1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * handling kvm guest interrupts 4 * 5 * Copyright IBM Corp. 2008, 2020 6 * 7 * Author(s): Carsten Otte <cotte@de.ibm.com> 8 */ 9 10 #define KMSG_COMPONENT "kvm-s390" 11 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 12 13 #include <linux/interrupt.h> 14 #include <linux/kvm_host.h> 15 #include <linux/hrtimer.h> 16 #include <linux/mmu_context.h> 17 #include <linux/nospec.h> 18 #include <linux/signal.h> 19 #include <linux/slab.h> 20 #include <linux/bitmap.h> 21 #include <linux/vmalloc.h> 22 #include <asm/asm-offsets.h> 23 #include <asm/dis.h> 24 #include <linux/uaccess.h> 25 #include <asm/sclp.h> 26 #include <asm/isc.h> 27 #include <asm/gmap.h> 28 #include <asm/switch_to.h> 29 #include <asm/nmi.h> 30 #include <asm/airq.h> 31 #include "kvm-s390.h" 32 #include "gaccess.h" 33 #include "trace-s390.h" 34 35 #define PFAULT_INIT 0x0600 36 #define PFAULT_DONE 0x0680 37 #define VIRTIO_PARAM 0x0d00 38 39 static struct kvm_s390_gib *gib; 40 41 /* handle external calls via sigp interpretation facility */ 42 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id) 43 { 44 int c, scn; 45 46 if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND)) 47 return 0; 48 49 BUG_ON(!kvm_s390_use_sca_entries()); 50 read_lock(&vcpu->kvm->arch.sca_lock); 51 if (vcpu->kvm->arch.use_esca) { 52 struct esca_block *sca = vcpu->kvm->arch.sca; 53 union esca_sigp_ctrl sigp_ctrl = 54 sca->cpu[vcpu->vcpu_id].sigp_ctrl; 55 56 c = sigp_ctrl.c; 57 scn = sigp_ctrl.scn; 58 } else { 59 struct bsca_block *sca = vcpu->kvm->arch.sca; 60 union bsca_sigp_ctrl sigp_ctrl = 61 sca->cpu[vcpu->vcpu_id].sigp_ctrl; 62 63 c = sigp_ctrl.c; 64 scn = sigp_ctrl.scn; 65 } 66 read_unlock(&vcpu->kvm->arch.sca_lock); 67 68 if (src_id) 69 *src_id = scn; 70 71 return c; 72 } 73 74 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id) 75 { 76 int expect, rc; 77 78 BUG_ON(!kvm_s390_use_sca_entries()); 79 read_lock(&vcpu->kvm->arch.sca_lock); 80 if (vcpu->kvm->arch.use_esca) { 81 struct esca_block *sca = vcpu->kvm->arch.sca; 82 union esca_sigp_ctrl *sigp_ctrl = 83 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl); 84 union esca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl; 85 86 new_val.scn = src_id; 87 new_val.c = 1; 88 old_val.c = 0; 89 90 expect = old_val.value; 91 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value); 92 } else { 93 struct bsca_block *sca = vcpu->kvm->arch.sca; 94 union bsca_sigp_ctrl *sigp_ctrl = 95 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl); 96 union bsca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl; 97 98 new_val.scn = src_id; 99 new_val.c = 1; 100 old_val.c = 0; 101 102 expect = old_val.value; 103 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value); 104 } 105 read_unlock(&vcpu->kvm->arch.sca_lock); 106 107 if (rc != expect) { 108 /* another external call is pending */ 109 return -EBUSY; 110 } 111 kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND); 112 return 0; 113 } 114 115 static void sca_clear_ext_call(struct kvm_vcpu *vcpu) 116 { 117 int rc, expect; 118 119 if (!kvm_s390_use_sca_entries()) 120 return; 121 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND); 122 read_lock(&vcpu->kvm->arch.sca_lock); 123 if (vcpu->kvm->arch.use_esca) { 124 struct esca_block *sca = vcpu->kvm->arch.sca; 125 union esca_sigp_ctrl *sigp_ctrl = 126 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl); 127 union esca_sigp_ctrl old = *sigp_ctrl; 128 129 expect = old.value; 130 rc = cmpxchg(&sigp_ctrl->value, old.value, 0); 131 } else { 132 struct bsca_block *sca = vcpu->kvm->arch.sca; 133 union bsca_sigp_ctrl *sigp_ctrl = 134 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl); 135 union bsca_sigp_ctrl old = *sigp_ctrl; 136 137 expect = old.value; 138 rc = cmpxchg(&sigp_ctrl->value, old.value, 0); 139 } 140 read_unlock(&vcpu->kvm->arch.sca_lock); 141 WARN_ON(rc != expect); /* cannot clear? */ 142 } 143 144 int psw_extint_disabled(struct kvm_vcpu *vcpu) 145 { 146 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT); 147 } 148 149 static int psw_ioint_disabled(struct kvm_vcpu *vcpu) 150 { 151 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO); 152 } 153 154 static int psw_mchk_disabled(struct kvm_vcpu *vcpu) 155 { 156 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK); 157 } 158 159 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu) 160 { 161 return psw_extint_disabled(vcpu) && 162 psw_ioint_disabled(vcpu) && 163 psw_mchk_disabled(vcpu); 164 } 165 166 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu) 167 { 168 if (psw_extint_disabled(vcpu) || 169 !