1 /* 2 * hosting zSeries kernel virtual machines 3 * 4 * Copyright IBM Corp. 2008, 2009 5 * 6 * This program is free software; you can redistribute it and/or modify 7 * it under the terms of the GNU General Public License (version 2 only) 8 * as published by the Free Software Foundation. 9 * 10 * Author(s): Carsten Otte <cotte@de.ibm.com> 11 * Christian Borntraeger <borntraeger@de.ibm.com> 12 * Heiko Carstens <heiko.carstens@de.ibm.com> 13 * Christian Ehrhardt <ehrhardt@de.ibm.com> 14 * Jason J. Herne <jjherne@us.ibm.com> 15 */ 16 17 #include <linux/compiler.h> 18 #include <linux/err.h> 19 #include <linux/fs.h> 20 #include <linux/hrtimer.h> 21 #include <linux/init.h> 22 #include <linux/kvm.h> 23 #include <linux/kvm_host.h> 24 #include <linux/module.h> 25 #include <linux/random.h> 26 #include <linux/slab.h> 27 #include <linux/timer.h> 28 #include <asm/asm-offsets.h> 29 #include <asm/lowcore.h> 30 #include <asm/pgtable.h> 31 #include <asm/nmi.h> 32 #include <asm/switch_to.h> 33 #include <asm/sclp.h> 34 #include "kvm-s390.h" 35 #include "gaccess.h" 36 37 #define CREATE_TRACE_POINTS 38 #include "trace.h" 39 #include "trace-s390.h" 40 41 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU 42 43 struct kvm_stats_debugfs_item debugfs_entries[] = { 44 { "userspace_handled", VCPU_STAT(exit_userspace) }, 45 { "exit_null", VCPU_STAT(exit_null) }, 46 { "exit_validity", VCPU_STAT(exit_validity) }, 47 { "exit_stop_request", VCPU_STAT(exit_stop_request) }, 48 { "exit_external_request", VCPU_STAT(exit_external_request) }, 49 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) }, 50 { "exit_instruction", VCPU_STAT(exit_instruction) }, 51 { "exit_program_interruption", VCPU_STAT(exit_program_interruption) }, 52 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) }, 53 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) }, 54 { "halt_wakeup", VCPU_STAT(halt_wakeup) }, 55 { "instruction_lctlg", VCPU_STAT(instruction_lctlg) }, 56 { "instruction_lctl", VCPU_STAT(instruction_lctl) }, 57 { "instruction_stctl", VCPU_STAT(instruction_stctl) }, 58 { "instruction_stctg", VCPU_STAT(instruction_stctg) }, 59 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) }, 60 { "deliver_external_call", VCPU_STAT(deliver_external_call) }, 61 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) }, 62 { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) }, 63 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) }, 64 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) }, 65 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) }, 66 { "deliver_program_interruption", VCPU_STAT(deliver_program_int) }, 67 { "exit_wait_state", VCPU_STAT(exit_wait_state) }, 68 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) }, 69 { "instruction_stidp", VCPU_STAT(instruction_stidp) }, 70 { "instruction_spx", VCPU_STAT(instruction_spx) }, 71 { "instruction_stpx", VCPU_STAT(instruction_stpx) }, 72 { "instruction_stap", VCPU_STAT(instruction_stap) }, 73 { "instruction_storage_key", VCPU_STAT(instruction_storage_key) }, 74 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) }, 75 { "instruction_stsch", VCPU_STAT(instruction_stsch) }, 76 { "instruction_chsc", VCPU_STAT(instruction_chsc) }, 77 { "instruction_essa", VCPU_STAT(instruction_essa) }, 78 { "instruction_stsi", VCPU_STAT(instruction_stsi) }, 79 { "instruction_stfl", VCPU_STAT(instruction_stfl) }, 80 { "instruction_tprot", VCPU_STAT(instruction_tprot) }, 81 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) }, 82 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) }, 83 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) }, 84 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) }, 85 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) }, 86 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) }, 87 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) }, 88 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) }, 89 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) }, 90 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) }, 91 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) }, 92 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) }, 93 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) }, 94 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) }, 95 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) }, 96 { "diagnose_10", VCPU_STAT(diagnose_10) }, 97 { "diagnose_44", VCPU_STAT(diagnose_44) }, 98 { "diagnose_9c", VCPU_STAT(diagnose_9c) }, 99 { NULL } 100 }; 101 102 /* upper facilities limit for kvm */ 103 unsigned long kvm_s390_fac_list_mask[] = { 104 0xff82fffbf4fc2000UL, 105 0x005c000000000000UL, 106 }; 107 108 unsigned long kvm_s390_fac_list_mask_size(void) 109 { 110 BUILD_BUG_ON(ARRAY_SIZE(kvm_s390_fac_list_mask) > S390_ARCH_FAC_MASK_SIZE_U64); 111 return ARRAY_SIZE(kvm_s390_fac_list_mask); 112 } 113 114 static struct gmap_notifier gmap_notifier; 115 116 /* Section: not file related */ 117 int kvm_arch_hardware_enable(void) 118 { 119 /* every s390 is virtualization enabled ;-) */ 120 return 0; 121 } 122 123 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address); 124 125 int kvm_arch_hardware_setup(void) 126 { 127 gmap_notifier.notifier_call = kvm_gmap_notifier; 128 gmap_register_ipte_notifier(&gmap_notifier); 129 return 0; 130 } 131 132 void kvm_arch_hardware_unsetup(void) 133 { 134 gmap_unregister_ipte_notifier(&gmap_notifier); 135 } 136 137 int kvm_arch_init(void *opaque) 138 { 139 /* Register floating interrupt controller interface. */ 140 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC); 141 } 142 143 /* Section: device related */ 144 long kvm_arch_dev_ioctl(struct file *filp, 145 unsigned int ioctl, unsigned long arg) 146 { 147 if (ioctl == KVM_S390_ENABLE_SIE) 148 return s390_enable_sie(); 149 return -EINVAL; 150 } 151 152 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 153 { 154 int r; 155 156 switch (ext) { 157 case KVM_CAP_S390_PSW: 158 case KVM_CAP_S390_GMAP: 159 case KVM_CAP_SYNC_MMU: 160 #ifdef CONFIG_KVM_S390_UCONTROL 161 case KVM_CAP_S390_UCONTROL: 162 #endif 163 case KVM_CAP_ASYNC_PF: 164 case KVM_CAP_SYNC_REGS: 165 case KVM_CAP_ONE_REG: 166 case KVM_CAP_ENABLE_CAP: 167 case KVM_CAP_S390_CSS_SUPPORT: 168 case KVM_CAP_IRQFD: 169 case KVM_CAP_IOEVENTFD: 170 case KVM_CAP_DEVICE_CTRL: 171 case KVM_CAP_ENABLE_CAP_VM: 172 case KVM_CAP_S390_IRQCHIP: 173 case KVM_CAP_VM_ATTRIBUTES: 174 case KVM_CAP_MP_STATE: 175 case KVM_CAP_S390_USER_SIGP: 176 r = 1; 177 break; 178 case KVM_CAP_NR_VCPUS: 179 case KVM_CAP_MAX_VCPUS: 180 r = KVM_MAX_VCPUS; 181 break; 182 case KVM_CAP_NR_MEMSLOTS: 183 r = KVM_USER_MEM_SLOTS; 184 break; 185 case KVM_CAP_S390_COW: 186 r = MACHINE_HAS_ESOP; 187 break; 188 default: 189 r = 0; 190 } 191 return r; 192 } 193 194 static void kvm_s390_sync_dirty_log(struct kvm *kvm, 195 struct kvm_memory_slot *memslot) 196 { 197 gfn_t cur_gfn, last_gfn; 198 unsigned long address; 199 struct gmap *gmap = kvm->arch.gmap; 200 201 down_read(&gmap->mm->mmap_sem); 202 /* Loop over all guest pages */ 203 last_gfn = memslot->base_gfn + memslot->npages; 204 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) { 205 address = gfn_to_hva_memslot(memslot, cur_gfn); 206 207 if (gmap_test_and_clear_dirty(address, gmap)) 208 mark_page_dirty(kvm, cur_gfn); 209 } 210 up_read(&gmap->mm->mmap_sem); 211 } 212 213 /* Section: vm related */ 214 /* 215 * Get (and clear) the dirty memory log for a memory slot. 216 */ 217 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, 218 struct kvm_dirty_log *log) 219 { 220 int r; 221 unsigned long n; 222 struct kvm_memory_slot *memslot; 223 int is_dirty = 0; 224 225 mutex_lock(&kvm->slots_lock); 226 227 r = -EINVAL; 228 if (log->slot >= KVM_USER_MEM_SLOTS) 229 goto out; 230 231 memslot = id_to_memslot(kvm->memslots, log->slot); 232 r = -ENOENT; 233 if (!memslot->dirty_bitmap) 234 goto out; 235 236 kvm_s390_sync_dirty_log(kvm, memslot); 237 r = kvm_get_dirty_log(kvm, log, &is_dirty); 238 if (r) 239 goto out; 240 241 /* Clear the dirty log */ 242 if (is_dirty) { 243 n = kvm_dirty_bitmap_bytes(memslot); 244 memset(memslot->dirty_bitmap, 0, n); 245 } 246 r = 0; 247 out: 248 