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 "kvm-s390.h" 54 #include "gaccess.h" 55 #include "gmap.h" 56 #include "faultin.h" 57 #include "pci.h" 58 59 #define CREATE_TRACE_POINTS 60 #include "trace.h" 61 #include "trace-s390.h" 62 63 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */ 64 #define LOCAL_IRQS 32 65 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \ 66 (KVM_MAX_VCPUS + LOCAL_IRQS)) 67 68 const struct _kvm_stats_desc kvm_vm_stats_desc[] = { 69 KVM_GENERIC_VM_STATS(), 70 STATS_DESC_COUNTER(VM, inject_io), 71 STATS_DESC_COUNTER(VM, inject_float_mchk), 72 STATS_DESC_COUNTER(VM, inject_pfault_done), 73 STATS_DESC_COUNTER(VM, inject_service_signal), 74 STATS_DESC_COUNTER(VM, inject_virtio), 75 STATS_DESC_COUNTER(VM, aen_forward), 76 STATS_DESC_COUNTER(VM, gmap_shadow_reuse), 77 STATS_DESC_COUNTER(VM, gmap_shadow_create), 78 STATS_DESC_COUNTER(VM, gmap_shadow_r1_entry), 79 STATS_DESC_COUNTER(VM, gmap_shadow_r2_entry), 80 STATS_DESC_COUNTER(VM, gmap_shadow_r3_entry), 81 STATS_DESC_COUNTER(VM, gmap_shadow_sg_entry), 82 STATS_DESC_COUNTER(VM, gmap_shadow_pg_entry), 83 }; 84 85 const struct kvm_stats_header kvm_vm_stats_header = { 86 .name_size = KVM_STATS_NAME_SIZE, 87 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc), 88 .id_offset = sizeof(struct kvm_stats_header), 89 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE, 90 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE + 91 sizeof(kvm_vm_stats_desc), 92 }; 93 94 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = { 95 KVM_GENERIC_VCPU_STATS(), 96 STATS_DESC_COUNTER(VCPU, exit_userspace), 97 STATS_DESC_COUNTER(VCPU, exit_null), 98 STATS_DESC_COUNTER(VCPU, exit_external_request), 99 STATS_DESC_COUNTER(VCPU, exit_io_request), 100 STATS_DESC_COUNTER(VCPU, exit_external_interrupt), 101 STATS_DESC_COUNTER(VCPU, exit_stop_request), 102 STATS_DESC_COUNTER(VCPU, exit_validity), 103 STATS_DESC_COUNTER(VCPU, exit_instruction), 104 STATS_DESC_COUNTER(VCPU, exit_pei), 105 STATS_DESC_COUNTER(VCPU, halt_no_poll_steal), 106 STATS_DESC_COUNTER(VCPU, instruction_lctl), 107 STATS_DESC_COUNTER(VCPU, instruction_lctlg), 108 STATS_DESC_COUNTER(VCPU, instruction_stctl), 109 STATS_DESC_COUNTER(VCPU, instruction_stctg), 110 STATS_DESC_COUNTER(VCPU, exit_program_interruption), 111 STATS_DESC_COUNTER(VCPU, exit_instr_and_program), 112 STATS_DESC_COUNTER(VCPU, exit_operation_exception), 113 STATS_DESC_COUNTER(VCPU, deliver_ckc), 114 STATS_DESC_COUNTER(VCPU, deliver_cputm), 115 STATS_DESC_COUNTER(VCPU, deliver_external_call), 116 STATS_DESC_COUNTER(VCPU, deliver_emergency_signal), 117 STATS_DESC_COUNTER(VCPU, deliver_service_signal), 118 STATS_DESC_COUNTER(VCPU, deliver_virtio), 119 STATS_DESC_COUNTER(VCPU, deliver_stop_signal), 120 STATS_DESC_COUNTER(VCPU, deliver_prefix_signal), 121 STATS_DESC_COUNTER(VCPU, deliver_restart_signal), 122 STATS_DESC_COUNTER(VCPU, deliver_program), 123 STATS_DESC_COUNTER(VCPU, deliver_io), 124 STATS_DESC_COUNTER(VCPU, deliver_machine_check), 125 STATS_DESC_COUNTER(VCPU, exit_wait_state), 126 STATS_DESC_COUNTER(VCPU, inject_ckc), 127 STATS_DESC_COUNTER(VCPU, inject_cputm), 128 STATS_DESC_COUNTER(VCPU, inject_external_call), 129 STATS_DESC_COUNTER(VCPU, inject_emergency_signal), 130 STATS_DESC_COUNTER(VCPU, inject_mchk), 131 STATS_DESC_COUNTER(VCPU, inject_pfault_init), 132 STATS_DESC_COUNTER(VCPU, inject_program), 133 STATS_DESC_COUNTER(VCPU, inject_restart), 134 STATS_DESC_COUNTER(VCPU, inject_set_prefix), 135 STATS_DESC_COUNTER(VCPU, inject_stop_signal), 136 STATS_DESC_COUNTER(VCPU, instruction_epsw), 137 STATS_DESC_COUNTER(VCPU, instruction_gs), 138 STATS_DESC_COUNTER(VCPU, instruction_io_other), 139 STATS_DESC_COUNTER(VCPU, instruction_lpsw), 140 STATS_DESC_COUNTER(VCPU, instruction_lpswe), 141 STATS_DESC_COUNTER(VCPU, instruction_lpswey), 142 STATS_DESC_COUNTER(VCPU, instruction_pfmf), 143 STATS_DESC_COUNTER(VCPU, instruction_ptff), 144 STATS_DESC_COUNTER(VCPU, instruction_sck), 145 STATS_DESC_COUNTER(VCPU, instruction_sckpf), 146 STATS_DESC_COUNTER(VCPU, instruction_stidp), 147 STATS_DESC_COUNTER(VCPU, instruction_spx), 148 STATS_DESC_COUNTER(VCPU, instruction_stpx), 149 STATS_DESC_COUNTER(VCPU, instruction_stap), 150 STATS_DESC_COUNTER(VCPU, instruction_iske), 151 STATS_DESC_COUNTER(VCPU, instruction_ri), 152 STATS_DESC_COUNTER(VCPU, instruction_rrbe), 153 STATS_DESC_COUNTER(VCPU, instruction_sske), 154 STATS_DESC_COUNTER(VCPU, instruction_ipte_interlock), 155 STATS_DESC_COUNTER(VCPU, instruction_stsi), 156 STATS_DESC_COUNTER(VCPU, instruction_stfl), 157 STATS_DESC_COUNTER(VCPU, instruction_tb), 158 STATS_DESC_COUNTER(VCPU, instruction_tpi), 159 STATS_DESC_COUNTER(VCPU, instruction_tprot), 160 STATS_DESC_COUNTER(VCPU, instruction_tsch), 161 STATS_DESC_COUNTER(VCPU, instruction_sie), 162 STATS_DESC_COUNTER(VCPU, instruction_essa), 163 STATS_DESC_COUNTER(VCPU, instruction_sthyi), 164 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense), 165 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense_running), 166 STATS_DESC_COUNTER(VCPU, instruction_sigp_external_call), 167 STATS_DESC_COUNTER(VCPU, instruction_sigp_emergency), 168 STATS_DESC_COUNTER(VCPU, instruction_sigp_cond_emergency), 169 STATS_DESC_COUNTER(VCPU, instruction_sigp_start), 170 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop), 171 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop_store_status), 172 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_status), 173 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_adtl_status), 174 STATS_DESC_COUNTER(VCPU, instruction_sigp_arch), 175 STATS_DESC_COUNTER(VCPU, instruction_sigp_prefix), 176 STATS_DESC_COUNTER(VCPU, instruction_sigp_restart), 177 STATS_DESC_COUNTER(VCPU, instruction_sigp_init_cpu_reset), 178 STATS_DESC_COUNTER(VCPU, instruction_sigp_cpu_reset), 179 STATS_DESC_COUNTER(VCPU, instruction_sigp_unknown), 180 STATS_DESC_COUNTER(VCPU, instruction_diagnose_10), 181 STATS_DESC_COUNTER(VCPU, instruction_diagnose_44), 182 STATS_DESC_COUNTER(VCPU, instruction_diagnose_9c), 183 STATS_DESC_COUNTER(VCPU, diag_9c_ignored), 184 STATS_DESC_COUNTER(VCPU, diag_9c_forward), 185 STATS_DESC_COUNTER(VCPU, instruction_diagnose_258), 186 STATS_DESC_COUNTER(VCPU, instruction_diagnose_308), 187 STATS_DESC_COUNTER(VCPU, instruction_diagnose_500), 188 STATS_DESC_COUNTER(VCPU, instruction_diagnose_other), 189 STATS_DESC_COUNTER(VCPU, pfault_sync), 190 STATS_DESC_COUNTER(VCPU, signal_exits) 191 }; 192 193 const struct kvm_stats_header kvm_vcpu_stats_header = { 194 .name_size = KVM_STATS_NAME_SIZE, 195 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc), 196 .id_offset = sizeof(struct kvm_stats_header), 197 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE, 198 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE + 199 sizeof(kvm_vcpu_stats_desc), 200 }; 201 202 /* allow nested virtualization in KVM (if enabled by user space) */ 203 static int nested; 204 module_param(nested, int, S_IRUGO); 205 MODULE_PARM_DESC(nested, "Nested virtualization support"); 206 207 /* allow 1m huge page guest backing, if !nested */ 208 static int hpage; 209 module_param(hpage, int, 0444); 210 MODULE_PARM_DESC(hpage, "1m huge page backing support"); 211 212 /* maximum percentage of steal time for polling. >100 is treated like 100 */ 213 static u8 halt_poll_max_steal = 10; 214 module_param(halt_poll_max_steal, byte, 0644); 215 MODULE_PARM_DESC(halt_poll_max_steal, "Maximum percentage of steal time to allow polling"); 216 217 /* if set to true, the GISA will be initialized and used if available */ 218 static bool use_gisa = true; 219 module_param(use_gisa, bool, 0644); 220 MODULE_PARM_DESC(use_gisa, "Use the GISA if the host supports it."); 221 222 /* maximum diag9c forwarding per second */ 223 unsigned int diag9c_forwarding_hz; 224 module_param(diag9c_forwarding_hz, uint, 0644); 225 MODULE_PARM_DESC(diag9c_forwarding_hz, "Maximum diag9c forwarding per second, 0 to turn off"); 226 227 /* 228 * allow asynchronous deinit for protected guests; enable by default since 229 * the feature is opt-in anyway 230 */ 231 static int async_destroy = 1; 232 module_param(async_destroy, int, 0444); 233 MODULE_PARM_DESC(async_destroy, "Asynchronous destroy for protected guests"); 234 235 /* 236 * For now we handle at most 16 double words as this is what the s390 base 237 * kernel handles and stores in the prefix page. If we ever need to go beyond 238 * this, this requires changes to code, but the external uapi can stay. 239 */ 240 #define SIZE_INTERNAL 16 241 242 /* 243 * Base feature mask that defines default mask for facilities. Consists of the 244 * defines in FACILITIES_KVM and the non-hypervisor managed bits. 245 */ 246 static unsigned long kvm_s390_fac_base[SIZE_INTERNAL] = { FACILITIES_KVM }; 247 /* 248 * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL 249 * and defines the facilities that can be enabled via a cpu model. 250 */ 251 static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL] = { FACILITIES_KVM_CPUMODEL }; 252 253 static unsigned long kvm_s390_fac_size(void) 254 { 255 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_MASK_SIZE_U64); 256 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_LIST_SIZE_U64); 257 BUILD_BUG_ON(SIZE_INTERNAL * sizeof(unsigned long) > 258 sizeof(stfle_fac_list)); 259 260 return SIZE_INTERNAL; 261 } 262 263 /* available cpu features supported by kvm */ 264 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS); 265 /* available subfunctions indicated via query / "test bit" */ 266 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc; 267 268 debug_info_t *kvm_s390_dbf; 269 debug_info_t *kvm_s390_dbf_uv; 270 271 /* Section: not file related */ 272 /* forward declarations */ 273 static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta) 274 { 275 u8 delta_idx = 0; 276 277 /* 278 * The TOD jumps by delta, we have to compensate this by adding 279 * -delta to the epoch. 280 */ 281 delta = -delta; 282 283 /* sign-extension - we're adding to signed values below */ 284 if ((s64)delta < 0) 285 delta_idx = -1; 286 287 scb->epoch += delta; 288 if (scb->ecd & ECD_MEF) { 289 scb->epdx += delta_idx; 290 if (scb->epoch < delta) 291 scb->epdx += 1; 292 } 293 } 294 295 /* 296 * This callback is executed during stop_machine(). All CPUs are therefore 297 * temporarily stopped. In order not to change guest behavior, we have to 298 * disable preemption whenever we touch the epoch of kvm and the VCPUs, 299 * so a CPU won't be stopped while calculating with the epoch. 300 */ 301 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val, 302 void *v) 303 { 304 struct kvm *kvm; 305 struct kvm_vcpu *vcpu; 306 unsigned long i; 307 unsigned long long *delta = v; 308 309 list_for_each_entry(kvm, &vm_list, vm_list) { 310 kvm_for_each_vcpu(i, vcpu, kvm) { 311 kvm_clock_sync_scb(vcpu->arch.sie_block, *delta); 312 if (i == 0) { 313 kvm->arch.epoch = vcpu->arch.sie_block->epoch; 314 kvm->arch.epdx = vcpu->arch.sie_block->epdx; 315 } 316 if (vcpu->arch.cputm_enabled) 317 vcpu->arch.cputm_start += *delta; 318 if (vcpu->arch.vsie_block) 319 kvm_clock_sync_scb(vcpu->arch.vsie_block, 320 *delta); 321 } 322 } 323 return NOTIFY_OK; 324 } 325 326 static struct notifier_block kvm_clock_notifier = { 327 .notifier_call = kvm_clock_sync, 328 }; 329 330 static void allow_cpu_feat(unsigned long nr) 331 { 332 set_bit_inv(nr, kvm_s390_available_cpu_feat); 333 } 334 335 static inline int plo_test_bit(unsigned char nr) 336 { 337 unsigned long function = (unsigned long)nr | 0x100; 338 int cc; 339 340 asm volatile( 341 " lgr 0,%[function]\n" 342 /* Parameter registers are ignored for "test bit" */ 343 " plo 0,0,0,0(0)\n" 344 CC_IPM(cc) 345 : CC_OUT(cc, cc) 346 : [function] "d" (function) 347 : CC_CLOBBER_LIST("0")); 348 return CC_TRANSFORM(cc) == 0; 349 } 350 351 static __always_inline void pfcr_query(u8 (*query)[16]) 352 { 353 asm volatile( 354 " lghi 0,0\n" 355 " .insn rsy,0xeb0000000016,0,0,%[query]" 356 : [query] "=QS" (*query) 357 : 358 : "cc", "0"); 359 } 360 361 static __always_inline void __sortl_query(u8 (*query)[32]) 362 { 363 asm volatile( 364 " lghi 0,0\n" 365 " la 1,%[query]\n" 366 /* Parameter registers are ignored */ 367 " .insn rre,0xb9380000,2,4" 368 : [query] "=R" (*query) 369 : 370 : "cc", "0", "1"); 371 } 372 373 static __always_inline void __dfltcc_query(u8 (*query)[32]) 374 { 375 asm volatile( 376 " lghi 0,0\n" 377 " la 1,%[query]\n" 378 /* Parameter registers are ignored */ 379 " .insn rrf,0xb9390000,2,4,6,0" 380 : [query] "=R" (*query) 381 : 382 : "cc", "0", "1"); 383 } 384 385 static void __init kvm_s390_cpu_feat_init(void) 386 { 387 int i; 388 389 for (i = 0; i < 256; ++i) { 390 if (plo_test_bit(i)) 391 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7); 392 } 393 394 if (test_facility(28)) /* TOD-clock steering */ 395 ptff(kvm_s390_available_subfunc.ptff, 396 sizeof(kvm_s390_available_subfunc.ptff), 397 PTFF_QAF); 398 399 if (test_facility(17)) { /* MSA */ 400 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *) 401 kvm_s390_available_subfunc.kmac); 402 __cpacf_query(CPACF_KMC, (cpacf_mask_t *) 403 kvm_s390_available_subfunc.kmc); 404 __cpacf_query(CPACF_KM, (cpacf_mask_t *) 405 kvm_s390_available_subfunc.km); 406 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *) 407 kvm_s390_available_subfunc.kimd); 408 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *) 409 kvm_s390_available_subfunc.klmd); 410 } 411 if (test_facility(76)) /* MSA3 */ 412 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *) 413 kvm_s390_available_subfunc.pckmo); 414 if (test_facility(77)) { /* MSA4 */ 415 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *) 416 kvm_s390_available_subfunc.kmctr); 417 __cpacf_query(CPACF_KMF, (cpacf_mask_t *) 418 kvm_s390_available_subfunc.kmf); 419 __cpacf_query(CPACF_KMO, (cpacf_mask_t *) 420 kvm_s390_available_subfunc.kmo); 421 __cpacf_query(CPACF_PCC, (cpacf_mask_t *) 422 kvm_s390_available_subfunc.pcc); 423 } 424 if (test_facility(57)) /* MSA5 */ 425 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *) 426 kvm_s390_available_subfunc.ppno); 427 428 if (test_facility(146)) /* MSA8 */ 429 __cpacf_query(CPACF_KMA, (cpacf_mask_t *) 430 kvm_s390_available_subfunc.kma); 431 432 if (test_facility(155)) /* MSA9 */ 433 __cpacf_query(CPACF_KDSA, (cpacf_mask_t *) 434 kvm_s390_available_subfunc.kdsa); 435 436 if (test_facility(150)) /* SORTL */ 437 __sortl_query(&kvm_s390_available_subfunc.sortl); 438 439 if (test_facility(151)) /* DFLTCC */ 440 __dfltcc_query(&kvm_s390_available_subfunc.dfltcc); 441 442 if (test_facility(201)) /* PFCR */ 443 pfcr_query(&kvm_s390_available_subfunc.pfcr); 444 445 if (machine_has_esop()) 446 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP); 447 /* 448 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow), 449 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing). 450 */ 451 if (!sclp.has_sief2 || !machine_has_esop() || !sclp.has_64bscao || 452 !test_facility(3) || !nested) 453 return; 454 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2); 455 if (sclp.has_64bscao) 456 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO); 457 if (sclp.has_siif) 458 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF); 459 if (sclp.has_gpere) 460 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE); 461 if (sclp.has_gsls) 462 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS); 463 if (sclp.has_ib) 464 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB); 465 if (sclp.has_cei) 466 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI); 467 if (sclp.has_ibs) 468 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS); 469 if (sclp.has_kss) 470 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS); 471 /* 472 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make 473 * all skey handling functions read/set the skey from the PGSTE 474 * instead of the real storage key. 475 * 476 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make 477 * pages being detected as preserved although they are resident. 478 * 479 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will 480 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY. 481 * 482 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and 483 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be 484 * correctly shadowed. We can do that for the PGSTE but not for PTE.I. 485 * 486 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We 487 * cannot easily shadow the SCA because of the ipte lock. 488 */ 489 } 490 491 static int __init __kvm_s390_init(void) 492 { 493 int rc = -ENOMEM; 494 495 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long)); 496 if (!kvm_s390_dbf) 497 return -ENOMEM; 498 499 kvm_s390_dbf_uv = debug_register("kvm-uv", 32, 1, 7 * sizeof(long)); 500 if (!kvm_s390_dbf_uv) 501 goto err_kvm_uv; 502 503 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view) || 504 debug_register_view(kvm_s390_dbf_uv, &debug_sprintf_view)) 505 goto err_debug_view; 506 507 kvm_s390_cpu_feat_init(); 508 509 /* Register floating interrupt controller interface. */ 510 rc = kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC); 511 if (rc) { 512 pr_err("A FLIC registration call failed with rc=%d\n", rc); 513 goto err_flic; 514 } 515 516 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) { 517 rc = kvm_s390_pci_init(); 518 if (rc) { 519 pr_err("Unable to allocate AIFT for PCI\n"); 520 goto err_pci; 521 } 522 } 523 524 rc = kvm_s390_gib_init(GAL_ISC); 525 if (rc) 526 goto err_gib; 527 528 atomic_notifier_chain_register(&s390_epoch_delta_notifier, 529 &kvm_clock_notifier); 530 531 return 0; 532 533 err_gib: 534 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) 535 kvm_s390_pci_exit(); 536 err_pci: 537 err_flic: 538 err_debug_view: 539 debug_unregister(kvm_s390_dbf_uv); 540 err_kvm_uv: 541 debug_unregister(kvm_s390_dbf); 542 return rc; 543 } 544 545 static void __kvm_s390_exit(void) 546 { 547 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier, 548 &kvm_clock_notifier); 549 550 kvm_s390_gib_destroy(); 551 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) 552 kvm_s390_pci_exit(); 553 debug_unregister(kvm_s390_dbf); 554 debug_unregister(kvm_s390_dbf_uv); 555 } 556 557 static int kvm_s390_keyop(struct kvm_s390_mmu_cache *mc, struct kvm *kvm, int op, 558 unsigned long addr, union skey skey) 559 { 560 union asce asce = kvm->arch.gmap->asce; 561 gfn_t gfn = gpa_to_gfn(addr); 562 int r; 563 564 guard(read_lock)(&kvm->mmu_lock); 565 566 switch (op) { 567 case KVM_S390_KEYOP_SSKE: 568 r = dat_cond_set_storage_key(mc, asce, gfn, skey, &skey, 0, 0, 0); 569 if (r >= 0) 570 return skey.skey; 571 break; 572 case KVM_S390_KEYOP_ISKE: 573 r = dat_get_storage_key(asce, gfn, &skey); 574 if (!r) 575 return skey.skey; 576 break; 577 case KVM_S390_KEYOP_RRBE: 578 r = dat_reset_reference_bit(asce, gfn); 579 if (r > 0) 580 return r << 1; 581 break; 582 default: 583 return -EINVAL; 584 } 585 return r; 586 } 587 588 /* Section: device related */ 589 long kvm_arch_dev_ioctl(struct file *filp, 590 unsigned int ioctl, unsigned long arg) 591 { 592 if (ioctl == KVM_S390_ENABLE_SIE) 593 return 0; 594 return -EINVAL; 595 } 596 597 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 598 { 599 int r; 600 601 switch (ext) { 602 case KVM_CAP_S390_PSW: 603 case KVM_CAP_S390_GMAP: 604 case KVM_CAP_SYNC_MMU: 605 #ifdef CONFIG_KVM_S390_UCONTROL 606 case KVM_CAP_S390_UCONTROL: 607 #endif 608 case KVM_CAP_ASYNC_PF: 609 case KVM_CAP_SYNC_REGS: 610 case KVM_CAP_ONE_REG: 611 case KVM_CAP_ENABLE_CAP: 612 case KVM_CAP_S390_CSS_SUPPORT: 613 case KVM_CAP_IOEVENTFD: 614 case KVM_CAP_S390_IRQCHIP: 615 case KVM_CAP_VM_ATTRIBUTES: 616 case KVM_CAP_MP_STATE: 617 case KVM_CAP_IMMEDIATE_EXIT: 618 case KVM_CAP_S390_INJECT_IRQ: 619 case KVM_CAP_S390_USER_SIGP: 620 case KVM_CAP_S390_USER_STSI: 621 case KVM_CAP_S390_SKEYS: 622 case KVM_CAP_S390_IRQ_STATE: 623 case KVM_CAP_S390_USER_INSTR0: 624 case KVM_CAP_S390_CMMA_MIGRATION: 625 case KVM_CAP_S390_AIS: 626 case KVM_CAP_S390_AIS_MIGRATION: 627 case KVM_CAP_S390_VCPU_RESETS: 628 case KVM_CAP_SET_GUEST_DEBUG: 629 case KVM_CAP_S390_DIAG318: 630 case KVM_CAP_IRQFD_RESAMPLE: 631 case KVM_CAP_S390_USER_OPEREXEC: 632 case KVM_CAP_S390_KEYOP: 633 r = 1; 634 break; 635 case KVM_CAP_SET_GUEST_DEBUG2: 636 r = KVM_GUESTDBG_VALID_MASK; 637 break; 638 case KVM_CAP_S390_HPAGE_1M: 639 r = 0; 640 if (hpage && !(kvm && kvm_is_ucontrol(kvm))) 641 r = 1; 642 break; 643 case KVM_CAP_S390_MEM_OP: 644 r = MEM_OP_MAX_SIZE; 645 break; 646 case KVM_CAP_S390_MEM_OP_EXTENSION: 647 /* 648 * Flag bits indicating which extensions are supported. 649 * If r > 0, the base extension must also be supported/indicated, 650 * in order to maintain backwards compatibility. 