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK)) 170 return 0; 171 if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu)) 172 /* No timer interrupts when single stepping */ 173 return 0; 174 return 1; 175 } 176 177 static int ckc_irq_pending(struct kvm_vcpu *vcpu) 178 { 179 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm); 180 const u64 ckc = vcpu->arch.sie_block->ckc; 181 182 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) { 183 if ((s64)ckc >= (s64)now) 184 return 0; 185 } else if (ckc >= now) { 186 return 0; 187 } 188 return ckc_interrupts_enabled(vcpu); 189 } 190 191 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu) 192 { 193 return !psw_extint_disabled(vcpu) && 194 (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK); 195 } 196 197 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu) 198 { 199 if (!cpu_timer_interrupts_enabled(vcpu)) 200 return 0; 201 return kvm_s390_get_cpu_timer(vcpu) >> 63; 202 } 203 204 static uint64_t isc_to_isc_bits(int isc) 205 { 206 return (0x80 >> isc) << 24; 207 } 208 209 static inline u32 isc_to_int_word(u8 isc) 210 { 211 return ((u32)isc << 27) | 0x80000000; 212 } 213 214 static inline u8 int_word_to_isc(u32 int_word) 215 { 216 return (int_word & 0x38000000) >> 27; 217 } 218 219 /* 220 * To use atomic bitmap functions, we have to provide a bitmap address 221 * that is u64 aligned. However, the ipm might be u32 aligned. 222 * Therefore, we logically start the bitmap at the very beginning of the 223 * struct and fixup the bit number. 224 */ 225 #define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE) 226 227 /** 228 * gisa_set_iam - change the GISA interruption alert mask 229 * 230 * @gisa: gisa to operate on 231 * @iam: new IAM value to use 232 * 233 * Change the IAM atomically with the next alert address and the IPM 234 * of the GISA if the GISA is not part of the GIB alert list. All three 235 * fields are located in the first long word of the GISA. 236 * 237 * Returns: 0 on success 238 * -EBUSY in case the gisa is part of the alert list 239 */ 240 static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam) 241 { 242 u64 word, _word; 243 244 do { 245 word = READ_ONCE(gisa->u64.word[0]); 246 if ((u64)gisa != word >> 32) 247 return -EBUSY; 248 _word = (word & ~0xffUL) | iam; 249 } while (cmpxchg(&gisa->u64.word[0], word, _word) != word); 250 251 return 0; 252 } 253 254 /** 255 * gisa_clear_ipm - clear the GISA interruption pending mask 256 * 257 * @gisa: gisa to operate on 258 * 259 * Clear the IPM atomically with the next alert address and the IAM 260 * of the GISA unconditionally. All three fields are located in the 261 * first long word of the GISA. 262 */ 263 static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa) 264 { 265 u64 word, _word; 266 267 do { 268 word = READ_ONCE(gisa->u64.word[0]); 269 _word = word & ~(0xffUL << 24); 270 } while (cmpxchg(&gisa->u64.word[0], word, _word) != word); 271 } 272 273 /** 274 * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM 275 * 276 * @gi: gisa interrupt struct to work on 277 * 278 * Atomically restores the interruption alert mask if none of the 279 * relevant ISCs are pending and return the IPM. 280 * 281 * Returns: the relevant pending ISCs 282 */ 283 static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi) 284 { 285 u8 pending_mask, alert_mask; 286 u64 word, _word; 287 288 do { 289 word = READ_ONCE(gi->origin->u64.word[0]); 290 alert_mask = READ_ONCE(gi->alert.mask); 291 pending_mask = (u8)(word >> 24) & alert_mask; 292 if (pending_mask) 293 return pending_mask; 294 _word = (word & ~0xffUL) | alert_mask; 295 } while (cmpxchg(&gi->origin->u64.word[0], word, _word) != word); 296 297 return 0; 298 } 299 300 static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa) 301 { 302 return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa; 303 } 304 305 static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc) 306 { 307 set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa); 308 } 309 310 static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa) 311 { 312 return READ_ONCE(gisa->ipm); 313 } 314 315 static inline void gisa_clear_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc) 316 { 317 clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa); 318 } 319 320 static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc) 321 { 322 return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa); 323 } 324 325 static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu) 326 { 327 unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs | 328 vcpu->arch.local_int.pending_irqs; 329 330 pending &= ~vcpu->kvm->arch.float_int.masked_irqs; 331 return pending; 332 } 333 334 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu) 335 { 336 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int; 337 unsigned long pending_mask; 338 339 pending_mask = pending_irqs_no_gisa(vcpu); 340 if (gi->origin) 341 pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7; 342 return pending_mask; 343 } 344 345 static inline int isc_to_irq_type(unsigned long isc) 346 { 347 return IRQ_PEND_IO_ISC_0 - isc; 348 } 349 350 static inline int irq_type_to_isc(unsigned long irq_type) 351 { 352 return IRQ_PEND_IO_ISC_0 - irq_type; 353 } 354 355 static