mutex_unlock(&kvm->slots_lock); 249 return r; 250 } 251 252 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap) 253 { 254 int r; 255 256 if (cap->flags) 257 return -EINVAL; 258 259 switch (cap->cap) { 260 case KVM_CAP_S390_IRQCHIP: 261 kvm->arch.use_irqchip = 1; 262 r = 0; 263 break; 264 case KVM_CAP_S390_USER_SIGP: 265 kvm->arch.user_sigp = 1; 266 r = 0; 267 break; 268 default: 269 r = -EINVAL; 270 break; 271 } 272 return r; 273 } 274 275 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr) 276 { 277 int ret; 278 279 switch (attr->attr) { 280 case KVM_S390_VM_MEM_LIMIT_SIZE: 281 ret = 0; 282 if (put_user(kvm->arch.gmap->asce_end, (u64 __user *)attr->addr)) 283 ret = -EFAULT; 284 break; 285 default: 286 ret = -ENXIO; 287 break; 288 } 289 return ret; 290 } 291 292 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr) 293 { 294 int ret; 295 unsigned int idx; 296 switch (attr->attr) { 297 case KVM_S390_VM_MEM_ENABLE_CMMA: 298 ret = -EBUSY; 299 mutex_lock(&kvm->lock); 300 if (atomic_read(&kvm->online_vcpus) == 0) { 301 kvm->arch.use_cmma = 1; 302 ret = 0; 303 } 304 mutex_unlock(&kvm->lock); 305 break; 306 case KVM_S390_VM_MEM_CLR_CMMA: 307 mutex_lock(&kvm->lock); 308 idx = srcu_read_lock(&kvm->srcu); 309 s390_reset_cmma(kvm->arch.gmap->mm); 310 srcu_read_unlock(&kvm->srcu, idx); 311 mutex_unlock(&kvm->lock); 312 ret = 0; 313 break; 314 case KVM_S390_VM_MEM_LIMIT_SIZE: { 315 unsigned long new_limit; 316 317 if (kvm_is_ucontrol(kvm)) 318 return -EINVAL; 319 320 if (get_user(new_limit, (u64 __user *)attr->addr)) 321 return -EFAULT; 322 323 if (new_limit > kvm->arch.gmap->asce_end) 324 return -E2BIG; 325 326 ret = -EBUSY; 327 mutex_lock(&kvm->lock); 328 if (atomic_read(&kvm->online_vcpus) == 0) { 329 /* gmap_alloc will round the limit up */ 330 struct gmap *new = gmap_alloc(current->mm, new_limit); 331 332 if (!new) { 333 ret = -ENOMEM; 334 } else { 335 gmap_free(kvm->arch.gmap); 336 new->private = kvm; 337 kvm->arch.gmap = new; 338 ret = 0; 339 } 340 } 341 mutex_unlock(&kvm->lock); 342 break; 343 } 344 default: 345 ret = -ENXIO; 346 break; 347 } 348 return ret; 349 } 350 351 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu); 352 353 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr) 354 { 355 struct kvm_vcpu *vcpu; 356 int i; 357 358 if (!test_kvm_facility(kvm, 76)) 359 return -EINVAL; 360 361 mutex_lock(&kvm->lock); 362 switch (attr->attr) { 363 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW: 364 get_random_bytes( 365 kvm->arch.crypto.crycb->aes_wrapping_key_mask, 366 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask)); 367 kvm->arch.crypto.aes_kw = 1; 368 break; 369 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW: 370 get_random_bytes( 371 kvm->arch.crypto.crycb->dea_wrapping_key_mask, 372 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask)); 373 kvm->arch.crypto.dea_kw = 1; 374 break; 375 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW: 376 kvm->arch.crypto.aes_kw = 0; 377 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0, 378 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask)); 379 break; 380 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW: 381 kvm->arch.crypto.dea_kw = 0; 382 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0, 383 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask)); 384 break; 385 default: 386 mutex_unlock(&kvm->lock); 387 return -ENXIO; 388 } 389 390 kvm_for_each_vcpu(i, vcpu, kvm) { 391 kvm_s390_vcpu_crypto_setup(vcpu); 392 exit_sie(vcpu); 393 } 394 mutex_unlock(&kvm->lock); 395 return 0; 396 } 397 398 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr) 399 { 400 u8 gtod_high; 401 402 if (copy_from_user(>od_high, (void __user *)attr->addr, 403 sizeof(gtod_high))) 404 return -EFAULT; 405 406 if (gtod_high != 0) 407 return -EINVAL; 408 409 return 0; 410 } 411 412 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr) 413 { 414 struct kvm_vcpu *cur_vcpu; 415 unsigned int vcpu_idx; 416 u64 host_tod, gtod; 417 int r; 418 419 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod))) 420 return -EFAULT; 421 422 r = store_tod_clock(&host_tod); 423 if (r) 424 return r; 425 426 mutex_lock(&kvm->lock); 427 kvm->arch.epoch = gtod - host_tod; 428 kvm_for_each_vcpu(vcpu_idx, cur_vcpu, kvm) { 429 cur_vcpu->arch.sie_block->epoch = kvm->arch.epoch; 430 exit_sie(cur_vcpu); 431 } 432 mutex_unlock(&kvm->lock); 433 return 0; 434 } 435 436 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr) 437 { 438 int ret; 439 440 if (attr->flags) 441 return -EINVAL; 442 443 switch (attr->attr) { 444 case KVM_S390_VM_TOD_HIGH: 445 ret = kvm_s390_set_tod_high(kvm, attr); 446 break; 447 case KVM_S390_VM_TOD_LOW: 448 ret = kvm_s390_set_tod_low(kvm, attr); 449 break; 450 default: 451 ret = -ENXIO; 452 break; 453 } 454 return ret; 455 } 456 457 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr) 458 { 459 u8 gtod_high = 0; 460 461 if (copy_to_user((void __user *)attr->addr, >od_high, 462 sizeof(gtod_high))) 463 return -EFAULT; 464 465 return 0; 466 } 467 468 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr) 469 { 470 u64 host_tod, gtod; 471 int r; 472 473 r = store_tod_clock(&host_tod); 474 if (r) 475 return r; 476 477 gtod = host_tod + kvm->arch.epoch; 478 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod))) 479 return -EFAULT; 480 481 return 0; 482 } 483 484 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr) 485 { 486 int ret; 487 488 if (attr->flags) 489 return -EINVAL; 490 491 switch (attr->attr) { 492 case KVM_S390_VM_TOD_HIGH: 493 ret = kvm_s390_get_tod_high(kvm, attr); 494 break; 495 case KVM_S390_VM_TOD_LOW: 496 ret = kvm_s390_get_tod_low(kvm, attr); 497 break; 498 default: 499 ret = -ENXIO; 500 break; 501 } 502 return ret; 503 } 504 505 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr) 506 { 507 struct kvm_s390_vm_cpu_processor *proc; 508 int ret = 0; 509 510 mutex_lock(&kvm->lock); 511 if (atomic_read(&kvm->online_vcpus)) { 512 ret = -EBUSY; 513 goto out; 514 } 515 proc = kzalloc(sizeof(*proc), GFP_KERNEL); 516 if (!proc) { 517 ret = -ENOMEM; 518 goto out; 519 } 520 if (!copy_from_user(proc, (void __user *)attr->addr, 521 sizeof(*proc))) { 522 memcpy(&kvm->arch.model.cpu_id, &proc->cpuid, 523 sizeof(struct cpuid)); 524 kvm->arch.model.ibc = proc->ibc; 525 memcpy(kvm->arch.model.fac->kvm, proc->fac_list, 526 S390_ARCH_FAC_LIST_SIZE_BYTE); 527 } else 528 ret = -EFAULT; 529 kfree(proc); 530 out: 531 mutex_unlock(&kvm->lock); 532 return ret; 533 } 534 535 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr) 536 { 537 int ret = -ENXIO; 538 539 switch (attr->attr) { 540 case KVM_S390_VM_CPU_PROCESSOR: 541 ret = kvm_s390_set_processor(kvm, attr); 542 break; 543 } 544 return ret; 545 } 546 547 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr) 548 { 549 struct kvm_s390_vm_cpu_processor *proc; 550 int ret = 0; 551 552 proc = kzalloc(sizeof(*proc), GFP_KERNEL); 553 if (!proc) { 554 ret = -ENOMEM; 555 goto out; 556 } 557 memcpy(&proc->cpuid, &kvm->arch.model.cpu_id, sizeof(struct cpuid)); 558 proc->ibc = kvm->arch.model.ibc; 559 memcpy(&proc->fac_list, kvm->arch.model.fac->kvm, S390_ARCH_FAC_LIST_SIZE_BYTE); 560 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc))) 561 ret = -EFAULT; 562 kfree(proc); 563 out: 564 return ret; 565 } 566 567 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr) 568 { 569 struct kvm_s390_vm_cpu_machine *mach; 570 int ret = 0; 571 572 mach = kzalloc(sizeof(*mach), GFP_KERNEL); 573 if (!mach) { 574 ret = -ENOMEM; 575 goto out; 576 } 577 get_cpu_id((struct cpuid *) &mach->cpuid); 578 mach->ibc = sclp_get_ibc(); 579 memcpy(&mach->fac_mask, kvm_s390_fac_list_mask, 580 kvm_s390_fac_list_mask_size() * sizeof(u64)); 581 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list, 582 S390_ARCH_FAC_LIST_SIZE_U64); 583 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach))) 584 ret = -EFAULT; 585 kfree(mach); 586 out: 587 return ret; 588 } 589 590 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr) 591 { 592 int ret = -ENXIO; 593 594 switch (attr->attr) { 595 case KVM_S390_VM_CPU_PROCESSOR: 596 ret = kvm_s390_get_processor(kvm, attr); 597 break; 598 case KVM_S390_VM_CPU_MACHINE: 599 ret = kvm_s390_get_machine(kvm, attr); 600 break; 601 } 602 return ret; 603 } 604 605 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr) 606 { 607 int ret; 608 609 switch (attr->group) { 610 case KVM_S390_VM_MEM_CTRL: 611 ret = kvm_s390_set_mem_control(kvm, attr); 612 break; 613 case KVM_S390_VM_TOD: 614 ret = kvm_s390_set_tod(kvm, attr); 615 break; 616 case KVM_S390_VM_CPU_MODEL: 617 ret = kvm_s390_set_cpu_model(kvm, attr); 618 break; 619 case KVM_S390_VM_CRYPTO: 620 ret = kvm_s390_vm_set_crypto(kvm, attr); 621 break; 622 default: 623 ret = -ENXIO; 624 break; 625 } 626 627 return ret; 628 } 629 630 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr) 631 { 632 int ret; 633 634 switch (attr->group) { 635 case KVM_S390_VM_MEM_CTRL: 636 ret = kvm_s390_get_mem_control(kvm, attr); 637 break; 638 case KVM_S390_VM_TOD: 639 ret = kvm_s390_get_tod(kvm, attr); 640 break; 641 case KVM_S390_VM_CPU_MODEL: 642 ret = kvm_s390_get_cpu_model(kvm, attr); 643 break; 644 default: 645 ret = -ENXIO; 646 break; 647 } 648 649 return ret; 650 } 651 652 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr) 653 { 654 int ret; 655 656 switch (attr->group) { 657 case KVM_S390_VM_MEM_CTRL: 658 switch (attr->attr) { 659 case KVM_S390_VM_MEM_ENABLE_CMMA: 660 case KVM_S390_VM_MEM_CLR_CMMA: 661 case KVM_S390_VM_MEM_LIMIT_SIZE: 662 ret = 0; 663 break; 664 default: 665 ret = -ENXIO; 666 break; 667 } 668 break; 669 case KVM_S390_VM_TOD: 670 switch (attr->attr) { 671 case KVM_S390_VM_TOD_LOW: 672 case KVM_S390_VM_TOD_HIGH: 673 ret = 0; 674 break; 675 default: 676 ret = -ENXIO; 677 break; 678 } 679 break; 680 case KVM_S390_VM_CPU_MODEL: 681 switch (attr->attr) { 682 case KVM_S390_VM_CPU_PROCESSOR: 683 case KVM_S390_VM_CPU_MACHINE: 684 ret = 0; 685 break; 686 default: 687 ret = -ENXIO; 688 break; 689 } 690 break; 691 case KVM_S390_VM_CRYPTO: 692 switch (attr->attr) { 693 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW: 694 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW: 695 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW: 696 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW: 697 ret = 0; 698 break; 699 default: 700 ret = -ENXIO; 701 break; 702 } 703 break; 704 default: 705 ret = -ENXIO; 706 break; 707 } 708 709 return ret; 710 } 711 712 long kvm_arch_vm_ioctl(struct file *filp, 713 unsigned int ioctl, unsigned long arg) 714 { 715 struct kvm *kvm = filp->private_data; 716 void __user *argp = (void __user *)arg; 717 struct kvm_device_attr attr; 718 int r; 719 720 switch (ioctl) { 721 case KVM_S390_INTERRUPT: { 722 struct kvm_s390_interrupt s390int; 723 724 r = -EFAULT; 725 if (copy_from_user(&s390int, argp, sizeof(s390int))) 726 break; 727 r = kvm_s390_inject_vm(kvm, &s390int); 728 break; 729 } 730 case KVM_ENABLE_CAP: { 731 struct kvm_enable_cap cap; 732 r = -EFAULT; 733 if (copy_from_user(&cap, argp, sizeof(cap))) 734 break; 735 r = kvm_vm_ioctl_enable_cap(kvm, &cap); 736 break; 737 } 738 case KVM_CREATE_IRQCHIP: { 739 struct kvm_irq_routing_entry routing; 740 741 r = -EINVAL; 742 if (kvm->arch.use_irqchip) { 743 /* Set up dummy routing. */ 744 memset(&routing, 0, sizeof(routing)); 745 kvm_set_irq_routing(kvm, &routing, 0, 0); 746 r = 0; 747 } 748 break; 749 } 750 case KVM_SET_DEVICE_ATTR: { 751 r = -EFAULT; 752 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) 753 break; 754 r = kvm_s390_vm_set_attr(kvm, &attr); 755 break; 756 } 757 case KVM_GET_DEVICE_ATTR: { 758 r = -EFAULT; 759 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) 760 break; 761 r = kvm_s390_vm_get_attr(kvm, &attr); 762 break; 763 } 764 case KVM_HAS_DEVICE_ATTR: { 765 r = -EFAULT; 766 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) 767 break; 768 r = kvm_s390_vm_has_attr(kvm, &attr); 769 break; 770 } 771 default: 772 r = -ENOTTY; 773 } 774 775 return r; 776 } 777 778 static int kvm_s390_query_ap_config(u8 *config) 779 { 780 u32 fcn_code = 0x04000000UL; 781 u32 cc; 782 783 asm volatile( 784 "lgr 0,%1\n" 785 "lgr 2,%2\n" 786 ".long 0xb2af0000\n" /* PQAP(QCI) */ 787 "ipm %0\n" 788 "srl %0,28\n" 789 : "=r" (cc) 790 : "r" (fcn_code), "r" (config) 791 : "cc", "0", "2", "memory" 792 ); 793 794 return cc; 795 } 796 797 static int kvm_s390_apxa_installed(void) 798 { 799 u8 config[128]; 800 int cc; 801 802 if (test_facility(2) && test_facility(12)) { 803 cc = kvm_s390_query_ap_config(config); 804 805 if (cc) 806 pr_err("PQAP(QCI) failed with cc=%d", cc); 807 else 808 return config[0] & 0x40; 809 } 810 811 return 0; 812 } 813 814 static void kvm_s390_set_crycb_format(struct kvm *kvm) 815 { 816 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb; 817 818 if (kvm_s390_apxa_installed()) 819 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2; 820 else 821 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1; 822 } 823 824 static void kvm_s390_get_cpu_id(struct cpuid *cpu_id) 825 { 826 get_cpu_id(cpu_id); 827 cpu_id->version = 0xff; 828 } 829 830 static int kvm_s390_crypto_init(struct kvm *kvm) 831 { 832 if (!test_kvm_facility(kvm, 76)) 833 return 0; 834 835 kvm->arch.crypto.crycb = kzalloc(sizeof(*kvm->arch.crypto.crycb), 836 GFP_KERNEL | GFP_DMA); 837 if (!kvm->arch.crypto.crycb) 838 return -ENOMEM; 839 840 kvm_s390_set_crycb_format(kvm); 841 842 /* Disable AES/DEA protected key functions by default */ 843 kvm->arch.crypto.aes_kw = 0; 844 kvm->arch.crypto.dea_kw = 0; 845 846 return 0; 847 } 848 849 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 850 { 851 int i, rc; 852 char debug_name[16]; 853 static unsigned long sca_offset; 854 855 rc = -EINVAL; 856 #ifdef CONFIG_KVM_S390_UCONTROL 857 if (type & ~KVM_VM_S390_UCONTROL) 858 goto out_err; 859 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN))) 860 goto out_err; 861 #else 862 if (type) 863 goto out_err; 864 #endif 865 866 rc = s390_enable_sie(); 867 if (rc) 868 goto out_err; 869 870 rc = -ENOMEM; 871 872 kvm->arch.sca = (struct sca_block *) get_zeroed_page(GFP_KERNEL); 873 if (!kvm->arch.sca) 874 goto out_err; 875 spin_lock(&kvm_lock); 876 sca_offset = (sca_offset + 16) & 0x7f0; 877 kvm->arch.sca = (struct sca_block *) ((char *) kvm->arch.sca + sca_offset); 878 spin_unlock(&kvm_lock); 879 880 sprintf(debug_name, "kvm-%u", current->pid); 881 882 kvm->arch.dbf = debug_register(debug_name, 8, 2, 8 * sizeof(long)); 883 if (!kvm->arch.dbf) 884 goto out_nodbf; 885 886 /* 887 * The architectural maximum amount of facilities is 16 kbit. To store 888 * this amount, 2 kbyte of memory is required. Thus we need a full 889 * page to hold the active copy (arch.model.fac->sie) and the current 890 * facilities set (arch.model.fac->kvm). Its address size has to be 891 * 31 bits and word aligned. 892 */ 893 kvm->arch.model.fac = 894 (struct s390_model_fac *) get_zeroed_page(GFP_KERNEL | GFP_DMA); 895 if (!kvm->arch.model.fac) 896 goto out_nofac; 897 898 memcpy(kvm->arch.model.fac->kvm, S390_lowcore.stfle_fac_list, 899 S390_ARCH_FAC_LIST_SIZE_U64); 900 901 /* 902 * If this KVM host runs *not* in a LPAR, relax the facility bits 903 * of the kvm facility mask by all missing facilities. This will allow 904 * to determine the right CPU model by means of the remaining facilities. 905 * Live guest migration must prohibit the migration of KVMs running in 906 * a LPAR to non LPAR hosts. 