guest_per_regs(vcpu); 4558 } 4559 4560 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask); 4561 4562 vcpu->arch.sie_block->icptcode = 0; 4563 current->thread.gmap_int_code = 0; 4564 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags); 4565 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags); 4566 trace_kvm_s390_sie_enter(vcpu, cpuflags); 4567 4568 return 0; 4569 } 4570 4571 static int vcpu_post_run_addressing_exception(struct kvm_vcpu *vcpu) 4572 { 4573 struct kvm_s390_pgm_info pgm_info = { 4574 .code = PGM_ADDRESSING, 4575 }; 4576 u8 opcode, ilen; 4577 int rc; 4578 4579 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction"); 4580 trace_kvm_s390_sie_fault(vcpu); 4581 4582 /* 4583 * We want to inject an addressing exception, which is defined as a 4584 * suppressing or terminating exception. However, since we came here 4585 * by a DAT access exception, the PSW still points to the faulting 4586 * instruction since DAT exceptions are nullifying. So we've got 4587 * to look up the current opcode to get the length of the instruction 4588 * to be able to forward the PSW. 4589 */ 4590 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1); 4591 ilen = insn_length(opcode); 4592 if (rc < 0) { 4593 return rc; 4594 } else if (rc) { 4595 /* Instruction-Fetching Exceptions - we can't detect the ilen. 4596 * Forward by arbitrary ilc, injection will take care of 4597 * nullification if necessary. 4598 */ 4599 pgm_info = vcpu->arch.pgm; 4600 ilen = 4; 4601 } 4602 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID; 4603 kvm_s390_forward_psw(vcpu, ilen); 4604 return kvm_s390_inject_prog_irq(vcpu, &pgm_info); 4605 } 4606 4607 static void kvm_s390_assert_primary_as(struct kvm_vcpu *vcpu) 4608 { 4609 KVM_BUG(current->thread.gmap_teid.as != PSW_BITS_AS_PRIMARY, vcpu->kvm, 4610 "Unexpected program interrupt 0x%x, TEID 0x%016lx", 4611 current->thread.gmap_int_code, current->thread.gmap_teid.val); 4612 } 4613 4614 static int vcpu_dat_fault_handler(struct kvm_vcpu *vcpu, gpa_t gaddr, bool wr) 4615 { 4616 struct guest_fault f = { 4617 .write_attempt = wr, 4618 .attempt_pfault = pfault_enabled(vcpu->arch.gmap), 4619 }; 4620 int rc; 4621 4622 if (vcpu_ucontrol_translate(vcpu, &gaddr)) 4623 return -EREMOTE; 4624 f.gfn = gpa_to_gfn(gaddr); 4625 4626 rc = kvm_s390_faultin_gfn(vcpu, NULL, &f); 4627 if (rc <= 0) 4628 return rc; 4629 if (rc == PGM_ADDRESSING) 4630 return vcpu_post_run_addressing_exception(vcpu); 4631 KVM_BUG_ON(rc, vcpu->kvm); 4632 return -EINVAL; 4633 } 4634 4635 static int vcpu_post_run_handle_fault(struct kvm_vcpu *vcpu) 4636 { 4637 unsigned int foll = 0; 4638 unsigned long gaddr; 4639 int rc; 4640 4641 gaddr = current->thread.gmap_teid.addr * PAGE_SIZE; 4642 if (kvm_s390_cur_gmap_fault_is_write()) 4643 foll = FOLL_WRITE; 4644 4645 switch (current->thread.gmap_int_code & PGM_INT_CODE_MASK) { 4646 case 0: 4647 vcpu->stat.exit_null++; 4648 break; 4649 case PGM_SECURE_STORAGE_ACCESS: 4650 case PGM_SECURE_STORAGE_VIOLATION: 4651 kvm_s390_assert_primary_as(vcpu); 4652 /* 4653 * This can happen after a reboot with asynchronous teardown; 4654 * the new guest (normal or protected) will run on top of the 4655 * previous protected guest. The old pages need to be destroyed 4656 * so the new guest can use them. 4657 */ 4658 if (kvm_s390_pv_destroy_page(vcpu->kvm, gaddr)) { 4659 /* 4660 * Either KVM messed up the secure guest mapping or the 4661 * same page is mapped into multiple secure guests. 4662 * 4663 * This exception is only triggered when a guest 2 is 4664 * running and can therefore never occur in kernel 4665 * context. 4666 */ 4667 pr_warn_ratelimited("Secure storage violation (%x) in task: %s, pid %d\n", 4668 current->thread.gmap_int_code, current->comm, 4669 current->pid); 4670 send_sig(SIGSEGV, current, 0); 4671 } 4672 break; 4673 case PGM_NON_SECURE_STORAGE_ACCESS: 4674 kvm_s390_assert_primary_as(vcpu); 4675 /* 4676 * This is normal operation; a page belonging to a protected 4677 * guest has not been imported yet. Try to import the page into 4678 * the protected guest. 4679 */ 4680 rc = kvm_s390_pv_convert_to_secure(vcpu->kvm, gaddr); 4681 if (rc == -EINVAL) 4682 send_sig(SIGSEGV, current, 0); 4683 if (rc != -ENXIO) 4684 break; 4685 foll = FOLL_WRITE; 4686 fallthrough; 4687 case PGM_PROTECTION: 4688 case PGM_SEGMENT_TRANSLATION: 4689 case PGM_PAGE_TRANSLATION: 4690 case PGM_ASCE_TYPE: 4691 case PGM_REGION_FIRST_TRANS: 4692 case PGM_REGION_SECOND_TRANS: 4693 case PGM_REGION_THIRD_TRANS: 4694 kvm_s390_assert_primary_as(vcpu); 4695 return vcpu_dat_fault_handler(vcpu, gaddr, foll); 4696 default: 4697 KVM_BUG(1, vcpu->kvm, "Unexpected program interrupt 0x%x, TEID 0x%016lx", 4698 current->thread.gmap_int_code, current->thread.gmap_teid.val); 4699 send_sig(SIGSEGV, current, 0); 4700 break; 4701 } 4702 return 0; 4703 } 4704 4705 static int vcpu_post_run(struct kvm_vcpu *vcpu, int sie_return) 4706 { 4707 struct mcck_volatile_info *mcck_info; 4708 struct sie_page *sie_page; 4709 int rc; 4710 4711 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d", 4712 vcpu->arch.sie_block->icptcode); 4713 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode); 4714 4715 if (guestdbg_enabled(vcpu)) 4716 kvm_s390_restore_guest_per_regs(vcpu); 4717 4718 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14; 4719 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15; 