651 */ 652 r = KVM_S390_MEMOP_EXTENSION_CAP_BASE | 653 KVM_S390_MEMOP_EXTENSION_CAP_CMPXCHG; 654 break; 655 case KVM_CAP_NR_VCPUS: 656 case KVM_CAP_MAX_VCPUS: 657 case KVM_CAP_MAX_VCPU_ID: 658 /* 659 * Return the same value for KVM_CAP_MAX_VCPUS and 660 * KVM_CAP_MAX_VCPU_ID to conform with the KVM API. 661 */ 662 r = KVM_S390_ESCA_CPU_SLOTS; 663 if (!kvm_s390_use_sca_entries()) 664 r = KVM_MAX_VCPUS; 665 if (ext == KVM_CAP_NR_VCPUS) 666 r = min_t(unsigned int, num_online_cpus(), r); 667 break; 668 case KVM_CAP_S390_COW: 669 r = machine_has_esop(); 670 break; 671 case KVM_CAP_S390_VECTOR_REGISTERS: 672 r = test_facility(129); 673 break; 674 case KVM_CAP_S390_RI: 675 r = test_facility(64); 676 break; 677 case KVM_CAP_S390_GS: 678 r = test_facility(133); 679 break; 680 case KVM_CAP_S390_BPB: 681 r = test_facility(82); 682 break; 683 case KVM_CAP_S390_PROTECTED_ASYNC_DISABLE: 684 r = async_destroy && is_prot_virt_host(); 685 break; 686 case KVM_CAP_S390_PROTECTED: 687 r = is_prot_virt_host(); 688 break; 689 case KVM_CAP_S390_PROTECTED_DUMP: { 690 u64 pv_cmds_dump[] = { 691 BIT_UVC_CMD_DUMP_INIT, 692 BIT_UVC_CMD_DUMP_CONFIG_STOR_STATE, 693 BIT_UVC_CMD_DUMP_CPU, 694 BIT_UVC_CMD_DUMP_COMPLETE, 695 }; 696 int i; 697 698 r = is_prot_virt_host(); 699 700 for (i = 0; i < ARRAY_SIZE(pv_cmds_dump); i++) { 701 if (!test_bit_inv(pv_cmds_dump[i], 702 (unsigned long *)&uv_info.inst_calls_list)) { 703 r = 0; 704 break; 705 } 706 } 707 break; 708 } 709 case KVM_CAP_S390_ZPCI_OP: 710 r = kvm_s390_pci_interp_allowed(); 711 break; 712 case KVM_CAP_S390_CPU_TOPOLOGY: 713 r = test_facility(11); 714 break; 715 default: 716 r = 0; 717 } 718 return r; 719 } 720 721 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot) 722 { 723 gfn_t last_gfn = memslot->base_gfn + memslot->npages; 724 725 scoped_guard(read_lock, &kvm->mmu_lock) 726 gmap_sync_dirty_log(kvm->arch.gmap, memslot->base_gfn, last_gfn); 727 } 728 729 /* Section: vm related */ 730 static void sca_del_vcpu(struct kvm_vcpu *vcpu); 731 732 /* 733 * Get (and clear) the dirty memory log for a memory slot. 734 */ 735 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, 736 struct kvm_dirty_log *log) 737 { 738 int r; 739 unsigned long n; 740 struct kvm_memory_slot *memslot; 741 int is_dirty; 742 743 if (kvm_is_ucontrol(kvm)) 744 return -EINVAL; 745 746 mutex_lock(&kvm->slots_lock); 747 748 r = -EINVAL; 749 if (log->slot >= KVM_USER_MEM_SLOTS) 750 goto out; 751 752 r = kvm_get_dirty_log(kvm, log, &is_dirty, &memslot); 753 if (r) 754 goto out; 755 756 /* Clear the dirty log */ 757 if (is_dirty) { 758 n = kvm_dirty_bitmap_bytes(memslot); 759 memset(memslot->dirty_bitmap, 0, n); 760 } 761 r = 0; 762 out: 763 mutex_unlock(&kvm->slots_lock); 764 return r; 765 } 766 767 static void icpt_operexc_on_all_vcpus(struct kvm *kvm) 768 { 769 unsigned long i; 770 struct kvm_vcpu *vcpu; 771 772 kvm_for_each_vcpu(i, vcpu, kvm) { 773 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu); 774 } 775 } 776 777 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap) 778 { 779 int r; 780 781 if (cap->flags) 782 return -EINVAL; 783 784 switch (cap->cap) { 785 case KVM_CAP_S390_IRQCHIP: 786 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP"); 787 kvm->arch.use_irqchip = 1; 788 r = 0; 789 break; 790 case KVM_CAP_S390_USER_SIGP: 791 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP"); 792 kvm->arch.user_sigp = 1; 793 r = 0; 794 break; 795 case KVM_CAP_S390_VECTOR_REGISTERS: 796 mutex_lock(&kvm->lock); 797 if (kvm->created_vcpus) { 798 r = -EBUSY; 799 } else if (cpu_has_vx()) { 800 set_kvm_facility(kvm->arch.model.fac_mask, 129); 801 set_kvm_facility(kvm->arch.model.fac_list, 129); 802 if (test_facility(134)) { 803 set_kvm_facility(kvm->arch.model.fac_mask, 134); 804 set_kvm_facility(kvm->arch.model.fac_list, 134); 805 } 806 if (test_facility(135)) { 807 set_kvm_facility(kvm->arch.model.fac_mask, 135); 808 set_kvm_facility(kvm->arch.model.fac_list, 135); 809 } 810 if (test_facility(148)) { 811 set_kvm_facility(kvm->arch.model.fac_mask, 148); 812 set_kvm_facility(kvm->arch.model.fac_list, 148); 813 } 814 if (test_facility(152)) { 815 set_kvm_facility(kvm->arch.model.fac_mask, 152); 816 set_kvm_facility(kvm->arch.model.fac_list, 152); 817 } 818 if (test_facility(192)) { 819 set_kvm_facility(kvm->arch.model.fac_mask, 192); 820 set_kvm_facility(kvm->arch.model.fac_list, 192); 821 } 822 if (test_facility(198)) { 823 set_kvm_facility(kvm->arch.model.fac_mask, 198); 824 set_kvm_facility(kvm->arch.model.fac_list, 198); 825 } 826 if (test_facility(199)) { 827 set_kvm_facility(kvm->arch.model.fac_mask, 199); 828 set_kvm_facility(kvm->arch.model.fac_list, 199); 829 } 830 r = 0; 831 } else 832 r = -EINVAL; 833 mutex_unlock(&kvm->lock); 834 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s", 835 r ? "(not available)" : "(success)"); 836 break; 837 case KVM_CAP_S390_RI: 838 r = -EINVAL; 839 mutex_lock(&kvm->lock); 840 if (kvm->created_vcpus) { 841 r = -EBUSY; 842 } else if (test_facility(64)) { 843 set_kvm_facility(kvm->arch.model.fac_mask, 64); 844 set_kvm_facility(kvm->arch.model.fac_list, 64); 845 r = 0; 846 } 847 mutex_unlock(&kvm->lock); 848 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s", 849 r ? "(not available)" : "(success)"); 850 break; 851 case KVM_CAP_S390_AIS: 852 mutex_lock(&kvm->lock); 853 if (kvm->created_vcpus) { 854 r = -EBUSY; 855 } else { 856 set_kvm_facility(kvm->arch.model.fac_mask, 72); 857 set_kvm_facility(kvm->arch.model.fac_list, 72); 858 r = 0; 859 } 860 mutex_unlock(&kvm->lock); 861 VM_EVENT(kvm, 3, "ENABLE: AIS %s", 862 r ? "(not available)" : "(success)"); 863 break; 864 case KVM_CAP_S390_GS: 865 r = -EINVAL; 866 mutex_lock(&kvm->lock); 867 if (kvm->created_vcpus) { 868 r = -EBUSY; 869 } else if (test_facility(133)) { 870 set_kvm_facility(kvm->arch.model.fac_mask, 133); 871 set_kvm_facility(kvm->arch.model.fac_list, 133); 872 r = 0; 873 } 874 mutex_unlock(&kvm->lock); 875 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s", 876 r ? "(not available)" : "(success)"); 877 break; 878 case KVM_CAP_S390_HPAGE_1M: 879 mutex_lock(&kvm->lock); 880 if (kvm->created_vcpus) 881 r = -EBUSY; 882 else if (!hpage || kvm->arch.use_cmma || kvm_is_ucontrol(kvm)) 883 r = -EINVAL; 884 else { 885 r = 0; 886 set_bit(GMAP_FLAG_ALLOW_HPAGE_1M, &kvm->arch.gmap->flags); 887 /* 888 * We might have to create fake 4k page 889 * tables. To avoid that the hardware works on 890 * stale PGSTEs, we emulate these instructions. 891 */ 892 kvm->arch.use_skf = 0; 893 kvm->arch.use_pfmfi = 0; 894 } 895 mutex_unlock(&kvm->lock); 896 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_HPAGE %s", 897 r ? "(not available)" : "(success)"); 898 break; 899 case KVM_CAP_S390_USER_STSI: 900 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI"); 901 kvm->arch.user_stsi = 1; 902 r = 0; 903 break; 904 case KVM_CAP_S390_USER_INSTR0: 905 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0"); 906 kvm->arch.user_instr0 = 1; 907 icpt_operexc_on_all_vcpus(kvm); 908 r = 0; 909 break; 910 case KVM_CAP_S390_CPU_TOPOLOGY: 911 r = -EINVAL; 912 mutex_lock(&kvm->lock); 913 if (kvm->created_vcpus) { 914 r = -EBUSY; 915 } else if (test_facility(11)) { 916 set_kvm_facility(kvm->arch.model.fac_mask, 11); 917 set_kvm_facility(kvm->arch.model.fac_list, 11); 918 r = 0; 919 } 920 mutex_unlock(&kvm->lock); 921 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_CPU_TOPOLOGY %s", 922 r ? "(not available)" : "(success)"); 923 break; 924 case KVM_CAP_S390_USER_OPEREXEC: 925 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_OPEREXEC"); 926 kvm->arch.user_operexec = 1; 927 icpt_operexc_on_all_vcpus(kvm); 928 r = 0; 929 break; 930 default: 931 r = -EINVAL; 932 break; 933 } 934 return r; 935 } 936 937 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr) 938 { 939 int ret; 940 941 switch (attr->attr) { 942 case KVM_S390_VM_MEM_LIMIT_SIZE: 943 ret = 0; 944 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes", 945 kvm->arch.mem_limit); 946 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr)) 947 ret = -EFAULT; 948 break; 949 default: 950 ret = -ENXIO; 951 break; 952 } 953 return ret; 954 } 955 956 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr) 957 { 958 int ret; 959 960 switch (attr->attr) { 961 case KVM_S390_VM_MEM_ENABLE_CMMA: 962 ret = -ENXIO; 963 if (!sclp.has_cmma) 964 break; 965 966 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support"); 967 mutex_lock(&kvm->lock); 968 if (kvm->created_vcpus) 969 ret = -EBUSY; 970 else { 971 kvm->arch.use_cmma = 1; 972 /* Not compatible with cmma. */ 973 kvm->arch.use_pfmfi = 0; 974 ret = 0; 975 } 976 mutex_unlock(&kvm->lock); 977 break; 978 case KVM_S390_VM_MEM_CLR_CMMA: { 979 gfn_t start_gfn = 0; 980 981 ret = -ENXIO; 982 if (!sclp.has_cmma) 983 break; 984 ret = -EINVAL; 985 if (!kvm->arch.use_cmma) 986 break; 987 988 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states"); 989 do { 990 start_gfn = dat_reset_cmma(kvm->arch.gmap->asce, start_gfn); 991 cond_resched(); 992 } while (start_gfn); 993 ret = 0; 994 break; 995 } 996 case KVM_S390_VM_MEM_LIMIT_SIZE: { 997 unsigned long new_limit; 998 999 if (kvm_is_ucontrol(kvm)) 1000 return -EINVAL; 1001 1002 if (get_user(new_limit, (u64 __user *)attr->addr)) 1003 return -EFAULT; 1004 1005 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT && 1006 new_limit > kvm->arch.mem_limit) 1007 return -E2BIG; 1008 1009 if (!new_limit) 1010 return -EINVAL; 1011 1012 ret = -EBUSY; 1013 if (!kvm->created_vcpus) 1014 ret = gmap_set_limit(kvm->arch.gmap, gpa_to_gfn(new_limit)); 1015 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit); 1016 VM_EVENT(kvm, 3, "New guest asce: 0x%p", 1017 (void *)kvm->arch.gmap->asce.val); 1018 break; 1019 } 1020 default: 1021 ret = -ENXIO; 1022 break; 1023 } 1024 return ret; 1025 } 1026 1027 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu); 1028 1029 void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm) 1030 { 1031 struct kvm_vcpu *vcpu; 1032 unsigned long i; 1033 1034 kvm_s390_vcpu_block_all(kvm); 1035 1036 kvm_for_each_vcpu(i, vcpu, kvm) { 1037 kvm_s390_vcpu_crypto_setup(vcpu); 1038 /* recreate the shadow crycb by leaving the VSIE handler */ 1039 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu); 1040 } 1041 1042 kvm_s390_vcpu_unblock_all(kvm); 1043 } 1044 1045 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr) 1046 { 1047 mutex_lock(&kvm->lock); 1048 switch (attr->attr) { 1049 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW: 1050 if (!test_kvm_facility(kvm, 76)) { 1051 mutex_unlock(&kvm->lock); 1052 return -EINVAL; 1053 } 1054 get_random_bytes( 1055 kvm->arch.crypto.crycb->aes_wrapping_key_mask, 1056 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask)); 1057 kvm->arch.crypto.aes_kw = 1; 1058 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support"); 1059 break; 1060 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW: 1061 if (!test_kvm_facility(kvm, 76)) { 1062 mutex_unlock(&kvm->lock); 1063 return -EINVAL; 1064 } 1065 get_random_bytes( 1066 kvm->arch.crypto.crycb->dea_wrapping_key_mask, 1067 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask)); 1068 kvm->arch.crypto.dea_kw = 1; 1069 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support"); 1070 break; 1071 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW: 1072 if (!test_kvm_facility(kvm, 76)) { 1073 mutex_unlock(&kvm->lock); 1074 return -EINVAL; 1075 } 1076 kvm->arch.crypto.aes_kw = 0; 1077 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0, 1078 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask)); 1079 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support"); 1080 break; 1081 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW: 1082 if (!test_kvm_facility(kvm, 76)) { 1083 mutex_unlock(&kvm->lock); 1084 return -EINVAL; 1085 } 1086 kvm->arch.crypto.dea_kw = 0; 1087 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0, 1088 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask)); 1089 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support"); 1090 break; 1091 case KVM_S390_VM_CRYPTO_ENABLE_APIE: 1092 if (!ap_instructions_available()) { 1093 mutex_unlock(&kvm->lock); 1094 return -EOPNOTSUPP; 1095 } 1096 kvm->arch.crypto.apie = 1; 1097 break; 1098 case KVM_S390_VM_CRYPTO_DISABLE_APIE: 1099 if (!ap_instructions_available()) { 1100 mutex_unlock(&kvm->lock); 1101 return -EOPNOTSUPP; 1102 } 1103 kvm->arch.crypto.apie = 0; 1104 break; 1105 default: 1106 mutex_unlock(&kvm->lock); 1107 return -ENXIO; 1108 } 1109 1110 kvm_s390_vcpu_crypto_reset_all(kvm); 1111 mutex_unlock(&kvm->lock); 1112 return 0; 1113 } 1114 1115 static void kvm_s390_vcpu_pci_setup(struct kvm_vcpu *vcpu) 1116 { 1117 /* Only set the ECB bits after guest requests zPCI interpretation */ 1118 if (!vcpu->kvm->arch.use_zpci_interp) 1119 return; 1120 1121 vcpu->arch.sie_block->ecb2 |= ECB2_ZPCI_LSI; 1122 vcpu->arch.sie_block->ecb3 |= ECB3_AISII + ECB3_AISI; 1123 } 1124 1125 void kvm_s390_vcpu_pci_enable_interp(struct kvm *kvm) 1126 { 1127 struct kvm_vcpu *vcpu; 1128 unsigned long i; 1129 1130 lockdep_assert_held(&kvm->lock); 1131 1132 if (!kvm_s390_pci_interp_allowed()) 1133 return; 1134 1135 /* 1136 * If host is configured for PCI and the necessary facilities are 1137 * available, turn on interpretation for the life of this guest 1138 */ 1139 kvm->arch.use_zpci_interp = 1; 1140 1141 kvm_s390_vcpu_block_all(kvm); 1142 1143 kvm_for_each_vcpu(i, vcpu, kvm) { 1144 kvm_s390_vcpu_pci_setup(vcpu); 1145 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu); 1146 } 1147 1148 kvm_s390_vcpu_unblock_all(kvm); 1149 } 1150 1151 static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req) 1152 { 1153 unsigned long cx; 1154 struct kvm_vcpu *vcpu; 1155 1156 kvm_for_each_vcpu(cx, vcpu, kvm) 1157 kvm_s390_sync_request(req, vcpu); 1158 } 1159 1160 /* 1161 * Must be called with kvm->srcu held to avoid races on memslots, and with 1162 * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration. 1163 */ 1164 static int kvm_s390_vm_start_migration(struct kvm *kvm) 1165 { 1166 struct kvm_memory_slot *ms; 1167 struct kvm_memslots *slots; 1168 unsigned long ram_pages = 0; 1169 int bkt; 1170 1171 /* migration mode already enabled */ 1172 if (kvm->arch.migration_mode) 1173 return 0; 1174 slots = kvm_memslots(kvm); 1175 if (!slots || kvm_memslots_empty(slots)) 1176 return -EINVAL; 1177 1178 if (!kvm->arch.use_cmma) { 1179 kvm->arch.migration_mode = 1; 1180 return 0; 1181 } 1182 kvm_for_each_memslot(ms, bkt, slots) { 1183 if (!ms->dirty_bitmap) 1184 return -EINVAL; 1185 ram_pages += ms->npages; 1186 } 1187 /* mark all the pages as dirty */ 1188 gmap_set_cmma_all_dirty(kvm->arch.gmap); 1189 atomic64_set(&kvm->arch.cmma_dirty_pages, ram_pages); 1190 kvm->arch.migration_mode = 1; 1191 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION); 1192 return 0; 1193 } 1194 1195 /* 1196 * Must be called with kvm->slots_lock to avoid races with ourselves and 1197 * kvm_s390_vm_start_migration. 1198 */ 1199 static int kvm_s390_vm_stop_migration(struct kvm *kvm) 1200 { 1201 /* migration mode already disabled */ 1202 if (!kvm->arch.migration_mode) 1203 return 0; 1204 kvm->arch.migration_mode = 0; 1205 if (kvm->arch.use_cmma) 1206 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION); 1207 return 0; 1208 } 1209 1210 static int kvm_s390_vm_set_migration(struct kvm *kvm, 1211 struct kvm_device_attr *attr) 1212 { 1213 int res = -ENXIO; 1214 1215 mutex_lock(&kvm->slots_lock); 1216 switch (attr->attr) { 1217 case KVM_S390_VM_MIGRATION_START: 1218 res = kvm_s390_vm_start_migration(kvm); 1219 break; 1220 case KVM_S390_VM_MIGRATION_STOP: 1221 res = kvm_s390_vm_stop_migration(kvm); 1222 break; 1223 default: 1224 break; 1225 } 1226 mutex_unlock(&kvm->slots_lock); 1227 1228 return res; 1229 } 1230 1231 static int kvm_s390_vm_get_migration(struct kvm *kvm, 1232 struct kvm_device_attr *attr) 1233 { 1234 u64 mig = kvm->arch.migration_mode; 1235 1236 if (attr->attr != KVM_S390_VM_MIGRATION_STATUS) 1237 return -ENXIO; 1238 1239 if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig))) 1240 return -EFAULT; 1241 return 0; 1242 } 1243 1244 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod); 1245 1246 static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr) 1247 { 1248 struct kvm_s390_vm_tod_clock gtod; 1249 1250 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod))) 1251 return -EFAULT; 1252 1253 if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx) 1254 return -EINVAL; 1255 __kvm_s390_set_tod_clock(kvm, >od); 1256 1257 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx", 1258 gtod.epoch_idx, gtod.tod); 1259 1260 return 0; 1261 } 1262 1263 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr) 1264 { 1265 u8 gtod_high; 1266 1267 if (copy_from_user(>od_high, (void __user *)attr->addr, 1268 sizeof(gtod_high))) 1269 return -EFAULT; 1270 1271 if (gtod_high != 0) 1272 return -EINVAL; 1273 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high); 1274 1275 return 0; 1276 } 1277 1278 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr) 1279 { 1280 struct kvm_s390_vm_tod_clock gtod = { 0 }; 1281 1282 if (copy_from_user(>od.tod, (void __user *)attr->addr, 1283 sizeof(gtod.tod))) 1284 return -EFAULT; 1285 1286 __kvm_s390_set_tod_clock(kvm, >od); 1287 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod); 1288 return 0; 1289 } 1290 1291 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr) 1292 { 1293 int ret; 1294 1295 if (attr->flags) 1296 return -EINVAL; 1297 1298 mutex_lock(&kvm->lock); 1299 /* 1300 * For protected guests, the TOD is managed by the ultravisor, so trying 1301 * to change it will never bring the expected results. 1302 */ 1303 if (kvm_s390_pv_is_protected(kvm)) { 1304 ret = -EOPNOTSUPP; 1305 goto out_unlock; 1306 } 1307 1308 switch (attr->attr) { 1309 case KVM_S390_VM_TOD_EXT: 1310 ret = kvm_s390_set_tod_ext(kvm, attr); 1311 break; 1312 case KVM_S390_VM_TOD_HIGH: 1313 ret = kvm_s390_set_tod_high(kvm, attr); 1314 break; 1315 case KVM_S390_VM_TOD_LOW: 1316 ret = kvm_s390_set_tod_low(kvm, attr); 1317 break; 1318 default: 1319 ret = -ENXIO; 1320 break; 1321 } 1322 1323 out_unlock: 1324 mutex_unlock(&kvm->lock); 1325 return ret; 1326 } 1327 1328 static void kvm_s390_get_tod_clock(struct kvm *kvm, 1329 struct kvm_s390_vm_tod_clock *gtod) 1330 { 1331 union tod_clock clk; 1332 1333 preempt_disable(); 1334 1335 store_tod_clock_ext(&clk); 1336 1337 gtod->tod = clk.tod + kvm->arch.epoch; 1338 gtod->epoch_idx = 0; 1339 if (test_kvm_facility(kvm, 139)) { 1340 gtod->epoch_idx = clk.ei + kvm->arch.epdx; 1341 if (gtod->tod < clk.tod) 1342 gtod->epoch_idx += 1; 1343 } 1344 1345 preempt_enable(); 1346 } 1347 1348 static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr) 1349 { 1350 struct kvm_s390_vm_tod_clock gtod; 1351 1352 memset(>od, 0, sizeof(gtod)); 1353 kvm_s390_get_tod_clock(kvm, >od); 1354 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod))) 1355 return -EFAULT; 1356 1357 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx", 1358 gtod.epoch_idx, gtod.tod); 1359 return 0; 1360 } 1361 1362 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr) 1363 { 1364 u8 gtod_high = 0; 1365 1366 if (copy_to_user((void __user *)attr->addr, >od_high, 1367 sizeof(gtod_high))) 1368 return -EFAULT; 1369 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high); 1370 1371 return 0; 1372 } 1373 1374 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr) 1375 { 1376 u64 gtod; 1377 1378 gtod = kvm_s390_get_tod_clock_fast(kvm); 1379 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod))) 1380 return -EFAULT; 1381 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod); 1382 1383 return 0; 1384 } 1385 1386 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr) 1387 { 1388 int ret; 1389 1390 if (attr->flags) 1391 return -EINVAL; 1392 1393 switch (attr->attr) { 1394 case KVM_S390_VM_TOD_EXT: 1395 ret = kvm_s390_get_tod_ext(kvm, attr); 1396 break; 1397 case KVM_S390_VM_TOD_HIGH: 1398 ret = kvm_s390_get_tod_high(kvm, attr); 1399 break; 1400 case KVM_S390_VM_TOD_LOW: 1401 ret = kvm_s390_get_tod_low(kvm, attr); 1402 break; 1403 default: 1404 ret = -ENXIO; 1405 break; 1406 } 1407 return ret; 1408 } 1409 1410 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr) 1411 { 1412 struct kvm_s390_vm_cpu_processor *proc; 1413 u16 lowest_ibc, unblocked_ibc; 1414 int ret = 0; 1415 1416 mutex_lock(&kvm->lock); 1417 if (kvm->created_vcpus) { 1418 ret = -EBUSY; 1419 goto out; 1420 } 1421 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT); 1422 if (!proc) { 1423 ret = -ENOMEM; 1424 goto out; 1425 } 1426 if (!copy_from_user(proc, (void __user *)attr->addr, 1427 sizeof(*proc))) { 1428 kvm->arch.model.cpuid = proc->cpuid; 1429 lowest_ibc = sclp.ibc >> 16 & 0xfff; 1430 unblocked_ibc = sclp.ibc & 0xfff; 1431 if (lowest_ibc && proc->ibc) { 1432 if (proc->ibc > unblocked_ibc) 1433 kvm->arch.model.ibc = unblocked_ibc; 1434 else if (proc->ibc < lowest_ibc) 1435 kvm->arch.model.ibc = lowest_ibc; 1436 else 1437 kvm->arch.model.ibc = proc->ibc; 1438 } 1439 memcpy(kvm->arch.model.fac_list, proc->fac_list, 1440 S390_ARCH_FAC_LIST_SIZE_BYTE); 1441 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx", 1442 kvm->arch.model.ibc, 1443 kvm->arch.model.cpuid); 1444 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx", 1445 kvm->arch.model.fac_list[0], 1446 kvm->arch.model.fac_list[1], 1447 kvm->arch.model.fac_list[2]); 1448 } else 1449 ret = -EFAULT; 1450 kfree(proc); 1451 out: 1452 mutex_unlock(&kvm->lock); 1453 return ret; 1454 } 1455 1456 static int kvm_s390_set_processor_feat(struct kvm *kvm, 1457 struct kvm_device_attr *attr) 1458 { 1459 struct kvm_s390_vm_cpu_feat data; 1460 1461 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data))) 1462 return -EFAULT; 1463 if (!bitmap_subset((unsigned long *) data.feat, 1464 kvm_s390_available_cpu_feat, 1465 KVM_S390_VM_CPU_FEAT_NR_BITS)) 1466 return -EINVAL; 1467 1468 mutex_lock(&kvm->lock); 1469 if (kvm->created_vcpus) { 1470 mutex_unlock(&kvm->lock); 1471 return -EBUSY; 1472 } 1473 bitmap_from_arr64(kvm->arch.cpu_feat, data.feat, KVM_S390_VM_CPU_FEAT_NR_BITS); 1474 mutex_unlock(&kvm->lock); 1475 VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx", 1476 data.feat[0], 1477 data.feat[1], 1478 data.feat[2]); 1479 return 0; 1480 } 1481 1482 static int kvm_s390_set_processor_subfunc(struct kvm *kvm, 1483 struct kvm_device_attr *attr) 1484 { 1485 mutex_lock(&kvm->lock); 1486 if (kvm->created_vcpus) { 1487 mutex_unlock(&kvm->lock); 1488 return -EBUSY; 1489 } 1490 1491 if (copy_from_user(&kvm->arch.model.subfuncs, (void __user *)attr->addr, 1492 sizeof(struct kvm_s390_vm_cpu_subfunc))) { 1493 mutex_unlock(&kvm->lock); 1494 return -EFAULT; 1495 } 1496 mutex_unlock(&kvm->lock); 1497 1498 VM_EVENT(kvm, 3, "SET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1499 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0], 1500 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1], 1501 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2], 1502 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]); 1503 VM_EVENT(kvm, 3, "SET: guest PTFF subfunc 0x%16.16lx.%16.16lx", 1504 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0], 1505 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]); 1506 VM_EVENT(kvm, 3, "SET: guest KMAC subfunc 0x%16.16lx.%16.16lx", 1507 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0], 1508 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]); 1509 VM_EVENT(kvm, 3, "SET: guest KMC subfunc 0x%16.16lx.%16.16lx", 1510 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0], 1511 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]); 1512 VM_EVENT(kvm, 3, "SET: guest KM subfunc 0x%16.16lx.%16.16lx", 1513 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0], 1514 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]); 1515 VM_EVENT(kvm, 3, "SET: guest KIMD subfunc 0x%16.16lx.%16.16lx", 1516 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0], 1517 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]); 1518 VM_EVENT(kvm, 3, "SET: guest KLMD subfunc 0x%16.16lx.%16.16lx", 1519 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0], 1520 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]); 1521 VM_EVENT(kvm, 3, "SET: guest PCKMO subfunc 0x%16.16lx.