907 */ 908 if (!MACHINE_IS_LPAR) 909 for (i = 0; i < kvm_s390_fac_list_mask_size(); i++) 910 kvm_s390_fac_list_mask[i] &= kvm->arch.model.fac->kvm[i]; 911 912 /* 913 * Apply the kvm facility mask to limit the kvm supported/tolerated 914 * facility list. 915 */ 916 for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) { 917 if (i < kvm_s390_fac_list_mask_size()) 918 kvm->arch.model.fac->kvm[i] &= kvm_s390_fac_list_mask[i]; 919 else 920 kvm->arch.model.fac->kvm[i] = 0UL; 921 } 922 923 kvm_s390_get_cpu_id(&kvm->arch.model.cpu_id); 924 kvm->arch.model.ibc = sclp_get_ibc() & 0x0fff; 925 926 if (kvm_s390_crypto_init(kvm) < 0) 927 goto out_crypto; 928 929 spin_lock_init(&kvm->arch.float_int.lock); 930 INIT_LIST_HEAD(&kvm->arch.float_int.list); 931 init_waitqueue_head(&kvm->arch.ipte_wq); 932 mutex_init(&kvm->arch.ipte_mutex); 933 934 debug_register_view(kvm->arch.dbf, &debug_sprintf_view); 935 VM_EVENT(kvm, 3, "%s", "vm created"); 936 937 if (type & KVM_VM_S390_UCONTROL) { 938 kvm->arch.gmap = NULL; 939 } else { 940 kvm->arch.gmap = gmap_alloc(current->mm, (1UL << 44) - 1); 941 if (!kvm->arch.gmap) 942 goto out_nogmap; 943 kvm->arch.gmap->private = kvm; 944 kvm->arch.gmap->pfault_enabled = 0; 945 } 946 947 kvm->arch.css_support = 0; 948 kvm->arch.use_irqchip = 0; 949 kvm->arch.epoch = 0; 950 951 spin_lock_init(&kvm->arch.start_stop_lock); 952 953 return 0; 954 out_nogmap: 955 kfree(kvm->arch.crypto.crycb); 956 out_crypto: 957 free_page((unsigned long)kvm->arch.model.fac); 958 out_nofac: 959 debug_unregister(kvm->arch.dbf); 960 out_nodbf: 961 free_page((unsigned long)(kvm->arch.sca)); 962 out_err: 963 return rc; 964 } 965 966 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 967 { 968 VCPU_EVENT(vcpu, 3, "%s", "free cpu"); 969 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id); 970 kvm_s390_clear_local_irqs(vcpu); 971 kvm_clear_async_pf_completion_queue(vcpu); 972 if (!kvm_is_ucontrol(vcpu->kvm)) { 973 clear_bit(63 - vcpu->vcpu_id, 974 (unsigned long *) &vcpu->kvm->arch.sca->mcn); 975 if (vcpu->kvm->arch.sca->cpu[vcpu->vcpu_id].sda == 976 (__u64) vcpu->arch.sie_block) 977 vcpu->kvm->arch.sca->cpu[vcpu->vcpu_id].sda = 0; 978 } 979 smp_mb(); 980 981 if (kvm_is_ucontrol(vcpu->kvm)) 982 gmap_free(vcpu->arch.gmap); 983 984 if (kvm_s390_cmma_enabled(vcpu->kvm)) 985 kvm_s390_vcpu_unsetup_cmma(vcpu); 986 free_page((unsigned long)(vcpu->arch.sie_block)); 987 988 kvm_vcpu_uninit(vcpu); 989 kmem_cache_free(kvm_vcpu_cache, vcpu); 990 } 991 992 static void kvm_free_vcpus(struct kvm *kvm) 993 { 994 unsigned int i; 995 struct kvm_vcpu *vcpu; 996 997 kvm_for_each_vcpu(i, vcpu, kvm) 998 kvm_arch_vcpu_destroy(vcpu); 999 1000 mutex_lock(&kvm->lock); 1001 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) 1002 kvm->vcpus[i] = NULL; 1003 1004 atomic_set(&kvm->online_vcpus, 0); 1005 mutex_unlock(&kvm->lock); 1006 } 1007 1008 void kvm_arch_destroy_vm(struct kvm *kvm) 1009 { 1010 kvm_free_vcpus(kvm); 1011 free_page((unsigned long)kvm->arch.model.fac); 1012 free_page((unsigned long)(kvm->arch.sca)); 1013 debug_unregister(kvm->arch.dbf); 1014 kfree(kvm->arch.crypto.crycb); 1015 if (!kvm_is_ucontrol(kvm)) 1016 gmap_free(kvm->arch.gmap); 1017 kvm_s390_destroy_adapters(kvm); 1018 kvm_s390_clear_float_irqs(kvm); 1019 } 1020 1021 /* Section: vcpu related */ 1022 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu) 1023 { 1024 vcpu->arch.gmap = gmap_alloc(current->mm, -1UL); 1025 if (!vcpu->arch.gmap) 1026 return -ENOMEM; 1027 vcpu->arch.gmap->private = vcpu->kvm; 1028 1029 return 0; 1030 } 1031 1032 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) 1033 { 1034 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID; 1035 kvm_clear_async_pf_completion_queue(vcpu); 1036 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX | 1037 KVM_SYNC_GPRS | 1038 KVM_SYNC_ACRS | 1039 KVM_SYNC_CRS | 1040 KVM_SYNC_ARCH0 | 1041 KVM_SYNC_PFAULT; 1042 1043 if (kvm_is_ucontrol(vcpu->kvm)) 1044 return __kvm_ucontrol_vcpu_init(vcpu); 1045 1046 return 0; 1047 } 1048 1049 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 1050 { 1051 save_fp_ctl(&vcpu->arch.host_fpregs.fpc); 1052 save_fp_regs(vcpu->arch.host_fpregs.fprs); 1053 save_access_regs(vcpu->arch.host_acrs); 1054 restore_fp_ctl(&vcpu->arch.guest_fpregs.fpc); 1055 restore_fp_regs(vcpu->arch.guest_fpregs.fprs); 1056 restore_access_regs(vcpu->run->s.regs.acrs); 1057 gmap_enable(vcpu->arch.gmap); 1058 atomic_set_mask(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags); 1059 } 1060 1061 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 1062 { 1063 atomic_clear_mask(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags); 1064 gmap_disable(vcpu->arch.gmap); 1065 save_fp_ctl(&vcpu->arch.guest_fpregs.fpc); 1066 save_fp_regs(vcpu->arch.guest_fpregs.fprs); 1067 save_access_regs(vcpu->run->s.regs.acrs); 1068 restore_fp_ctl(&vcpu->arch.host_fpregs.fpc); 1069 restore_fp_regs(vcpu->arch.host_fpregs.fprs); 1070 restore_access_regs(vcpu->arch.host_acrs); 1071 } 1072 1073 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu) 1074 { 1075 /* this equals initial cpu reset in pop, but we don't switch to ESA */ 1076 vcpu->arch.sie_block->gpsw.mask = 0UL; 1077 vcpu->arch.sie_block->gpsw.addr = 0UL; 1078 kvm_s390_set_prefix(vcpu, 0); 1079 vcpu->arch.sie_block->cputm = 0UL; 1080 vcpu->arch.sie_block->ckc = 0UL; 1081 vcpu->arch.sie_block->todpr = 0; 1082 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64)); 1083 vcpu->arch.sie_block->gcr[0] = 0xE0UL; 1084 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL; 1085 vcpu->arch.guest_fpregs.fpc = 0; 1086 asm volatile("lfpc %0" : : "Q" (vcpu->arch.guest_fpregs.fpc)); 1087 vcpu->arch.sie_block->gbea = 1; 1088 vcpu->arch.sie_block->pp = 0; 1089 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID; 1090 kvm_clear_async_pf_completion_queue(vcpu); 1091 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) 1092 kvm_s390_vcpu_stop(vcpu); 1093 kvm_s390_clear_local_irqs(vcpu); 1094 } 1095 1096 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 1097 { 1098 mutex_lock(&vcpu->kvm->lock); 1099 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch; 1100 mutex_unlock(&vcpu->kvm->lock); 1101 if (!kvm_is_ucontrol(vcpu->kvm)) 1102 vcpu->arch.gmap = vcpu->kvm->arch.gmap; 1103 } 1104 1105 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu) 1106 { 1107 if (!test_kvm_facility(vcpu->kvm, 76)) 1108 return; 1109 1110 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA); 1111 1112 if (vcpu->kvm->arch.crypto.aes_kw) 1113 vcpu->arch.sie_block->ecb3 |= ECB3_AES; 1114 if (vcpu->kvm->arch.crypto.dea_kw) 1115 vcpu->arch.sie_block->ecb3 |= ECB3_DEA; 1116 1117 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd; 1118 } 1119 1120 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu) 1121 { 1122 free_page(vcpu->arch.sie_block->cbrlo); 1123 vcpu->arch.sie_block->cbrlo = 0; 1124 } 1125 1126 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu) 1127 { 1128 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL); 1129 if (!vcpu->arch.sie_block->cbrlo) 1130 return -ENOMEM; 1131 1132 vcpu->arch.sie_block->ecb2 |= 0x80; 1133 vcpu->arch.sie_block->ecb2 &= ~0x08; 1134 return 0; 1135 } 1136 1137 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu) 1138 { 1139 int rc = 0; 1140 1141 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH | 1142 CPUSTAT_SM | 1143 CPUSTAT_STOPPED | 1144 CPUSTAT_GED); 1145 vcpu->arch.sie_block->ecb = 6; 1146 if (test_kvm_facility(vcpu->kvm, 50) && test_kvm_facility(vcpu->kvm, 73)) 1147 vcpu->arch.sie_block->ecb |= 0x10; 1148 1149 vcpu->arch.sie_block->ecb2 = 8; 1150 vcpu->arch.sie_block->eca = 0xC1002000U; 1151 if (sclp_has_siif()) 1152 vcpu->arch.sie_block->eca |= 1; 1153 if (sclp_has_sigpif()) 1154 vcpu->arch.sie_block->eca |= 0x10000000U; 1155 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE | 1156 ICTL_TPROT; 1157 1158 if (kvm_s390_cmma_enabled(vcpu->kvm)) { 1159 rc = kvm_s390_vcpu_setup_cmma(vcpu); 1160 if (rc) 1161 return rc; 1162 } 1163 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1164 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup; 1165 1166 mutex_lock(&vcpu->kvm->lock); 1167 vcpu->arch.cpu_id = vcpu->kvm->arch.model.cpu_id; 1168 memcpy(vcpu->kvm->arch.model.fac->sie, vcpu->kvm->arch.model.fac->kvm, 1169 S390_ARCH_FAC_LIST_SIZE_BYTE); 1170 vcpu->arch.sie_block->ibc = vcpu->kvm->arch.model.ibc; 1171 mutex_unlock(&vcpu->kvm->lock); 1172 1173 kvm_s390_vcpu_crypto_setup(vcpu); 1174 1175 return rc; 1176 } 1177 1178 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, 1179 unsigned int id) 1180 { 1181 struct kvm_vcpu *vcpu; 1182 struct sie_page *sie_page; 1183 int rc = -EINVAL; 1184 1185 if (id >= KVM_MAX_VCPUS) 1186 goto out; 1187 1188 rc = -ENOMEM; 1189 1190 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); 1191 if (!vcpu) 1192 goto out; 1193 1194 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL); 1195 if (!sie_page) 1196 goto out_free_cpu; 1197 1198 vcpu->arch.sie_block = &sie_page->sie_block; 1199 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb; 1200 1201 vcpu->arch.sie_block->icpua = id; 1202 if (!kvm_is_ucontrol(kvm)) { 1203 if (!kvm->arch.sca) { 1204 WARN_ON_ONCE(1); 1205 goto out_free_cpu; 1206 } 1207 if (!kvm->arch.sca->cpu[id].sda) 1208 kvm->arch.sca->cpu[id].sda = 1209 (__u64) vcpu->arch.sie_block; 1210 vcpu->arch.sie_block->scaoh = 1211 (__u32)(((__u64)kvm->arch.sca) >> 32); 1212 vcpu->arch.sie_block->scaol = (__u32)(__u64)kvm->arch.sca; 1213 set_bit(63 - id, (unsigned long *) &kvm->arch.sca->mcn); 1214 } 1215 vcpu->arch.sie_block->fac = (int) (long) kvm->arch.model.fac->sie; 1216 1217 spin_lock_init(&vcpu->arch.local_int.lock); 1218 vcpu->arch.local_int.float_int = &kvm->arch.float_int; 1219 vcpu->arch.local_int.wq = &vcpu->wq; 1220 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags; 1221 1222 rc = kvm_vcpu_init(vcpu, kvm, id); 1223 if (rc) 1224 goto out_free_sie_block; 1225 VM_EVENT(kvm, 3, "create cpu %d at %p, sie block at %p", id, vcpu, 1226 vcpu->arch.sie_block); 1227 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block); 1228 1229 return vcpu; 1230 out_free_sie_block: 1231 free_page((unsigned long)(vcpu->arch.sie_block)); 1232 out_free_cpu: 1233 kmem_cache_free(kvm_vcpu_cache, vcpu); 1234 out: 1235 return ERR_PTR(rc); 1236 } 1237 1238 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) 1239 { 1240 return kvm_s390_vcpu_has_irq(vcpu, 0); 1241 } 1242 1243 void s390_vcpu_block(struct kvm_vcpu *vcpu) 1244 { 1245 atomic_set_mask(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20); 1246 } 1247 1248 void s390_vcpu_unblock(struct kvm_vcpu *vcpu) 1249 { 1250 atomic_clear_mask(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20); 1251 } 1252 1253 /* 1254 * Kick a guest cpu out of SIE and wait until SIE is not running. 1255 * If the CPU is not running (e.g. waiting as idle) the function will 1256 * return immediately. */ 1257 void exit_sie(struct kvm_vcpu *vcpu) 1258 { 1259 atomic_set_mask(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags); 1260 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE) 1261 cpu_relax(); 1262 } 1263 1264 /* Kick a guest cpu out of SIE and prevent SIE-reentry */ 1265 void exit_sie_sync(struct kvm_vcpu *vcpu) 1266 { 1267 s390_vcpu_block(vcpu); 1268 exit_sie(vcpu); 1269 } 1270 1271 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address) 1272 { 1273 int i; 1274 struct kvm *kvm = gmap->private; 1275 struct kvm_vcpu *vcpu; 1276 1277 kvm_for_each_vcpu(i, vcpu, kvm) { 1278 /* match against both prefix pages */ 1279 if (kvm_s390_get_prefix(vcpu) == (address & ~0x1000UL)) { 1280 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx", address); 1281 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu); 1282 exit_sie_sync(vcpu); 1283 } 1284 } 1285 } 1286 1287 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 1288 { 1289 /* kvm common code refers to this, but never calls it */ 1290 BUG(); 1291 return 0; 1292 } 1293 1294 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu, 1295 struct kvm_one_reg *reg) 1296 { 1297 int r = -EINVAL; 1298 1299 switch (reg->id) { 1300 case KVM_REG_S390_TODPR: 1301 r = put_user(vcpu->arch.sie_block->todpr, 1302 (u32 __user *)reg->addr); 1303 break; 1304 case KVM_REG_S390_EPOCHDIFF: 1305 r = put_user(vcpu->arch.sie_block->epoch, 1306 (u64 __user *)reg->addr); 1307 break; 1308 case KVM_REG_S390_CPU_TIMER: 1309 r = put_user(vcpu->arch.sie_block->cputm, 1310 (u64 __user *)reg->addr); 1311 break; 1312 case KVM_REG_S390_CLOCK_COMP: 1313 r = put_user(vcpu->arch.sie_block->ckc, 1314 (u64 __user *)reg->addr); 1315 break; 1316 case KVM_REG_S390_PFTOKEN: 1317 r = put_user(vcpu->arch.pfault_token, 1318 (u64 __user *)reg->addr); 1319 break; 1320 case KVM_REG_S390_PFCOMPARE: 1321 r = put_user(vcpu->arch.pfault_compare, 1322 (u64 __user *)reg->addr); 1323 break; 1324 case KVM_REG_S390_PFSELECT: 1325 r = put_user(vcpu->arch.pfault_select, 1326 (u64 __user *)reg->addr); 1327 break; 1328 case KVM_REG_S390_PP: 1329 r = put_user(vcpu->arch.sie_block->pp, 1330 (u64 __user *)reg->addr); 1331 break; 1332 case KVM_REG_S390_GBEA: 1333 r = put_user(vcpu->arch.sie_block->gbea, 1334 (u64 __user *)reg->addr); 1335 break; 1336 default: 1337 break; 1338 } 1339 1340 return r; 1341 } 1342 1343 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu, 1344 struct kvm_one_reg *reg) 1345 { 1346 int r = -EINVAL; 1347 1348 switch (reg->id) { 1349 case KVM_REG_S390_TODPR: 1350 r = get_user(vcpu->arch.sie_block->todpr, 1351 (u32 __user *)reg->addr); 1352 break; 1353 case KVM_REG_S390_EPOCHDIFF: 1354 r = get_user(vcpu->arch.sie_block->epoch, 1355 (u64 __user *)reg->addr); 1356 break; 1357 case KVM_REG_S390_CPU_TIMER: 1358 r = get_user(vcpu->arch.sie_block->cputm, 1359 (u64 __user *)reg->addr); 1360 break; 1361 case KVM_REG_S390_CLOCK_COMP: 1362 r = get_user(vcpu->arch.sie_block->ckc, 1363 (u64 __user *)reg->addr); 1364 break; 1365 case KVM_REG_S390_PFTOKEN: 1366 r = get_user(vcpu->arch.pfault_token, 1367 (u64 __user *)reg->addr); 1368 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) 1369 kvm_clear_async_pf_completion_queue(vcpu); 1370 break; 1371 case KVM_REG_S390_PFCOMPARE: 1372 r = get_user(vcpu->arch.pfault_compare, 1373 (u64 __user *)reg->addr); 1374 break; 1375 case KVM_REG_S390_PFSELECT: 1376 r = get_user(vcpu->arch.pfault_select, 1377 (u64 __user *)reg->addr); 1378 break; 1379 case KVM_REG_S390_PP: 1380 r = get_user(vcpu->arch.sie_block->pp, 1381 (u64 __user *)reg->addr); 1382 break; 1383 case KVM_REG_S390_GBEA: 1384 r = get_user(vcpu->arch.sie_block->gbea, 1385 (u64 __user *)reg->addr); 1386 break; 1387 default: 1388 break; 1389 } 1390 1391 return r; 1392 } 1393 1394 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu) 1395 { 1396 kvm_s390_vcpu_initial_reset(vcpu); 1397 return 0; 1398 } 1399 1400 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) 1401 { 1402 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs)); 1403 return 0; 1404 } 1405 1406 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) 1407 { 1408 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs)); 1409 return 0; 1410 } 1411 1412 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, 1413 struct kvm_sregs *sregs) 1414 { 1415 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs)); 1416 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs)); 1417 restore_access_regs(vcpu->run->s.regs.acrs); 1418 return 0; 1419 } 1420 1421 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, 1422 struct kvm_sregs *sregs) 1423 { 1424 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs)); 1425 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs)); 1426 return 0; 1427 } 1428 1429 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) 1430 { 1431 if (test_fp_ctl(fpu->fpc)) 1432 return -EINVAL; 1433 memcpy(&vcpu->arch.guest_fpregs.fprs, &fpu->fprs, sizeof(fpu->fprs)); 1434 vcpu->arch.guest_fpregs.fpc = fpu->fpc; 1435 restore_fp_ctl(&vcpu->arch.guest_fpregs.fpc); 1436 restore_fp_regs(vcpu->arch.guest_fpregs.fprs); 1437 return 0; 1438 } 1439 1440 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) 1441 { 1442 memcpy(&fpu->fprs, &vcpu->arch.guest_fpregs.fprs, sizeof(fpu->fprs)); 1443 fpu->fpc = vcpu->arch.guest_fpregs.fpc; 1444 return 0; 1445 } 1446 1447 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw) 1448 { 1449 int rc = 0; 1450 1451 if (!is_vcpu_stopped(vcpu)) 1452 rc = -EBUSY; 1453 else { 1454 vcpu->run->psw_mask = psw.mask; 1455 vcpu->run->psw_addr = psw.addr; 1456 } 1457 return rc; 1458 } 1459 1460 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, 1461 struct kvm_translation *tr) 1462 { 1463 return -EINVAL; /* not implemented yet */ 1464 } 1465 1466 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \ 1467 KVM_GUESTDBG_USE_HW_BP | \ 1468 KVM_GUESTDBG_ENABLE) 1469 1470 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, 1471 struct kvm_guest_debug *dbg) 1472 { 1473 int rc = 0; 1474 1475 vcpu->guest_debug = 0; 1476 kvm_s390_clear_bp_data(vcpu); 1477 1478 if (dbg->control & ~VALID_GUESTDBG_FLAGS) 1479 return -EINVAL; 1480 1481 if (dbg->control & KVM_GUESTDBG_ENABLE) { 1482 vcpu->guest_debug = dbg->control; 1483 /* enforce guest PER */ 1484 atomic_set_mask(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags); 1485 1486 if (dbg->control & KVM_GUESTDBG_USE_HW_BP) 