4720 4721 if (sie_return == SIE64_RETURN_MCCK) { 4722 sie_page = container_of(vcpu->arch.sie_block, 4723 struct sie_page, sie_block); 4724 mcck_info = &sie_page->mcck_info; 4725 kvm_s390_reinject_machine_check(vcpu, mcck_info); 4726 return 0; 4727 } 4728 WARN_ON_ONCE(sie_return != SIE64_RETURN_NORMAL); 4729 4730 if (vcpu->arch.sie_block->icptcode > 0) { 4731 rc = kvm_handle_sie_intercept(vcpu); 4732 4733 if (rc != -EOPNOTSUPP) 4734 return rc; 4735 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC; 4736 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode; 4737 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa; 4738 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb; 4739 return -EREMOTE; 4740 } 4741 4742 return vcpu_post_run_handle_fault(vcpu); 4743 } 4744 4745 int noinstr kvm_s390_enter_exit_sie(struct kvm_s390_sie_block *scb, 4746 u64 *gprs, unsigned long gasce) 4747 { 4748 int ret; 4749 4750 guest_state_enter_irqoff(); 4751 4752 /* 4753 * The guest_state_{enter,exit}_irqoff() functions inform lockdep and 4754 * tracing that entry to the guest will enable host IRQs, and exit from 4755 * the guest will disable host IRQs. 4756 */ 4757 ret = sie64a(scb, gprs, gasce); 4758 4759 guest_state_exit_irqoff(); 4760 4761 return ret; 4762 } 4763 4764 #define PSW_INT_MASK (PSW_MASK_EXT | PSW_MASK_IO | PSW_MASK_MCHECK) 4765 static int __vcpu_run(struct kvm_vcpu *vcpu) 4766 { 4767 int rc, sie_return; 4768 struct sie_page *sie_page = (struct sie_page *)vcpu->arch.sie_block; 4769 4770 /* 4771 * We try to hold kvm->srcu during most of vcpu_run (except when run- 4772 * ning the guest), so that memslots (and other stuff) are protected 4773 */ 4774 kvm_vcpu_srcu_read_lock(vcpu); 4775 4776 while (true) { 4777 rc = vcpu_pre_run(vcpu); 4778 kvm_vcpu_srcu_read_unlock(vcpu); 4779 if (rc || guestdbg_exit_pending(vcpu)) 4780 break; 4781 4782 /* 4783 * As PF_VCPU will be used in fault handler, between 4784 * guest_timing_enter_irqoff and guest_timing_exit_irqoff 4785 * should be no uaccess. 4786 */ 4787 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 4788 memcpy(sie_page->pv_grregs, 4789 vcpu->run->s.regs.gprs, 4790 sizeof(sie_page->pv_grregs)); 4791 } 4792 4793 xfer_to_guest_mode_check: 4794 local_irq_disable(); 4795 xfer_to_guest_mode_prepare(); 4796 if (xfer_to_guest_mode_work_pending()) { 4797 local_irq_enable(); 4798 rc = kvm_xfer_to_guest_mode_handle_work(vcpu); 4799 if (rc) 4800 break; 4801 goto xfer_to_guest_mode_check; 4802 } 4803 4804 guest_timing_enter_irqoff(); 4805 __disable_cpu_timer_accounting(vcpu); 4806 4807 sie_return = kvm_s390_enter_exit_sie(vcpu->arch.sie_block, 4808 vcpu->run->s.regs.gprs, 4809 vcpu->arch.gmap->asce.val); 4810 4811 __enable_cpu_timer_accounting(vcpu); 4812 guest_timing_exit_irqoff(); 4813 local_irq_enable(); 4814 4815 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 4816 memcpy(vcpu->run->s.regs.gprs, 4817 sie_page->pv_grregs, 4818 sizeof(sie_page->pv_grregs)); 4819 /* 4820 * We're not allowed to inject interrupts on intercepts 4821 * that leave the guest state in an "in-between" state 4822 * where the next SIE entry will do a continuation. 4823 * Fence interrupts in our "internal" PSW. 4824 */ 4825 if (vcpu->arch.sie_block->icptcode == ICPT_PV_INSTR || 4826 vcpu->arch.sie_block->icptcode == ICPT_PV_PREF) { 4827 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK; 4828 } 4829 } 4830 kvm_vcpu_srcu_read_lock(vcpu); 4831 4832 rc = vcpu_post_run(vcpu, sie_return); 4833 if (rc || guestdbg_exit_pending(vcpu)) { 4834 kvm_vcpu_srcu_read_unlock(vcpu); 4835 break; 4836 } 4837 } 4838 4839 return rc; 4840 } 4841 4842 static void sync_regs_fmt2(struct kvm_vcpu *vcpu) 4843 { 4844 struct kvm_run *kvm_run = vcpu->run; 4845 struct runtime_instr_cb *riccb; 4846 struct gs_cb *gscb; 4847 4848 riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb; 4849 gscb = (struct gs_cb *) &kvm_run->s.regs.gscb; 4850 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask; 4851 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr; 4852 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) { 4853 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr; 4854 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp; 4855 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea; 4856 } 4857 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) { 4858 vcpu->arch.pfault_token = kvm_run->s.regs.pft; 4859 vcpu->arch.pfault_select = kvm_run->s.regs.pfs; 4860 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc; 4861 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) 4862 kvm_clear_async_pf_completion_queue(vcpu); 4863 } 4864 if (kvm_run->kvm_dirty_regs & KVM_SYNC_DIAG318) { 4865 vcpu->arch.diag318_info.val = kvm_run->s.regs.diag318; 4866 vcpu->arch.sie_block->cpnc = vcpu->arch.diag318_info.cpnc; 4867 VCPU_EVENT(vcpu, 3, "setting cpnc to %d", vcpu->arch.diag318_info.cpnc); 4868 } 4869 /* 4870 * If userspace sets the riccb (e.g. after migration) to a valid state, 4871 * we should enable RI here instead of doing the lazy enablement. 4872 */ 4873 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) && 4874 test_kvm_facility(vcpu->kvm, 64) && 4875 riccb->v && 4876 !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) { 4877 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)"); 4878 vcpu->arch.sie_block->ecb3 |= ECB3_RI; 4879 } 4880 /* 4881 * If userspace sets the gscb (e.g. after migration) to non-zero, 4882 * we should enable GS here instead of doing the lazy enablement. 