unsigned long disable_iscs(struct kvm_vcpu *vcpu, 356 unsigned long active_mask) 357 { 358 int i; 359 360 for (i = 0; i <= MAX_ISC; i++) 361 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i))) 362 active_mask &= ~(1UL << (isc_to_irq_type(i))); 363 364 return active_mask; 365 } 366 367 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu) 368 { 369 unsigned long active_mask; 370 371 active_mask = pending_irqs(vcpu); 372 if (!active_mask) 373 return 0; 374 375 if (psw_extint_disabled(vcpu)) 376 active_mask &= ~IRQ_PEND_EXT_MASK; 377 if (psw_ioint_disabled(vcpu)) 378 active_mask &= ~IRQ_PEND_IO_MASK; 379 else 380 active_mask = disable_iscs(vcpu, active_mask); 381 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK)) 382 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask); 383 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK)) 384 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask); 385 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK)) 386 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask); 387 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK)) 388 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask); 389 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) { 390 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask); 391 __clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask); 392 } 393 if (psw_mchk_disabled(vcpu)) 394 active_mask &= ~IRQ_PEND_MCHK_MASK; 395 /* PV guest cpus can have a single interruption injected at a time. */ 396 if (kvm_s390_pv_cpu_get_handle(vcpu) && 397 vcpu->arch.sie_block->iictl != IICTL_CODE_NONE) 398 active_mask &= ~(IRQ_PEND_EXT_II_MASK | 399 IRQ_PEND_IO_MASK | 400 IRQ_PEND_MCHK_MASK); 401 /* 402 * Check both floating and local interrupt's cr14 because 403 * bit IRQ_PEND_MCHK_REP could be set in both cases. 404 */ 405 if (!(vcpu->arch.sie_block->gcr[14] & 406 (vcpu->kvm->arch.float_int.mchk.cr14 | 407 vcpu->arch.local_int.irq.mchk.cr14))) 408 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask); 409 410 /* 411 * STOP irqs will never be actively delivered. They are triggered via 412 * intercept requests and cleared when the stop intercept is performed. 413 */ 414 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask); 415 416 return active_mask; 417 } 418 419 static void __set_cpu_idle(struct kvm_vcpu *vcpu) 420 { 421 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT); 422 set_bit(vcpu->vcpu_id, vcpu->kvm->arch.idle_mask); 423 } 424 425 static void __unset_cpu_idle(struct kvm_vcpu *vcpu) 426 { 427 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT); 428 clear_bit(vcpu->vcpu_id, vcpu->kvm->arch.idle_mask); 429 } 430 431 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu) 432 { 433 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT | 434 CPUSTAT_STOP_INT); 435 vcpu->arch.sie_block->lctl = 0x0000; 436 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT); 437 438 if (guestdbg_enabled(vcpu)) { 439 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 | 440 LCTL_CR10 | LCTL_CR11); 441 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT); 442 } 443 } 444 445 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu) 446 { 447 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK)) 448 return; 449 if (psw_ioint_disabled(vcpu)) 450 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT); 451 else 452 vcpu->arch.sie_block->lctl |= LCTL_CR6; 453 } 454 455 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu) 456 { 457 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK)) 458 return; 459 if (psw_extint_disabled(vcpu)) 460 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 461 else 462 vcpu->arch.sie_block->lctl |= LCTL_CR0; 463 } 464 465 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu) 466 { 467 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK)) 468 return; 469 if (psw_mchk_disabled(vcpu)) 470 vcpu->arch.sie_block->ictl |= ICTL_LPSW; 471 else 472 vcpu->arch.sie_block->lctl |= LCTL_CR14; 473 } 474 475 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu) 476 { 477 if (kvm_s390_is_stop_irq_pending(vcpu)) 478 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT); 479 } 480 481 /* Set interception request for non-deliverable interrupts */ 482 static void set_intercept_indicators(struct kvm_vcpu *vcpu) 483 { 484 set_intercept_indicators_io(vcpu); 485 set_intercept_indicators_ext(vcpu); 486 set_intercept_indicators_mchk(vcpu); 487 set_intercept_indicators_stop(vcpu); 488 } 489 490 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu) 491 { 492 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 493 int rc = 0; 494 495 vcpu->stat.deliver_cputm++; 496 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER, 497 0, 0); 498 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 499 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT; 500 vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER; 501 } else { 502 