1487 rc = kvm_s390_import_bp_data(vcpu, dbg); 1488 } else { 1489 atomic_clear_mask(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags); 1490 vcpu->arch.guestdbg.last_bp = 0; 1491 } 1492 1493 if (rc) { 1494 vcpu->guest_debug = 0; 1495 kvm_s390_clear_bp_data(vcpu); 1496 atomic_clear_mask(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags); 1497 } 1498 1499 return rc; 1500 } 1501 1502 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 1503 struct kvm_mp_state *mp_state) 1504 { 1505 /* CHECK_STOP and LOAD are not supported yet */ 1506 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED : 1507 KVM_MP_STATE_OPERATING; 1508 } 1509 1510 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 1511 struct kvm_mp_state *mp_state) 1512 { 1513 int rc = 0; 1514 1515 /* user space knows about this interface - let it control the state */ 1516 vcpu->kvm->arch.user_cpu_state_ctrl = 1; 1517 1518 switch (mp_state->mp_state) { 1519 case KVM_MP_STATE_STOPPED: 1520 kvm_s390_vcpu_stop(vcpu); 1521 break; 1522 case KVM_MP_STATE_OPERATING: 1523 kvm_s390_vcpu_start(vcpu); 1524 break; 1525 case KVM_MP_STATE_LOAD: 1526 case KVM_MP_STATE_CHECK_STOP: 1527 /* fall through - CHECK_STOP and LOAD are not supported yet */ 1528 default: 1529 rc = -ENXIO; 1530 } 1531 1532 return rc; 1533 } 1534 1535 bool kvm_s390_cmma_enabled(struct kvm *kvm) 1536 { 1537 if (!MACHINE_IS_LPAR) 1538 return false; 1539 /* only enable for z10 and later */ 1540 if (!MACHINE_HAS_EDAT1) 1541 return false; 1542 if (!kvm->arch.use_cmma) 1543 return false; 1544 return true; 1545 } 1546 1547 static bool ibs_enabled(struct kvm_vcpu *vcpu) 1548 { 1549 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS; 1550 } 1551 1552 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu) 1553 { 1554 retry: 1555 s390_vcpu_unblock(vcpu); 1556 /* 1557 * We use MMU_RELOAD just to re-arm the ipte notifier for the 1558 * guest prefix page. gmap_ipte_notify will wait on the ptl lock. 1559 * This ensures that the ipte instruction for this request has 1560 * already finished. We might race against a second unmapper that 1561 * wants to set the blocking bit. Lets just retry the request loop. 1562 */ 1563 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) { 1564 int rc; 1565 rc = gmap_ipte_notify(vcpu->arch.gmap, 1566 kvm_s390_get_prefix(vcpu), 1567 PAGE_SIZE * 2); 1568 if (rc) 1569 return rc; 1570 goto retry; 1571 } 1572 1573 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) { 1574 vcpu->arch.sie_block->ihcpu = 0xffff; 1575 goto retry; 1576 } 1577 1578 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) { 1579 if (!ibs_enabled(vcpu)) { 1580 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1); 1581 atomic_set_mask(CPUSTAT_IBS, 1582 &vcpu->arch.sie_block->cpuflags); 1583 } 1584 goto retry; 1585 } 1586 1587 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) { 1588 if (ibs_enabled(vcpu)) { 1589 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0); 1590 atomic_clear_mask(CPUSTAT_IBS, 1591 &vcpu->arch.sie_block->cpuflags); 1592 } 1593 goto retry; 1594 } 1595 1596 /* nothing to do, just clear the request */ 1597 clear_bit(KVM_REQ_UNHALT, &vcpu->requests); 1598 1599 return 0; 1600 } 1601 1602 /** 1603 * kvm_arch_fault_in_page - fault-in guest page if necessary 1604 * @vcpu: The corresponding virtual cpu 1605 * @gpa: Guest physical address 1606 * @writable: Whether the page should be writable or not 1607 * 1608 * Make sure that a guest page has been faulted-in on the host. 1609 * 1610 * Return: Zero on success, negative error code otherwise. 1611 */ 1612 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable) 1613 { 1614 return gmap_fault(vcpu->arch.gmap, gpa, 1615 writable ? FAULT_FLAG_WRITE : 0); 1616 } 1617 1618 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token, 1619 unsigned long token) 1620 { 1621 struct kvm_s390_interrupt inti; 1622 struct kvm_s390_irq irq; 1623 1624 if (start_token) { 1625 irq.u.ext.ext_params2 = token; 1626 irq.type = KVM_S390_INT_PFAULT_INIT; 1627 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq)); 1628 } else { 1629 inti.type = KVM_S390_INT_PFAULT_DONE; 1630 inti.parm64 = token; 1631 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti)); 1632 } 1633 } 1634 1635 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, 1636 struct kvm_async_pf *work) 1637 { 1638 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token); 1639 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token); 1640 } 1641 1642 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, 1643 struct kvm_async_pf *work) 1644 { 1645 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token); 1646 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token); 1647 } 1648 1649 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, 1650 struct kvm_async_pf *work) 1651 { 1652 /* s390 will always inject the page directly */ 1653 } 1654 1655 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu) 1656 { 1657 /* 1658 * s390 will always inject the page directly, 1659 * but we still want check_async_completion to cleanup 1660 */ 1661 return true; 1662 } 1663 1664 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu) 1665 { 1666 hva_t hva; 1667 struct kvm_arch_async_pf arch; 1668 int rc; 1669 1670 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) 1671 return 0; 1672 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) != 1673 vcpu->arch.pfault_compare) 1674 return 0; 1675 if (psw_extint_disabled(vcpu)) 1676 return 0; 1677 if (kvm_s390_vcpu_has_irq(vcpu, 0)) 1678 return 0; 1679 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul)) 1680 return 0; 1681 if (!vcpu->arch.gmap->pfault_enabled) 1682 return 0; 1683 1684 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr)); 1685 hva += current->thread.gmap_addr & ~PAGE_MASK; 1686 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8)) 1687 return 0; 1688 1689 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch); 1690 return rc; 1691 } 1692 1693 static int vcpu_pre_run(struct kvm_vcpu *vcpu) 1694 { 1695 int rc, cpuflags; 1696 1697 /* 1698 * On s390 notifications for arriving pages will be delivered directly 1699 * to the guest but the house keeping for completed pfaults is 1700 * handled outside the worker. 1701 */ 1702 kvm_check_async_pf_completion(vcpu); 1703 1704 memcpy(&vcpu->arch.sie_block->gg14, &vcpu->run->s.regs.gprs[14], 16); 1705 1706 if (need_resched()) 1707 schedule(); 1708 1709 if (test_cpu_flag(CIF_MCCK_PENDING)) 1710 s390_handle_mcck(); 1711 1712 if (!kvm_is_ucontrol(vcpu->kvm)) { 1713 rc = kvm_s390_deliver_pending_interrupts(vcpu); 1714 if (rc) 1715 return rc; 1716 } 1717 1718 rc = kvm_s390_handle_requests(vcpu); 1719 if (rc) 1720 return rc; 1721 1722 if (guestdbg_enabled(vcpu)) { 1723 kvm_s390_backup_guest_per_regs(vcpu); 1724 kvm_s390_patch_guest_per_regs(vcpu); 1725 } 1726 1727 vcpu->arch.sie_block->icptcode = 0; 1728 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags); 1729 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags); 1730 trace_kvm_s390_sie_enter(vcpu, cpuflags); 1731 1732 return 0; 1733 } 1734 1735 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason) 1736 { 1737 int rc = -1; 1738 1739 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d", 1740 vcpu->arch.sie_block->icptcode); 1741 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode); 1742 1743 if (guestdbg_enabled(vcpu)) 1744 kvm_s390_restore_guest_per_regs(vcpu); 1745 1746 if (exit_reason >= 0) { 1747 rc = 0; 1748 } else if (kvm_is_ucontrol(vcpu->kvm)) { 1749 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL; 1750 vcpu->run->s390_ucontrol.trans_exc_code = 1751 current->thread.gmap_addr; 1752 vcpu->run->s390_ucontrol.pgm_code = 0x10; 1753 rc = -EREMOTE; 1754 1755 } else if (current->thread.gmap_pfault) { 1756 trace_kvm_s390_major_guest_pfault(vcpu); 1757 current->thread.gmap_pfault = 0; 1758 if (kvm_arch_setup_async_pf(vcpu)) { 1759 rc = 0; 1760 } else { 1761 gpa_t gpa = current->thread.gmap_addr; 1762 rc = kvm_arch_fault_in_page(vcpu, gpa, 1); 1763 } 1764 } 1765 1766 if (rc == -1) { 1767 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction"); 1768 trace_kvm_s390_sie_fault(vcpu); 1769 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 1770 } 1771 1772 memcpy(&vcpu->run->s.regs.gprs[14], &vcpu->arch.sie_block->gg14, 16); 1773 1774 if (rc == 0) { 1775 if (kvm_is_ucontrol(vcpu->kvm)) 1776 /* Don't exit for host interrupts. */ 1777 rc = vcpu->arch.sie_block->icptcode ? -EOPNOTSUPP : 0; 1778 else 1779 rc = kvm_handle_sie_intercept(vcpu); 1780 } 1781 1782 return rc; 1783 } 1784 1785 static int __vcpu_run(struct kvm_vcpu *vcpu) 1786 { 1787 int rc, exit_reason; 1788 1789 /* 1790 * We try to hold kvm->srcu during most of vcpu_run (except when run- 1791 * ning the guest), so that memslots (and other stuff) are protected 1792 */ 1793 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 1794 1795 do { 1796 rc = vcpu_pre_run(vcpu); 1797 if (rc) 1798 break; 1799 1800 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); 1801 /* 1802 * As PF_VCPU will be used in fault handler, between 1803 * guest_enter and guest_exit should be no uaccess. 