%16.16lx", 1522 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0], 1523 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]); 1524 VM_EVENT(kvm, 3, "SET: guest KMCTR subfunc 0x%16.16lx.%16.16lx", 1525 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0], 1526 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]); 1527 VM_EVENT(kvm, 3, "SET: guest KMF subfunc 0x%16.16lx.%16.16lx", 1528 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0], 1529 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]); 1530 VM_EVENT(kvm, 3, "SET: guest KMO subfunc 0x%16.16lx.%16.16lx", 1531 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0], 1532 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]); 1533 VM_EVENT(kvm, 3, "SET: guest PCC subfunc 0x%16.16lx.%16.16lx", 1534 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0], 1535 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]); 1536 VM_EVENT(kvm, 3, "SET: guest PPNO subfunc 0x%16.16lx.%16.16lx", 1537 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0], 1538 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]); 1539 VM_EVENT(kvm, 3, "SET: guest KMA subfunc 0x%16.16lx.%16.16lx", 1540 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0], 1541 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]); 1542 VM_EVENT(kvm, 3, "SET: guest KDSA subfunc 0x%16.16lx.%16.16lx", 1543 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0], 1544 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]); 1545 VM_EVENT(kvm, 3, "SET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1546 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0], 1547 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1], 1548 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2], 1549 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]); 1550 VM_EVENT(kvm, 3, "SET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1551 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0], 1552 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1], 1553 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2], 1554 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]); 1555 VM_EVENT(kvm, 3, "GET: guest PFCR subfunc 0x%16.16lx.%16.16lx", 1556 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[0], 1557 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[1]); 1558 1559 return 0; 1560 } 1561 1562 #define KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK \ 1563 ( \ 1564 ((struct kvm_s390_vm_cpu_uv_feat){ \ 1565 .ap = 1, \ 1566 .ap_intr = 1, \ 1567 }) \ 1568 .feat \ 1569 ) 1570 1571 static int kvm_s390_set_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr) 1572 { 1573 struct kvm_s390_vm_cpu_uv_feat __user *ptr = (void __user *)attr->addr; 1574 unsigned long data, filter; 1575 1576 filter = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK; 1577 if (get_user(data, &ptr->feat)) 1578 return -EFAULT; 1579 if (!bitmap_subset(&data, &filter, KVM_S390_VM_CPU_UV_FEAT_NR_BITS)) 1580 return -EINVAL; 1581 1582 mutex_lock(&kvm->lock); 1583 if (kvm->created_vcpus) { 1584 mutex_unlock(&kvm->lock); 1585 return -EBUSY; 1586 } 1587 kvm->arch.model.uv_feat_guest.feat = data; 1588 mutex_unlock(&kvm->lock); 1589 1590 VM_EVENT(kvm, 3, "SET: guest UV-feat: 0x%16.16lx", data); 1591 1592 return 0; 1593 } 1594 1595 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr) 1596 { 1597 int ret = -ENXIO; 1598 1599 switch (attr->attr) { 1600 case KVM_S390_VM_CPU_PROCESSOR: 1601 ret = kvm_s390_set_processor(kvm, attr); 1602 break; 1603 case KVM_S390_VM_CPU_PROCESSOR_FEAT: 1604 ret = kvm_s390_set_processor_feat(kvm, attr); 1605 break; 1606 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC: 1607 ret = kvm_s390_set_processor_subfunc(kvm, attr); 1608 break; 1609 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST: 1610 ret = kvm_s390_set_uv_feat(kvm, attr); 1611 break; 1612 } 1613 return ret; 1614 } 1615 1616 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr) 1617 { 1618 struct kvm_s390_vm_cpu_processor *proc; 1619 int ret = 0; 1620 1621 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT); 1622 if (!proc) { 1623 ret = -ENOMEM; 1624 goto out; 1625 } 1626 proc->cpuid = kvm->arch.model.cpuid; 1627 proc->ibc = kvm->arch.model.ibc; 1628 memcpy(&proc->fac_list, kvm->arch.model.fac_list, 1629 S390_ARCH_FAC_LIST_SIZE_BYTE); 1630 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx", 1631 kvm->arch.model.ibc, 1632 kvm->arch.model.cpuid); 1633 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx", 1634 kvm->arch.model.fac_list[0], 1635 kvm->arch.model.fac_list[1], 1636 kvm->arch.model.fac_list[2]); 1637 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc))) 1638 ret = -EFAULT; 1639 kfree(proc); 1640 out: 1641 return ret; 1642 } 1643 1644 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr) 1645 { 1646 struct kvm_s390_vm_cpu_machine *mach; 1647 int ret = 0; 1648 1649 mach = kzalloc(sizeof(*mach), GFP_KERNEL_ACCOUNT); 1650 if (!mach) { 1651 ret = -ENOMEM; 1652 goto out; 1653 } 1654 get_cpu_id((struct cpuid *) &mach->cpuid); 1655 mach->ibc = sclp.ibc; 1656 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask, 1657 S390_ARCH_FAC_LIST_SIZE_BYTE); 1658 memcpy((unsigned long *)&mach->fac_list, stfle_fac_list, 1659 sizeof(stfle_fac_list)); 1660 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx", 1661 kvm->arch.model.ibc, 1662 kvm->arch.model.cpuid); 1663 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx", 1664 mach->fac_mask[0], 1665 mach->fac_mask[1], 1666 mach->fac_mask[2]); 1667 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx", 1668 mach->fac_list[0], 1669 mach->fac_list[1], 1670 mach->fac_list[2]); 1671 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach))) 1672 ret = -EFAULT; 1673 kfree(mach); 1674 out: 1675 return ret; 1676 } 1677 1678 static int kvm_s390_get_processor_feat(struct kvm *kvm, 1679 struct kvm_device_attr *attr) 1680 { 1681 struct kvm_s390_vm_cpu_feat data; 1682 1683 bitmap_to_arr64(data.feat, kvm->arch.cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS); 1684 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data))) 1685 return -EFAULT; 1686 VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx", 1687 data.feat[0], 1688 data.feat[1], 1689 data.feat[2]); 1690 return 0; 1691 } 1692 1693 static int kvm_s390_get_machine_feat(struct kvm *kvm, 1694 struct kvm_device_attr *attr) 1695 { 1696 struct kvm_s390_vm_cpu_feat data; 1697 1698 bitmap_to_arr64(data.feat, kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS); 1699 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data))) 1700 return -EFAULT; 1701 VM_EVENT(kvm, 3, "GET: host feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx", 1702 data.feat[0], 1703 data.feat[1], 1704 data.feat[2]); 1705 return 0; 1706 } 1707 1708 static int kvm_s390_get_processor_subfunc(struct kvm *kvm, 1709 struct kvm_device_attr *attr) 1710 { 1711 if (copy_to_user((void __user *)attr->addr, &kvm->arch.model.subfuncs, 1712 sizeof(struct kvm_s390_vm_cpu_subfunc))) 1713 return -EFAULT; 1714 1715 VM_EVENT(kvm, 3, "GET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1716 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0], 1717 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1], 1718 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2], 1719 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]); 1720 VM_EVENT(kvm, 3, "GET: guest PTFF subfunc 0x%16.16lx.%16.16lx", 1721 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0], 1722 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]); 1723 VM_EVENT(kvm, 3, "GET: guest KMAC subfunc 0x%16.16lx.%16.16lx", 1724 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0], 1725 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]); 1726 VM_EVENT(kvm, 3, "GET: guest KMC subfunc 0x%16.16lx.%16.16lx", 1727 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0], 1728 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]); 1729 VM_EVENT(kvm, 3, "GET: guest KM subfunc 0x%16.16lx.%16.16lx", 1730 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0], 1731 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]); 1732 VM_EVENT(kvm, 3, "GET: guest KIMD subfunc 0x%16.16lx.%16.16lx", 1733 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0], 1734 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]); 1735 VM_EVENT(kvm, 3, "GET: guest KLMD subfunc 0x%16.16lx.%16.16lx", 1736 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0], 1737 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]); 1738 VM_EVENT(kvm, 3, "GET: guest PCKMO subfunc 0x%16.16lx.%16.16lx", 1739 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0], 1740 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]); 1741 VM_EVENT(kvm, 3, "GET: guest KMCTR subfunc 0x%16.16lx.%16.16lx", 1742 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0], 1743 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]); 1744 VM_EVENT(kvm, 3, "GET: guest KMF subfunc 0x%16.16lx.%16.16lx", 1745 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0], 1746 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]); 1747 VM_EVENT(kvm, 3, "GET: guest KMO subfunc 0x%16.16lx.%16.16lx", 1748 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0], 1749 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]); 1750 VM_EVENT(kvm, 3, "GET: guest PCC subfunc 0x%16.16lx.%16.16lx", 1751 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0], 1752 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]); 1753 VM_EVENT(kvm, 3, "GET: guest PPNO subfunc 0x%16.16lx.%16.16lx", 1754 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0], 1755 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]); 1756 VM_EVENT(kvm, 3, "GET: guest KMA subfunc 0x%16.16lx.%16.16lx", 1757 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0], 1758 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]); 1759 VM_EVENT(kvm, 3, "GET: guest KDSA subfunc 0x%16.16lx.%16.16lx", 1760 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0], 1761 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]); 1762 VM_EVENT(kvm, 3, "GET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1763 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0], 1764 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1], 1765 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2], 1766 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]); 1767 VM_EVENT(kvm, 3, "GET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1768 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0], 1769 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1], 1770 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2], 1771 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]); 1772 VM_EVENT(kvm, 3, "GET: guest PFCR subfunc 0x%16.16lx.%16.16lx", 1773 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[0], 1774 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[1]); 1775 1776 return 0; 1777 } 1778 1779 static int kvm_s390_get_machine_subfunc(struct kvm *kvm, 1780 struct kvm_device_attr *attr) 1781 { 1782 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc, 1783 sizeof(struct kvm_s390_vm_cpu_subfunc))) 1784 return -EFAULT; 1785 1786 VM_EVENT(kvm, 3, "GET: host PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1787 ((unsigned long *) &kvm_s390_available_subfunc.plo)[0], 1788 ((unsigned long *) &kvm_s390_available_subfunc.plo)[1], 1789 ((unsigned long *) &kvm_s390_available_subfunc.plo)[2], 1790 ((unsigned long *) &kvm_s390_available_subfunc.plo)[3]); 1791 VM_EVENT(kvm, 3, "GET: host PTFF subfunc 0x%16.16lx.%16.16lx", 1792 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[0], 1793 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[1]); 1794 VM_EVENT(kvm, 3, "GET: host KMAC subfunc 0x%16.16lx.%16.16lx", 1795 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[0], 1796 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[1]); 1797 VM_EVENT(kvm, 3, "GET: host KMC subfunc 0x%16.16lx.%16.16lx", 1798 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[0], 1799 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[1]); 1800 VM_EVENT(kvm, 3, "GET: host KM subfunc 0x%16.16lx.%16.16lx", 1801 ((unsigned long *) &kvm_s390_available_subfunc.km)[0], 1802 ((unsigned long *) &kvm_s390_available_subfunc.km)[1]); 1803 VM_EVENT(kvm, 3, "GET: host KIMD subfunc 0x%16.16lx.%16.16lx", 1804 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[0], 1805 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[1]); 1806 VM_EVENT(kvm, 3, "GET: host KLMD subfunc 0x%16.16lx.%16.16lx", 1807 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[0], 1808 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[1]); 1809 VM_EVENT(kvm, 3, "GET: host PCKMO subfunc 0x%16.16lx.%16.16lx", 1810 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[0], 1811 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[1]); 1812 VM_EVENT(kvm, 3, "GET: host KMCTR subfunc 0x%16.16lx.%16.16lx", 1813 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[0], 1814 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[1]); 1815 VM_EVENT(kvm, 3, "GET: host KMF subfunc 0x%16.16lx.%16.16lx", 1816 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[0], 1817 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[1]); 1818 VM_EVENT(kvm, 3, "GET: host KMO subfunc 0x%16.16lx.%16.16lx", 1819 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[0], 1820 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[1]); 1821 VM_EVENT(kvm, 3, "GET: host PCC subfunc 0x%16.16lx.%16.16lx", 1822 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[0], 1823 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[1]); 1824 VM_EVENT(kvm, 3, "GET: host PPNO subfunc 0x%16.16lx.%16.16lx", 1825 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[0], 1826 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[1]); 1827 VM_EVENT(kvm, 3, "GET: host KMA subfunc 0x%16.16lx.%16.16lx", 1828 ((unsigned long *) &kvm_s390_available_subfunc.kma)[0], 1829 ((unsigned long *) &kvm_s390_available_subfunc.kma)[1]); 1830 VM_EVENT(kvm, 3, "GET: host KDSA subfunc 0x%16.16lx.%16.16lx", 1831 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[0], 1832 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[1]); 1833 VM_EVENT(kvm, 3, "GET: host SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1834 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[0], 1835 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[1], 1836 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[2], 1837 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[3]); 1838 VM_EVENT(kvm, 3, "GET: host DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx", 1839 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[0], 1840 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[1], 1841 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[2], 1842 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[3]); 1843 VM_EVENT(kvm, 3, "GET: host PFCR subfunc 0x%16.16lx.%16.16lx", 1844 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[0], 1845 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[1]); 1846 1847 return 0; 1848 } 1849 1850 static int kvm_s390_get_processor_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr) 1851 { 1852 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr; 1853 unsigned long feat = kvm->arch.model.uv_feat_guest.feat; 1854 1855 if (put_user(feat, &dst->feat)) 1856 return -EFAULT; 1857 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat); 1858 1859 return 0; 1860 } 1861 1862 static int kvm_s390_get_machine_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr) 1863 { 1864 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr; 1865 unsigned long feat; 1866 1867 BUILD_BUG_ON(sizeof(*dst) != sizeof(uv_info.uv_feature_indications)); 1868 1869 feat = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK; 1870 if (put_user(feat, &dst->feat)) 1871 return -EFAULT; 1872 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat); 1873 1874 return 0; 1875 } 1876 1877 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr) 1878 { 1879 int ret = -ENXIO; 1880 1881 switch (attr->attr) { 1882 case KVM_S390_VM_CPU_PROCESSOR: 1883 ret = kvm_s390_get_processor(kvm, attr); 1884 break; 1885 case KVM_S390_VM_CPU_MACHINE: 1886 ret = kvm_s390_get_machine(kvm, attr); 1887 break; 1888 case KVM_S390_VM_CPU_PROCESSOR_FEAT: 1889 ret = kvm_s390_get_processor_feat(kvm, attr); 1890 break; 1891 case KVM_S390_VM_CPU_MACHINE_FEAT: 1892 ret = kvm_s390_get_machine_feat(kvm, attr); 1893 break; 1894 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC: 1895 ret = kvm_s390_get_processor_subfunc(kvm, attr); 1896 break; 1897 case KVM_S390_VM_CPU_MACHINE_SUBFUNC: 1898 ret = kvm_s390_get_machine_subfunc(kvm, attr); 1899 break; 1900 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST: 1901 ret = kvm_s390_get_processor_uv_feat(kvm, attr); 1902 break; 1903 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST: 1904 ret = kvm_s390_get_machine_uv_feat(kvm, attr); 1905 break; 1906 } 1907 return ret; 1908 } 1909 1910 /** 1911 * kvm_s390_update_topology_change_report - update CPU topology change report 1912 * @kvm: guest KVM description 1913 * @val: set or clear the MTCR bit 1914 * 1915 * Updates the Multiprocessor Topology-Change-Report bit to signal 1916 * the guest with a topology change. 1917 * This is only relevant if the topology facility is present. 1918 */ 1919 static void kvm_s390_update_topology_change_report(struct kvm *kvm, bool val) 1920 { 1921 union sca_utility new, old; 1922 struct esca_block *sca; 1923 1924 sca = kvm->arch.sca; 1925 old = READ_ONCE(sca->utility); 1926 do { 1927 new = old; 1928 new.mtcr = val; 1929 } while (!try_cmpxchg(&sca->utility.val, &old.val, new.val)); 1930 } 1931 1932 static int kvm_s390_set_topo_change_indication(struct kvm *kvm, 1933 struct kvm_device_attr *attr) 1934 { 1935 if (!test_kvm_facility(kvm, 11)) 1936 return -ENXIO; 1937 1938 kvm_s390_update_topology_change_report(kvm, !!attr->attr); 1939 return 0; 1940 } 1941 1942 static int kvm_s390_get_topo_change_indication(struct kvm *kvm, 1943 struct kvm_device_attr *attr) 1944 { 1945 u8 topo; 1946 1947 if (!test_kvm_facility(kvm, 11)) 1948 return -ENXIO; 1949 1950 topo = kvm->arch.sca->utility.mtcr; 1951 1952 return put_user(topo, (u8 __user *)attr->addr); 1953 } 1954 1955 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr) 1956 { 1957 int ret; 1958 1959 switch (attr->group) { 1960 case KVM_S390_VM_MEM_CTRL: 1961 ret = kvm_s390_set_mem_control(kvm, attr); 1962 break; 1963 case KVM_S390_VM_TOD: 1964 ret = kvm_s390_set_tod(kvm, attr); 1965 break; 1966 case KVM_S390_VM_CPU_MODEL: 1967 ret = kvm_s390_set_cpu_model(kvm, attr); 1968 break; 1969 case KVM_S390_VM_CRYPTO: 1970 ret = kvm_s390_vm_set_crypto(kvm, attr); 1971 break; 1972 case KVM_S390_VM_MIGRATION: 1973 ret = kvm_s390_vm_set_migration(kvm, attr); 1974 break; 1975 case KVM_S390_VM_CPU_TOPOLOGY: 1976 ret = kvm_s390_set_topo_change_indication(kvm, attr); 1977 break; 1978 default: 1979 ret = -ENXIO; 1980 break; 1981 } 1982 1983 return ret; 1984 } 1985 1986 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr) 1987 { 1988 int ret; 1989 1990 switch (attr->group) { 1991 case KVM_S390_VM_MEM_CTRL: 1992 ret = kvm_s390_get_mem_control(kvm, attr); 1993 break; 1994 case KVM_S390_VM_TOD: 1995 ret = kvm_s390_get_tod(kvm, attr); 1996 break; 1997 case KVM_S390_VM_CPU_MODEL: 1998 ret = kvm_s390_get_cpu_model(kvm, attr); 1999 break; 2000 case KVM_S390_VM_MIGRATION: 2001 ret = kvm_s390_vm_get_migration(kvm, attr); 2002 break; 2003 case KVM_S390_VM_CPU_TOPOLOGY: 2004 ret = kvm_s390_get_topo_change_indication(kvm, attr); 2005 break; 2006 default: 2007 ret = -ENXIO; 2008 break; 2009 } 2010 2011 return ret; 2012 } 2013 2014 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr) 2015 { 2016 int ret; 2017 2018 switch (attr->group) { 2019 case KVM_S390_VM_MEM_CTRL: 2020 switch (attr->attr) { 2021 case KVM_S390_VM_MEM_ENABLE_CMMA: 2022 case KVM_S390_VM_MEM_CLR_CMMA: 2023 ret = sclp.has_cmma ? 0 : -ENXIO; 2024 break; 2025 case KVM_S390_VM_MEM_LIMIT_SIZE: 2026 ret = 0; 2027 break; 2028 default: 2029 ret = -ENXIO; 2030 break; 2031 } 2032 break; 2033 case KVM_S390_VM_TOD: 2034 switch (attr->attr) { 2035 case KVM_S390_VM_TOD_LOW: 2036 case KVM_S390_VM_TOD_HIGH: 2037 ret = 0; 2038 break; 2039 default: 2040 ret = -ENXIO; 2041 break; 2042 } 2043 break; 2044 case KVM_S390_VM_CPU_MODEL: 2045 switch (attr->attr) { 2046 case KVM_S390_VM_CPU_PROCESSOR: 2047 case KVM_S390_VM_CPU_MACHINE: 2048 case KVM_S390_VM_CPU_PROCESSOR_FEAT: 2049 case KVM_S390_VM_CPU_MACHINE_FEAT: 2050 case KVM_S390_VM_CPU_MACHINE_SUBFUNC: 2051 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC: 2052 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST: 2053 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST: 2054 ret = 0; 2055 break; 2056 default: 2057 ret = -ENXIO; 2058 break; 2059 } 2060 break; 2061 case KVM_S390_VM_CRYPTO: 2062 switch (attr->attr) { 2063 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW: 2064 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW: 2065 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW: 2066 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW: 2067 ret = 0; 2068 break; 2069 case KVM_S390_VM_CRYPTO_ENABLE_APIE: 2070 case KVM_S390_VM_CRYPTO_DISABLE_APIE: 2071 ret = ap_instructions_available() ? 0 : -ENXIO; 2072 break; 2073 default: 2074 ret = -ENXIO; 2075 break; 2076 } 2077 break; 2078 case KVM_S390_VM_MIGRATION: 2079 ret = 0; 2080 break; 2081 case KVM_S390_VM_CPU_TOPOLOGY: 2082 ret = test_kvm_facility(kvm, 11) ? 