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER, 503 (u16 *)__LC_EXT_INT_CODE); 504 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR); 505 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 506 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 507 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 508 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 509 } 510 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 511 return rc ? -EFAULT : 0; 512 } 513 514 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu) 515 { 516 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 517 int rc = 0; 518 519 vcpu->stat.deliver_ckc++; 520 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP, 521 0, 0); 522 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 523 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT; 524 vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP; 525 } else { 526 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP, 527 (u16 __user *)__LC_EXT_INT_CODE); 528 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR); 529 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 530 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 531 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 532 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 533 } 534 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 535 return rc ? -EFAULT : 0; 536 } 537 538 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu) 539 { 540 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 541 struct kvm_s390_ext_info ext; 542 int rc; 543 544 spin_lock(&li->lock); 545 ext = li->irq.ext; 546 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs); 547 li->irq.ext.ext_params2 = 0; 548 spin_unlock(&li->lock); 549 550 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx", 551 ext.ext_params2); 552 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 553 KVM_S390_INT_PFAULT_INIT, 554 0, ext.ext_params2); 555 556 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE); 557 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR); 558 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 559 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 560 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 561 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 562 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2); 563 return rc ? -EFAULT : 0; 564 } 565 566 static int __write_machine_check(struct kvm_vcpu *vcpu, 567 struct kvm_s390_mchk_info *mchk) 568 { 569 unsigned long ext_sa_addr; 570 unsigned long lc; 571 freg_t fprs[NUM_FPRS]; 572 union mci mci; 573 int rc; 574 575 /* 576 * All other possible payload for a machine check (e.g. the register 577 * contents in the save area) will be handled by the ultravisor, as 578 * the hypervisor does not not have the needed information for 579 * protected guests. 580 */ 581 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 582 vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK; 583 vcpu->arch.sie_block->mcic = mchk->mcic; 584 vcpu->arch.sie_block->faddr = mchk->failing_storage_address; 585 vcpu->arch.sie_block->edc = mchk->ext_damage_code; 586 return 0; 587 } 588 589 mci.val = mchk->mcic; 590 /* take care of lazy register loading */ 591 save_fpu_regs(); 592 save_access_regs(vcpu->run->s.regs.acrs); 593 if (MACHINE_HAS_GS && vcpu->arch.gs_enabled) 594 save_gs_cb(current->thread.gs_cb); 595 596 /* Extended save area */ 597 rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr, 598 sizeof(unsigned long)); 599 /* Only bits 0 through 63-LC are used for address formation */ 600 lc = ext_sa_addr & MCESA_LC_MASK; 601 if (test_kvm_facility(vcpu->kvm, 133)) { 602 switch (lc) { 603 case 0: 604 case 10: 605 ext_sa_addr &= ~0x3ffUL; 606 break; 607 case 11: 608 ext_sa_addr &= ~0x7ffUL; 609 break; 610 case 12: 611 ext_sa_addr &= ~0xfffUL; 612 break; 613 default: 614 ext_sa_addr = 0; 615 break; 616 } 617 } else { 618 ext_sa_addr &= ~0x3ffUL; 619 } 620 621 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) { 622 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs, 623 512)) 624 mci.vr = 0; 625 } else { 626 mci.vr = 0; 627 } 628 if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133) 629 && (lc == 11 || lc == 12)) { 630 if (write_guest_abs(vcpu, ext_sa_addr + 1024, 631 &vcpu->run->s.regs.gscb, 32)) 632 mci.gs = 0; 633 } else { 634 mci.gs = 0; 635 } 636 637 /* General interruption information */ 638 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID); 639 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW, 640 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 641 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW, 642 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 643 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE); 644 645 /* Register-save areas */ 646 if (MACHINE_HAS_VX) { 647 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs); 648 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128); 