4883 */ 4884 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) && 4885 test_kvm_facility(vcpu->kvm, 133) && 4886 gscb->gssm && 4887 !vcpu->arch.gs_enabled) { 4888 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)"); 4889 vcpu->arch.sie_block->ecb |= ECB_GS; 4890 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT; 4891 vcpu->arch.gs_enabled = 1; 4892 } 4893 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) && 4894 test_kvm_facility(vcpu->kvm, 82)) { 4895 vcpu->arch.sie_block->fpf &= ~FPF_BPBC; 4896 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0; 4897 } 4898 if (cpu_has_gs()) { 4899 preempt_disable(); 4900 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT); 4901 if (current->thread.gs_cb) { 4902 vcpu->arch.host_gscb = current->thread.gs_cb; 4903 save_gs_cb(vcpu->arch.host_gscb); 4904 } 4905 if (vcpu->arch.gs_enabled) { 4906 current->thread.gs_cb = (struct gs_cb *) 4907 &vcpu->run->s.regs.gscb; 4908 restore_gs_cb(current->thread.gs_cb); 4909 } 4910 preempt_enable(); 4911 } 4912 /* SIE will load etoken directly from SDNX and therefore kvm_run */ 4913 } 4914 4915 static void sync_regs(struct kvm_vcpu *vcpu) 4916 { 4917 struct kvm_run *kvm_run = vcpu->run; 4918 4919 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX) 4920 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix); 4921 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) { 4922 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128); 4923 /* some control register changes require a tlb flush */ 4924 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 4925 } 4926 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) { 4927 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm); 4928 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc; 4929 } 4930 save_access_regs(vcpu->arch.host_acrs); 4931 restore_access_regs(vcpu->run->s.regs.acrs); 4932 vcpu->arch.acrs_loaded = true; 4933 kvm_s390_fpu_load(vcpu->run); 4934 /* Sync fmt2 only data */ 4935 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu))) { 4936 sync_regs_fmt2(vcpu); 4937 } else { 4938 /* 4939 * In several places we have to modify our internal view to 4940 * not do things that are disallowed by the ultravisor. For 4941 * example we must not inject interrupts after specific exits 4942 * (e.g. 112 prefix page not secure). We do this by turning 4943 * off the machine check, external and I/O interrupt bits 4944 * of our PSW copy. To avoid getting validity intercepts, we 4945 * do only accept the condition code from userspace. 4946 */ 4947 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_CC; 4948 vcpu->arch.sie_block->gpsw.mask |= kvm_run->psw_mask & 4949 PSW_MASK_CC; 4950 } 4951 4952 kvm_run->kvm_dirty_regs = 0; 4953 } 4954 4955 static void store_regs_fmt2(struct kvm_vcpu *vcpu) 4956 { 4957 struct kvm_run *kvm_run = vcpu->run; 4958 4959 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr; 4960 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp; 4961 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea; 4962 kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC; 4963 kvm_run->s.regs.diag318 = vcpu->arch.diag318_info.val; 4964 if (cpu_has_gs()) { 4965 preempt_disable(); 4966 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT); 4967 if (vcpu->arch.gs_enabled) 4968 save_gs_cb(current->thread.gs_cb); 4969 current->thread.gs_cb = vcpu->arch.host_gscb; 4970 restore_gs_cb(vcpu->arch.host_gscb); 4971 if (!vcpu->arch.host_gscb) 4972 local_ctl_clear_bit(2, CR2_GUARDED_STORAGE_BIT); 4973 vcpu->arch.host_gscb = NULL; 4974 preempt_enable(); 4975 } 4976 /* SIE will save etoken directly into SDNX and therefore kvm_run */ 4977 } 4978 4979 static void store_regs(struct kvm_vcpu *vcpu) 4980 { 4981 struct kvm_run *kvm_run = vcpu->run; 4982 4983 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask; 4984 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr; 4985 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu); 4986 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128); 4987 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu); 4988 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc; 4989 kvm_run->s.regs.pft = vcpu->arch.pfault_token; 4990 kvm_run->s.regs.pfs = vcpu->arch.pfault_select; 4991 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare; 4992 save_access_regs(vcpu->run->s.regs.acrs); 4993 restore_access_regs(vcpu->arch.host_acrs); 4994 vcpu->arch.acrs_loaded = false; 4995 kvm_s390_fpu_store(vcpu->run); 4996 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu))) 4997 store_regs_fmt2(vcpu); 4998 } 4999 5000 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) 5001 { 5002 struct kvm_run *kvm_run = vcpu->run; 5003 DECLARE_KERNEL_FPU_ONSTACK32(fpu); 5004 int rc; 5005 5006 /* 5007 * Running a VM while dumping always has the potential to 5008 * produce inconsistent dump data. But for PV vcpus a SIE 5009 * entry while dumping could also lead to a fatal validity 5010 * intercept which we absolutely want to avoid. 