0 : -ENXIO; 2083 break; 2084 default: 2085 ret = -ENXIO; 2086 break; 2087 } 2088 2089 return ret; 2090 } 2091 2092 static int kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args) 2093 { 2094 union skey *keys; 2095 int i, r = 0; 2096 2097 if (args->flags != 0) 2098 return -EINVAL; 2099 2100 /* Is this guest using storage keys? */ 2101 if (!uses_skeys(kvm->arch.gmap)) 2102 return KVM_S390_GET_SKEYS_NONE; 2103 2104 /* Enforce sane limit on memory allocation */ 2105 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX) 2106 return -EINVAL; 2107 2108 keys = kvmalloc_array(args->count, sizeof(*keys), GFP_KERNEL_ACCOUNT); 2109 if (!keys) 2110 return -ENOMEM; 2111 2112 scoped_guard(read_lock, &kvm->mmu_lock) { 2113 for (i = 0; i < args->count; i++) { 2114 r = dat_get_storage_key(kvm->arch.gmap->asce, 2115 args->start_gfn + i, keys + i); 2116 if (r) 2117 break; 2118 } 2119 } 2120 2121 if (!r) { 2122 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys, 2123 sizeof(uint8_t) * args->count); 2124 if (r) 2125 r = -EFAULT; 2126 } 2127 2128 kvfree(keys); 2129 return r; 2130 } 2131 2132 static int kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args) 2133 { 2134 struct kvm_s390_mmu_cache *mc; 2135 union skey *keys; 2136 int i, r = 0; 2137 2138 if (args->flags != 0) 2139 return -EINVAL; 2140 2141 /* Enforce sane limit on memory allocation */ 2142 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX) 2143 return -EINVAL; 2144 2145 keys = kvmalloc_array(args->count, sizeof(*keys), GFP_KERNEL_ACCOUNT); 2146 if (!keys) 2147 return -ENOMEM; 2148 2149 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr, 2150 sizeof(uint8_t) * args->count); 2151 if (r) { 2152 r = -EFAULT; 2153 goto out; 2154 } 2155 2156 /* Enable storage key handling for the guest */ 2157 r = gmap_enable_skeys(kvm->arch.gmap); 2158 if (r) 2159 goto out; 2160 2161 r = -EINVAL; 2162 for (i = 0; i < args->count; i++) { 2163 /* Lowest order bit is reserved */ 2164 if (keys[i].zero) 2165 goto out; 2166 } 2167 2168 mc = kvm_s390_new_mmu_cache(); 2169 if (!mc) { 2170 r = -ENOMEM; 2171 goto out; 2172 } 2173 2174 r = 0; 2175 do { 2176 r = kvm_s390_mmu_cache_topup(mc); 2177 if (r == -ENOMEM) 2178 break; 2179 scoped_guard(read_lock, &kvm->mmu_lock) { 2180 for (i = 0 ; i < args->count; i++) { 2181 r = dat_set_storage_key(mc, kvm->arch.gmap->asce, 2182 args->start_gfn + i, keys[i], 0); 2183 if (r) 2184 break; 2185 } 2186 } 2187 } while (r == -ENOMEM); 2188 kvm_s390_free_mmu_cache(mc); 2189 out: 2190 kvfree(keys); 2191 return r; 2192 } 2193 2194 /* 2195 * This function searches for the next page with dirty CMMA attributes, and 2196 * saves the attributes in the buffer up to either the end of the buffer or 2197 * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found; 2198 * no trailing clean bytes are saved. 2199 * In case no dirty bits were found, or if CMMA was not enabled or used, the 2200 * output buffer will indicate 0 as length. 2201 */ 2202 static int kvm_s390_get_cmma_bits(struct kvm *kvm, 2203 struct kvm_s390_cmma_log *args) 2204 { 2205 int peek, ret; 2206 u8 *values; 2207 2208 if (!kvm->arch.use_cmma) 2209 return -ENXIO; 2210 /* Invalid/unsupported flags were specified */ 2211 if (args->flags & ~KVM_S390_CMMA_PEEK) 2212 return -EINVAL; 2213 /* Migration mode query, and we are not doing a migration */ 2214 peek = !!(args->flags & KVM_S390_CMMA_PEEK); 2215 if (!peek && !kvm->arch.migration_mode) 2216 return -EINVAL; 2217 /* CMMA is disabled or was not used, or the buffer has length zero */ 2218 args->count = min(args->count, KVM_S390_CMMA_SIZE_MAX); 2219 if (!args->count || !uses_cmm(kvm->arch.gmap)) { 2220 memset(args, 0, sizeof(*args)); 2221 return 0; 2222 } 2223 /* We are not peeking, and there are no dirty pages */ 2224 if (!peek && !atomic64_read(&kvm->arch.cmma_dirty_pages)) { 2225 memset(args, 0, sizeof(*args)); 2226 return 0; 2227 } 2228 2229 values = vmalloc(args->count); 2230 if (!values) 2231 return -ENOMEM; 2232 2233 scoped_guard(read_lock, &kvm->mmu_lock) { 2234 if (peek) 2235 ret = dat_peek_cmma(args->start_gfn, kvm->arch.gmap->asce, &args->count, 2236 values); 2237 else 2238 ret = dat_get_cmma(kvm->arch.gmap->asce, &args->start_gfn, &args->count, 2239 values, &kvm->arch.cmma_dirty_pages); 2240 } 2241 2242 if (kvm->arch.migration_mode) 2243 args->remaining = atomic64_read(&kvm->arch.cmma_dirty_pages); 2244 else 2245 args->remaining = 0; 2246 2247 if (copy_to_user((void __user *)args->values, values, args->count)) 2248 ret = -EFAULT; 2249 2250 vfree(values); 2251 return ret; 2252 } 2253 2254 /* 2255 * This function sets the CMMA attributes for the given pages. If the input 2256 * buffer has zero length, no action is taken, otherwise the attributes are 2257 * set and the mm->context.uses_cmm flag is set. 2258 */ 2259 static int kvm_s390_set_cmma_bits(struct kvm *kvm, 2260 const struct kvm_s390_cmma_log *args) 2261 { 2262 struct kvm_s390_mmu_cache *mc; 2263 u8 *bits = NULL; 2264 int r = 0; 2265 2266 if (!kvm->arch.use_cmma) 2267 return -ENXIO; 2268 /* invalid/unsupported flags */ 2269 if (args->flags != 0) 2270 return -EINVAL; 2271 /* Enforce sane limit on memory allocation */ 2272 if (args->count > KVM_S390_CMMA_SIZE_MAX) 2273 return -EINVAL; 2274 /* Nothing to do */ 2275 if (args->count == 0) 2276 return 0; 2277 2278 mc = kvm_s390_new_mmu_cache(); 2279 if (!mc) 2280 return -ENOMEM; 2281 bits = vmalloc(array_size(sizeof(*bits), args->count)); 2282 if (!bits) 2283 goto out; 2284 2285 r = copy_from_user(bits, (void __user *)args->values, args->count); 2286 if (r) { 2287 r = -EFAULT; 2288 goto out; 2289 } 2290 2291 do { 2292 r = kvm_s390_mmu_cache_topup(mc); 2293 if (r) 2294 break; 2295 scoped_guard(read_lock, &kvm->mmu_lock) { 2296 r = dat_set_cmma_bits(mc, kvm->arch.gmap->asce, args->start_gfn, 2297 args->count, args->mask, bits); 2298 } 2299 } while (r == -ENOMEM); 2300 2301 set_bit(GMAP_FLAG_USES_CMM, &kvm->arch.gmap->flags); 2302 out: 2303 kvm_s390_free_mmu_cache(mc); 2304 vfree(bits); 2305 return r; 2306 } 2307 2308 /** 2309 * kvm_s390_cpus_from_pv - Convert all protected vCPUs in a protected VM to 2310 * non protected. 2311 * @kvm: the VM whose protected vCPUs are to be converted 2312 * @rc: return value for the RC field of the UVC (in case of error) 2313 * @rrc: return value for the RRC field of the UVC (in case of error) 2314 * 2315 * Does not stop in case of error, tries to convert as many 2316 * CPUs as possible. In case of error, the RC and RRC of the last error are 2317 * returned. 2318 * 2319 * Return: 0 in case of success, otherwise -EIO 2320 */ 2321 int kvm_s390_cpus_from_pv(struct kvm *kvm, u16 *rc, u16 *rrc) 2322 { 2323 struct kvm_vcpu *vcpu; 2324 unsigned long i; 2325 u16 _rc, _rrc; 2326 int ret = 0; 2327 2328 /* 2329 * We ignore failures and try to destroy as many CPUs as possible. 2330 * At the same time we must not free the assigned resources when 2331 * this fails, as the ultravisor has still access to that memory. 2332 * So kvm_s390_pv_destroy_cpu can leave a "wanted" memory leak 2333 * behind. 2334 * We want to return the first failure rc and rrc, though. 2335 */ 2336 kvm_for_each_vcpu(i, vcpu, kvm) { 2337 mutex_lock(&vcpu->mutex); 2338 if (kvm_s390_pv_destroy_cpu(vcpu, &_rc, &_rrc) && !ret) { 2339 *rc = _rc; 2340 *rrc = _rrc; 2341 ret = -EIO; 2342 } 2343 mutex_unlock(&vcpu->mutex); 2344 } 2345 /* Ensure that we re-enable gisa if the non-PV guest used it but the PV guest did not. */ 2346 if (use_gisa) 2347 kvm_s390_gisa_enable(kvm); 2348 return ret; 2349 } 2350 2351 /** 2352 * kvm_s390_cpus_to_pv - Convert all non-protected vCPUs in a protected VM 2353 * to protected. 2354 * @kvm: the VM whose protected vCPUs are to be converted 2355 * @rc: return value for the RC field of the UVC (in case of error) 2356 * @rrc: return value for the RRC field of the UVC (in case of error) 2357 * 2358 * Tries to undo the conversion in case of error. 2359 * 2360 * Return: 0 in case of success, otherwise -EIO 2361 */ 2362 static int kvm_s390_cpus_to_pv(struct kvm *kvm, u16 *rc, u16 *rrc) 2363 { 2364 unsigned long i; 2365 int r = 0; 2366 u16 dummy; 2367 2368 struct kvm_vcpu *vcpu; 2369 2370 /* Disable the GISA if the ultravisor does not support AIV. */ 2371 if (!uv_has_feature(BIT_UV_FEAT_AIV)) 2372 kvm_s390_gisa_disable(kvm); 2373 2374 kvm_for_each_vcpu(i, vcpu, kvm) { 2375 mutex_lock(&vcpu->mutex); 2376 r = kvm_s390_pv_create_cpu(vcpu, rc, rrc); 2377 mutex_unlock(&vcpu->mutex); 2378 if (r) 2379 break; 2380 } 2381 if (r) 2382 kvm_s390_cpus_from_pv(kvm, &dummy, &dummy); 2383 return r; 2384 } 2385 2386 /* 2387 * Here we provide user space with a direct interface to query UV 2388 * related data like UV maxima and available features as well as 2389 * feature specific data. 2390 * 2391 * To facilitate future extension of the data structures we'll try to 2392 * write data up to the maximum requested length. 2393 */ 2394 static ssize_t kvm_s390_handle_pv_info(struct kvm_s390_pv_info *info) 2395 { 2396 ssize_t len_min; 2397 2398 switch (info->header.id) { 2399 case KVM_PV_INFO_VM: { 2400 len_min = sizeof(info->header) + sizeof(info->vm); 2401 2402 if (info->header.len_max < len_min) 2403 return -EINVAL; 2404 2405 memcpy(info->vm.inst_calls_list, 2406 uv_info.inst_calls_list, 2407 sizeof(uv_info.inst_calls_list)); 2408 2409 /* It's max cpuid not max cpus, so it's off by one */ 2410 info->vm.max_cpus = uv_info.max_guest_cpu_id + 1; 2411 info->vm.max_guests = uv_info.max_num_sec_conf; 2412 info->vm.max_guest_addr = uv_info.max_sec_stor_addr; 2413 info->vm.feature_indication = uv_info.uv_feature_indications; 2414 2415 return len_min; 2416 } 2417 case KVM_PV_INFO_DUMP: { 2418 len_min = sizeof(info->header) + sizeof(info->dump); 2419 2420 if (info->header.len_max < len_min) 2421 return -EINVAL; 2422 2423 info->dump.dump_cpu_buffer_len = uv_info.guest_cpu_stor_len; 2424 info->dump.dump_config_mem_buffer_per_1m = uv_info.conf_dump_storage_state_len; 2425 info->dump.dump_config_finalize_len = uv_info.conf_dump_finalize_len; 2426 return len_min; 2427 } 2428 default: 2429 return -EINVAL; 2430 } 2431 } 2432 2433 static int kvm_s390_pv_dmp(struct kvm *kvm, struct kvm_pv_cmd *cmd, 2434 struct kvm_s390_pv_dmp dmp) 2435 { 2436 int r = -EINVAL; 2437 void __user *result_buff = (void __user *)dmp.buff_addr; 2438 2439 switch (dmp.subcmd) { 2440 case KVM_PV_DUMP_INIT: { 2441 if (kvm->arch.pv.dumping) 2442 break; 2443 2444 /* 2445 * Block SIE entry as concurrent dump UVCs could lead 2446 * to validities. 2447 */ 2448 kvm_s390_vcpu_block_all(kvm); 2449 2450 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm), 2451 UVC_CMD_DUMP_INIT, &cmd->rc, &cmd->rrc); 2452 KVM_UV_EVENT(kvm, 3, "PROTVIRT DUMP INIT: rc %x rrc %x", 2453 cmd->rc, cmd->rrc); 2454 if (!r) { 2455 kvm->arch.pv.dumping = true; 2456 } else { 2457 kvm_s390_vcpu_unblock_all(kvm); 2458 r = -EINVAL; 2459 } 2460 break; 2461 } 2462 case KVM_PV_DUMP_CONFIG_STOR_STATE: { 2463 if (!kvm->arch.pv.dumping) 2464 break; 2465 2466 /* 2467 * gaddr is an output parameter since we might stop 2468 * early. As dmp will be copied back in our caller, we 2469 * don't need to do it ourselves. 2470 */ 2471 r = kvm_s390_pv_dump_stor_state(kvm, result_buff, &dmp.gaddr, dmp.buff_len, 2472 &cmd->rc, &cmd->rrc); 2473 break; 2474 } 2475 case KVM_PV_DUMP_COMPLETE: { 2476 if (!kvm->arch.pv.dumping) 2477 break; 2478 2479 r = -EINVAL; 2480 if (dmp.buff_len < uv_info.conf_dump_finalize_len) 2481 break; 2482 2483 r = kvm_s390_pv_dump_complete(kvm, result_buff, 2484 &cmd->rc, &cmd->rrc); 2485 break; 2486 } 2487 default: 2488 r = -ENOTTY; 2489 break; 2490 } 2491 2492 return r; 2493 } 2494 2495 static int kvm_s390_handle_pv(struct kvm *kvm, struct kvm_pv_cmd *cmd) 2496 { 2497 const bool need_lock = (cmd->cmd != KVM_PV_ASYNC_CLEANUP_PERFORM); 2498 void __user *argp = (void __user *)cmd->data; 2499 int r = 0; 2500 u16 dummy; 2501 2502 if (need_lock) 2503 mutex_lock(&kvm->lock); 2504 2505 switch (cmd->cmd) { 2506 case KVM_PV_ENABLE: { 2507 r = -EINVAL; 2508 if (kvm_s390_pv_is_protected(kvm)) 2509 break; 2510 2511 mmap_write_lock(kvm->mm); 2512 /* 2513 * Disable creation of new THPs. Existing THPs can stay, they 2514 * will be split when any part of them gets imported. 2515 */ 2516 mm_flags_clear(MMF_DISABLE_THP_EXCEPT_ADVISED, kvm->mm); 2517 mm_flags_set(MMF_DISABLE_THP_COMPLETELY, kvm->mm); 2518 set_bit(GMAP_FLAG_EXPORT_ON_UNMAP, &kvm->arch.gmap->flags); 2519 r = gmap_helper_disable_cow_sharing(); 2520 mmap_write_unlock(kvm->mm); 2521 if (r) 2522 break; 2523 2524 r = kvm_s390_pv_init_vm(kvm, &cmd->rc, &cmd->rrc); 2525 if (r) 2526 break; 2527 2528 r = kvm_s390_cpus_to_pv(kvm, &cmd->rc, &cmd->rrc); 2529 if (r) 2530 kvm_s390_pv_deinit_vm(kvm, &dummy, &dummy); 2531 2532 /* we need to block service interrupts from now on */ 2533 set_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs); 2534 break; 2535 } 2536 case KVM_PV_ASYNC_CLEANUP_PREPARE: 2537 r = -EINVAL; 2538 if (!kvm_s390_pv_is_protected(kvm) || !async_destroy) 2539 break; 2540 2541 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc); 2542 /* 2543 * If a CPU could not be destroyed, destroy VM will also fail. 2544 * There is no point in trying to destroy it. Instead return 2545 * the rc and rrc from the first CPU that failed destroying. 2546 */ 2547 if (r) 2548 break; 2549 r = kvm_s390_pv_set_aside(kvm, &cmd->rc, &cmd->rrc); 2550 2551 /* no need to block service interrupts any more */ 2552 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs); 2553 break; 2554 case KVM_PV_ASYNC_CLEANUP_PERFORM: 2555 r = -EINVAL; 2556 if (!async_destroy) 2557 break; 2558 /* kvm->lock must not be held; this is asserted inside the function. */ 2559 r = kvm_s390_pv_deinit_aside_vm(kvm, &cmd->rc, &cmd->rrc); 2560 break; 2561 case KVM_PV_DISABLE: { 2562 r = -EINVAL; 2563 if (!kvm_s390_pv_is_protected(kvm)) 2564 break; 2565 2566 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc); 2567 /* 2568 * If a CPU could not be destroyed, destroy VM will also fail. 2569 * There is no point in trying to destroy it. Instead return 2570 * the rc and rrc from the first CPU that failed destroying. 2571 */ 2572 if (r) 2573 break; 2574 r = kvm_s390_pv_deinit_cleanup_all(kvm, &cmd->rc, &cmd->rrc); 2575 2576 /* no need to block service interrupts any more */ 2577 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs); 2578 break; 2579 } 2580 case KVM_PV_SET_SEC_PARMS: { 2581 struct kvm_s390_pv_sec_parm parms = {}; 2582 void *hdr; 2583 2584 r = -EINVAL; 2585 if (!kvm_s390_pv_is_protected(kvm)) 2586 break; 2587 2588 r = -EFAULT; 2589 if (copy_from_user(&parms, argp, sizeof(parms))) 2590 break; 2591 2592 /* Currently restricted to 1MiB */ 2593 r = -EINVAL; 2594 if (parms.length > SZ_1M) 2595 break; 2596 2597 r = -ENOMEM; 2598 hdr = vmalloc(parms.length); 2599 if (!hdr) 2600 break; 2601 2602 r = -EFAULT; 2603 if (!copy_from_user(hdr, (void __user *)parms.origin, 2604 parms.length)) 2605 r = kvm_s390_pv_set_sec_parms(kvm, hdr, parms.length, 2606 &cmd->rc, &cmd->rrc); 2607 2608 vfree(hdr); 2609 break; 2610 } 2611 case KVM_PV_UNPACK: { 2612 struct kvm_s390_pv_unp unp = {}; 2613 2614 r = -EINVAL; 2615 if (!kvm_s390_pv_is_protected(kvm) || !mm_is_protected(kvm->mm)) 2616 break; 2617 2618 r = -EFAULT; 2619 if (copy_from_user(&unp, argp, sizeof(unp))) 2620 break; 2621 2622 r = kvm_s390_pv_unpack(kvm, unp.addr, unp.size, unp.tweak, 2623 &cmd->rc, &cmd->rrc); 2624 break; 2625 } 2626 case KVM_PV_VERIFY: { 2627 r = -EINVAL; 2628 if (!kvm_s390_pv_is_protected(kvm)) 2629 break; 2630 2631 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm), 2632 UVC_CMD_VERIFY_IMG, &cmd->rc, &cmd->rrc); 2633 KVM_UV_EVENT(kvm, 3, "PROTVIRT VERIFY: rc %x rrc %x", cmd->rc, 2634 cmd->rrc); 2635 break; 2636 } 2637 case KVM_PV_PREP_RESET: { 2638 r = -EINVAL; 2639 if (!kvm_s390_pv_is_protected(kvm)) 2640 break; 2641 2642 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm), 2643 UVC_CMD_PREPARE_RESET, &cmd->rc, &cmd->rrc); 2644 KVM_UV_EVENT(kvm, 3, "PROTVIRT PREP RESET: rc %x rrc %x", 2645 cmd->rc, cmd->rrc); 2646 break; 2647 } 2648 case KVM_PV_UNSHARE_ALL: { 2649 r = -EINVAL; 2650 if (!kvm_s390_pv_is_protected(kvm)) 2651 break; 2652 2653 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm), 2654 UVC_CMD_SET_UNSHARE_ALL, &cmd->rc, &cmd->rrc); 2655 KVM_UV_EVENT(kvm, 3, "PROTVIRT UNSHARE: rc %x rrc %x", 2656 cmd->rc, cmd->rrc); 2657 break; 2658 } 2659 case KVM_PV_INFO: { 2660 struct kvm_s390_pv_info info = {}; 2661 ssize_t data_len; 2662 2663 /* 2664 * No need to check the VM protection here. 2665 * 2666 * Maybe user space wants to query some of the data 2667 * when the VM is still unprotected. If we see the 2668 * need to fence a new data command we can still 2669 * return an error in the info handler. 2670 */ 2671 2672 r = -EFAULT; 2673 if (copy_from_user(&info, argp, sizeof(info.header))) 2674 break; 2675 2676 r = -EINVAL; 2677 if (info.header.len_max < sizeof(info.header)) 2678 break; 2679 2680 data_len = kvm_s390_handle_pv_info(&info); 2681 if (data_len < 0) { 2682 r = data_len; 2683 break; 2684 } 2685 /* 2686 * If a data command struct is extended (multiple 2687 * times) this can be used to determine how much of it 2688 * is valid. 2689 */ 2690 info.header.len_written = data_len; 2691 2692 r = -EFAULT; 2693 if (copy_to_user(argp, &info, data_len)) 2694 break; 2695 2696 r = 0; 2697 break; 2698 } 2699 case KVM_PV_DUMP: { 2700 struct kvm_s390_pv_dmp dmp; 2701 2702 r = -EINVAL; 2703 if (!kvm_s390_pv_is_protected(kvm)) 2704 break; 2705 2706 r = -EFAULT; 2707 if (copy_from_user(&dmp, argp, sizeof(dmp))) 2708 break; 2709 2710 r = kvm_s390_pv_dmp(kvm, cmd, dmp); 2711 if (r) 2712 break; 2713 2714 if (copy_to_user(argp, &dmp, sizeof(dmp))) { 2715 r = -EFAULT; 2716 break; 2717 } 2718 2719 break; 2720 } 2721 default: 2722 r = -ENOTTY; 2723 } 2724 if (need_lock) 2725 mutex_unlock(&kvm->lock); 2726 2727 return r; 2728 } 2729 2730 static int mem_op_validate_common(struct kvm_s390_mem_op *mop, u64 supported_flags) 2731 { 2732 if (mop->flags & ~supported_flags || !mop->size) 2733 return -EINVAL; 2734 if (mop->size > MEM_OP_MAX_SIZE) 2735 return -E2BIG; 2736 if (mop->flags & KVM_S390_MEMOP_F_SKEY_PROTECTION) { 2737 if (mop->key > 0xf) 2738 return -EINVAL; 2739 } else { 2740 mop->key = 0; 2741 } 2742 return 0; 2743 } 2744 2745 static int kvm_s390_vm_mem_op_abs(struct kvm *kvm, struct kvm_s390_mem_op *mop) 2746 { 2747 void __user *uaddr = (void __user *)mop->buf; 2748 void *tmpbuf __free(kvfree) = NULL; 2749 enum gacc_mode acc_mode; 2750 int r; 2751 2752 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION | 2753 KVM_S390_MEMOP_F_CHECK_ONLY); 2754 if (r) 2755 return r; 2756 2757 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) { 2758 tmpbuf = vmalloc(mop->size); 2759 if (!tmpbuf) 2760 return -ENOMEM; 2761 } 2762 2763 acc_mode = mop->op == KVM_S390_MEMOP_ABSOLUTE_READ ? GACC_FETCH : GACC_STORE; 2764 2765 scoped_guard(srcu, &kvm->srcu) { 2766 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) 2767 return check_gpa_range(kvm, mop->gaddr, mop->size, acc_mode, mop->key); 2768 2769 if (acc_mode == GACC_STORE && copy_from_user(tmpbuf, uaddr, mop->size)) 2770 return -EFAULT; 2771 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf, 2772 mop->size, acc_mode, mop->key); 2773 if (r) 2774 return r; 2775 if (acc_mode != GACC_STORE && copy_to_user(uaddr, tmpbuf, mop->size)) 2776 return -EFAULT; 2777 } 2778 return 0; 2779 } 2780 2781 static int kvm_s390_vm_mem_op_cmpxchg(struct kvm *kvm, struct kvm_s390_mem_op *mop) 2782 { 2783 void __user *uaddr = (void __user *)mop->buf; 2784 void __user *old_addr = (void __user *)mop->old_addr; 2785 union kvm_s390_quad old = { .sixteen = 0 }; 2786 union kvm_s390_quad new = { .sixteen = 0 }; 2787 bool success = false; 2788 int r; 2789 2790 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION); 2791 if (r) 2792 return r; 2793 /* 2794 * This validates off_in_quad. Checking that size is a power 2795 * of two is not necessary, as cmpxchg_guest_abs_with_key 2796 * takes care of that 2797 */ 2798 if (mop->size > sizeof(new)) 2799 return -EINVAL; 2800 if (copy_from_user(&new, uaddr, mop->size)) 2801 return -EFAULT; 2802 if (copy_from_user(&old, old_addr, mop->size)) 2803 return -EFAULT; 2804 2805 scoped_guard(srcu, &kvm->srcu) { 2806 r = cmpxchg_guest_abs_with_key(kvm, mop->gaddr, mop->size, &old, new, 2807 mop->key, &success); 2808 2809 if (!success && copy_to_user(old_addr, &old, mop->size)) 2810 return -EFAULT; 2811 } 2812 return r; 2813 } 2814 2815 static int kvm_s390_vm_mem_op(struct kvm *kvm, struct kvm_s390_mem_op *mop) 2816 { 2817 /* 2818 * This is technically a heuristic only, if the kvm->lock is not 2819 * taken, it is not guaranteed that the vm is/remains non-protected. 2820 * This is ok from a kernel perspective, wrongdoing is detected 2821 * on the access, -EFAULT is returned and the vm may crash the 2822 * next time it accesses the memory in question. 2823 * There is no sane usecase to do switching and a memop on two 2824 * different CPUs at the same time. 2825 */ 2826 if (kvm_s390_pv_get_handle(kvm)) 2827 return -EINVAL; 2828 2829 switch (mop->op) { 2830 case KVM_S390_MEMOP_ABSOLUTE_READ: 2831 case KVM_S390_MEMOP_ABSOLUTE_WRITE: 2832 return kvm_s390_vm_mem_op_abs(kvm, mop); 2833 case KVM_S390_MEMOP_ABSOLUTE_CMPXCHG: 2834 return kvm_s390_vm_mem_op_cmpxchg(kvm, mop); 2835 default: 2836 return -EINVAL; 2837 } 2838 } 2839 2840 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg) 2841 { 2842 struct kvm *kvm = filp->private_data; 2843 void __user *argp = (void __user *)arg; 2844 struct kvm_device_attr attr; 2845 int r; 2846 2847 switch (ioctl) { 2848 case KVM_S390_INTERRUPT: { 2849 struct kvm_s390_interrupt s390int; 2850 2851 r = -EFAULT; 2852 if (copy_from_user(&s390int, argp, sizeof(s390int))) 2853 break; 2854 r = kvm_s390_inject_vm(kvm, &s390int); 2855 break; 2856 } 2857 case KVM_CREATE_IRQCHIP: { 2858 r = -EINVAL; 2859 if (kvm->arch.use_irqchip) 2860 r = 0; 2861 break; 2862 } 2863 case KVM_SET_DEVICE_ATTR: { 2864 r = -EFAULT; 2865 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) 2866 break; 2867 r = kvm_s390_vm_set_attr(kvm, &attr); 2868 break; 2869 } 2870 case KVM_GET_DEVICE_ATTR: { 2871 r = -EFAULT; 2872 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) 2873 break; 2874 r = kvm_s390_vm_get_attr(kvm, &attr); 2875 break; 2876 } 2877 case KVM_HAS_DEVICE_ATTR: { 2878 r = -EFAULT; 2879 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) 2880 break; 2881 r = kvm_s390_vm_has_attr(kvm, &attr); 2882 break; 2883 } 2884 case KVM_S390_GET_SKEYS: { 2885 struct kvm_s390_skeys args; 2886 2887 r = -EFAULT; 2888 if (copy_from_user(&args, argp, 2889 sizeof(struct kvm_s390_skeys))) 2890 break; 2891 r = kvm_s390_get_skeys(kvm, &args); 2892 break; 2893 } 2894 case KVM_S390_SET_SKEYS: { 2895 struct kvm_s390_skeys args; 2896 2897 r = -EFAULT; 2898 if (copy_from_user(&args, argp, 2899 sizeof(struct kvm_s390_skeys))) 2900 break; 2901 r = kvm_s390_set_skeys(kvm, &args); 2902 break; 2903 } 2904 case KVM_S390_GET_CMMA_BITS: { 2905 struct kvm_s390_cmma_log args; 2906 2907 r = -EFAULT; 2908 if (copy_from_user(&args, argp, sizeof(args))) 2909 break; 2910 mutex_lock(&kvm->slots_lock); 2911 r = kvm_s390_get_cmma_bits(kvm, &args); 2912 mutex_unlock(&kvm->slots_lock); 2913 if (!r) { 2914 r = copy_to_user(argp, &args, sizeof(args)); 2915 if (r) 2916 r = -EFAULT; 2917 } 2918 break; 2919 } 2920 case KVM_S390_SET_CMMA_BITS: { 2921 struct kvm_s390_cmma_log args; 2922 2923 r = -EFAULT; 2924 if (copy_from_user(&args, argp, sizeof(args))) 2925 break; 2926 mutex_lock(&kvm->slots_lock); 2927 r = kvm_s390_set_cmma_bits(kvm, &args); 2928 mutex_unlock(&kvm->slots_lock); 2929 break; 2930 } 2931 case KVM_S390_PV_COMMAND: { 2932 struct kvm_pv_cmd args; 2933 2934 /* protvirt means user cpu state */ 2935 kvm_s390_set_user_cpu_state_ctrl(kvm); 2936 r = 0; 2937 if (!is_prot_virt_host()) { 2938 r = -EINVAL; 2939 break; 2940 } 2941 if (copy_from_user(&args, argp, sizeof(args))) { 2942 r = -EFAULT; 2943 break; 2944 } 2945 if (args.flags) { 2946 r = -EINVAL; 2947 break; 2948 } 2949 /* must be called without kvm->lock */ 2950 r = kvm_s390_handle_pv(kvm, &args); 2951 if (copy_to_user(argp, &args, sizeof(args))) { 2952 r = -EFAULT; 2953 break; 2954 } 2955 break; 2956 } 2957 case KVM_S390_MEM_OP: { 2958 struct kvm_s390_mem_op mem_op; 2959 2960 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0) 2961 r = kvm_s390_vm_mem_op(kvm, &mem_op); 2962 else 2963 r = -EFAULT; 2964 break; 2965 } 2966 case KVM_S390_KEYOP: { 2967 struct kvm_s390_mmu_cache *mc; 2968 struct kvm_s390_keyop kop; 2969 union skey skey; 2970 2971 if (copy_from_user(&kop, argp, sizeof(kop))) { 2972 r = -EFAULT; 2973 break; 2974 } 2975 skey.skey = kop.key; 2976 2977 mc = kvm_s390_new_mmu_cache(); 2978 if (!mc) 2979 return -ENOMEM; 2980 2981 r = kvm_s390_keyop(mc, kvm, kop.operation, kop.guest_addr, skey); 2982 kvm_s390_free_mmu_cache(mc); 2983 if (r < 0) 2984 break; 2985 2986 kop.key = r; 2987 r = 0; 2988 if (copy_to_user(argp, &kop, sizeof(kop))) 2989 r = -EFAULT; 2990 break; 2991 } 2992 case KVM_S390_ZPCI_OP: { 2993 struct kvm_s390_zpci_op args; 2994 2995 r = -EINVAL; 2996 if (!IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) 2997 break; 2998 if (copy_from_user(&args, argp, sizeof(args))) { 2999 r = -EFAULT; 3000 break; 3001 } 3002 r = kvm_s390_pci_zpci_op(kvm, &args); 3003 break; 3004 } 3005 default: 3006 r = -ENOTTY; 3007 } 3008 3009 return r; 3010 } 3011 3012 static int kvm_s390_apxa_installed(void) 3013 { 3014 struct ap_config_info info; 3015 3016 if (ap_instructions_available()) { 3017 if (ap_qci(&info) == 0) 3018 return info.apxa; 3019 } 3020 3021 return 0; 3022 } 3023 3024 /* 3025 * The format of the crypto control block (CRYCB) is specified in the 3 low 3026 * order bits of the CRYCB designation (CRYCBD) field as follows: 3027 * Format 0: Neither the message security assist extension 3 (MSAX3) nor the 3028 * AP extended addressing (APXA) facility are installed. 