1804 */ 1805 preempt_disable(); 1806 kvm_guest_enter(); 1807 preempt_enable(); 1808 exit_reason = sie64a(vcpu->arch.sie_block, 1809 vcpu->run->s.regs.gprs); 1810 kvm_guest_exit(); 1811 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 1812 1813 rc = vcpu_post_run(vcpu, exit_reason); 1814 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc); 1815 1816 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); 1817 return rc; 1818 } 1819 1820 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) 1821 { 1822 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask; 1823 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr; 1824 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX) 1825 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix); 1826 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) { 1827 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128); 1828 /* some control register changes require a tlb flush */ 1829 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 1830 } 1831 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) { 1832 vcpu->arch.sie_block->cputm = kvm_run->s.regs.cputm; 1833 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc; 1834 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr; 1835 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp; 1836 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea; 1837 } 1838 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) { 1839 vcpu->arch.pfault_token = kvm_run->s.regs.pft; 1840 vcpu->arch.pfault_select = kvm_run->s.regs.pfs; 1841 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc; 1842 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID) 1843 kvm_clear_async_pf_completion_queue(vcpu); 1844 } 1845 kvm_run->kvm_dirty_regs = 0; 1846 } 1847 1848 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) 1849 { 1850 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask; 1851 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr; 1852 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu); 1853 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128); 1854 kvm_run->s.regs.cputm = vcpu->arch.sie_block->cputm; 1855 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc; 1856 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr; 1857 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp; 1858 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea; 1859 kvm_run->s.regs.pft = vcpu->arch.pfault_token; 1860 kvm_run->s.regs.pfs = vcpu->arch.pfault_select; 1861 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare; 1862 } 1863 1864 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) 1865 { 1866 int rc; 1867 sigset_t sigsaved; 1868 1869 if (guestdbg_exit_pending(vcpu)) { 1870 kvm_s390_prepare_debug_exit(vcpu); 1871 return 0; 1872 } 1873 1874 if (vcpu->sigset_active) 1875 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); 1876 1877 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) { 1878 kvm_s390_vcpu_start(vcpu); 1879 } else if (is_vcpu_stopped(vcpu)) { 1880 pr_err_ratelimited("kvm-s390: can't run stopped vcpu %d\n", 1881 vcpu->vcpu_id); 1882 return -EINVAL; 1883 } 1884 1885 sync_regs(vcpu, kvm_run); 1886 1887 might_fault(); 1888 rc = __vcpu_run(vcpu); 1889 1890 if (signal_pending(current) && !rc) { 1891 kvm_run->exit_reason = KVM_EXIT_INTR; 1892 rc = -EINTR; 1893 } 1894 1895 if (guestdbg_exit_pending(vcpu) && !rc) { 1896 kvm_s390_prepare_debug_exit(vcpu); 1897 rc = 0; 1898 } 1899 1900 if (rc == -EOPNOTSUPP) { 1901 /* intercept cannot be handled in-kernel, prepare kvm-run */ 1902 kvm_run->exit_reason = KVM_EXIT_S390_SIEIC; 1903 kvm_run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode; 1904 kvm_run->s390_sieic.ipa = vcpu->arch.sie_block->ipa; 1905 kvm_run->s390_sieic.ipb = vcpu->arch.sie_block->ipb; 1906 rc = 0; 1907 } 1908 1909 if (rc == -EREMOTE) { 1910 /* intercept was handled, but userspace support is needed 1911 * kvm_run has been prepared by the handler */ 1912 rc = 0; 1913 } 1914 1915 store_regs(vcpu, kvm_run); 1916 1917 if (vcpu->sigset_active) 1918 sigprocmask(SIG_SETMASK, &sigsaved, NULL); 1919 1920 vcpu->stat.exit_userspace++; 1921 return rc; 1922 } 1923 1924 /* 1925 * store status at address 1926 * we use have two special cases: 1927 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit 1928 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix 1929 */ 1930 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa) 1931 { 1932 unsigned char archmode = 1; 1933 unsigned int px; 1934 u64 clkcomp; 1935 int rc; 1936 1937 if (gpa == KVM_S390_STORE_STATUS_NOADDR) { 1938 if (write_guest_abs(vcpu, 163, &archmode, 1)) 1939 return -EFAULT; 1940 gpa = SAVE_AREA_BASE; 1941 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) { 1942 if (write_guest_real(vcpu, 163, &archmode, 1)) 1943 return -EFAULT; 1944 gpa = kvm_s390_real_to_abs(vcpu, SAVE_AREA_BASE); 1945 } 1946 rc = write_guest_abs(vcpu, gpa + offsetof(struct save_area, fp_regs), 1947 vcpu->arch.guest_fpregs.fprs, 128); 1948 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, gp_regs), 1949 vcpu->run->s.regs.gprs, 128); 1950 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, psw), 1951 &vcpu->arch.sie_block->gpsw, 16); 1952 px = kvm_s390_get_prefix(vcpu); 1953 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, pref_reg), 1954 &px, 4); 1955 rc |= write_guest_abs(vcpu, 1956 gpa + offsetof(struct save_area, fp_ctrl_reg), 1957 &vcpu->arch.guest_fpregs.fpc, 4); 1958 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, tod_reg), 1959 &vcpu->arch.sie_block->todpr, 4); 1960 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, timer), 1961 &vcpu->arch.sie_block->cputm, 8); 1962 clkcomp = vcpu->arch.sie_block->ckc >> 8; 1963 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, clk_cmp), 1964 &clkcomp, 8); 1965 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, acc_regs), 1966 &vcpu->run->s.regs.acrs, 64); 1967 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, ctrl_regs), 1968 &vcpu->arch.sie_block->gcr, 128); 1969 return rc ? -EFAULT : 0; 1970 } 1971 1972 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr) 1973 { 1974 /* 1975 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy 1976 * copying in vcpu load/put. Lets update our copies before we save 1977 * it into the save area 1978 */ 1979 save_fp_ctl(&vcpu->arch.guest_fpregs.fpc); 1980 save_fp_regs(vcpu->arch.guest_fpregs.fprs); 1981 save_access_regs(vcpu->run->s.regs.acrs); 1982 1983 return kvm_s390_store_status_unloaded(vcpu, addr); 1984 } 1985 1986 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu) 1987 { 1988 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu); 1989 kvm_make_request(KVM_REQ_DISABLE_IBS, vcpu); 1990 exit_sie_sync(vcpu); 1991 } 1992 1993 static void __disable_ibs_on_all_vcpus(struct kvm *kvm) 1994 { 1995 unsigned int i; 1996 struct kvm_vcpu *vcpu; 1997 1998 kvm_for_each_vcpu(i, vcpu, kvm) { 1999 __disable_ibs_on_vcpu(vcpu); 2000 } 2001 } 2002 2003 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu) 2004 { 2005 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu); 2006 kvm_make_request(KVM_REQ_ENABLE_IBS, vcpu); 2007 exit_sie_sync(vcpu); 2008 } 2009 2010 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu) 2011 { 2012 int i, online_vcpus, started_vcpus = 0; 2013 2014 if (!is_vcpu_stopped(vcpu)) 2015 return; 2016 2017 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1); 2018 /* Only one cpu at a time may enter/leave the STOPPED state. */ 2019 spin_lock(&vcpu->kvm->arch.start_stop_lock); 2020 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus); 2021 2022 for (i = 0; i < online_vcpus; i++) { 2023 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) 2024 started_vcpus++; 2025 } 2026 2027 if (started_vcpus == 0) { 2028 /* we're the only active VCPU -> speed it up */ 2029 __enable_ibs_on_vcpu(vcpu); 2030 } else if (started_vcpus == 1) { 2031 /* 2032 * As we are starting a second VCPU, we have to disable 2033 * the IBS facility on all VCPUs to remove potentially 2034 * oustanding ENABLE requests. 