5011 */ 5012 if (vcpu->kvm->arch.pv.dumping) 5013 return -EINVAL; 5014 5015 if (!vcpu->wants_to_run) 5016 return -EINTR; 5017 5018 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_S390_VALID_FIELDS || 5019 kvm_run->kvm_dirty_regs & ~KVM_SYNC_S390_VALID_FIELDS) 5020 return -EINVAL; 5021 5022 /* Pairs with smp_store_release() in kvm_arch_vcpu_postcreate() */ 5023 if (!smp_load_acquire(&vcpu->arch.initialized)) 5024 return -EINVAL; 5025 5026 vcpu_load(vcpu); 5027 5028 if (guestdbg_exit_pending(vcpu)) { 5029 kvm_s390_prepare_debug_exit(vcpu); 5030 rc = 0; 5031 goto out; 5032 } 5033 5034 kvm_sigset_activate(vcpu); 5035 5036 /* 5037 * no need to check the return value of vcpu_start as it can only have 5038 * an error for protvirt, but protvirt means user cpu state 5039 */ 5040 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) { 5041 kvm_s390_vcpu_start(vcpu); 5042 } else if (is_vcpu_stopped(vcpu)) { 5043 pr_err_ratelimited("can't run stopped vcpu %d\n", 5044 vcpu->vcpu_id); 5045 rc = -EINVAL; 5046 goto out_sigset; 5047 } 5048 5049 kernel_fpu_begin(&fpu, KERNEL_FPC | KERNEL_VXR); 5050 sync_regs(vcpu); 5051 enable_cpu_timer_accounting(vcpu); 5052 5053 might_fault(); 5054 rc = __vcpu_run(vcpu); 5055 5056 if (signal_pending(current) && !rc) { 5057 kvm_run->exit_reason = KVM_EXIT_INTR; 5058 vcpu->stat.signal_exits++; 5059 rc = -EINTR; 5060 } 5061 5062 if (guestdbg_exit_pending(vcpu) && !rc) { 5063 kvm_s390_prepare_debug_exit(vcpu); 5064 rc = 0; 5065 } 5066 5067 if (rc == -EREMOTE) { 5068 /* userspace support is needed, kvm_run has been prepared */ 5069 rc = 0; 5070 } 5071 5072 disable_cpu_timer_accounting(vcpu); 5073 store_regs(vcpu); 5074 kernel_fpu_end(&fpu, KERNEL_FPC | KERNEL_VXR); 5075 5076 vcpu->stat.exit_userspace++; 5077 5078 out_sigset: 5079 kvm_sigset_deactivate(vcpu); 5080 5081 out: 5082 vcpu_put(vcpu); 5083 return rc; 5084 } 5085 5086 /* 5087 * store status at address 5088 * we use have two special cases: 5089 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit 5090 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix 5091 */ 5092 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa) 5093 { 5094 unsigned char archmode = 1; 5095 freg_t fprs[NUM_FPRS]; 5096 unsigned int px; 5097 u64 clkcomp, cputm; 5098 int rc; 5099 5100 px = kvm_s390_get_prefix(vcpu); 5101 if (gpa == KVM_S390_STORE_STATUS_NOADDR) { 5102 if (write_guest_abs(vcpu, 163, &archmode, 1)) 5103 return -EFAULT; 5104 gpa = 0; 5105 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) { 5106 if (write_guest_real(vcpu, 163, &archmode, 1)) 5107 return -EFAULT; 5108 gpa = px; 5109 } else 5110 gpa -= __LC_FPREGS_SAVE_AREA; 5111 5112 /* manually convert vector registers if necessary */ 5113 if (cpu_has_vx()) { 5114 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs); 5115 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA, 5116 fprs, 128); 5117 } else { 5118 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA, 5119 vcpu->run->s.regs.fprs, 128); 5120 } 5121 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA, 5122 vcpu->run->s.regs.gprs, 128); 5123 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA, 5124 &vcpu->arch.sie_block->gpsw, 16); 5125 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA, 5126 &px, 4); 5127 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA, 5128 &vcpu->run->s.regs.fpc, 4); 5129 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA, 5130 &vcpu->arch.sie_block->todpr, 4); 5131 cputm = kvm_s390_get_cpu_timer(vcpu); 5132 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA, 5133 &cputm, 8); 5134 clkcomp = vcpu->arch.sie_block->ckc >> 8; 5135 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA, 5136 &clkcomp, 8); 5137 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA, 5138 &vcpu->run->s.regs.acrs, 64); 5139 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA, 5140 &vcpu->arch.sie_block->gcr, 128); 5141 return rc ? -EFAULT : 0; 5142 } 5143 5144 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr) 5145 { 5146 /* 5147 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy 5148 * switch in the run ioctl. Let's update our copies before we save 5149 * it into the save area 5150 */ 5151 kvm_s390_fpu_store(vcpu->run); 5152 save_access_regs(vcpu->run->s.regs.acrs); 5153 5154 return kvm_s390_store_status_unloaded(vcpu, addr); 5155 } 5156 5157 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu) 5158 { 5159 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu); 5160 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu); 5161 } 5162 5163 static void __disable_ibs_on_all_vcpus(struct kvm *kvm) 5164 { 5165 unsigned long i; 5166 struct kvm_vcpu *vcpu; 5167 5168 kvm_for_each_vcpu(i, vcpu, kvm) { 5169 __disable_ibs_on_vcpu(vcpu); 5170 } 5171 } 5172 5173 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu) 5174 { 5175 if (!sclp.has_ibs) 5176 return; 5177 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu); 5178 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu); 5179 } 5180 5181 int kvm_s390_vcpu_start(struct kvm_vcpu *vcpu) 5182 { 5183 int i, online_vcpus, r = 0, started_vcpus = 0; 5184 5185 if (!is_vcpu_stopped(vcpu)) 5186 return 0; 5187 5188 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1); 5189 /* Only one cpu at a time may enter/leave the STOPPED state. */ 5190 spin_lock(&vcpu->kvm->arch.start_stop_lock); 5191 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus); 5192 5193 /* Let's tell the UV that we want to change into the operating state */ 5194 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 5195 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR); 5196 if (r) { 5197 spin_unlock(&vcpu->kvm->arch.start_stop_lock); 5198 return r; 5199 } 5200 } 5201 5202 for (i = 0; i < online_vcpus; i++) { 5203 if (!is_vcpu_stopped(kvm_get_vcpu(vcpu->kvm, i))) 5204 started_vcpus++; 5205 } 5206 5207 if (started_vcpus == 0) { 5208 /* we're the only active VCPU -> speed it up */ 5209 __enable_ibs_on_vcpu(vcpu); 5210 } else if (started_vcpus == 1) { 5211 /* 5212 * As we are starting a second VCPU, we have to disable 5213 * the IBS facility on all VCPUs to remove potentially 5214 * outstanding ENABLE requests. 