3029 * Format 1: The APXA facility is not installed but the MSAX3 facility is. 3030 * Format 2: Both the APXA and MSAX3 facilities are installed 3031 */ 3032 static void kvm_s390_set_crycb_format(struct kvm *kvm) 3033 { 3034 kvm->arch.crypto.crycbd = virt_to_phys(kvm->arch.crypto.crycb); 3035 3036 /* Clear the CRYCB format bits - i.e., set format 0 by default */ 3037 kvm->arch.crypto.crycbd &= ~(CRYCB_FORMAT_MASK); 3038 3039 /* Check whether MSAX3 is installed */ 3040 if (!test_kvm_facility(kvm, 76)) 3041 return; 3042 3043 if (kvm_s390_apxa_installed()) 3044 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2; 3045 else 3046 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1; 3047 } 3048 3049 /* 3050 * kvm_arch_crypto_set_masks 3051 * 3052 * @kvm: pointer to the target guest's KVM struct containing the crypto masks 3053 * to be set. 3054 * @apm: the mask identifying the accessible AP adapters 3055 * @aqm: the mask identifying the accessible AP domains 3056 * @adm: the mask identifying the accessible AP control domains 3057 * 3058 * Set the masks that identify the adapters, domains and control domains to 3059 * which the KVM guest is granted access. 3060 * 3061 * Note: The kvm->lock mutex must be locked by the caller before invoking this 3062 * function. 3063 */ 3064 void kvm_arch_crypto_set_masks(struct kvm *kvm, unsigned long *apm, 3065 unsigned long *aqm, unsigned long *adm) 3066 { 3067 struct kvm_s390_crypto_cb *crycb = kvm->arch.crypto.crycb; 3068 3069 kvm_s390_vcpu_block_all(kvm); 3070 3071 switch (kvm->arch.crypto.crycbd & CRYCB_FORMAT_MASK) { 3072 case CRYCB_FORMAT2: /* APCB1 use 256 bits */ 3073 memcpy(crycb->apcb1.apm, apm, 32); 3074 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx %016lx %016lx %016lx", 3075 apm[0], apm[1], apm[2], apm[3]); 3076 memcpy(crycb->apcb1.aqm, aqm, 32); 3077 VM_EVENT(kvm, 3, "SET CRYCB: aqm %016lx %016lx %016lx %016lx", 3078 aqm[0], aqm[1], aqm[2], aqm[3]); 3079 memcpy(crycb->apcb1.adm, adm, 32); 3080 VM_EVENT(kvm, 3, "SET CRYCB: adm %016lx %016lx %016lx %016lx", 3081 adm[0], adm[1], adm[2], adm[3]); 3082 break; 3083 case CRYCB_FORMAT1: 3084 case CRYCB_FORMAT0: /* Fall through both use APCB0 */ 3085 memcpy(crycb->apcb0.apm, apm, 8); 3086 memcpy(crycb->apcb0.aqm, aqm, 2); 3087 memcpy(crycb->apcb0.adm, adm, 2); 3088 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx aqm %04x adm %04x", 3089 apm[0], *((unsigned short *)aqm), 3090 *((unsigned short *)adm)); 3091 break; 3092 default: /* Can not happen */ 3093 break; 3094 } 3095 3096 /* recreate the shadow crycb for each vcpu */ 3097 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART); 3098 kvm_s390_vcpu_unblock_all(kvm); 3099 } 3100 EXPORT_SYMBOL_GPL(kvm_arch_crypto_set_masks); 3101 3102 /* 3103 * kvm_arch_crypto_clear_masks 3104 * 3105 * @kvm: pointer to the target guest's KVM struct containing the crypto masks 3106 * to be cleared. 3107 * 3108 * Clear the masks that identify the adapters, domains and control domains to 3109 * which the KVM guest is granted access. 3110 * 3111 * Note: The kvm->lock mutex must be locked by the caller before invoking this 3112 * function. 3113 */ 3114 void kvm_arch_crypto_clear_masks(struct kvm *kvm) 3115 { 3116 kvm_s390_vcpu_block_all(kvm); 3117 3118 memset(&kvm->arch.crypto.crycb->apcb0, 0, 3119 sizeof(kvm->arch.crypto.crycb->apcb0)); 3120 memset(&kvm->arch.crypto.crycb->apcb1, 0, 3121 sizeof(kvm->arch.crypto.crycb->apcb1)); 3122 3123 VM_EVENT(kvm, 3, "%s", "CLR CRYCB:"); 3124 /* recreate the shadow crycb for each vcpu */ 3125 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART); 3126 kvm_s390_vcpu_unblock_all(kvm); 3127 } 3128 EXPORT_SYMBOL_GPL(kvm_arch_crypto_clear_masks); 3129 3130 static u64 kvm_s390_get_initial_cpuid(void) 3131 { 3132 struct cpuid cpuid; 3133 3134 get_cpu_id(&cpuid); 3135 cpuid.version = 0xff; 3136 return *((u64 *) &cpuid); 3137 } 3138 3139 static void kvm_s390_crypto_init(struct kvm *kvm) 3140 { 3141 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb; 3142 kvm_s390_set_crycb_format(kvm); 3143 init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem); 3144 3145 if (!test_kvm_facility(kvm, 76)) 3146 return; 3147 3148 /* Enable AES/DEA protected key functions by default */ 3149 kvm->arch.crypto.aes_kw = 1; 3150 kvm->arch.crypto.dea_kw = 1; 3151 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 3152 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask)); 3153 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 3154 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask)); 3155 } 3156 3157 static void sca_dispose(struct kvm *kvm) 3158 { 3159 free_pages_exact(kvm->arch.sca, sizeof(*kvm->arch.sca)); 3160 kvm->arch.sca = NULL; 3161 } 3162 3163 void kvm_arch_free_vm(struct kvm *kvm) 3164 { 3165 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) 3166 kvm_s390_pci_clear_list(kvm); 3167 3168 __kvm_arch_free_vm(kvm); 3169 } 3170 3171 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 3172 { 3173 gfp_t alloc_flags = GFP_KERNEL_ACCOUNT | __GFP_ZERO; 3174 char debug_name[16]; 3175 int i, rc; 3176 3177 mutex_init(&kvm->arch.pv.import_lock); 3178 3179 rc = -EINVAL; 3180 #ifdef CONFIG_KVM_S390_UCONTROL 3181 if (type & ~KVM_VM_S390_UCONTROL) 3182 goto out_err; 3183 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN))) 3184 goto out_err; 3185 #else 3186 if (type) 3187 goto out_err; 3188 #endif 3189 rc = -ENOMEM; 3190 3191 if (!sclp.has_64bscao) 3192 alloc_flags |= GFP_DMA; 3193 mutex_lock(&kvm_lock); 3194 3195 kvm->arch.sca = alloc_pages_exact(sizeof(*kvm->arch.sca), alloc_flags); 3196 mutex_unlock(&kvm_lock); 3197 if (!kvm->arch.sca) 3198 goto out_err; 3199 3200 snprintf(debug_name, sizeof(debug_name), "kvm-%u", current->pid); 3201 3202 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long)); 3203 if (!kvm->arch.dbf) 3204 goto out_err; 3205 3206 BUILD_BUG_ON(sizeof(struct sie_page2) != 4096); 3207 kvm->arch.sie_page2 = 3208 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA); 3209 if (!kvm->arch.sie_page2) 3210 goto out_err; 3211 3212 kvm->arch.sie_page2->kvm = kvm; 3213 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list; 3214 3215 for (i = 0; i < kvm_s390_fac_size(); i++) { 3216 kvm->arch.model.fac_mask[i] = stfle_fac_list[i] & 3217 (kvm_s390_fac_base[i] | 3218 kvm_s390_fac_ext[i]); 3219 kvm->arch.model.fac_list[i] = stfle_fac_list[i] & 3220 kvm_s390_fac_base[i]; 3221 } 3222 kvm->arch.model.subfuncs = kvm_s390_available_subfunc; 3223 3224 /* we are always in czam mode - even on pre z14 machines */ 3225 set_kvm_facility(kvm->arch.model.fac_mask, 138); 3226 set_kvm_facility(kvm->arch.model.fac_list, 138); 3227 /* we emulate STHYI in kvm */ 3228 set_kvm_facility(kvm->arch.model.fac_mask, 74); 3229 set_kvm_facility(kvm->arch.model.fac_list, 74); 3230 if (machine_has_tlb_guest()) { 3231 set_kvm_facility(kvm->arch.model.fac_mask, 147); 3232 set_kvm_facility(kvm->arch.model.fac_list, 147); 3233 } 3234 3235 if (css_general_characteristics.aiv && test_facility(65)) 3236 set_kvm_facility(kvm->arch.model.fac_mask, 65); 3237 3238 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid(); 3239 kvm->arch.model.ibc = sclp.ibc & 0x0fff; 3240 3241 kvm->arch.model.uv_feat_guest.feat = 0; 3242 3243 kvm_s390_crypto_init(kvm); 3244 3245 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) { 3246 mutex_lock(&kvm->lock); 3247 kvm_s390_pci_init_list(kvm); 3248 kvm_s390_vcpu_pci_enable_interp(kvm); 3249 mutex_unlock(&kvm->lock); 3250 } 3251 3252 mutex_init(&kvm->arch.float_int.ais_lock); 3253 spin_lock_init(&kvm->arch.float_int.lock); 3254 for (i = 0; i < FIRQ_LIST_COUNT; i++) 3255 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]); 3256 init_waitqueue_head(&kvm->arch.ipte_wq); 3257 mutex_init(&kvm->arch.ipte_mutex); 3258 3259 debug_register_view(kvm->arch.dbf, &debug_sprintf_view); 3260 VM_EVENT(kvm, 3, "vm created with type %lu", type); 3261 3262 kvm->arch.mem_limit = type & KVM_VM_S390_UCONTROL ? KVM_S390_NO_MEM_LIMIT : sclp.hamax + 1; 3263 kvm->arch.gmap = gmap_new(kvm, gpa_to_gfn(kvm->arch.mem_limit)); 3264 if (!kvm->arch.gmap) 3265 goto out_err; 3266 clear_bit(GMAP_FLAG_PFAULT_ENABLED, &kvm->arch.gmap->flags); 3267 3268 if (type & KVM_VM_S390_UCONTROL) { 3269 struct kvm_userspace_memory_region2 fake_memslot = { 3270 .slot = KVM_S390_UCONTROL_MEMSLOT, 3271 .guest_phys_addr = 0, 3272 .userspace_addr = 0, 3273 .memory_size = ALIGN_DOWN(TASK_SIZE, _SEGMENT_SIZE), 3274 .flags = 0, 3275 }; 3276 3277 /* one flat fake memslot covering the whole address-space */ 3278 mutex_lock(&kvm->slots_lock); 3279 KVM_BUG_ON(kvm_set_internal_memslot(kvm, &fake_memslot), kvm); 3280 mutex_unlock(&kvm->slots_lock); 3281 set_bit(GMAP_FLAG_IS_UCONTROL, &kvm->arch.gmap->flags); 3282 } else { 3283 struct crst_table *table = dereference_asce(kvm->arch.gmap->asce); 3284 3285 crst_table_init((void *)table, _CRSTE_HOLE(table->crstes[0].h.tt).val); 3286 } 3287 3288 kvm->arch.use_pfmfi = sclp.has_pfmfi; 3289 kvm->arch.use_skf = sclp.has_skey; 3290 spin_lock_init(&kvm->arch.start_stop_lock); 3291 kvm_s390_vsie_init(kvm); 3292 if (use_gisa) 3293 kvm_s390_gisa_init(kvm); 3294 INIT_LIST_HEAD(&kvm->arch.pv.need_cleanup); 3295 kvm->arch.pv.set_aside = NULL; 3296 KVM_EVENT(3, "vm 0x%p created by pid %u", kvm, current->pid); 3297 3298 return 0; 3299 out_err: 3300 free_page((unsigned long)kvm->arch.sie_page2); 3301 debug_unregister(kvm->arch.dbf); 3302 sca_dispose(kvm); 3303 KVM_EVENT(3, "creation of vm failed: %d", rc); 3304 return rc; 3305 } 3306 3307 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 3308 { 3309 u16 rc, rrc; 3310 3311 VCPU_EVENT(vcpu, 3, "%s", "free cpu"); 3312 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id); 3313 kvm_s390_clear_local_irqs(vcpu); 3314 kvm_clear_async_pf_completion_queue(vcpu); 3315 if (!kvm_is_ucontrol(vcpu->kvm)) 3316 sca_del_vcpu(vcpu); 3317 kvm_s390_update_topology_change_report(vcpu->kvm, 1); 3318 3319 if (kvm_is_ucontrol(vcpu->kvm)) { 3320 scoped_guard(spinlock, &vcpu->kvm->arch.gmap->children_lock) 3321 gmap_remove_child(vcpu->arch.gmap); 3322 vcpu->arch.gmap = gmap_put(vcpu->arch.gmap); 3323 } 3324 3325 if (vcpu->kvm->arch.use_cmma) 3326 kvm_s390_vcpu_unsetup_cmma(vcpu); 3327 /* We can not hold the vcpu mutex here, we are already dying */ 3328 if (kvm_s390_pv_cpu_get_handle(vcpu)) 3329 kvm_s390_pv_destroy_cpu(vcpu, &rc, &rrc); 3330 free_page((unsigned long)(vcpu->arch.sie_block)); 3331 kvm_s390_free_mmu_cache(vcpu->arch.mc); 3332 } 3333 3334 void kvm_arch_destroy_vm(struct kvm *kvm) 3335 { 3336 u16 rc, rrc; 3337 3338 kvm_destroy_vcpus(kvm); 3339 sca_dispose(kvm); 3340 kvm_s390_gisa_destroy(kvm); 3341 /* 3342 * We are already at the end of life and kvm->lock is not taken. 3343 * This is ok as the file descriptor is closed by now and nobody 3344 * can mess with the pv state. 3345 */ 3346 kvm_s390_pv_deinit_cleanup_all(kvm, &rc, &rrc); 3347 /* 3348 * Remove the mmu notifier only when the whole KVM VM is torn down, 3349 * and only if one was registered to begin with. If the VM is 3350 * currently not protected, but has been previously been protected, 3351 * then it's possible that the notifier is still registered. 3352 */ 3353 if (kvm->arch.pv.mmu_notifier.ops) 3354 mmu_notifier_unregister(&kvm->arch.pv.mmu_notifier, kvm->mm); 3355 3356 debug_unregister(kvm->arch.dbf); 3357 free_page((unsigned long)kvm->arch.sie_page2); 3358 kvm_s390_destroy_adapters(kvm); 3359 kvm_s390_clear_float_irqs(kvm); 3360 kvm_s390_vsie_destroy(kvm); 3361 kvm->arch.gmap = gmap_put(kvm->arch.gmap); 3362 KVM_EVENT(3, "vm 0x%p destroyed", kvm); 3363 } 3364 3365 /* Section: vcpu related */ 3366 static void sca_del_vcpu(struct kvm_vcpu *vcpu) 3367 { 3368 struct esca_block *sca = vcpu->kvm->arch.sca; 3369 3370 if (!kvm_s390_use_sca_entries()) 3371 return; 3372 3373 clear_bit_inv(vcpu->vcpu_id, (unsigned long *)sca->mcn); 3374 sca->cpu[vcpu->vcpu_id].sda = 0; 3375 } 3376 3377 static void sca_add_vcpu(struct kvm_vcpu *vcpu) 3378 { 3379 struct esca_block *sca = vcpu->kvm->arch.sca; 3380 phys_addr_t sca_phys = virt_to_phys(sca); 3381 3382 /* we still need the sca header for the ipte control */ 3383 vcpu->arch.sie_block->scaoh = sca_phys >> 32; 3384 vcpu->arch.sie_block->scaol = sca_phys & ESCA_SCAOL_MASK; 3385 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA; 3386 3387 if (!kvm_s390_use_sca_entries()) 3388 return; 3389 3390 set_bit_inv(vcpu->vcpu_id, (unsigned long *)sca->mcn); 3391 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block); 3392 } 3393 3394 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id) 3395 { 3396 if (!kvm_s390_use_sca_entries()) 3397 return id < KVM_MAX_VCPUS; 3398 3399 return id < KVM_S390_ESCA_CPU_SLOTS; 3400 } 3401 3402 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */ 3403 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3404 { 3405 WARN_ON_ONCE(vcpu->arch.cputm_start != 0); 3406 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount); 3407 vcpu->arch.cputm_start = get_tod_clock_fast(); 3408 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount); 3409 } 3410 3411 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */ 3412 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3413 { 3414 WARN_ON_ONCE(vcpu->arch.cputm_start == 0); 3415 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount); 3416 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start; 3417 vcpu->arch.cputm_start = 0; 3418 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount); 3419 } 3420 3421 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */ 3422 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3423 { 3424 WARN_ON_ONCE(vcpu->arch.cputm_enabled); 3425 vcpu->arch.cputm_enabled = true; 3426 __start_cpu_timer_accounting(vcpu); 3427 } 3428 3429 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */ 3430 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3431 { 3432 WARN_ON_ONCE(!vcpu->arch.cputm_enabled); 3433 __stop_cpu_timer_accounting(vcpu); 3434 vcpu->arch.cputm_enabled = false; 3435 } 3436 3437 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3438 { 3439 preempt_disable(); /* protect from TOD sync and vcpu_load/put */ 3440 __enable_cpu_timer_accounting(vcpu); 3441 preempt_enable(); 3442 } 3443 3444 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu) 3445 { 3446 preempt_disable(); /* protect from TOD sync and vcpu_load/put */ 3447 __disable_cpu_timer_accounting(vcpu); 3448 preempt_enable(); 3449 } 3450 3451 /* set the cpu timer - may only be called from the VCPU thread itself */ 3452 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm) 3453 { 3454 preempt_disable(); /* protect from TOD sync and vcpu_load/put */ 3455 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount); 3456 if (vcpu->arch.cputm_enabled) 3457 vcpu->arch.cputm_start = get_tod_clock_fast(); 3458 vcpu->arch.sie_block->cputm = cputm; 3459 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount); 3460 preempt_enable(); 3461 } 3462 3463 /* update and get the cpu timer - can also be called from other VCPU threads */ 3464 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu) 3465 { 3466 unsigned int seq; 3467 __u64 value; 3468 3469 if (unlikely(!vcpu->arch.cputm_enabled)) 3470 return vcpu->arch.sie_block->cputm; 3471 3472 preempt_disable(); /* protect from TOD sync and vcpu_load/put */ 3473 do { 3474 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount); 3475 /* 3476 * If the writer would ever execute a read in the critical 3477 * section, e.g. in irq context, we have a deadlock. 3478 */ 3479 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu); 3480 value = vcpu->arch.sie_block->cputm; 3481 /* if cputm_start is 0, accounting is being started/stopped */ 3482 if (likely(vcpu->arch.cputm_start)) 3483 value -= get_tod_clock_fast() - vcpu->arch.cputm_start; 3484 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1)); 3485 preempt_enable(); 3486 return value; 3487 } 3488 3489 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 3490 { 3491 3492 kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING); 3493 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu)) 3494 __start_cpu_timer_accounting(vcpu); 3495 vcpu->cpu = cpu; 3496 } 3497 3498 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 3499 { 3500 vcpu->cpu = -1; 3501 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu)) 3502 __stop_cpu_timer_accounting(vcpu); 3503 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING); 3504 3505 } 3506 3507 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 3508 { 3509 mutex_lock(&vcpu->kvm->lock); 3510 preempt_disable(); 3511 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch; 3512 vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx; 3513 preempt_enable(); 3514 mutex_unlock(&vcpu->kvm->lock); 3515 if (!kvm_is_ucontrol(vcpu->kvm)) { 3516 vcpu->arch.gmap = vcpu->kvm->arch.gmap; 3517 sca_add_vcpu(vcpu); 3518 } 3519 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0) 3520 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC; 3521 } 3522 3523 static bool kvm_has_pckmo_subfunc(struct kvm *kvm, unsigned long nr) 3524 { 3525 if (test_bit_inv(nr, (unsigned long *)&kvm->arch.model.subfuncs.pckmo) && 3526 test_bit_inv(nr, (unsigned long *)&kvm_s390_available_subfunc.pckmo)) 3527 return true; 3528 return false; 3529 } 3530 3531 static bool kvm_has_pckmo_ecc(struct kvm *kvm) 3532 { 3533 /* At least one ECC subfunction must be present */ 3534 return kvm_has_pckmo_subfunc(kvm, 32) || 3535 kvm_has_pckmo_subfunc(kvm, 33) || 3536 kvm_has_pckmo_subfunc(kvm, 34) || 3537 kvm_has_pckmo_subfunc(kvm, 40) || 3538 kvm_has_pckmo_subfunc(kvm, 41); 3539 3540 } 3541 3542 static bool kvm_has_pckmo_hmac(struct kvm *kvm) 3543 { 3544 /* At least one HMAC subfunction must be present */ 3545 return kvm_has_pckmo_subfunc(kvm, 118) || 3546 kvm_has_pckmo_subfunc(kvm, 122); 3547 } 3548 3549 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu) 3550 { 3551 /* 3552 * If the AP instructions are not being interpreted and the MSAX3 3553 * facility is not configured for the guest, there is nothing to set up. 3554 */ 3555 if (!vcpu->kvm->arch.crypto.apie && !test_kvm_facility(vcpu->kvm, 76)) 3556 return; 3557 3558 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd; 3559 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA); 3560 vcpu->arch.sie_block->eca &= ~ECA_APIE; 3561 vcpu->arch.sie_block->ecd &= ~(ECD_ECC | ECD_HMAC); 3562 3563 if (vcpu->kvm->arch.crypto.apie) 3564 vcpu->arch.sie_block->eca |= ECA_APIE; 3565 3566 /* Set up protected key support */ 3567 if (vcpu->kvm->arch.crypto.aes_kw) { 3568 vcpu->arch.sie_block->ecb3 |= ECB3_AES; 3569 /* ecc/hmac is also wrapped with AES key */ 3570 if (kvm_has_pckmo_ecc(vcpu->kvm)) 3571 vcpu->arch.sie_block->ecd |= ECD_ECC; 3572 if (kvm_has_pckmo_hmac(vcpu->kvm)) 3573 vcpu->arch.sie_block->ecd |= ECD_HMAC; 3574 } 3575 3576 if (vcpu->kvm->arch.crypto.dea_kw) 3577 vcpu->arch.sie_block->ecb3 |= ECB3_DEA; 3578 } 3579 3580 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu) 3581 { 3582 free_page((unsigned long)phys_to_virt(vcpu->arch.sie_block->cbrlo)); 3583 vcpu->arch.sie_block->cbrlo = 0; 3584 } 3585 3586 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu) 3587 { 3588 void *cbrlo_page = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT); 3589 3590 if (!cbrlo_page) 3591 return -ENOMEM; 3592 3593 vcpu->arch.sie_block->cbrlo = virt_to_phys(cbrlo_page); 3594 return 0; 3595 } 3596 3597 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu) 3598 { 3599 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model; 3600 3601 vcpu->arch.sie_block->ibc = model->ibc; 3602 if (test_kvm_facility(vcpu->kvm, 7)) 3603 vcpu->arch.sie_block->fac = virt_to_phys(model->fac_list); 3604 } 3605 3606 static int kvm_s390_vcpu_setup(struct kvm_vcpu *vcpu) 3607 { 3608 int rc = 0; 3609 u16 uvrc, uvrrc; 3610 3611 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH | 3612 CPUSTAT_SM | 3613 CPUSTAT_STOPPED); 3614 3615 if (test_kvm_facility(vcpu->kvm, 78)) 3616 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2); 