2035 */ 2036 __disable_ibs_on_all_vcpus(vcpu->kvm); 2037 } 2038 2039 atomic_clear_mask(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags); 2040 /* 2041 * Another VCPU might have used IBS while we were offline. 2042 * Let's play safe and flush the VCPU at startup. 2043 */ 2044 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 2045 spin_unlock(&vcpu->kvm->arch.start_stop_lock); 2046 return; 2047 } 2048 2049 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu) 2050 { 2051 int i, online_vcpus, started_vcpus = 0; 2052 struct kvm_vcpu *started_vcpu = NULL; 2053 2054 if (is_vcpu_stopped(vcpu)) 2055 return; 2056 2057 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0); 2058 /* Only one cpu at a time may enter/leave the STOPPED state. */ 2059 spin_lock(&vcpu->kvm->arch.start_stop_lock); 2060 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus); 2061 2062 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */ 2063 kvm_s390_clear_stop_irq(vcpu); 2064 2065 atomic_set_mask(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags); 2066 __disable_ibs_on_vcpu(vcpu); 2067 2068 for (i = 0; i < online_vcpus; i++) { 2069 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) { 2070 started_vcpus++; 2071 started_vcpu = vcpu->kvm->vcpus[i]; 2072 } 2073 } 2074 2075 if (started_vcpus == 1) { 2076 /* 2077 * As we only have one VCPU left, we want to enable the 2078 * IBS facility for that VCPU to speed it up. 2079 */ 2080 __enable_ibs_on_vcpu(started_vcpu); 2081 } 2082 2083 spin_unlock(&vcpu->kvm->arch.start_stop_lock); 2084 return; 2085 } 2086 2087 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, 2088 struct kvm_enable_cap *cap) 2089 { 2090 int r; 2091 2092 if (cap->flags) 2093 return -EINVAL; 2094 2095 switch (cap->cap) { 2096 case KVM_CAP_S390_CSS_SUPPORT: 2097 if (!vcpu->kvm->arch.css_support) { 2098 vcpu->kvm->arch.css_support = 1; 2099 trace_kvm_s390_enable_css(vcpu->kvm); 2100 } 2101 r = 0; 2102 break; 2103 default: 2104 r = -EINVAL; 2105 break; 2106 } 2107 return r; 2108 } 2109 2110 long kvm_arch_vcpu_ioctl(struct file *filp, 2111 unsigned int ioctl, unsigned long arg) 2112 { 2113 struct kvm_vcpu *vcpu = filp->private_data; 2114 void __user *argp = (void __user *)arg; 2115 int idx; 2116 long r; 2117 2118 switch (ioctl) { 2119 case KVM_S390_INTERRUPT: { 2120 struct kvm_s390_interrupt s390int; 2121 struct kvm_s390_irq s390irq; 2122 2123 r = -EFAULT; 2124 if (copy_from_user(&s390int, argp, sizeof(s390int))) 2125 break; 2126 if (s390int_to_s390irq(&s390int, &s390irq)) 2127 return -EINVAL; 2128 r = kvm_s390_inject_vcpu(vcpu, &s390irq); 2129 break; 2130 } 2131 case KVM_S390_STORE_STATUS: 2132 idx = srcu_read_lock(&vcpu->kvm->srcu); 2133 r = kvm_s390_vcpu_store_status(vcpu, arg); 2134 srcu_read_unlock(&vcpu->kvm->srcu, idx); 2135 break; 2136 case KVM_S390_SET_INITIAL_PSW: { 2137 psw_t psw; 2138 2139 r = -EFAULT; 2140 if (copy_from_user(&psw, argp, sizeof(psw))) 2141 break; 2142 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw); 2143 break; 2144 } 2145 case KVM_S390_INITIAL_RESET: 2146 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu); 2147 break; 2148 case KVM_SET_ONE_REG: 2149 case KVM_GET_ONE_REG: { 2150 struct kvm_one_reg reg; 2151 r = -EFAULT; 2152 if (copy_from_user(®, argp, sizeof(reg))) 2153 break; 2154 if (ioctl == KVM_SET_ONE_REG) 2155 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®); 2156 else 2157 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®); 2158 break; 2159 } 2160 #ifdef CONFIG_KVM_S390_UCONTROL 2161 case KVM_S390_UCAS_MAP: { 2162 struct kvm_s390_ucas_mapping ucasmap; 2163 2164 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) { 2165 r = -EFAULT; 2166 break; 2167 } 2168 2169 if (!kvm_is_ucontrol(vcpu->kvm)) { 2170 r = -EINVAL; 2171 break; 2172 } 2173 2174 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr, 2175 ucasmap.vcpu_addr, ucasmap.length); 2176 break; 2177 } 2178 case KVM_S390_UCAS_UNMAP: { 2179 struct kvm_s390_ucas_mapping ucasmap; 2180 2181 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) { 2182 r = -EFAULT; 2183 break; 2184 } 2185 2186 if (!kvm_is_ucontrol(vcpu->kvm)) { 2187 r = -EINVAL; 2188 break; 2189 } 2190 2191 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr, 2192 ucasmap.length); 2193 break; 2194 } 2195 #endif 2196 case KVM_S390_VCPU_FAULT: { 2197 r = gmap_fault(vcpu->arch.gmap, arg, 0); 2198 break; 2199 } 2200 case KVM_ENABLE_CAP: 2201 { 2202 struct kvm_enable_cap cap; 2203 r = -EFAULT; 2204 if (copy_from_user(&cap, argp, sizeof(cap))) 2205 break; 2206 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); 2207 break; 2208 } 2209 default: 2210 r = -ENOTTY; 2211 } 2212 return r; 2213 } 2214 2215 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 2216 { 2217 #ifdef CONFIG_KVM_S390_UCONTROL 2218 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET) 2219 && (kvm_is_ucontrol(vcpu->kvm))) { 2220 vmf->page = virt_to_page(vcpu->arch.sie_block); 2221 get_page(vmf->page); 2222 return 0; 2223 } 2224 #endif 2225 return VM_FAULT_SIGBUS; 2226 } 2227 2228 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, 2229 unsigned long npages) 2230 { 2231 return 0; 2232 } 2233 2234 /* Section: memory related */ 2235 int kvm_arch_prepare_memory_region(struct kvm *kvm, 2236 struct kvm_memory_slot *memslot, 2237 struct kvm_userspace_memory_region *mem, 2238 enum kvm_mr_change change) 2239 { 2240 /* A few sanity checks. We can have memory slots which have to be 2241 located/ended at a segment boundary (1MB). The memory in userland is 2242 ok to be fragmented into various different vmas. It is okay to mmap() 2243 and munmap() stuff in this slot after doing this call at any time */ 2244 2245 if (mem->userspace_addr & 0xffffful) 2246 return -EINVAL; 2247 2248 if (mem->memory_size & 0xffffful) 2249 return -EINVAL; 2250 2251 return 0; 2252 } 2253 2254 void kvm_arch_commit_memory_region(struct kvm *kvm, 2255 struct kvm_userspace_memory_region *mem, 2256 const struct kvm_memory_slot *old, 2257 enum kvm_mr_change change) 2258 { 2259 int rc; 2260 2261 /* If the basics of the memslot do not change, we do not want 2262 * to update the gmap. Every update causes several unnecessary 2263 * segment translation exceptions. This is usually handled just 2264 * fine by the normal fault handler + gmap, but it will also 2265 * cause faults on the prefix page of running guest CPUs. 2266 */ 2267 if (old->userspace_addr == mem->userspace_addr && 2268 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr && 2269 old->npages * PAGE_SIZE == mem->memory_size) 2270 return; 2271 2272 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr, 2273 mem->guest_phys_addr, mem->memory_size); 2274 if (rc) 2275 printk(KERN_WARNING "kvm-s390: failed to commit memory region\n"); 2276 return; 2277 } 2278 2279 static int __init kvm_s390_init(void) 2280 { 2281 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); 2282 } 2283 2284 static void __exit kvm_s390_exit(void) 2285 { 2286 kvm_exit(); 2287 } 2288 2289 module_init(kvm_s390_init); 2290 module_exit(kvm_s390_exit); 2291 2292 /* 2293 * Enable autoloading of the kvm module. 2294 * Note that we add the module alias here instead of virt/kvm/kvm_main.c 2295 * since x86 takes a different approach. 2296 */ 2297 #include <linux/miscdevice.h> 2298 MODULE_ALIAS_MISCDEV(KVM_MINOR); 2299 MODULE_ALIAS("devname:kvm"); 2300