5215 */ 5216 __disable_ibs_on_all_vcpus(vcpu->kvm); 5217 } 5218 5219 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED); 5220 /* 5221 * The real PSW might have changed due to a RESTART interpreted by the 5222 * ultravisor. We block all interrupts and let the next sie exit 5223 * refresh our view. 5224 */ 5225 if (kvm_s390_pv_cpu_is_protected(vcpu)) 5226 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK; 5227 /* 5228 * Another VCPU might have used IBS while we were offline. 5229 * Let's play safe and flush the VCPU at startup. 5230 */ 5231 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 5232 spin_unlock(&vcpu->kvm->arch.start_stop_lock); 5233 return 0; 5234 } 5235 5236 int kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu) 5237 { 5238 int i, online_vcpus, r = 0, started_vcpus = 0; 5239 struct kvm_vcpu *started_vcpu = NULL; 5240 5241 if (is_vcpu_stopped(vcpu)) 5242 return 0; 5243 5244 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0); 5245 /* Only one cpu at a time may enter/leave the STOPPED state. */ 5246 spin_lock(&vcpu->kvm->arch.start_stop_lock); 5247 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus); 5248 5249 /* Let's tell the UV that we want to change into the stopped state */ 5250 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 5251 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_STP); 5252 if (r) { 5253 spin_unlock(&vcpu->kvm->arch.start_stop_lock); 5254 return r; 5255 } 5256 } 5257 5258 /* 5259 * Set the VCPU to STOPPED and THEN clear the interrupt flag, 5260 * now that the SIGP STOP and SIGP STOP AND STORE STATUS orders 5261 * have been fully processed. This will ensure that the VCPU 5262 * is kept BUSY if another VCPU is inquiring with SIGP SENSE. 5263 */ 5264 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED); 5265 kvm_s390_clear_stop_irq(vcpu); 5266 5267 __disable_ibs_on_vcpu(vcpu); 5268 5269 for (i = 0; i < online_vcpus; i++) { 5270 struct kvm_vcpu *tmp = kvm_get_vcpu(vcpu->kvm, i); 5271 5272 if (!is_vcpu_stopped(tmp)) { 5273 started_vcpus++; 5274 started_vcpu = tmp; 5275 } 5276 } 5277 5278 if (started_vcpus == 1) { 5279 /* 5280 * As we only have one VCPU left, we want to enable the 5281 * IBS facility for that VCPU to speed it up. 5282 */ 5283 __enable_ibs_on_vcpu(started_vcpu); 5284 } 5285 5286 spin_unlock(&vcpu->kvm->arch.start_stop_lock); 5287 return 0; 5288 } 5289 5290 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, 5291 struct kvm_enable_cap *cap) 5292 { 5293 int r; 5294 5295 if (cap->flags) 5296 return -EINVAL; 5297 5298 switch (cap->cap) { 5299 case KVM_CAP_S390_CSS_SUPPORT: 5300 if (!vcpu->kvm->arch.css_support) { 5301 vcpu->kvm->arch.css_support = 1; 5302 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support"); 5303 trace_kvm_s390_enable_css(vcpu->kvm); 5304 } 5305 r = 0; 5306 break; 5307 default: 5308 r = -EINVAL; 5309 break; 5310 } 5311 return r; 5312 } 5313 5314 static long kvm_s390_vcpu_sida_op(struct kvm_vcpu *vcpu, 5315 struct kvm_s390_mem_op *mop) 5316 { 5317 void __user *uaddr = (void __user *)mop->buf; 5318 void *sida_addr; 5319 int r = 0; 5320 5321 if (mop->flags || !mop->size) 5322 return -EINVAL; 5323 if (mop->size + mop->sida_offset < mop->size) 5324 return -EINVAL; 5325 if (mop->size + mop->sida_offset > sida_size(vcpu->arch.sie_block)) 5326 return -E2BIG; 5327 if (!kvm_s390_pv_cpu_is_protected(vcpu)) 5328 return -EINVAL; 5329 5330 sida_addr = (char *)sida_addr(vcpu->arch.sie_block) + mop->sida_offset; 5331 5332 switch (mop->op) { 5333 case KVM_S390_MEMOP_SIDA_READ: 5334 if (copy_to_user(uaddr, sida_addr, mop->size)) 5335 r = -EFAULT; 5336 5337 break; 5338 case KVM_S390_MEMOP_SIDA_WRITE: 5339 if (copy_from_user(sida_addr, uaddr, mop->size)) 5340 r = -EFAULT; 5341 break; 5342 } 5343 return r; 5344 } 5345 5346 static long kvm_s390_vcpu_mem_op(struct kvm_vcpu *vcpu, 5347 struct kvm_s390_mem_op *mop) 5348 { 5349 void __user *uaddr = (void __user *)mop->buf; 5350 void *tmpbuf __free(kvfree) = NULL; 5351 enum gacc_mode acc_mode; 5352 int r; 5353 5354 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_INJECT_EXCEPTION | 5355 KVM_S390_MEMOP_F_CHECK_ONLY | 5356 KVM_S390_MEMOP_F_SKEY_PROTECTION); 5357 if (r) 5358 return r; 5359 if (mop->ar >= NUM_ACRS) 5360 return -EINVAL; 5361 if (kvm_s390_pv_cpu_is_protected(vcpu)) 5362 return -EINVAL; 5363 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) { 5364 tmpbuf = vmalloc(mop->size); 5365 if (!tmpbuf) 5366 return -ENOMEM; 5367 } 5368 5369 acc_mode = mop->op == KVM_S390_MEMOP_LOGICAL_READ ? GACC_FETCH : GACC_STORE; 5370 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) { 5371 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size, 5372 acc_mode, mop->key); 5373 } else if (acc_mode == GACC_FETCH) { 5374 r = read_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf, 5375 mop->size, mop->key); 5376 if (!r && copy_to_user(uaddr, tmpbuf, mop->size)) 5377 return -EFAULT; 5378 } else { 5379 if (copy_from_user(tmpbuf, uaddr, mop->size)) 5380 return -EFAULT; 5381 r = write_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf, 5382 mop->size, mop->key); 5383 } 5384 5385 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0) 5386 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm); 5387 5388 return r; 5389 } 5390 5391 static long kvm_s390_vcpu_memsida_op(struct kvm_vcpu *vcpu, 5392 struct kvm_s390_mem_op *mop) 5393 { 5394 int r, srcu_idx; 5395 5396 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 5397 5398 switch (mop->op) { 5399 case KVM_S390_MEMOP_LOGICAL_READ: 5400 case KVM_S390_MEMOP_LOGICAL_WRITE: 5401 r = kvm_s390_vcpu_mem_op(vcpu, mop); 5402 break; 5403 case KVM_S390_MEMOP_SIDA_READ: 5404 case KVM_S390_MEMOP_SIDA_WRITE: 5405 /* we are locked against sida going away by the vcpu->mutex */ 5406 r = kvm_s390_vcpu_sida_op(vcpu, mop); 5407 break; 5408 default: 5409 r = -EINVAL; 5410 } 5411 5412 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx); 5413 return r; 5414 } 5415 5416 long kvm_arch_vcpu_unlocked_ioctl(struct file *filp, unsigned int ioctl, 5417 unsigned long arg) 5418 { 5419 struct kvm_vcpu *vcpu = filp->private_data; 5420 void __user *argp = (void __user *)arg; 5421 int rc; 5422 5423 switch (ioctl) { 5424 case KVM_S390_IRQ: { 5425 struct kvm_s390_irq s390irq; 5426 5427 if (copy_from_user(&s390irq, argp, sizeof(s390irq))) 5428 return -EFAULT; 5429 scoped_guard(srcu, &vcpu->kvm->srcu) 5430 rc = kvm_s390_inject_vcpu(vcpu, &s390irq); 5431 break; 5432 } 5433 case KVM_S390_INTERRUPT: { 5434 struct kvm_s390_interrupt s390int; 5435 struct kvm_s390_irq s390irq = {}; 5436 5437 if (kvm_is_ucontrol(vcpu->kvm)) 5438 return -EINVAL; 5439 if (copy_from_user(&s390int, argp, sizeof(s390int))) 5440 return -EFAULT; 5441 if (s390int_to_s390irq(&s390int, &s390irq)) 5442 return -EINVAL; 5443 scoped_guard(srcu, &vcpu->kvm->srcu) 5444 rc = kvm_s390_inject_vcpu(vcpu, &s390irq); 5445 break; 5446 } 5447 default: 5448 rc = -ENOIOCTLCMD; 5449 break; 5450 } 5451 5452 /* 5453 * To simplify single stepping of userspace-emulated instructions, 5454 * KVM_EXIT_S390_SIEIC exit sets KVM_GUESTDBG_EXIT_PENDING (see 5455 * should_handle_per_ifetch()). However, if userspace emulation injects 5456 * an interrupt, it needs to be cleared, so that KVM_EXIT_DEBUG happens 5457 * after (and not before) the interrupt delivery. 5458 */ 5459 if (!rc) 5460 vcpu->guest_debug &= ~KVM_GUESTDBG_EXIT_PENDING; 5461 5462 return rc; 5463 } 5464 5465 static int kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu *vcpu, 5466 struct kvm_pv_cmd *cmd) 5467 { 5468 struct kvm_s390_pv_dmp dmp; 5469 void *data; 5470 int ret; 5471 5472 /* Dump initialization is a prerequisite */ 5473 if (!vcpu->kvm->arch.pv.dumping) 5474 return -EINVAL; 5475 5476 if (copy_from_user(&dmp, (__u8 __user *)cmd->data, sizeof(dmp))) 5477 return -EFAULT; 5478 5479 /* We only handle this subcmd right now */ 5480 if (dmp.subcmd != KVM_PV_DUMP_CPU) 5481 return -EINVAL; 5482 5483 /* CPU dump length is the same as create cpu storage donation. */ 5484 if (dmp.buff_len != uv_info.guest_cpu_stor_len) 5485 return -EINVAL; 5486 5487 data = kvzalloc(uv_info.guest_cpu_stor_len, GFP_KERNEL); 5488 if (!data) 5489 return -ENOMEM; 5490 5491 ret = kvm_s390_pv_dump_cpu(vcpu, data, &cmd->rc, &cmd->rrc); 5492 5493 VCPU_EVENT(vcpu, 3, "PROTVIRT DUMP CPU %d rc %x rrc %x", 5494 vcpu->vcpu_id, cmd->rc, cmd->rrc); 5495 5496 if (ret) 5497 ret = -EINVAL; 5498 5499 /* On success copy over the dump data */ 5500 if (!ret && copy_to_user((__u8 __user *)dmp.buff_addr, data, uv_info.guest_cpu_stor_len)) 5501 ret = -EFAULT; 5502 5503 kvfree(data); 5504 return ret; 5505 } 5506 5507 long kvm_arch_vcpu_ioctl(struct file *filp, 5508 unsigned int ioctl, unsigned long arg) 5509 { 5510 struct kvm_vcpu *vcpu = filp->private_data; 5511 void __user *argp = (void __user *)arg; 5512 int idx; 5513 long r; 5514 u16 rc, rrc; 5515 5516 /* Pairs with smp_store_release() in kvm_arch_vcpu_postcreate() */ 5517 if (!smp_load_acquire(&vcpu->arch.initialized)) 5518 return -EINVAL; 5519 5520 vcpu_load(vcpu); 5521 5522 switch (ioctl) { 5523 case KVM_S390_STORE_STATUS: 5524 idx = srcu_read_lock(&vcpu->kvm->srcu); 5525 r = kvm_s390_store_status_unloaded(vcpu, arg); 5526 srcu_read_unlock(&vcpu->kvm->srcu, idx); 5527 break; 5528 case KVM_S390_SET_INITIAL_PSW: { 5529 psw_t psw; 5530 5531 r = -EFAULT; 5532 if (copy_from_user(&psw, argp, sizeof(psw))) 5533 break; 5534 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw); 5535 break; 5536 } 5537 case KVM_S390_CLEAR_RESET: 5538 r = 0; 5539 kvm_arch_vcpu_ioctl_clear_reset(vcpu); 5540 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 5541 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu), 5542 UVC_CMD_CPU_RESET_CLEAR, &rc, &rrc); 5543 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET CLEAR VCPU: rc %x rrc %x", 5544 rc, rrc); 5545 } 5546 break; 5547 case KVM_S390_INITIAL_RESET: 5548 r = 0; 5549 kvm_arch_vcpu_ioctl_initial_reset(vcpu); 5550 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 5551 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu), 5552 UVC_CMD_CPU_RESET_INITIAL, 5553 &rc, &rrc); 5554 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET INITIAL VCPU: rc %x rrc %x", 5555 rc, rrc); 5556 } 5557 break; 5558 case KVM_S390_NORMAL_RESET: 5559 r = 0; 5560 kvm_arch_vcpu_ioctl_normal_reset(vcpu); 5561 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 5562 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu), 5563 UVC_CMD_CPU_RESET, &rc, &rrc); 5564 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET NORMAL VCPU: rc %x rrc %x", 5565 rc, rrc); 5566 } 5567 break; 5568 case KVM_SET_ONE_REG: 5569 case KVM_GET_ONE_REG: { 5570 struct kvm_one_reg reg; 5571 r = -EINVAL; 5572 if (kvm_s390_pv_cpu_is_protected(vcpu)) 5573 break; 5574 r = -EFAULT; 5575 if (copy_from_user(®, argp, sizeof(reg))) 5576 break; 5577 if (ioctl == KVM_SET_ONE_REG) 5578 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®); 5579 else 5580 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®); 5581 break; 5582 } 5583 #ifdef CONFIG_KVM_S390_UCONTROL 5584 case KVM_S390_UCAS_MAP: { 5585 struct kvm_s390_ucas_mapping ucas; 5586 5587 r = -EFAULT; 5588 if (copy_from_user(&ucas, argp, sizeof(ucas))) 5589 break; 5590 5591 r = -EINVAL; 5592 if (!kvm_is_ucontrol(vcpu->kvm)) 5593 break; 5594 if (!IS_ALIGNED(ucas.user_addr | ucas.vcpu_addr | ucas.length, _SEGMENT_SIZE)) 5595 break; 5596 5597 r = gmap_ucas_map(vcpu->arch.gmap, gpa_to_gfn(ucas.user_addr), 5598 gpa_to_gfn(ucas.vcpu_addr), 5599 ucas.length >> _SEGMENT_SHIFT); 5600 break; 5601 } 5602 case KVM_S390_UCAS_UNMAP: { 5603 struct kvm_s390_ucas_mapping ucas; 5604 5605 r = -EFAULT; 5606 if (copy_from_user(&ucas, argp, sizeof(ucas))) 5607 break; 5608 5609 r = -EINVAL; 5610 if (!kvm_is_ucontrol(vcpu->kvm)) 5611 break; 5612 if (!IS_ALIGNED(ucas.vcpu_addr | ucas.length, _SEGMENT_SIZE)) 5613 break; 5614 5615 gmap_ucas_unmap(vcpu->arch.gmap, gpa_to_gfn(ucas.vcpu_addr), 5616 ucas.length >> _SEGMENT_SHIFT); 5617 r = 0; 5618 break; 5619 } 5620 #endif 5621 case KVM_S390_VCPU_FAULT: { 5622 gpa_t gaddr = arg; 5623 5624 scoped_guard(srcu, &vcpu->kvm->srcu) { 5625 r = vcpu_ucontrol_translate(vcpu, &gaddr); 5626 if (r) 5627 break; 5628 5629 r = kvm_s390_faultin_gfn_simple(vcpu, NULL, gpa_to_gfn(gaddr), false); 5630 if (r == PGM_ADDRESSING) 5631 r = -EFAULT; 5632 if (r <= 0) 5633 break; 5634 r = -EIO; 5635 KVM_BUG_ON(r, vcpu->kvm); 5636 } 5637 break; 5638 } 5639 case KVM_ENABLE_CAP: 5640 { 5641 struct kvm_enable_cap cap; 5642 r = -EFAULT; 5643 if (copy_from_user(&cap, argp, sizeof(cap))) 5644 break; 5645 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); 5646 break; 5647 } 5648 case KVM_S390_MEM_OP: { 5649 struct kvm_s390_mem_op mem_op; 5650 5651 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0) 5652 r = kvm_s390_vcpu_memsida_op(vcpu, &mem_op); 5653 else 5654 r = -EFAULT; 5655 break; 5656 } 5657 case KVM_S390_SET_IRQ_STATE: { 5658 struct kvm_s390_irq_state irq_state; 5659 5660 r = -EFAULT; 5661 if (copy_from_user(&irq_state, argp, sizeof(irq_state))) 5662 break; 5663 if (irq_state.len > VCPU_IRQS_MAX_BUF || 5664 irq_state.len == 0 || 5665 irq_state.len % sizeof(struct kvm_s390_irq) > 0) { 5666 r = -EINVAL; 5667 break; 5668 } 5669 /* do not use irq_state.flags, it will break old QEMUs */ 5670 r = kvm_s390_set_irq_state(vcpu, 5671 (void __user *) irq_state.buf, 5672 irq_state.len); 5673 break; 5674 } 5675 case KVM_S390_GET_IRQ_STATE: { 5676 struct kvm_s390_irq_state irq_state; 5677 5678 r = -EFAULT; 5679 if (copy_from_user(&irq_state, argp, sizeof(irq_state))) 5680 break; 5681 if (irq_state.len == 0) { 5682 r = -EINVAL; 5683 break; 5684 } 5685 /* do not use irq_state.flags, it will break old QEMUs */ 5686 r = kvm_s390_get_irq_state(vcpu, 5687 (__u8 __user *) irq_state.buf, 5688 irq_state.len); 5689 break; 5690 } 5691 case KVM_S390_PV_CPU_COMMAND: { 5692 struct kvm_pv_cmd cmd; 5693 5694 r = -EINVAL; 5695 if (!is_prot_virt_host()) 5696 break; 5697 5698 r = -EFAULT; 5699 if (copy_from_user(&cmd, argp, sizeof(cmd))) 5700 break; 5701 5702 r = -EINVAL; 5703 if (cmd.flags) 5704 break; 5705 5706 /* We only handle this cmd right now */ 5707 if (cmd.cmd != KVM_PV_DUMP) 5708 break; 5709 5710 r = kvm_s390_handle_pv_vcpu_dump(vcpu, &cmd); 5711 5712 /* Always copy over UV rc / rrc data */ 5713 if (copy_to_user(argp + offsetof(struct kvm_pv_cmd, rc), &cmd.rc, 5714 sizeof(cmd.rc) + sizeof(cmd.rrc))) 5715 r = -EFAULT; 5716 break; 5717 } 5718 default: 5719 r = -ENOTTY; 5720 } 5721 5722 vcpu_put(vcpu); 5723 return r; 5724 } 5725 5726 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 5727 { 5728 #ifdef CONFIG_KVM_S390_UCONTROL 5729 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET) 5730 && (kvm_is_ucontrol(vcpu->kvm))) { 5731 vmf->page = virt_to_page(vcpu->arch.sie_block); 5732 get_page(vmf->page); 5733 return 0; 5734 } 5735 #endif 5736 return VM_FAULT_SIGBUS; 5737 } 5738 5739 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm) 5740 { 5741 return true; 5742 } 5743 5744 /* Section: memory related */ 5745 int kvm_arch_prepare_memory_region(struct kvm *kvm, 5746 const struct kvm_memory_slot *old, 5747 struct kvm_memory_slot *new, 5748 enum kvm_mr_change change) 5749 { 5750 return s390_kvm_mmu_prepare_memory_region(kvm, old, new, change); 5751 } 5752 5753 static long cmma_d_count_pte(union pte *ptep, gfn_t gfn, gfn_t next, struct dat_walk *walk) 5754 { 5755 union pgste pgste; 5756 5757 pgste = pgste_get_lock(ptep); 5758 if (pgste.cmma_d) { 5759 pgste.cmma_d = 0; 5760 