3617 else if (test_kvm_facility(vcpu->kvm, 8)) 3618 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED); 3619 3620 kvm_s390_vcpu_setup_model(vcpu); 3621 3622 /* pgste_set_pte has special handling for !machine_has_esop() */ 3623 if (machine_has_esop()) 3624 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT; 3625 if (test_kvm_facility(vcpu->kvm, 9)) 3626 vcpu->arch.sie_block->ecb |= ECB_SRSI; 3627 if (test_kvm_facility(vcpu->kvm, 11)) 3628 vcpu->arch.sie_block->ecb |= ECB_PTF; 3629 if (test_kvm_facility(vcpu->kvm, 73)) 3630 vcpu->arch.sie_block->ecb |= ECB_TE; 3631 if (!kvm_is_ucontrol(vcpu->kvm)) 3632 vcpu->arch.sie_block->ecb |= ECB_SPECI; 3633 3634 if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi) 3635 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI; 3636 if (test_kvm_facility(vcpu->kvm, 130)) 3637 vcpu->arch.sie_block->ecb2 |= ECB2_IEP; 3638 vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI; 3639 if (sclp.has_cei) 3640 vcpu->arch.sie_block->eca |= ECA_CEI; 3641 if (sclp.has_ib) 3642 vcpu->arch.sie_block->eca |= ECA_IB; 3643 if (sclp.has_siif) 3644 vcpu->arch.sie_block->eca |= ECA_SII; 3645 if (kvm_s390_use_sca_entries()) 3646 vcpu->arch.sie_block->eca |= ECA_SIGPI; 3647 if (test_kvm_facility(vcpu->kvm, 129)) { 3648 vcpu->arch.sie_block->eca |= ECA_VX; 3649 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT; 3650 } 3651 if (test_kvm_facility(vcpu->kvm, 139)) 3652 vcpu->arch.sie_block->ecd |= ECD_MEF; 3653 if (test_kvm_facility(vcpu->kvm, 156)) 3654 vcpu->arch.sie_block->ecd |= ECD_ETOKENF; 3655 if (vcpu->arch.sie_block->gd) { 3656 vcpu->arch.sie_block->eca |= ECA_AIV; 3657 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u", 3658 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id); 3659 } 3660 vcpu->arch.sie_block->sdnxo = virt_to_phys(&vcpu->run->s.regs.sdnx) | SDNXC; 3661 vcpu->arch.sie_block->riccbd = virt_to_phys(&vcpu->run->s.regs.riccb); 3662 3663 if (sclp.has_kss) 3664 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS); 3665 else 3666 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE; 3667 3668 if (vcpu->kvm->arch.use_cmma) { 3669 rc = kvm_s390_vcpu_setup_cmma(vcpu); 3670 if (rc) 3671 return rc; 3672 } 3673 hrtimer_setup(&vcpu->arch.ckc_timer, kvm_s390_idle_wakeup, CLOCK_MONOTONIC, 3674 HRTIMER_MODE_REL); 3675 3676 vcpu->arch.sie_block->hpid = HPID_KVM; 3677 3678 kvm_s390_vcpu_crypto_setup(vcpu); 3679 3680 kvm_s390_vcpu_pci_setup(vcpu); 3681 3682 mutex_lock(&vcpu->kvm->lock); 3683 if (kvm_s390_pv_is_protected(vcpu->kvm)) { 3684 rc = kvm_s390_pv_create_cpu(vcpu, &uvrc, &uvrrc); 3685 if (rc) 3686 kvm_s390_vcpu_unsetup_cmma(vcpu); 3687 } 3688 mutex_unlock(&vcpu->kvm->lock); 3689 3690 return rc; 3691 } 3692 3693 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) 3694 { 3695 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id)) 3696 return -EINVAL; 3697 return 0; 3698 } 3699 3700 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) 3701 { 3702 struct sie_page *sie_page; 3703 int rc; 3704 3705 BUILD_BUG_ON(sizeof(struct sie_page) != 4096); 3706 vcpu->arch.mc = kvm_s390_new_mmu_cache(); 3707 if (!vcpu->arch.mc) 3708 return -ENOMEM; 3709 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL_ACCOUNT); 3710 if (!sie_page) { 3711 kvm_s390_free_mmu_cache(vcpu->arch.mc); 3712 vcpu->arch.mc = NULL; 3713 return -ENOMEM; 3714 } 3715 3716 vcpu->arch.sie_block = &sie_page->sie_block; 3717 vcpu->arch.sie_block->itdba = virt_to_phys(&sie_page->itdb); 3718 3719 /* the real guest size will always be smaller than msl */ 3720 vcpu->arch.sie_block->mso = 0; 3721 vcpu->arch.sie_block->msl = sclp.hamax; 3722 3723 vcpu->arch.sie_block->icpua = vcpu->vcpu_id; 3724 spin_lock_init(&vcpu->arch.local_int.lock); 3725 vcpu->arch.sie_block->gd = kvm_s390_get_gisa_desc(vcpu->kvm); 3726 seqcount_init(&vcpu->arch.cputm_seqcount); 3727 3728 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID; 3729 kvm_clear_async_pf_completion_queue(vcpu); 3730 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX | 3731 KVM_SYNC_GPRS | 3732 KVM_SYNC_ACRS | 3733 KVM_SYNC_CRS | 3734 KVM_SYNC_ARCH0 | 3735 KVM_SYNC_PFAULT | 3736 KVM_SYNC_DIAG318; 3737 vcpu->arch.acrs_loaded = false; 3738 kvm_s390_set_prefix(vcpu, 0); 3739 if (test_kvm_facility(vcpu->kvm, 64)) 3740 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB; 3741 if (test_kvm_facility(vcpu->kvm, 82)) 3742 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC; 3743 if (test_kvm_facility(vcpu->kvm, 133)) 3744 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB; 3745 if (test_kvm_facility(vcpu->kvm, 156)) 3746 vcpu->run->kvm_valid_regs |= KVM_SYNC_ETOKEN; 3747 /* fprs can be synchronized via vrs, even if the guest has no vx. With 3748 * cpu_has_vx(), (load|store)_fpu_regs() will work with vrs format. 3749 */ 3750 if (cpu_has_vx()) 3751 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS; 3752 else 3753 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS; 3754 3755 if (kvm_is_ucontrol(vcpu->kvm)) { 3756 rc = -ENOMEM; 3757 vcpu->arch.gmap = gmap_new_child(vcpu->kvm->arch.gmap, -1UL); 3758 if (!vcpu->arch.gmap) 3759 goto out_free_sie_block; 3760 } 3761 3762 VM_EVENT(vcpu->kvm, 3, "create cpu %d at 0x%p, sie block at 0x%p", 3763 vcpu->vcpu_id, vcpu, vcpu->arch.sie_block); 3764 trace_kvm_s390_create_vcpu(vcpu->vcpu_id, vcpu, vcpu->arch.sie_block); 3765 3766 rc = kvm_s390_vcpu_setup(vcpu); 3767 if (rc) 3768 goto out_ucontrol_uninit; 3769 3770 kvm_s390_update_topology_change_report(vcpu->kvm, 1); 3771 return 0; 3772 3773 out_ucontrol_uninit: 3774 if (kvm_is_ucontrol(vcpu->kvm)) { 3775 gmap_remove_child(vcpu->arch.gmap); 3776 vcpu->arch.gmap = gmap_put(vcpu->arch.gmap); 3777 } 3778 out_free_sie_block: 3779 free_page((unsigned long)(vcpu->arch.sie_block)); 3780 return rc; 3781 } 3782 3783 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) 3784 { 3785 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask); 3786 return kvm_s390_vcpu_has_irq(vcpu, 0); 3787 } 3788 3789 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 3790 { 3791 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE); 3792 } 3793 3794 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu) 3795 { 3796 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20); 3797 exit_sie(vcpu); 3798 } 3799 3800 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu) 3801 { 3802 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20); 3803 } 3804 3805 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu) 3806 { 3807 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20); 3808 exit_sie(vcpu); 3809 } 3810 3811 bool kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu *vcpu) 3812 { 3813 return atomic_read(&vcpu->arch.sie_block->prog20) & 3814 (PROG_BLOCK_SIE | PROG_REQUEST); 3815 } 3816 3817 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu) 3818 { 3819 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20); 3820 } 3821 3822 /* 3823 * Kick a guest cpu out of (v)SIE and wait until (v)SIE is not running. 3824 * If the CPU is not running (e.g. waiting as idle) the function will 3825 * return immediately. */ 3826 void exit_sie(struct kvm_vcpu *vcpu) 3827 { 3828 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT); 3829 kvm_s390_vsie_kick(vcpu); 3830 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE) 3831 cpu_relax(); 3832 } 3833 3834 /* Kick a guest cpu out of SIE to process a request synchronously */ 3835 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu) 3836 { 3837 __kvm_make_request(req, vcpu); 3838 kvm_s390_vcpu_request(vcpu); 3839 } 3840 3841 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu) 3842 { 3843 /* do not poll with more than halt_poll_max_steal percent of steal time */ 3844 if (get_lowcore()->avg_steal_timer * 100 / (TICK_USEC << 12) >= 3845 READ_ONCE(halt_poll_max_steal)) { 3846 vcpu->stat.halt_no_poll_steal++; 3847 return true; 3848 } 3849 return false; 3850 } 3851 3852 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 3853 { 3854 /* kvm common code refers to this, but never calls it */ 3855 BUG(); 3856 return 0; 3857 } 3858 3859 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, 3860 struct kvm_one_reg *reg) 3861 { 3862 int r = -EINVAL; 3863 3864 switch (reg->id) { 3865 case KVM_REG_S390_TODPR: 3866 r = put_user(vcpu->arch.sie_block->todpr, 3867 (u32 __user *)reg->addr); 3868 break; 3869 case KVM_REG_S390_EPOCHDIFF: 3870 r = put_user(vcpu->arch.sie_block->epoch, 3871 (u64 __user *)reg->addr); 3872 break; 3873 case KVM_REG_S390_CPU_TIMER: 3874 r = put_user(kvm_s390_get_cpu_timer(vcpu), 3875 (u64 __user *)reg->addr); 3876 break; 3877 case KVM_REG_S390_CLOCK_COMP: 3878 r = put_user(vcpu->arch.sie_block->ckc, 3879 (u64 __user *)reg->addr); 3880 break; 3881 case KVM_REG_S390_PFTOKEN: 3882 r = put_user(vcpu->arch.pfault_token, 3883 (u64 __user *)reg->addr); 3884 break; 3885 case KVM_REG_S390_PFCOMPARE: 3886 r = put_user(vcpu->arch.pfault_compare, 3887 (u64 __user *)reg->addr); 3888 break; 3889 case KVM_REG_S390_PFSELECT: 3890 r = put_user(vcpu->arch.pfault_select, 3891 (u64 __user *)reg->addr); 3892 break; 3893 case KVM_REG_S390_PP: 3894 r = put_user(vcpu->arch.sie_block->pp, 3895 (u64 __user *)reg->addr); 3896 break; 3897 case KVM_REG_S390_GBEA: 3898 r = put_user(vcpu->arch.sie_block->gbea, 3899 (u64 __user *)reg->addr); 3900 break; 3901 default: 3902 break; 3903 } 3904 3905 return r; 3906 } 3907 3908 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, 3909 struct kvm_one_reg *reg) 3910 { 3911 int r = -EINVAL; 3912 __u64 val; 3913 3914 switch (reg->id) { 3915 case KVM_REG_S390_TODPR: 3916 r = get_user(vcpu->arch.sie_block->todpr, 3917 (u32 __user *)reg->addr); 3918 break; 3919 case KVM_REG_S390_EPOCHDIFF: 3920 r = get_user(vcpu->arch.sie_block->epoch, 3921 (u64 __user *)reg->addr); 3922 break; 3923 case KVM_REG_S390_CPU_TIMER: 3924 r = get_user(val, (u64 __user *)reg->addr); 3925 if (!r) 3926 kvm_s390_set_cpu_timer(vcpu, val); 3927 break; 3928 case KVM_REG_S390_CLOCK_COMP: 3929 r = get_user(vcpu->arch.sie_block->ckc, 3930 (u64 __user *)reg->addr); 3931 break; 3932 case KVM_REG_S390_PFTOKEN: 3933 r = get_user(vcpu->arch.pfault_token, 3934 (u64 __user *)reg->addr); 3935 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) 3936 kvm_clear_async_pf_completion_queue(vcpu); 3937 break; 3938 case KVM_REG_S390_PFCOMPARE: 3939 r = get_user(vcpu->arch.pfault_compare, 3940 (u64 __user *)reg->addr); 3941 break; 3942 case KVM_REG_S390_PFSELECT: 3943 r = get_user(vcpu->arch.pfault_select, 3944 (u64 __user *)reg->addr); 3945 break; 3946 case KVM_REG_S390_PP: 3947 r = get_user(vcpu->arch.sie_block->pp, 3948 (u64 __user *)reg->addr); 3949 break; 3950 case KVM_REG_S390_GBEA: 3951 r = get_user(vcpu->arch.sie_block->gbea, 3952 (u64 __user *)reg->addr); 3953 break; 3954 default: 3955 break; 3956 } 3957 3958 return r; 3959 } 3960 3961 static void kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu *vcpu) 3962 { 3963 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_RI; 3964 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID; 3965 memset(vcpu->run->s.regs.riccb, 0, sizeof(vcpu->run->s.regs.riccb)); 3966 3967 kvm_clear_async_pf_completion_queue(vcpu); 3968 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) 3969 kvm_s390_vcpu_stop(vcpu); 3970 kvm_s390_clear_local_irqs(vcpu); 3971 } 3972 3973 static void kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu) 3974 { 3975 /* Initial reset is a superset of the normal reset */ 3976 kvm_arch_vcpu_ioctl_normal_reset(vcpu); 3977 3978 /* 3979 * This equals initial cpu reset in pop, but we don't switch to ESA. 3980 * We do not only reset the internal data, but also ... 3981 */ 3982 vcpu->arch.sie_block->gpsw.mask = 0; 3983 vcpu->arch.sie_block->gpsw.addr = 0; 3984 kvm_s390_set_prefix(vcpu, 0); 3985 kvm_s390_set_cpu_timer(vcpu, 0); 3986 vcpu->arch.sie_block->ckc = 0; 3987 memset(vcpu->arch.sie_block->gcr, 0, sizeof(vcpu->arch.sie_block->gcr)); 3988 vcpu->arch.sie_block->gcr[0] = CR0_INITIAL_MASK; 3989 vcpu->arch.sie_block->gcr[14] = CR14_INITIAL_MASK; 3990 3991 /* ... the data in sync regs */ 3992 memset(vcpu->run->s.regs.crs, 0, sizeof(vcpu->run->s.regs.crs)); 3993 vcpu->run->s.regs.ckc = 0; 3994 vcpu->run->s.regs.crs[0] = CR0_INITIAL_MASK; 3995 vcpu->run->s.regs.crs[14] = CR14_INITIAL_MASK; 3996 vcpu->run->psw_addr = 0; 3997 vcpu->run->psw_mask = 0; 3998 vcpu->run->s.regs.todpr = 0; 3999 vcpu->run->s.regs.cputm = 0; 4000 vcpu->run->s.regs.ckc = 0; 4001 vcpu->run->s.regs.pp = 0; 4002 vcpu->run->s.regs.gbea = 1; 4003 vcpu->run->s.regs.fpc = 0; 4004 /* 4005 * Do not reset these registers in the protected case, as some of 4006 * them are overlaid and they are not accessible in this case 4007 * anyway. 4008 */ 4009 if (!kvm_s390_pv_cpu_is_protected(vcpu)) { 4010 vcpu->arch.sie_block->gbea = 1; 4011 vcpu->arch.sie_block->pp = 0; 4012 vcpu->arch.sie_block->fpf &= ~FPF_BPBC; 4013 vcpu->arch.sie_block->todpr = 0; 4014 } 4015 } 4016 4017 static void kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu *vcpu) 4018 { 4019 struct kvm_sync_regs *regs = &vcpu->run->s.regs; 4020 4021 /* Clear reset is a superset of the initial reset */ 4022 kvm_arch_vcpu_ioctl_initial_reset(vcpu); 4023 4024 memset(®s->gprs, 0, sizeof(regs->gprs)); 4025 memset(®s->vrs, 0, sizeof(regs->vrs)); 4026 memset(®s->acrs, 0, sizeof(regs->acrs)); 4027 memset(®s->gscb, 0, sizeof(regs->gscb)); 4028 4029 regs->etoken = 0; 4030 regs->etoken_extension = 0; 4031 } 4032 4033 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) 4034 { 4035 vcpu_load(vcpu); 4036 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs)); 4037 vcpu_put(vcpu); 4038 return 0; 4039 } 4040 4041 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) 4042 { 4043 vcpu_load(vcpu); 4044 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs)); 4045 vcpu_put(vcpu); 4046 return 0; 4047 } 4048 4049 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, 4050 struct kvm_sregs *sregs) 4051 { 4052 vcpu_load(vcpu); 4053 4054 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs)); 4055 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs)); 4056 4057 vcpu_put(vcpu); 4058 return 0; 4059 } 4060 4061 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, 4062 struct kvm_sregs *sregs) 4063 { 4064 vcpu_load(vcpu); 4065 4066 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs)); 4067 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs)); 4068 4069 vcpu_put(vcpu); 4070 return 0; 4071 } 4072 4073 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) 4074 { 4075 vcpu_load(vcpu); 4076 4077 vcpu->run->s.regs.fpc = fpu->fpc; 4078 if (cpu_has_vx()) 4079 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs, 4080 (freg_t *) fpu->fprs); 4081 else 4082 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs)); 4083 4084 vcpu_put(vcpu); 4085 return 0; 4086 } 4087 4088 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) 4089 { 4090 vcpu_load(vcpu); 4091 4092 if (cpu_has_vx()) 4093 convert_vx_to_fp((freg_t *) fpu->fprs, 4094 (__vector128 *) vcpu->run->s.regs.vrs); 4095 else 4096 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs)); 4097 fpu->fpc = vcpu->run->s.regs.fpc; 4098 4099 vcpu_put(vcpu); 4100 return 0; 4101 } 4102 4103 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw) 4104 { 4105 int rc = 0; 4106 4107 if (!is_vcpu_stopped(vcpu)) 4108 rc = -EBUSY; 4109 else { 4110 vcpu->run->psw_mask = psw.mask; 4111 vcpu->run->psw_addr = psw.addr; 4112 } 4113 return rc; 4114 } 4115 4116 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, 4117 struct kvm_translation *tr) 4118 { 4119 return -EINVAL; /* not implemented yet */ 4120 } 4121 4122 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \ 4123 KVM_GUESTDBG_USE_HW_BP | \ 4124 KVM_GUESTDBG_ENABLE) 4125 4126 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, 4127 struct kvm_guest_debug *dbg) 4128 { 4129 int rc = 0; 4130 4131 vcpu_load(vcpu); 4132 4133 vcpu->guest_debug = 0; 4134 kvm_s390_clear_bp_data(vcpu); 4135 4136 if (dbg->control & ~VALID_GUESTDBG_FLAGS) { 4137 rc = -EINVAL; 4138 goto out; 4139 } 4140 if (!sclp.has_gpere) { 4141 rc = -EINVAL; 4142 goto out; 4143 } 4144 4145 if (dbg->control & KVM_GUESTDBG_ENABLE) { 4146 vcpu->guest_debug = dbg->control; 4147 /* enforce guest PER */ 4148 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P); 4149 4150 if (dbg->control & KVM_GUESTDBG_USE_HW_BP) 4151 rc = kvm_s390_import_bp_data(vcpu, dbg); 4152 } else { 4153 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P); 4154 vcpu->arch.guestdbg.last_bp = 0; 4155 } 4156 4157 if (rc) { 4158 vcpu->guest_debug = 0; 4159 kvm_s390_clear_bp_data(vcpu); 4160 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P); 4161 } 4162 4163 out: 4164 vcpu_put(vcpu); 4165 return rc; 4166 } 4167 4168 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 4169 struct kvm_mp_state *mp_state) 4170 { 4171 int ret; 4172 4173 vcpu_load(vcpu); 4174 4175 /* CHECK_STOP and LOAD are not supported yet */ 4176 ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED : 4177 KVM_MP_STATE_OPERATING; 4178 4179 vcpu_put(vcpu); 4180 return ret; 4181 } 4182 4183 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 4184 struct kvm_mp_state *mp_state) 4185 { 4186 int rc = 0; 4187 4188 vcpu_load(vcpu); 4189 4190 /* user space knows about this interface - let it control the state */ 4191 kvm_s390_set_user_cpu_state_ctrl(vcpu->kvm); 4192 4193 switch (mp_state->mp_state) { 4194 case KVM_MP_STATE_STOPPED: 4195 rc = kvm_s390_vcpu_stop(vcpu); 4196 break; 4197 case KVM_MP_STATE_OPERATING: 4198 rc = kvm_s390_vcpu_start(vcpu); 4199 break; 4200 case KVM_MP_STATE_LOAD: 4201 if (!kvm_s390_pv_cpu_is_protected(vcpu)) { 4202 rc = -ENXIO; 4203 break; 4204 } 4205 rc = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR_LOAD); 4206 break; 4207 case KVM_MP_STATE_CHECK_STOP: 4208 fallthrough; /* CHECK_STOP and LOAD are not supported yet */ 4209 default: 4210 rc = -ENXIO; 4211 } 4212 4213 vcpu_put(vcpu); 4214 return rc; 4215 } 4216 4217 static bool ibs_enabled(struct kvm_vcpu *vcpu) 4218 { 4219 return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS); 4220 } 4221 4222 static int vcpu_ucontrol_translate(struct kvm_vcpu *vcpu, gpa_t *gaddr) 4223 { 4224 int rc; 4225 4226 if (kvm_is_ucontrol(vcpu->kvm)) { 4227 rc = gmap_ucas_translate(vcpu->arch.mc, vcpu->arch.gmap, gaddr); 4228 if (rc == -EREMOTE) { 4229 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL; 4230 vcpu->run->s390_ucontrol.trans_exc_code = *gaddr; 4231 vcpu->run->s390_ucontrol.pgm_code = PGM_SEGMENT_TRANSLATION; 4232 } 4233 return rc; 4234 } 4235 return 0; 4236 } 4237 4238 static int kvm_s390_fixup_prefix(struct kvm_vcpu *vcpu) 4239 { 4240 gpa_t gaddr = kvm_s390_get_prefix(vcpu); 4241 gfn_t gfn; 4242 int rc; 4243 4244 if (vcpu_ucontrol_translate(vcpu, &gaddr)) 4245 return -EREMOTE; 4246 gfn = gpa_to_gfn(gaddr); 4247 4248 rc = kvm_s390_faultin_gfn_simple(vcpu, NULL, gfn, true); 4249 if (rc) 4250 return rc; 4251 rc = kvm_s390_faultin_gfn_simple(vcpu, NULL, gfn + 1, true); 4252 if (rc) 4253 return rc; 4254 4255 scoped_guard(write_lock, &vcpu->kvm->mmu_lock) 4256 rc = dat_set_prefix_notif_bit(vcpu->kvm->arch.gmap->asce, gfn); 4257 return rc; 4258 } 4259 4260 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu) 4261 { 4262 retry: 4263 kvm_s390_vcpu_request_handled(vcpu); 4264 if (!kvm_request_pending(vcpu)) 4265 return 0; 4266 /* 4267 * If the guest prefix changed, re-arm the ipte notifier for the 4268 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock. 4269 * This ensures that the ipte instruction for this request has 4270 * already finished. We might race against a second unmapper that 4271 * wants to set the blocking bit. Lets just retry the request loop. 4272 */ 4273 if (kvm_check_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu)) { 4274 int rc; 4275 4276 rc = kvm_s390_fixup_prefix(vcpu); 4277 if (rc) { 4278 kvm_make_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu); 4279 return rc; 4280 } 4281 goto retry; 4282 } 4283 4284 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) { 4285 vcpu->arch.sie_block->ihcpu = 0xffff; 4286 goto retry; 4287 } 4288 4289 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) { 4290 if (!ibs_enabled(vcpu)) { 4291 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1); 4292 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS); 4293 } 4294 goto retry; 4295 } 4296 4297 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) { 4298 if (ibs_enabled(vcpu)) { 4299 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0); 4300 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS); 4301 } 4302 goto retry; 4303 } 4304 4305 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) { 4306 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC; 4307 goto retry; 4308 } 4309 4310 if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) { 4311 /* 4312 * Disable CMM virtualization; we will emulate the ESSA 4313 * instruction manually, in order to provide additional 4314 * functionalities needed for live migration. 