649 } else { 650 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, 651 vcpu->run->s.regs.fprs, 128); 652 } 653 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA, 654 vcpu->run->s.regs.gprs, 128); 655 rc |= put_guest_lc(vcpu, current->thread.fpu.fpc, 656 (u32 __user *) __LC_FP_CREG_SAVE_AREA); 657 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr, 658 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA); 659 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu), 660 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA); 661 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8, 662 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA); 663 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA, 664 &vcpu->run->s.regs.acrs, 64); 665 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA, 666 &vcpu->arch.sie_block->gcr, 128); 667 668 /* Extended interruption information */ 669 rc |= put_guest_lc(vcpu, mchk->ext_damage_code, 670 (u32 __user *) __LC_EXT_DAMAGE_CODE); 671 rc |= put_guest_lc(vcpu, mchk->failing_storage_address, 672 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR); 673 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout, 674 sizeof(mchk->fixed_logout)); 675 return rc ? -EFAULT : 0; 676 } 677 678 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu) 679 { 680 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; 681 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 682 struct kvm_s390_mchk_info mchk = {}; 683 int deliver = 0; 684 int rc = 0; 685 686 spin_lock(&fi->lock); 687 spin_lock(&li->lock); 688 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) || 689 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) { 690 /* 691 * If there was an exigent machine check pending, then any 692 * repressible machine checks that might have been pending 693 * are indicated along with it, so always clear bits for 694 * repressible and exigent interrupts 695 */ 696 mchk = li->irq.mchk; 697 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs); 698 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs); 699 memset(&li->irq.mchk, 0, sizeof(mchk)); 700 deliver = 1; 701 } 702 /* 703 * We indicate floating repressible conditions along with 704 * other pending conditions. Channel Report Pending and Channel 705 * Subsystem damage are the only two and and are indicated by 706 * bits in mcic and masked in cr14. 707 */ 708 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) { 709 mchk.mcic |= fi->mchk.mcic; 710 mchk.cr14 |= fi->mchk.cr14; 711 memset(&fi->mchk, 0, sizeof(mchk)); 712 deliver = 1; 713 } 714 spin_unlock(&li->lock); 715 spin_unlock(&fi->lock); 716 717 if (deliver) { 718 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx", 719 mchk.mcic); 720 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 721 KVM_S390_MCHK, 722 mchk.cr14, mchk.mcic); 723 vcpu->stat.deliver_machine_check++; 724 rc = __write_machine_check(vcpu, &mchk); 725 } 726 return rc; 727 } 728 729 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu) 730 { 731 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 732 int rc = 0; 733 734 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart"); 735 vcpu->stat.deliver_restart_signal++; 736 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0); 737 738 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 739 vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART; 740 } else { 741 rc = write_guest_lc(vcpu, 742 offsetof(struct lowcore, restart_old_psw), 743 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 744 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw), 745 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 746 } 747 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs); 748 return rc ? -EFAULT : 0; 749 } 750 751 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu) 752 { 753 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 754 struct kvm_s390_prefix_info prefix; 755 756 spin_lock(&li->lock); 757 prefix = li->irq.prefix; 758 li->irq.prefix.address = 0; 759 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs); 760 spin_unlock(&li->lock); 761 762 vcpu->stat.deliver_prefix_signal++; 763 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 764 KVM_S390_SIGP_SET_PREFIX, 765 prefix.address, 0); 766 767 kvm_s390_set_prefix(vcpu, prefix.address); 768 return 0; 769 } 770 771 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu) 772 { 773 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 774 int rc; 775 int cpu_addr; 776 777 spin_lock(&li->lock); 778 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS); 779 clear_bit(cpu_addr, li->sigp_emerg_pending); 780 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS)) 781 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs); 782 spin_unlock(&li->lock); 783 784 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg"); 785 vcpu->stat.deliver_emergency_signal++; 786 