atomic64_dec(walk->priv); 5761 } 5762 pgste_set_unlock(ptep, pgste); 5763 return 0; 5764 } 5765 5766 void kvm_s390_update_cmma_dirty(struct kvm *kvm, const struct kvm_memory_slot *old) 5767 { 5768 const struct dat_walk_ops ops = { .pte_entry = cmma_d_count_pte, }; 5769 5770 if (kvm->arch.migration_mode && kvm->arch.use_cmma && old) { 5771 _dat_walk_gfn_range(old->base_gfn, old->base_gfn + old->npages, 5772 kvm->arch.gmap->asce, &ops, DAT_WALK_IGN_HOLES, 5773 &kvm->arch.cmma_dirty_pages); 5774 } 5775 } 5776 5777 void kvm_arch_commit_memory_region(struct kvm *kvm, struct kvm_memory_slot *old, 5778 const struct kvm_memory_slot *new, 5779 enum kvm_mr_change change) 5780 { 5781 } 5782 5783 /** 5784 * kvm_arch_vcpu_pre_fault_memory() -- pre-fault and link gmap dat tables 5785 * @vcpu: the vcpu that shall appear to have generated the fault-in. 5786 * @range: the range that needs to be faulted in. 5787 * 5788 * The first page of the given range is faulted in and the corresponding gmap 5789 * page tables are created, as if the given vCPU had performed a read 5790 * operation. 5791 * If the range starts outside any memslots, an error is returned. An error is 5792 * also returned for UCONTROL VMs, which should instead use the 5793 * KVM_S390_VCPU_FAULT ioctl. 5794 * 5795 * Return: 5796 * * %-ENOENT if the range lies outside of a memslot. 5797 * * %-EINVAL in case of invalid state (for example if the VM is UCONTROL). 5798 * * %-EIO if errors happen while faulting-in the page (will trigger a warning 5799 * in the caller). 5800 * * other error codes < 0 in case of other errors. 5801 * * otherwise a number > 0 of bytes that have been faulted in successfully. 5802 */ 5803 long kvm_arch_vcpu_pre_fault_memory(struct kvm_vcpu *vcpu, struct kvm_pre_fault_memory *range) 5804 { 5805 struct guest_fault f = { .gfn = gpa_to_gfn(range->gpa), }; 5806 gpa_t end; 5807 int rc; 5808 5809 if (kvm_is_ucontrol(vcpu->kvm)) 5810 return -EINVAL; 5811 5812 rc = kvm_s390_faultin_gfn(vcpu, NULL, &f); 5813 if (rc == PGM_ADDRESSING) 5814 return -ENOENT; 5815 if (rc > 0) 5816 return -EIO; 5817 if (rc < 0) 5818 return rc; 5819 5820 if (f.ptep) 5821 return PAGE_SIZE; 5822 5823 end = ALIGN(range->gpa + PAGE_SIZE, f.crste_region3 ? _REGION3_SIZE : HPAGE_SIZE); 5824 return min(range->size, end - range->gpa); 5825 } 5826 5827 /** 5828 * kvm_test_age_gfn() - test young 5829 * @kvm: the kvm instance 5830 * @range: the range of guest addresses whose young status needs to be cleared 5831 * 5832 * Context: called by KVM common code without holding the kvm mmu lock 5833 * Return: true if any page in the given range is young, otherwise 0. 5834 */ 5835 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range) 5836 { 5837 scoped_guard(read_lock, &kvm->mmu_lock) 5838 return dat_test_age_gfn(kvm->arch.gmap->asce, range->start, range->end); 5839 } 5840 5841 /** 5842 * kvm_age_gfn() - clear young 5843 * @kvm: the kvm instance 5844 * @range: the range of guest addresses whose young status needs to be cleared 5845 * 5846 * Context: called by KVM common code without holding the kvm mmu lock 5847 * Return: true if any page in the given range was young, otherwise 0. 5848 */ 5849 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range) 5850 { 5851 scoped_guard(read_lock, &kvm->mmu_lock) 5852 return gmap_age_gfn(kvm->arch.gmap, range->start, range->end); 5853 } 5854 5855 /** 5856 * kvm_unmap_gfn_range() - Unmap a range of guest addresses 5857 * @kvm: the kvm instance 5858 * @range: the range of guest page frames to invalidate 5859 * 5860 * This function always returns false because every DAT table modification 5861 * has to use the appropriate DAT table manipulation instructions, which will 5862 * keep the TLB coherent, hence no additional TLB flush is ever required. 5863 * 5864 * Context: called by KVM common code with the kvm mmu write lock held 5865 * Return: false 5866 */ 5867 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range) 5868 { 5869 return gmap_unmap_gfn_range(kvm->arch.gmap, range->slot, range->start, range->end); 5870 } 5871 5872 static inline unsigned long nonhyp_mask(int i) 5873 { 5874 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30; 5875 5876 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4); 5877 } 5878 5879 static int __init kvm_s390_init(void) 5880 { 5881 int i, r; 5882 5883 if (!sclp.has_sief2) { 5884 pr_info("SIE is not available\n"); 5885 return -ENODEV; 5886 } 5887 5888 if (hpage_2g && !hpage) { 5889 hpage_2g = 0; 5890 pr_info("Disabling 2G hugepage support, since 1M hugepage support is not enabled.\n"); 5891 } 5892 5893 for (i = 0; i < HMFAI_DWORDS; i++) 5894 kvm_s390_fac_base[i] |= nonhyp_mask(i); 5895 5896 r = __kvm_s390_init(); 5897 if (r) 5898 return r; 5899 5900 r = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE); 5901 if (r) { 5902 __kvm_s390_exit(); 5903 return r; 5904 } 5905 return 0; 5906 } 5907 5908 static void __exit kvm_s390_exit(void) 5909 { 5910 kvm_exit(); 5911 5912 __kvm_s390_exit(); 5913 } 5914 5915 module_init(kvm_s390_init); 5916 module_exit(kvm_s390_exit); 5917 5918 /* 5919 * Enable autoloading of the kvm module. 5920 * Note that we add the module alias here instead of virt/kvm/kvm_main.c 5921 * since x86 takes a different approach. 5922 */ 5923 #include <linux/miscdevice.h> 5924 MODULE_ALIAS_MISCDEV(KVM_MINOR); 5925 MODULE_ALIAS("devname:kvm"); 5926