4315 */ 4316 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA; 4317 goto retry; 4318 } 4319 4320 if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) { 4321 /* 4322 * Re-enable CMM virtualization if CMMA is available and 4323 * CMM has been used. 4324 */ 4325 if (vcpu->kvm->arch.use_cmma && uses_cmm(vcpu->arch.gmap)) 4326 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA; 4327 goto retry; 4328 } 4329 4330 /* we left the vsie handler, nothing to do, just clear the request */ 4331 kvm_clear_request(KVM_REQ_VSIE_RESTART, vcpu); 4332 4333 return 0; 4334 } 4335 4336 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod) 4337 { 4338 struct kvm_vcpu *vcpu; 4339 union tod_clock clk; 4340 unsigned long i; 4341 4342 preempt_disable(); 4343 4344 store_tod_clock_ext(&clk); 4345 4346 kvm->arch.epoch = gtod->tod - clk.tod; 4347 kvm->arch.epdx = 0; 4348 if (test_kvm_facility(kvm, 139)) { 4349 kvm->arch.epdx = gtod->epoch_idx - clk.ei; 4350 if (kvm->arch.epoch > gtod->tod) 4351 kvm->arch.epdx -= 1; 4352 } 4353 4354 kvm_s390_vcpu_block_all(kvm); 4355 kvm_for_each_vcpu(i, vcpu, kvm) { 4356 vcpu->arch.sie_block->epoch = kvm->arch.epoch; 4357 vcpu->arch.sie_block->epdx = kvm->arch.epdx; 4358 } 4359 4360 kvm_s390_vcpu_unblock_all(kvm); 4361 preempt_enable(); 4362 } 4363 4364 int kvm_s390_try_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod) 4365 { 4366 if (!mutex_trylock(&kvm->lock)) 4367 return 0; 4368 __kvm_s390_set_tod_clock(kvm, gtod); 4369 mutex_unlock(&kvm->lock); 4370 return 1; 4371 } 4372 4373 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token, 4374 unsigned long token) 4375 { 4376 struct kvm_s390_interrupt inti; 4377 struct kvm_s390_irq irq; 4378 4379 if (start_token) { 4380 irq.u.ext.ext_params2 = token; 4381 irq.type = KVM_S390_INT_PFAULT_INIT; 4382 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq)); 4383 } else { 4384 inti.type = KVM_S390_INT_PFAULT_DONE; 4385 inti.parm64 = token; 4386 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti)); 4387 } 4388 } 4389 4390 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, 4391 struct kvm_async_pf *work) 4392 { 4393 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token); 4394 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token); 4395 4396 return true; 4397 } 4398 4399 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, 4400 struct kvm_async_pf *work) 4401 { 4402 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token); 4403 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token); 4404 } 4405 4406 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, 4407 struct kvm_async_pf *work) 4408 { 4409 /* s390 will always inject the page directly */ 4410 } 4411 4412 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu) 4413 { 4414 /* 4415 * s390 will always inject the page directly, 4416 * but we still want check_async_completion to cleanup 4417 */ 4418 return true; 4419 } 4420 4421 bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu) 4422 { 4423 hva_t hva; 4424 struct kvm_arch_async_pf arch; 4425 4426 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) 4427 return false; 4428 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) != 4429 vcpu->arch.pfault_compare) 4430 return false; 4431 if (psw_extint_disabled(vcpu)) 4432 return false; 4433 if (kvm_s390_vcpu_has_irq(vcpu, 0)) 4434 return false; 4435 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) 4436 return false; 4437 if (!pfault_enabled(vcpu->arch.gmap)) 4438 return false; 4439 4440 hva = gfn_to_hva(vcpu->kvm, current->thread.gmap_teid.addr); 4441 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8)) 4442 return false; 4443 4444 return kvm_setup_async_pf(vcpu, current->thread.gmap_teid.addr * PAGE_SIZE, hva, &arch); 4445 } 4446 4447 static int vcpu_pre_run(struct kvm_vcpu *vcpu) 4448 { 4449 int rc, cpuflags; 4450 4451 /* 4452 * On s390 notifications for arriving pages will be delivered directly 4453 * to the guest but the house keeping for completed pfaults is 4454 * handled outside the worker. 4455 */ 4456 kvm_check_async_pf_completion(vcpu); 4457 4458 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14]; 4459 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15]; 4460 4461 if (!kvm_is_ucontrol(vcpu->kvm)) { 4462 rc = kvm_s390_deliver_pending_interrupts(vcpu); 4463 if (rc || guestdbg_exit_pending(vcpu)) 4464 return rc; 4465 } 4466 4467 rc = kvm_s390_handle_requests(vcpu); 4468 if (rc) 4469 return rc; 4470 4471 if (guestdbg_enabled(vcpu)) { 4472 kvm_s390_backup_guest_per_regs(vcpu); 4473 kvm_s390_patch_guest_per_regs(vcpu); 4474 } 4475 4476 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask); 4477 4478 vcpu->arch.sie_block->icptcode = 0; 4479 current->thread.gmap_int_code = 0; 4480 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags); 4481 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags); 4482 trace_kvm_s390_sie_enter(vcpu, cpuflags); 4483 4484 return 0; 4485 } 4486 4487 static int vcpu_post_run_addressing_exception(struct kvm_vcpu *vcpu) 4488 { 4489 struct kvm_s390_pgm_info pgm_info = { 4490 .code = PGM_ADDRESSING, 4491 }; 4492 u8 opcode, ilen; 4493 int rc; 4494 4495 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction"); 4496 trace_kvm_s390_sie_fault(vcpu); 4497 4498 /* 4499 * We want to inject an addressing exception, which is defined as a 4500 * suppressing or terminating exception. However, since we came here 4501 * by a DAT access exception, the PSW still points to the faulting 4502 * instruction since DAT exceptions are nullifying. So we've got 4503 * to look up the current opcode to get the length of the instruction 4504 * to be able to forward the PSW. 4505 */ 4506 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1); 4507 ilen = insn_length(opcode); 4508 if (rc < 0) { 4509 return rc; 4510 } else if (rc) { 4511 /* Instruction-Fetching Exceptions - we can't detect the ilen. 4512 * Forward by arbitrary ilc, injection will take care of 4513 * nullification if necessary. 4514 */ 4515 pgm_info = vcpu->arch.pgm; 4516 ilen = 4; 4517 } 4518 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID; 4519 kvm_s390_forward_psw(vcpu, ilen); 4520 return kvm_s390_inject_prog_irq(vcpu, &pgm_info); 4521 } 4522 4523 static void kvm_s390_assert_primary_as(struct kvm_vcpu *vcpu) 4524 { 4525 KVM_BUG(current->thread.gmap_teid.as != PSW_BITS_AS_PRIMARY, vcpu->kvm, 4526 "Unexpected program interrupt 0x%x, TEID 0x%016lx", 4527 current->thread.gmap_int_code, current->thread.gmap_teid.val); 4528 } 4529 4530 static int vcpu_dat_fault_handler(struct kvm_vcpu *vcpu, gpa_t gaddr, bool wr) 4531 { 4532 struct guest_fault f = { 4533 .write_attempt = wr, 4534 .attempt_pfault = pfault_enabled(vcpu->arch.gmap), 4535 }; 4536 int rc; 4537 4538 if (vcpu_ucontrol_translate(vcpu, &gaddr)) 4539 return -EREMOTE; 4540 f.gfn = gpa_to_gfn(gaddr); 4541 4542 rc = kvm_s390_faultin_gfn(vcpu, NULL, &f); 4543 if (rc <= 0) 4544 return rc; 4545 if (rc == PGM_ADDRESSING) 4546 return vcpu_post_run_addressing_exception(vcpu); 4547 KVM_BUG_ON(rc, vcpu->kvm); 4548 return -EINVAL; 4549 } 4550 4551 static int vcpu_post_run_handle_fault(struct kvm_vcpu *vcpu) 4552 { 4553 unsigned int foll = 0; 4554 unsigned long gaddr; 4555 int rc; 4556 4557 gaddr = current->thread.gmap_teid.addr * PAGE_SIZE; 4558 if (kvm_s390_cur_gmap_fault_is_write()) 4559 foll = FOLL_WRITE; 4560 4561 switch (current->thread.gmap_int_code & PGM_INT_CODE_MASK) { 4562 case 0: 4563 vcpu->stat.exit_null++; 4564 break; 4565 case PGM_SECURE_STORAGE_ACCESS: 4566 case PGM_SECURE_STORAGE_VIOLATION: 4567 kvm_s390_assert_primary_as(vcpu); 4568 /* 4569 * This can happen after a reboot with asynchronous teardown; 4570 * the new guest (normal or protected) will run on top of the 4571 * previous protected guest. The old pages need to be destroyed 4572 * so the new guest can use them. 4573 */ 4574 if (kvm_s390_pv_destroy_page(vcpu->kvm, gaddr)) { 4575 /* 4576 * Either KVM messed up the secure guest mapping or the 4577 * same page is mapped into multiple secure guests. 4578 * 4579 * This exception is only triggered when a guest 2 is 4580 * running and can therefore never occur in kernel 4581 * context. 4582 */ 4583 pr_warn_ratelimited("Secure storage violation (%x) in task: %s, pid %d\n", 4584 current->thread.gmap_int_code, current->comm, 4585 current->pid); 4586 send_sig(SIGSEGV, current, 0); 4587 } 4588 break; 4589 case PGM_NON_SECURE_STORAGE_ACCESS: 4590 kvm_s390_assert_primary_as(vcpu); 4591 /* 4592 * This is normal operation; a page belonging to a protected 4593 * guest has not been imported yet. Try to import the page into 4594 * the protected guest. 4595 */ 4596 rc = kvm_s390_pv_convert_to_secure(vcpu->kvm, gaddr); 4597 if (rc == -EINVAL) 4598 send_sig(SIGSEGV, current, 0); 4599 if (rc != -ENXIO) 4600 break; 4601 foll = FOLL_WRITE; 4602 fallthrough; 4603 case PGM_PROTECTION: 4604 case PGM_SEGMENT_TRANSLATION: 4605 case PGM_PAGE_TRANSLATION: 4606 case PGM_ASCE_TYPE: 4607 case PGM_REGION_FIRST_TRANS: 4608 case PGM_REGION_SECOND_TRANS: 4609 case PGM_REGION_THIRD_TRANS: 4610 kvm_s390_assert_primary_as(vcpu); 4611 return vcpu_dat_fault_handler(vcpu, gaddr, foll); 4612 default: 4613 KVM_BUG(1, vcpu->kvm, "Unexpected program interrupt 0x%x, TEID 0x%016lx", 4614 current->thread.gmap_int_code, current->thread.gmap_teid.val); 4615 send_sig(SIGSEGV, current, 0); 4616 break; 4617 } 4618 return 0; 4619 } 4620 4621 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason) 4622 { 4623 struct mcck_volatile_info *mcck_info; 4624 struct sie_page *sie_page; 4625 int rc; 4626 4627 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d", 4628 vcpu->arch.sie_block->icptcode); 4629 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode); 4630 4631 if (guestdbg_enabled(vcpu)) 4632 kvm_s390_restore_guest_per_regs(vcpu); 4633 4634 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14; 4635 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15; 4636 4637 if (exit_reason == -EINTR) { 4638 VCPU_EVENT(vcpu, 3, "%s", "machine check"); 4639 sie_page = container_of(vcpu->arch.sie_block, 4640 struct sie_page, sie_block); 4641 mcck_info = &sie_page->mcck_info; 4642 kvm_s390_reinject_machine_check(vcpu, mcck_info); 4643 return 0; 4644 } 4645 4646 if (vcpu->arch.sie_block->icptcode > 0) { 4647 rc = kvm_handle_sie_intercept(vcpu); 4648 4649 if (rc != -EOPNOTSUPP) 4650 return rc; 4651 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC; 4652 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode; 4653 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa; 4654 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb; 4655 return -EREMOTE; 4656 } 4657 4658 return vcpu_post_run_handle_fault(vcpu); 4659 } 4660 4661 int noinstr kvm_s390_enter_exit_sie(struct kvm_s390_sie_block *scb, 4662 u64 *gprs, unsigned long gasce) 4663 { 4664 int ret; 4665 4666 guest_state_enter_irqoff(); 4667 4668 /* 4669 * The guest_state_{enter,exit}_irqoff() functions inform lockdep and 4670 * tracing that entry to the guest will enable host IRQs, and exit from 4671 * the guest will disable host IRQs. 4672 */ 4673 ret = sie64a(scb, gprs, gasce); 4674 4675 guest_state_exit_irqoff(); 4676 4677 return ret; 4678 } 4679 4680 #define PSW_INT_MASK (PSW_MASK_EXT | PSW_MASK_IO | PSW_MASK_MCHECK) 4681 static int __vcpu_run(struct kvm_vcpu *vcpu) 4682 { 4683 int rc, exit_reason; 4684 struct sie_page *sie_page = (struct sie_page *)vcpu->arch.sie_block; 4685 4686 /* 4687 * We try to hold kvm->srcu during most of vcpu_run (except when run- 4688 * ning the guest), so that memslots (and other stuff) are protected 4689 */ 4690 kvm_vcpu_srcu_read_lock(vcpu); 4691 4692 while (true) { 4693 rc = vcpu_pre_run(vcpu); 4694 kvm_vcpu_srcu_read_unlock(vcpu); 4695 if (rc || guestdbg_exit_pending(vcpu)) 4696 break; 4697 4698 /* 4699 * As PF_VCPU will be used in fault handler, between 4700 * guest_timing_enter_irqoff and guest_timing_exit_irqoff 4701 * should be no uaccess. 4702 */ 4703 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 4704 memcpy(sie_page->pv_grregs, 4705 vcpu->run->s.regs.gprs, 4706 sizeof(sie_page->pv_grregs)); 4707 } 4708 4709 xfer_to_guest_mode_check: 4710 local_irq_disable(); 4711 xfer_to_guest_mode_prepare(); 4712 if (xfer_to_guest_mode_work_pending()) { 4713 local_irq_enable(); 4714 rc = kvm_xfer_to_guest_mode_handle_work(vcpu); 4715 if (rc) 4716 break; 4717 goto xfer_to_guest_mode_check; 4718 } 4719 4720 guest_timing_enter_irqoff(); 4721 __disable_cpu_timer_accounting(vcpu); 4722 4723 exit_reason = kvm_s390_enter_exit_sie(vcpu->arch.sie_block, 4724 vcpu->run->s.regs.gprs, 4725 vcpu->arch.gmap->asce.val); 4726 4727 __enable_cpu_timer_accounting(vcpu); 4728 guest_timing_exit_irqoff(); 4729 local_irq_enable(); 4730 4731 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 4732 memcpy(vcpu->run->s.regs.gprs, 4733 sie_page->pv_grregs, 4734 sizeof(sie_page->pv_grregs)); 4735 /* 4736 * We're not allowed to inject interrupts on intercepts 4737 * that leave the guest state in an "in-between" state 4738 * where the next SIE entry will do a continuation. 4739 * Fence interrupts in our "internal" PSW. 4740 */ 4741 if (vcpu->arch.sie_block->icptcode == ICPT_PV_INSTR || 4742 vcpu->arch.sie_block->icptcode == ICPT_PV_PREF) { 4743 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK; 4744 } 4745 } 4746 kvm_vcpu_srcu_read_lock(vcpu); 4747 4748 rc = vcpu_post_run(vcpu, exit_reason); 4749 if (rc || guestdbg_exit_pending(vcpu)) { 4750 kvm_vcpu_srcu_read_unlock(vcpu); 4751 break; 4752 } 4753 } 4754 4755 return rc; 4756 } 4757 4758 static void sync_regs_fmt2(struct kvm_vcpu *vcpu) 4759 { 4760 struct kvm_run *kvm_run = vcpu->run; 4761 struct runtime_instr_cb *riccb; 4762 struct gs_cb *gscb; 4763 4764 riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb; 4765 gscb = (struct gs_cb *) &kvm_run->s.regs.gscb; 4766 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask; 4767 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr; 4768 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) { 4769 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr; 4770 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp; 4771 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea; 4772 } 4773 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) { 4774 vcpu->arch.pfault_token = kvm_run->s.regs.pft; 4775 vcpu->arch.pfault_select = kvm_run->s.regs.pfs; 4776 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc; 4777 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) 4778 kvm_clear_async_pf_completion_queue(vcpu); 4779 } 4780 if (kvm_run->kvm_dirty_regs & KVM_SYNC_DIAG318) { 4781 vcpu->arch.diag318_info.val = kvm_run->s.regs.diag318; 4782 vcpu->arch.sie_block->cpnc = vcpu->arch.diag318_info.cpnc; 4783 VCPU_EVENT(vcpu, 3, "setting cpnc to %d", vcpu->arch.diag318_info.cpnc); 4784 } 4785 /* 4786 * If userspace sets the riccb (e.g. after migration) to a valid state, 4787 * we should enable RI here instead of doing the lazy enablement. 4788 */ 4789 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) && 4790 test_kvm_facility(vcpu->kvm, 64) && 4791 riccb->v && 4792 !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) { 4793 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)"); 4794 vcpu->arch.sie_block->ecb3 |= ECB3_RI; 4795 } 4796 /* 4797 * If userspace sets the gscb (e.g. after migration) to non-zero, 4798 * we should enable GS here instead of doing the lazy enablement. 4799 */ 4800 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) && 4801 test_kvm_facility(vcpu->kvm, 133) && 4802 gscb->gssm && 4803 !vcpu->arch.gs_enabled) { 4804 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)"); 4805 vcpu->arch.sie_block->ecb |= ECB_GS; 4806 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT; 4807 vcpu->arch.gs_enabled = 1; 4808 } 4809 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) && 4810 test_kvm_facility(vcpu->kvm, 82)) { 4811 vcpu->arch.sie_block->fpf &= ~FPF_BPBC; 4812 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0; 4813 } 4814 if (cpu_has_gs()) { 4815 preempt_disable(); 4816 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT); 4817 if (current->thread.gs_cb) { 4818 vcpu->arch.host_gscb = current->thread.gs_cb; 4819 save_gs_cb(vcpu->arch.host_gscb); 4820 } 4821 if (vcpu->arch.gs_enabled) { 4822 current->thread.gs_cb = (struct gs_cb *) 4823 &vcpu->run->s.regs.gscb; 4824 restore_gs_cb(current->thread.gs_cb); 4825 } 4826 preempt_enable(); 4827 } 4828 /* SIE will load etoken directly from SDNX and therefore kvm_run */ 4829 } 4830 4831 static void sync_regs(struct kvm_vcpu *vcpu) 4832 { 4833 struct kvm_run *kvm_run = vcpu->run; 4834 4835 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX) 4836 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix); 4837 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) { 4838 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128); 4839 /* some control register changes require a tlb flush */ 4840 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 4841 } 4842 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) { 4843 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm); 4844 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc; 4845 } 4846 save_access_regs(vcpu->arch.host_acrs); 4847 restore_access_regs(vcpu->run->s.regs.acrs); 4848 vcpu->arch.acrs_loaded = true; 4849 kvm_s390_fpu_load(vcpu->run); 4850 /* Sync fmt2 only data */ 4851 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu))) { 4852 sync_regs_fmt2(vcpu); 4853 } else { 4854 /* 4855 * In several places we have to modify our internal view to 4856 * not do things that are disallowed by the ultravisor. For 4857 * example we must not inject interrupts after specific exits 4858 * (e.g. 112 prefix page not secure). We do this by turning 4859 * off the machine check, external and I/O interrupt bits 4860 * of our PSW copy. To avoid getting validity intercepts, we 4861 * do only accept the condition code from userspace. 4862 */ 4863 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_CC; 4864 vcpu->arch.sie_block->gpsw.mask |= kvm_run->psw_mask & 4865 PSW_MASK_CC; 4866 } 4867 4868 kvm_run->kvm_dirty_regs = 0; 4869 } 4870 4871 static void store_regs_fmt2(struct kvm_vcpu *vcpu) 4872 { 4873 struct kvm_run *kvm_run = vcpu->run; 4874 4875 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr; 4876 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp; 4877 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea; 4878 kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC; 4879 kvm_run->s.regs.diag318 = vcpu->arch.diag318_info.val; 4880 if (cpu_has_gs()) { 4881 preempt_disable(); 4882 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT); 4883 if (vcpu->arch.gs_enabled) 4884 save_gs_cb(current->thread.gs_cb); 4885 current->thread.gs_cb = vcpu->arch.host_gscb; 4886 restore_gs_cb(vcpu->arch.host_gscb); 4887 if (!vcpu->arch.host_gscb) 4888 local_ctl_clear_bit(2, CR2_GUARDED_STORAGE_BIT); 4889 vcpu->arch.host_gscb = NULL; 4890 preempt_enable(); 4891 } 4892 /* SIE will save etoken directly into SDNX and therefore kvm_run */ 4893 } 4894 4895 static void store_regs(struct kvm_vcpu *vcpu) 4896 { 4897 struct kvm_run *kvm_run = vcpu->run; 4898 4899 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask; 4900 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr; 4901 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu); 4902 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128); 4903 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu); 4904 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc; 4905 kvm_run->s.regs.pft = vcpu->arch.pfault_token; 4906 kvm_run->s.regs.pfs = vcpu->arch.pfault_select; 4907 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare; 4908 save_access_regs(vcpu->run->s.regs.acrs); 4909 restore_access_regs(vcpu->arch.host_acrs); 4910 vcpu->arch.acrs_loaded = false; 4911 kvm_s390_fpu_store(vcpu->run); 4912 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu))) 4913 store_regs_fmt2(vcpu); 4914 } 4915 4916 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) 4917 { 4918 struct kvm_run *kvm_run = vcpu->run; 4919 DECLARE_KERNEL_FPU_ONSTACK32(fpu); 4920 int rc; 4921 4922 /* 4923 * Running a VM while dumping always has the potential to 4924 * produce inconsistent dump data. But for PV vcpus a SIE 4925 * entry while dumping could also lead to a fatal validity 4926 * intercept which we absolutely want to avoid. 4927 */ 4928 if (vcpu->kvm->arch.pv.dumping) 4929 return -EINVAL; 4930 4931 if (!vcpu->wants_to_run) 4932 return -EINTR; 4933 4934 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_S390_VALID_FIELDS || 4935 kvm_run->kvm_dirty_regs & ~KVM_SYNC_S390_VALID_FIELDS) 4936 return -EINVAL; 4937 4938 vcpu_load(vcpu); 4939 4940 if (guestdbg_exit_pending(vcpu)) { 4941 kvm_s390_prepare_debug_exit(vcpu); 4942 rc = 0; 4943 goto out; 4944 } 4945 4946 kvm_sigset_activate(vcpu); 4947 4948 /* 4949 * no need to check the return value of vcpu_start as it can only have 4950 * an error for protvirt, but protvirt means user cpu state 4951 */ 4952 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) { 4953 kvm_s390_vcpu_start(vcpu); 4954 } else if (is_vcpu_stopped(vcpu)) { 4955 pr_err_ratelimited("can't run stopped vcpu %d\n", 4956 vcpu->vcpu_id); 4957 rc = -EINVAL; 4958 goto out; 4959 } 4960 4961 kernel_fpu_begin(&fpu, KERNEL_FPC | KERNEL_VXR); 4962 sync_regs(vcpu); 4963 enable_cpu_timer_accounting(vcpu); 4964 4965 might_fault(); 4966 rc = __vcpu_run(vcpu); 4967 4968 if (signal_pending(current) && !rc) { 4969 kvm_run->exit_reason = KVM_EXIT_INTR; 4970 vcpu->stat.signal_exits++; 4971 rc = -EINTR; 4972 } 4973 4974 if (guestdbg_exit_pending(vcpu) && !rc) { 4975 kvm_s390_prepare_debug_exit(vcpu); 4976 rc = 0; 4977 } 4978 4979 if (rc == -EREMOTE) { 4980 /* userspace support is needed, kvm_run has been prepared */ 4981 rc = 0; 4982 } 4983 4984 disable_cpu_timer_accounting(vcpu); 4985 store_regs(vcpu); 4986 kernel_fpu_end(&fpu, KERNEL_FPC | KERNEL_VXR); 4987 4988 kvm_sigset_deactivate(vcpu); 4989 4990 vcpu->stat.exit_userspace++; 4991 out: 4992 vcpu_put(vcpu); 4993 return rc; 4994 } 4995 4996 /* 4997 * store status at address 4998 * we use have two special cases: 4999 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit 5000 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix 5001 */ 5002 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa) 5003 { 5004 unsigned char archmode = 1; 5005 freg_t fprs[NUM_FPRS]; 5006 unsigned int px; 5007 u64 clkcomp, cputm; 5008 int rc; 5009 5010 px = kvm_s390_get_prefix(vcpu); 5011 if (gpa == KVM_S390_STORE_STATUS_NOADDR) { 5012 if (write_guest_abs(vcpu, 163, &archmode, 1)) 5013 return -EFAULT; 5014 gpa = 0; 5015 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) { 5016 if (write_guest_real(vcpu, 163, &archmode, 1)) 5017 return -EFAULT; 5018 gpa = px; 5019 } else 5020 gpa -= __LC_FPREGS_SAVE_AREA; 5021 5022 /* manually convert vector registers if necessary */ 5023 if (cpu_has_vx()) { 5024 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs); 5025 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA, 5026 fprs, 128); 5027 } else { 5028 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA, 5029 vcpu->run->s.regs.fprs, 128); 5030 } 5031 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA, 5032 vcpu->run->s.regs.gprs, 128); 5033 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA, 5034 &vcpu->arch.sie_block->gpsw, 16); 5035 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA, 5036 &px, 4); 5037 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA, 5038 &vcpu->run->s.regs.fpc, 4); 5039 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA, 5040 &vcpu->arch.sie_block->todpr, 4); 5041 cputm = kvm_s390_get_cpu_timer(vcpu); 5042 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA, 5043 &cputm, 8); 5044 clkcomp = vcpu->arch.sie_block->ckc >> 8; 5045 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA, 5046 &clkcomp, 8); 5047 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA, 5048 &vcpu->run->s.regs.acrs, 64); 5049 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA, 5050 &vcpu->arch.sie_block->gcr, 128); 5051 return rc ? -EFAULT : 0; 5052 } 5053 5054 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr) 5055 { 5056 /* 5057 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy 5058 * switch in the run ioctl. Let's update our copies before we save 5059 * it into the save area 5060 */ 5061 kvm_s390_fpu_store(vcpu->run); 5062 save_access_regs(vcpu->run->s.regs.acrs); 5063 5064 return kvm_s390_store_status_unloaded(vcpu, addr); 5065 } 5066 5067 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu) 5068 { 5069 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu); 5070 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu); 5071 } 5072 5073 static void __disable_ibs_on_all_vcpus(struct kvm *kvm) 5074 { 5075 unsigned long i; 5076 struct kvm_vcpu *vcpu; 5077 5078 kvm_for_each_vcpu(i, vcpu, kvm) { 5079 __disable_ibs_on_vcpu(vcpu); 5080 } 5081 } 5082 5083 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu) 5084 { 5085 if (!sclp.has_ibs) 5086 return; 5087 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu); 5088 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu); 5089 } 5090 5091 int kvm_s390_vcpu_start(struct kvm_vcpu *vcpu) 5092 { 5093 int i, online_vcpus, r = 0, started_vcpus = 0; 5094 5095 if (!is_vcpu_stopped(vcpu)) 5096 return 0; 5097 5098 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1); 5099 /* Only one cpu at a time may enter/leave the STOPPED state. */ 5100 spin_lock(&vcpu->kvm->arch.start_stop_lock); 5101 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus); 5102 5103 /* Let's tell the UV that we want to change into the operating state */ 5104 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 5105 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR); 5106 if (r) { 5107 spin_unlock(&vcpu->kvm->arch.start_stop_lock); 5108 return r; 5109 } 5110 } 5111 5112 for (i = 0; i < online_vcpus; i++) { 5113 if (!is_vcpu_stopped(kvm_get_vcpu(vcpu->kvm, i))) 5114 started_vcpus++; 5115 } 5116 5117 if (started_vcpus == 0) { 5118 /* we're the only active VCPU -> speed it up */ 5119 __enable_ibs_on_vcpu(vcpu); 5120 } else if (started_vcpus == 1) { 5121 /* 5122 * As we are starting a second VCPU, we have to disable 5123 * the IBS facility on all VCPUs to remove potentially 5124 * outstanding ENABLE requests. 