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY, 787 cpu_addr, 0); 788 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 789 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT; 790 vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG; 791 vcpu->arch.sie_block->extcpuaddr = cpu_addr; 792 return 0; 793 } 794 795 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG, 796 (u16 *)__LC_EXT_INT_CODE); 797 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR); 798 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 799 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 800 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 801 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 802 return rc ? -EFAULT : 0; 803 } 804 805 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu) 806 { 807 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 808 struct kvm_s390_extcall_info extcall; 809 int rc; 810 811 spin_lock(&li->lock); 812 extcall = li->irq.extcall; 813 li->irq.extcall.code = 0; 814 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs); 815 spin_unlock(&li->lock); 816 817 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call"); 818 vcpu->stat.deliver_external_call++; 819 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 820 KVM_S390_INT_EXTERNAL_CALL, 821 extcall.code, 0); 822 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 823 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT; 824 vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL; 825 vcpu->arch.sie_block->extcpuaddr = extcall.code; 826 return 0; 827 } 828 829 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL, 830 (u16 *)__LC_EXT_INT_CODE); 831 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR); 832 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 833 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 834 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw, 835 sizeof(psw_t)); 836 return rc ? -EFAULT : 0; 837 } 838 839 static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code) 840 { 841 switch (code) { 842 case PGM_SPECIFICATION: 843 vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION; 844 break; 845 case PGM_OPERAND: 846 vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND; 847 break; 848 default: 849 return -EINVAL; 850 } 851 return 0; 852 } 853 854 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu) 855 { 856 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 857 struct kvm_s390_pgm_info pgm_info; 858 int rc = 0, nullifying = false; 859 u16 ilen; 860 861 spin_lock(&li->lock); 862 pgm_info = li->irq.pgm; 863 clear_bit(IRQ_PEND_PROG, &li->pending_irqs); 864 memset(&li->irq.pgm, 0, sizeof(pgm_info)); 865 spin_unlock(&li->lock); 866 867 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK; 868 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d", 869 pgm_info.code, ilen); 870 vcpu->stat.deliver_program++; 871 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT, 872 pgm_info.code, 0); 873 874 /* PER is handled by the ultravisor */ 875 if (kvm_s390_pv_cpu_is_protected(vcpu)) 876 return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER); 877 878 switch (pgm_info.code & ~PGM_PER) { 879 case PGM_AFX_TRANSLATION: 880 case PGM_ASX_TRANSLATION: 881 case PGM_EX_TRANSLATION: 882 case PGM_LFX_TRANSLATION: 883 case PGM_LSTE_SEQUENCE: 884 case PGM_LSX_TRANSLATION: 885 case PGM_LX_TRANSLATION: 886 case PGM_PRIMARY_AUTHORITY: 887 case PGM_SECONDARY_AUTHORITY: 888 nullifying = true; 889 fallthrough; 890 case PGM_SPACE_SWITCH: 891 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code, 892 (u64 *)__LC_TRANS_EXC_CODE); 893 break; 894 case PGM_ALEN_TRANSLATION: 895 case PGM_ALE_SEQUENCE: 896 case PGM_ASTE_INSTANCE: 897 case PGM_ASTE_SEQUENCE: 898 case PGM_ASTE_VALIDITY: 899 case PGM_EXTENDED_AUTHORITY: 900 rc = put_guest_lc(vcpu, pgm_info.exc_access_id, 901 (u8 *)__LC_EXC_ACCESS_ID); 902 nullifying = true; 903 break; 904 case PGM_ASCE_TYPE: 905 case PGM_PAGE_TRANSLATION: 906 case PGM_REGION_FIRST_TRANS: 907 case PGM_REGION_SECOND_TRANS: 908 case PGM_REGION_THIRD_TRANS: 909 case PGM_SEGMENT_TRANSLATION: 910 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code, 911 (u64 *)__LC_TRANS_EXC_CODE); 912 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id, 913 (u8 *)__LC_EXC_ACCESS_ID); 914 rc |= put_guest_lc(vcpu, pgm_info.op_access_id, 915 (u8 *)__LC_OP_ACCESS_ID); 916 nullifying = true; 917 break; 918 case PGM_MONITOR: 919 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr, 920 (u16 *)__LC_MON_CLASS_NR); 921 rc |= put_guest_lc(vcpu, pgm_info.mon_code, 922 (u64 *)__LC_MON_CODE); 923 break; 924 case PGM_VECTOR_PROCESSING: 925 case PGM_DATA: 926 rc = put_guest_lc(vcpu, pgm_info.data_exc_code, 927 (u32 *)__LC_DATA_EXC_CODE); 928 break; 929 case PGM_PROTECTION: 930 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code, 931 (u64 *)__LC_TRANS_EXC_CODE); 932 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id, 933 (u8 *)__LC_EXC_ACCESS_ID); 934 break; 935 case PGM_STACK_FULL: 936 case PGM_STACK_EMPTY: 937 case PGM_STACK_SPECIFICATION: 938 case PGM_STACK_TYPE: 939 case PGM_STACK_OPERATION: 940 case PGM_TRACE_TABEL: 941 case PGM_CRYPTO_OPERATION: 942 nullifying = true; 943 break; 944 } 945 946 if (pgm_info.code & PGM_PER) { 947 rc |= put_guest_lc(vcpu, pgm_info.per_code, 948 (u8 *) __LC_PER_CODE); 949 rc |= put_guest_lc(vcpu, pgm_info.per_atmid, 950 (u8 *)__LC_PER_ATMID); 951 rc |= put_guest_lc(vcpu, pgm_info.per_address, 952 (u64 *) __LC_PER_ADDRESS); 953 rc |= put_guest_lc(vcpu, pgm_info.per_access_id, 954 (u8 *) __LC_PER_ACCESS_ID); 955 } 956 957 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND)) 958 kvm_s390_rewind_psw(vcpu, ilen); 959 960 /* bit 1+2 of the target are the ilc, so we can directly use ilen */ 961 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC); 962 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea, 963 (u64 *) __LC_LAST_BREAK); 964 rc |= put_guest_lc(vcpu, pgm_info.code, 965 (u16 *)__LC_PGM_INT_CODE); 966 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW, 967 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 968 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW, 969 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 970 return rc ? -EFAULT : 0; 971 } 972 973 #define SCCB_MASK 0xFFFFFFF8 974 #define SCCB_EVENT_PENDING 0x3 975 976 static int write_sclp(struct kvm_vcpu *vcpu, u32 parm) 977 { 978 int rc; 979 980 if (kvm_s390_pv_cpu_get_handle(vcpu)) { 981 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT; 982 vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG; 983 vcpu->arch.sie_block->eiparams = parm; 984 return 0; 985 } 986 987 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE); 988 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR); 989 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 990 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 991 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 992 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 993 rc |= put_guest_lc(vcpu, parm, 994 (u32 *)__LC_EXT_PARAMS); 995 996 return rc ? -EFAULT : 0; 997 } 998 999 static int __must_check __deliver_service(struct kvm_vcpu *vcpu) 1000 { 1001 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; 1002 struct kvm_s390_ext_info ext; 1003 1004 spin_lock(&fi->lock); 1005 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) || 1006 !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) { 1007 spin_unlock(&fi->lock); 1008 return 0; 1009 } 1010 ext = fi->srv_signal; 1011 memset(&fi->srv_signal, 0, sizeof(ext)); 1012 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs); 1013 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs); 1014 if (kvm_s390_pv_cpu_is_protected(vcpu)) 1015 set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs); 1016 spin_unlock(&fi->lock); 1017 1018 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x", 1019 ext.ext_params); 1020 vcpu->stat.deliver_service_signal++; 1021 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE, 1022 ext.ext_params, 0); 1023 1024 return write_sclp(vcpu, ext.ext_params); 1025 } 1026 1027 static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu) 1028 { 1029 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; 1030 struct kvm_s390_ext_info ext; 1031 1032 spin_lock(&fi->lock); 1033 if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) { 1034 spin_unlock(&fi->lock); 1035 return 0; 1036 } 1037 ext = fi->srv_signal; 1038 /* only clear the event bit */ 1039 fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING; 1040 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs); 1041 spin_unlock(&fi->lock); 1042 1043 VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event"); 1044 vcpu->stat.deliver_service_signal++; 1045 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE, 1046 ext.ext_params, 0); 1047 1048 return write_sclp(vcpu, SCCB_EVENT_PENDING); 1049 } 1050 1051 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu) 1052 { 1053 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; 1054 struct kvm_s390_interrupt_info *inti; 1055 int rc = 0; 1056 1057 spin_lock(&fi->lock); 1058 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT], 1059 struct kvm_s390_interrupt_info, 1060 list); 1061 if (inti) { 1062 list_del(&inti->list); 1063 fi->counters[FIRQ_CNTR_PFAULT] -= 1; 1064 } 1065 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT])) 1066 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs); 1067 spin_unlock(&fi->lock); 1068 1069 if (inti) { 1070 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 1071 KVM_S390_INT_PFAULT_DONE, 0, 1072 inti->ext.ext_params2); 1073 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx", 1074 inti->ext.ext_params2); 1075 1076 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, 1077 (u16 *)__LC_EXT_INT_CODE); 1078 rc |= put_guest_lc(vcpu, PFAULT_DONE, 1079 (u16 *)__LC_EXT_CPU_ADDR); 1080 