5125 */ 5126 __disable_ibs_on_all_vcpus(vcpu->kvm); 5127 } 5128 5129 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED); 5130 /* 5131 * The real PSW might have changed due to a RESTART interpreted by the 5132 * ultravisor. We block all interrupts and let the next sie exit 5133 * refresh our view. 5134 */ 5135 if (kvm_s390_pv_cpu_is_protected(vcpu)) 5136 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK; 5137 /* 5138 * Another VCPU might have used IBS while we were offline. 5139 * Let's play safe and flush the VCPU at startup. 5140 */ 5141 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 5142 spin_unlock(&vcpu->kvm->arch.start_stop_lock); 5143 return 0; 5144 } 5145 5146 int kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu) 5147 { 5148 int i, online_vcpus, r = 0, started_vcpus = 0; 5149 struct kvm_vcpu *started_vcpu = NULL; 5150 5151 if (is_vcpu_stopped(vcpu)) 5152 return 0; 5153 5154 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0); 5155 /* Only one cpu at a time may enter/leave the STOPPED state. */ 5156 spin_lock(&vcpu->kvm->arch.start_stop_lock); 5157 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus); 5158 5159 /* Let's tell the UV that we want to change into the stopped state */ 5160 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 5161 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_STP); 5162 if (r) { 5163 spin_unlock(&vcpu->kvm->arch.start_stop_lock); 5164 return r; 5165 } 5166 } 5167 5168 /* 5169 * Set the VCPU to STOPPED and THEN clear the interrupt flag, 5170 * now that the SIGP STOP and SIGP STOP AND STORE STATUS orders 5171 * have been fully processed. This will ensure that the VCPU 5172 * is kept BUSY if another VCPU is inquiring with SIGP SENSE. 5173 */ 5174 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED); 5175 kvm_s390_clear_stop_irq(vcpu); 5176 5177 __disable_ibs_on_vcpu(vcpu); 5178 5179 for (i = 0; i < online_vcpus; i++) { 5180 struct kvm_vcpu *tmp = kvm_get_vcpu(vcpu->kvm, i); 5181 5182 if (!is_vcpu_stopped(tmp)) { 5183 started_vcpus++; 5184 started_vcpu = tmp; 5185 } 5186 } 5187 5188 if (started_vcpus == 1) { 5189 /* 5190 * As we only have one VCPU left, we want to enable the 5191 * IBS facility for that VCPU to speed it up. 5192 */ 5193 __enable_ibs_on_vcpu(started_vcpu); 5194 } 5195 5196 spin_unlock(&vcpu->kvm->arch.start_stop_lock); 5197 return 0; 5198 } 5199 5200 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, 5201 struct kvm_enable_cap *cap) 5202 { 5203 int r; 5204 5205 if (cap->flags) 5206 return -EINVAL; 5207 5208 switch (cap->cap) { 5209 case KVM_CAP_S390_CSS_SUPPORT: 5210 if (!vcpu->kvm->arch.css_support) { 5211 vcpu->kvm->arch.css_support = 1; 5212 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support"); 5213 trace_kvm_s390_enable_css(vcpu->kvm); 5214 } 5215 r = 0; 5216 break; 5217 default: 5218 r = -EINVAL; 5219 break; 5220 } 5221 return r; 5222 } 5223 5224 static long kvm_s390_vcpu_sida_op(struct kvm_vcpu *vcpu, 5225 struct kvm_s390_mem_op *mop) 5226 { 5227 void __user *uaddr = (void __user *)mop->buf; 5228 void *sida_addr; 5229 int r = 0; 5230 5231 if (mop->flags || !mop->size) 5232 return -EINVAL; 5233 if (mop->size + mop->sida_offset < mop->size) 5234 return -EINVAL; 5235 if (mop->size + mop->sida_offset > sida_size(vcpu->arch.sie_block)) 5236 return -E2BIG; 5237 if (!kvm_s390_pv_cpu_is_protected(vcpu)) 5238 return -EINVAL; 5239 5240 sida_addr = (char *)sida_addr(vcpu->arch.sie_block) + mop->sida_offset; 5241 5242 switch (mop->op) { 5243 case KVM_S390_MEMOP_SIDA_READ: 5244 if (copy_to_user(uaddr, sida_addr, mop->size)) 5245 r = -EFAULT; 5246 5247 break; 5248 case KVM_S390_MEMOP_SIDA_WRITE: 5249 if (copy_from_user(sida_addr, uaddr, mop->size)) 5250 r = -EFAULT; 5251 break; 5252 } 5253 return r; 5254 } 5255 5256 static long kvm_s390_vcpu_mem_op(struct kvm_vcpu *vcpu, 5257 struct kvm_s390_mem_op *mop) 5258 { 5259 void __user *uaddr = (void __user *)mop->buf; 5260 void *tmpbuf __free(kvfree) = NULL; 5261 enum gacc_mode acc_mode; 5262 int r; 5263 5264 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_INJECT_EXCEPTION | 5265 KVM_S390_MEMOP_F_CHECK_ONLY | 5266 KVM_S390_MEMOP_F_SKEY_PROTECTION); 5267 if (r) 5268 return r; 5269 if (mop->ar >= NUM_ACRS) 5270 return -EINVAL; 5271 if (kvm_s390_pv_cpu_is_protected(vcpu)) 5272 return -EINVAL; 5273 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) { 5274 tmpbuf = vmalloc(mop->size); 5275 if (!tmpbuf) 5276 return -ENOMEM; 5277 } 5278 5279 acc_mode = mop->op == KVM_S390_MEMOP_LOGICAL_READ ? GACC_FETCH : GACC_STORE; 5280 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) { 5281 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size, 5282 acc_mode, mop->key); 5283 } else if (acc_mode == GACC_FETCH) { 5284 r = read_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf, 5285 mop->size, mop->key); 5286 if (!r && copy_to_user(uaddr, tmpbuf, mop->size)) 5287 return -EFAULT; 5288 } else { 5289 if (copy_from_user(tmpbuf, uaddr, mop->size)) 5290 return -EFAULT; 5291 r = write_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf, 5292 mop->size, mop->key); 5293 } 5294 5295 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0) 5296 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm); 5297 5298 return r; 5299 } 5300 5301 static long kvm_s390_vcpu_memsida_op(struct kvm_vcpu *vcpu, 5302 struct kvm_s390_mem_op *mop) 5303 { 5304 int r, srcu_idx; 5305 5306 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 5307 5308 switch (mop->op) { 5309 case KVM_S390_MEMOP_LOGICAL_READ: 5310 case KVM_S390_MEMOP_LOGICAL_WRITE: 5311 r = kvm_s390_vcpu_mem_op(vcpu, mop); 5312 break; 5313 case KVM_S390_MEMOP_SIDA_READ: 5314 case KVM_S390_MEMOP_SIDA_WRITE: 5315 /* we are locked against sida going away by the vcpu->mutex */ 5316 r = kvm_s390_vcpu_sida_op(vcpu, mop); 5317 break; 5318 default: 5319 r = -EINVAL; 5320 } 5321 5322 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx); 5323 return r; 5324 } 5325 5326 long kvm_arch_vcpu_unlocked_ioctl(struct file *filp, unsigned int ioctl, 5327 unsigned long arg) 5328 { 5329 struct kvm_vcpu *vcpu = filp->private_data; 5330 void __user *argp = (void __user *)arg; 5331 int rc; 5332 5333 switch (ioctl) { 5334 case KVM_S390_IRQ: { 5335 struct kvm_s390_irq s390irq; 5336 5337 if (copy_from_user(&s390irq, argp, sizeof(s390irq))) 5338 return -EFAULT; 5339 rc = kvm_s390_inject_vcpu(vcpu, &s390irq); 5340 break; 5341 } 5342 case KVM_S390_INTERRUPT: { 5343 struct kvm_s390_interrupt s390int; 5344 struct kvm_s390_irq s390irq = {}; 5345 5346 if (copy_from_user(&s390int, argp, sizeof(s390int))) 5347 return -EFAULT; 5348 if (s390int_to_s390irq(&s390int, &s390irq)) 5349 return -EINVAL; 5350 rc = kvm_s390_inject_vcpu(vcpu, &s390irq); 5351 break; 5352 } 5353 default: 5354 rc = -ENOIOCTLCMD; 5355 break; 5356 } 5357 5358 /* 5359 * To simplify single stepping of userspace-emulated instructions, 5360 * KVM_EXIT_S390_SIEIC exit sets KVM_GUESTDBG_EXIT_PENDING (see 5361 * should_handle_per_ifetch()). However, if userspace emulation injects 5362 * an interrupt, it needs to be cleared, so that KVM_EXIT_DEBUG happens 5363 * after (and not before) the interrupt delivery. 5364 */ 5365 if (!rc) 5366 vcpu->guest_debug &= ~KVM_GUESTDBG_EXIT_PENDING; 5367 5368 return rc; 5369 } 5370 5371 static int kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu *vcpu, 5372 struct kvm_pv_cmd *cmd) 5373 { 5374 struct kvm_s390_pv_dmp dmp; 5375 void *data; 5376 int ret; 5377 5378 /* Dump initialization is a prerequisite */ 5379 if (!vcpu->kvm->arch.pv.dumping) 5380 return -EINVAL; 5381 5382 if (copy_from_user(&dmp, (__u8 __user *)cmd->data, sizeof(dmp))) 5383 return -EFAULT; 5384 5385 /* We only handle this subcmd right now */ 5386 if (dmp.subcmd != KVM_PV_DUMP_CPU) 5387 return -EINVAL; 5388 5389 /* CPU dump length is the same as create cpu storage donation. */ 5390 if (dmp.buff_len != uv_info.guest_cpu_stor_len) 5391 return -EINVAL; 5392 5393 data = kvzalloc(uv_info.guest_cpu_stor_len, GFP_KERNEL); 5394 if (!data) 5395 return -ENOMEM; 5396 5397 ret = kvm_s390_pv_dump_cpu(vcpu, data, &cmd->rc, &cmd->rrc); 5398 5399 VCPU_EVENT(vcpu, 3, "PROTVIRT DUMP CPU %d rc %x rrc %x", 5400 vcpu->vcpu_id, cmd->rc, cmd->rrc); 5401 5402 if (ret) 5403 ret = -EINVAL; 5404 5405 /* On success copy over the dump data */ 5406 if (!ret && copy_to_user((__u8 __user *)dmp.buff_addr, data, uv_info.guest_cpu_stor_len)) 5407 ret = -EFAULT; 5408 5409 kvfree(data); 5410 return ret; 5411 } 5412 5413 long kvm_arch_vcpu_ioctl(struct file *filp, 5414 unsigned int ioctl, unsigned long arg) 5415 { 5416 struct kvm_vcpu *vcpu = filp->private_data; 5417 void __user *argp = (void __user *)arg; 5418 int idx; 5419 long r; 5420 u16 rc, rrc; 5421 5422 vcpu_load(vcpu); 5423 5424 switch (ioctl) { 5425 case KVM_S390_STORE_STATUS: 5426 idx = srcu_read_lock(&vcpu->kvm->srcu); 5427 r = kvm_s390_store_status_unloaded(vcpu, arg); 5428 srcu_read_unlock(&vcpu->kvm->srcu, idx); 5429 break; 5430 case KVM_S390_SET_INITIAL_PSW: { 5431 psw_t psw; 5432 5433 r = -EFAULT; 5434 if (copy_from_user(&psw, argp, sizeof(psw))) 5435 break; 5436 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw); 5437 break; 5438 } 5439 case KVM_S390_CLEAR_RESET: 5440 r = 0; 5441 kvm_arch_vcpu_ioctl_clear_reset(vcpu); 5442 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 5443 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu), 5444 UVC_CMD_CPU_RESET_CLEAR, &rc, &rrc); 5445 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET CLEAR VCPU: rc %x rrc %x", 5446 rc, rrc); 5447 } 5448 break; 5449 case KVM_S390_INITIAL_RESET: 5450 r = 0; 5451 kvm_arch_vcpu_ioctl_initial_reset(vcpu); 5452 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 5453 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu), 5454 UVC_CMD_CPU_RESET_INITIAL, 5455 &rc, &rrc); 5456 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET INITIAL VCPU: rc %x rrc %x", 5457 rc, rrc); 5458 } 5459 break; 5460 case KVM_S390_NORMAL_RESET: 5461 r = 0; 5462 kvm_arch_vcpu_ioctl_normal_reset(vcpu); 5463 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 5464 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu), 5465 UVC_CMD_CPU_RESET, &rc, &rrc); 5466 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET NORMAL VCPU: rc %x rrc %x", 5467 rc, rrc); 5468 } 5469 break; 5470 case KVM_SET_ONE_REG: 5471 case KVM_GET_ONE_REG: { 5472 struct kvm_one_reg reg; 5473 r = -EINVAL; 5474 if (kvm_s390_pv_cpu_is_protected(vcpu)) 5475 break; 5476 r = -EFAULT; 5477 if (copy_from_user(®, argp, sizeof(reg))) 5478 break; 5479 if (ioctl == KVM_SET_ONE_REG) 5480 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®); 5481 else 5482 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®); 5483 break; 5484 } 5485 #ifdef CONFIG_KVM_S390_UCONTROL 5486 case KVM_S390_UCAS_MAP: { 5487 struct kvm_s390_ucas_mapping ucas; 5488 5489 r = -EFAULT; 5490 if (copy_from_user(&ucas, argp, sizeof(ucas))) 5491 break; 5492 5493 r = -EINVAL; 5494 if (!kvm_is_ucontrol(vcpu->kvm)) 5495 break; 5496 if (!IS_ALIGNED(ucas.user_addr | ucas.vcpu_addr | ucas.length, _SEGMENT_SIZE)) 5497 break; 5498 5499 r = gmap_ucas_map(vcpu->arch.gmap, gpa_to_gfn(ucas.user_addr), 5500 gpa_to_gfn(ucas.vcpu_addr), 5501 ucas.length >> _SEGMENT_SHIFT); 5502 break; 5503 } 5504 case KVM_S390_UCAS_UNMAP: { 5505 struct kvm_s390_ucas_mapping ucas; 5506 5507 r = -EFAULT; 5508 if (copy_from_user(&ucas, argp, sizeof(ucas))) 5509 break; 5510 5511 r = -EINVAL; 5512 if (!kvm_is_ucontrol(vcpu->kvm)) 5513 break; 5514 if (!IS_ALIGNED(ucas.vcpu_addr | ucas.length, _SEGMENT_SIZE)) 5515 break; 5516 5517 gmap_ucas_unmap(vcpu->arch.gmap, gpa_to_gfn(ucas.vcpu_addr), 5518 ucas.length >> _SEGMENT_SHIFT); 5519 r = 0; 5520 break; 5521 } 5522 #endif 5523 case KVM_S390_VCPU_FAULT: { 5524 idx = srcu_read_lock(&vcpu->kvm->srcu); 5525 r = vcpu_dat_fault_handler(vcpu, arg, 0); 5526 srcu_read_unlock(&vcpu->kvm->srcu, idx); 5527 break; 5528 } 5529 case KVM_ENABLE_CAP: 5530 { 5531 struct kvm_enable_cap cap; 5532 r = -EFAULT; 5533 if (copy_from_user(&cap, argp, sizeof(cap))) 5534 break; 5535 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); 5536 break; 5537 } 5538 case KVM_S390_MEM_OP: { 5539 struct kvm_s390_mem_op mem_op; 5540 5541 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0) 5542 r = kvm_s390_vcpu_memsida_op(vcpu, &mem_op); 5543 else 5544 r = -EFAULT; 5545 break; 5546 } 5547 case KVM_S390_SET_IRQ_STATE: { 5548 struct kvm_s390_irq_state irq_state; 5549 5550 r = -EFAULT; 5551 if (copy_from_user(&irq_state, argp, sizeof(irq_state))) 5552 break; 5553 if (irq_state.len > VCPU_IRQS_MAX_BUF || 5554 irq_state.len == 0 || 5555 irq_state.len % sizeof(struct kvm_s390_irq) > 0) { 5556 r = -EINVAL; 5557 break; 5558 } 5559 /* do not use irq_state.flags, it will break old QEMUs */ 5560 r = kvm_s390_set_irq_state(vcpu, 5561 (void __user *) irq_state.buf, 5562 irq_state.len); 5563 break; 5564 } 5565 case KVM_S390_GET_IRQ_STATE: { 5566 struct kvm_s390_irq_state irq_state; 5567 5568 r = -EFAULT; 5569 if (copy_from_user(&irq_state, argp, sizeof(irq_state))) 5570 break; 5571 if (irq_state.len == 0) { 5572 r = -EINVAL; 5573 break; 5574 } 5575 /* do not use irq_state.flags, it will break old QEMUs */ 5576 r = kvm_s390_get_irq_state(vcpu, 5577 (__u8 __user *) irq_state.buf, 5578 irq_state.len); 5579 break; 5580 } 5581 case KVM_S390_PV_CPU_COMMAND: { 5582 struct kvm_pv_cmd cmd; 5583 5584 r = -EINVAL; 5585 if (!is_prot_virt_host()) 5586 break; 5587 5588 r = -EFAULT; 5589 if (copy_from_user(&cmd, argp, sizeof(cmd))) 5590 break; 5591 5592 r = -EINVAL; 5593 if (cmd.flags) 5594 break; 5595 5596 /* We only handle this cmd right now */ 5597 if (cmd.cmd != KVM_PV_DUMP) 5598 break; 5599 5600 r = kvm_s390_handle_pv_vcpu_dump(vcpu, &cmd); 5601 5602 /* Always copy over UV rc / rrc data */ 5603 if (copy_to_user((__u8 __user *)argp, &cmd.rc, 5604 sizeof(cmd.rc) + sizeof(cmd.rrc))) 5605 r = -EFAULT; 5606 break; 5607 } 5608 default: 5609 r = -ENOTTY; 5610 } 5611 5612 vcpu_put(vcpu); 5613 return r; 5614 } 5615 5616 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 5617 { 5618 #ifdef CONFIG_KVM_S390_UCONTROL 5619 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET) 5620 && (kvm_is_ucontrol(vcpu->kvm))) { 5621 vmf->page = virt_to_page(vcpu->arch.sie_block); 5622 get_page(vmf->page); 5623 return 0; 5624 } 5625 #endif 5626 return VM_FAULT_SIGBUS; 5627 } 5628 5629 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm) 5630 { 5631 return true; 5632 } 5633 5634 /* Section: memory related */ 5635 int kvm_arch_prepare_memory_region(struct kvm *kvm, 5636 const struct kvm_memory_slot *old, 5637 struct kvm_memory_slot *new, 5638 enum kvm_mr_change change) 5639 { 5640 gpa_t size; 5641 5642 if (kvm_is_ucontrol(kvm) && new->id < KVM_USER_MEM_SLOTS) 5643 return -EINVAL; 5644 5645 /* When we are protected, we should not change the memory slots */ 5646 if (kvm_s390_pv_get_handle(kvm)) 5647 return -EINVAL; 5648 5649 if (change != KVM_MR_DELETE && change != KVM_MR_FLAGS_ONLY) { 5650 /* 5651 * A few sanity checks. We can have memory slots which have to be 5652 * located/ended at a segment boundary (1MB). The memory in userland is 5653 * ok to be fragmented into various different vmas. It is okay to mmap() 5654 * and munmap() stuff in this slot after doing this call at any time 5655 */ 5656 5657 if (new->userspace_addr & 0xffffful) 5658 return -EINVAL; 5659 5660 size = new->npages * PAGE_SIZE; 5661 if (size & 0xffffful) 5662 return -EINVAL; 5663 5664 if ((new->base_gfn * PAGE_SIZE) + size > kvm->arch.mem_limit) 5665 return -EINVAL; 5666 } 5667 5668 if (!kvm->arch.migration_mode) 5669 return 0; 5670 5671 /* 5672 * Turn off migration mode when: 5673 * - userspace creates a new memslot with dirty logging off, 5674 * - userspace modifies an existing memslot (MOVE or FLAGS_ONLY) and 5675 * dirty logging is turned off. 5676 * Migration mode expects dirty page logging being enabled to store 5677 * its dirty bitmap. 5678 */ 5679 if (change != KVM_MR_DELETE && 5680 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) 5681 WARN(kvm_s390_vm_stop_migration(kvm), 5682 "Failed to stop migration mode"); 5683 5684 return 0; 5685 } 5686 5687 void kvm_arch_commit_memory_region(struct kvm *kvm, 5688 struct kvm_memory_slot *old, 5689 const struct kvm_memory_slot *new, 5690 enum kvm_mr_change change) 5691 { 5692 struct kvm_s390_mmu_cache *mc = NULL; 5693 int rc = 0; 5694 5695 if (change == KVM_MR_FLAGS_ONLY) 5696 return; 5697 5698 mc = kvm_s390_new_mmu_cache(); 5699 if (!mc) { 5700 rc = -ENOMEM; 5701 goto out; 5702 } 5703 5704 scoped_guard(write_lock, &kvm->mmu_lock) { 5705 switch (change) { 5706 case KVM_MR_DELETE: 5707 rc = dat_delete_slot(mc, kvm->arch.gmap->asce, old->base_gfn, old->npages); 5708 break; 5709 case KVM_MR_MOVE: 5710 rc = dat_delete_slot(mc, kvm->arch.gmap->asce, old->base_gfn, old->npages); 5711 if (rc) 5712 break; 5713 fallthrough; 5714 case KVM_MR_CREATE: 5715 rc = dat_create_slot(mc, kvm->arch.gmap->asce, new->base_gfn, new->npages); 5716 break; 5717 case KVM_MR_FLAGS_ONLY: 5718 break; 5719 default: 5720 WARN(1, "Unknown KVM MR CHANGE: %d\n", change); 5721 } 5722 } 5723 out: 5724 if (rc) 5725 pr_warn("failed to commit memory region\n"); 5726 kvm_s390_free_mmu_cache(mc); 5727 return; 5728 } 5729 5730 /** 5731 * kvm_test_age_gfn() - test young 5732 * @kvm: the kvm instance 5733 * @range: the range of guest addresses whose young status needs to be cleared 5734 * 5735 * Context: called by KVM common code without holding the kvm mmu lock 5736 * Return: true if any page in the given range is young, otherwise 0. 5737 */ 5738 bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range) 5739 { 5740 scoped_guard(read_lock, &kvm->mmu_lock) 5741 return dat_test_age_gfn(kvm->arch.gmap->asce, range->start, range->end); 5742 } 5743 5744 /** 5745 * kvm_age_gfn() - clear young 5746 * @kvm: the kvm instance 5747 * @range: the range of guest addresses whose young status needs to be cleared 5748 * 5749 * Context: called by KVM common code without holding the kvm mmu lock 5750 * Return: true if any page in the given range was young, otherwise 0. 5751 */ 5752 bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range) 5753 { 5754 scoped_guard(read_lock, &kvm->mmu_lock) 5755 return gmap_age_gfn(kvm->arch.gmap, range->start, range->end); 5756 } 5757 5758 /** 5759 * kvm_unmap_gfn_range() - Unmap a range of guest addresses 5760 * @kvm: the kvm instance 5761 * @range: the range of guest page frames to invalidate 5762 * 5763 * This function always returns false because every DAT table modification 5764 * has to use the appropriate DAT table manipulation instructions, which will 5765 * keep the TLB coherent, hence no additional TLB flush is ever required. 5766 * 5767 * Context: called by KVM common code with the kvm mmu write lock held 5768 * Return: false 5769 */ 5770 bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range) 5771 { 5772 return gmap_unmap_gfn_range(kvm->arch.gmap, range->slot, range->start, range->end); 5773 } 5774 5775 static inline unsigned long nonhyp_mask(int i) 5776 { 5777 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30; 5778 5779 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4); 5780 } 5781 5782 static int __init kvm_s390_init(void) 5783 { 5784 int i, r; 5785 5786 if (!sclp.has_sief2) { 5787 pr_info("SIE is not available\n"); 5788 return -ENODEV; 5789 } 5790 5791 for (i = 0; i < 16; i++) 5792 kvm_s390_fac_base[i] |= 5793 stfle_fac_list[i] & nonhyp_mask(i); 5794 5795 r = __kvm_s390_init(); 5796 if (r) 5797 return r; 5798 5799 r = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE); 5800 if (r) { 5801 __kvm_s390_exit(); 5802 return r; 5803 } 5804 return 0; 5805 } 5806 5807 static void __exit kvm_s390_exit(void) 5808 { 5809 kvm_exit(); 5810 5811 __kvm_s390_exit(); 5812 } 5813 5814 module_init(kvm_s390_init); 5815 module_exit(kvm_s390_exit); 5816 5817 /* 5818 * Enable autoloading of the kvm module. 5819 * Note that we add the module alias here instead of virt/kvm/kvm_main.c 5820 * since x86 takes a different approach. 5821 */ 5822 #include <linux/miscdevice.h> 5823 MODULE_ALIAS_MISCDEV(KVM_MINOR); 5824 MODULE_ALIAS("devname:kvm"); 5825