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 1081 &vcpu->arch.sie_block->gpsw, 1082 sizeof(psw_t)); 1083 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 1084 &vcpu->arch.sie_block->gpsw, 1085 sizeof(psw_t)); 1086 rc |= put_guest_lc(vcpu, inti->ext.ext_params2, 1087 (u64 *)__LC_EXT_PARAMS2); 1088 kfree(inti); 1089 } 1090 return rc ? -EFAULT : 0; 1091 } 1092 1093 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu) 1094 { 1095 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; 1096 struct kvm_s390_interrupt_info *inti; 1097 int rc = 0; 1098 1099 spin_lock(&fi->lock); 1100 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO], 1101 struct kvm_s390_interrupt_info, 1102 list); 1103 if (inti) { 1104 VCPU_EVENT(vcpu, 4, 1105 "deliver: virtio parm: 0x%x,parm64: 0x%llx", 1106 inti->ext.ext_params, inti->ext.ext_params2); 1107 vcpu->stat.deliver_virtio++; 1108 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 1109 inti->type, 1110 inti->ext.ext_params, 1111 inti->ext.ext_params2); 1112 list_del(&inti->list); 1113 fi->counters[FIRQ_CNTR_VIRTIO] -= 1; 1114 } 1115 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO])) 1116 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs); 1117 spin_unlock(&fi->lock); 1118 1119 if (inti) { 1120 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, 1121 (u16 *)__LC_EXT_INT_CODE); 1122 rc |= put_guest_lc(vcpu, VIRTIO_PARAM, 1123 (u16 *)__LC_EXT_CPU_ADDR); 1124 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 1125 &vcpu->arch.sie_block->gpsw, 1126 sizeof(psw_t)); 1127 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 1128 &vcpu->arch.sie_block->gpsw, 1129 sizeof(psw_t)); 1130 rc |= put_guest_lc(vcpu, inti->ext.ext_params, 1131 (u32 *)__LC_EXT_PARAMS); 1132 rc |= put_guest_lc(vcpu, inti->ext.ext_params2, 1133 (u64 *)__LC_EXT_PARAMS2); 1134 kfree(inti); 1135 } 1136 return rc ? -EFAULT : 0; 1137 } 1138 1139 static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io) 1140 { 1141 int rc; 1142 1143 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 1144 vcpu->arch.sie_block->iictl = IICTL_CODE_IO; 1145 vcpu->arch.sie_block->subchannel_id = io->subchannel_id; 1146 vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr; 1147 vcpu->arch.sie_block->io_int_parm = io->io_int_parm; 1148 vcpu->arch.sie_block->io_int_word = io->io_int_word; 1149 return 0; 1150 } 1151 1152 rc = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID); 1153 rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR); 1154 rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM); 1155 rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD); 1156 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW, 1157 &vcpu->arch.sie_block->gpsw, 1158 sizeof(psw_t)); 1159 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW, 1160 &vcpu->arch.sie_block->gpsw, 1161 sizeof(psw_t)); 1162 return rc ? -EFAULT : 0; 1163 } 1164 1165 static int __must_check __deliver_io(struct kvm_vcpu *vcpu, 1166 unsigned long irq_type) 1167 { 1168 struct list_head *isc_list; 1169 struct kvm_s390_float_interrupt *fi; 1170 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int; 1171 struct kvm_s390_interrupt_info *inti = NULL; 1172 struct kvm_s390_io_info io; 1173 u32 isc; 1174 int rc = 0; 1175 1176 fi = &vcpu->kvm->arch.float_int; 1177 1178 spin_lock(&fi->lock); 1179 isc = irq_type_to_isc(irq_type); 1180 isc_list = &fi->lists[isc]; 1181 inti = list_first_entry_or_null(isc_list, 1182 struct kvm_s390_interrupt_info, 1183 list); 1184 if (inti) { 1185 if (inti->type & KVM_S390_INT_IO_AI_MASK) 1186 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)"); 1187 else 1188 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x", 1189 inti->io.subchannel_id >> 8, 1190 inti->io.subchannel_id >> 1 & 0x3, 1191 inti->io.subchannel_nr); 1192 1193 vcpu->stat.deliver_io++; 1194 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 1195 inti->type, 1196 ((__u32)inti->io.subchannel_id << 16) | 1197 inti->io.subchannel_nr, 1198 ((__u64)inti->io.io_int_parm << 32) | 1199 inti->io.io_int_word); 1200 list_del(&inti->list); 1201 fi->counters[FIRQ_CNTR_IO] -= 1; 1202 } 1203 if (list_empty(isc_list)) 1204 clear_bit(irq_type, &fi->pending_irqs); 1205 spin_unlock(&fi->lock); 1206 1207 if (inti) { 1208 rc = __do_deliver_io(vcpu, &(inti->io)); 1209 kfree(inti); 1210 goto out; 1211 } 1212 1213 if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) { 1214 /* 1215 * in case an adapter interrupt was not delivered 1216 * in SIE context KVM will handle the delivery 1217 */ 1218 VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc); 1219 memset(&io, 0, sizeof(io)); 1220 io.io_int_word = isc_to_int_word(isc); 1221 vcpu->stat.deliver_io++; 1222 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 1223 KVM_S390_INT_IO(1, 0, 0, 0), 1224 ((__u32)io.subchannel_id << 16) | 1225 io.subchannel_nr, 1226 ((__u64)io.io_int_parm << 32) | 1227 io.io_int_word); 1228 rc = __do_deliver_io(vcpu, &io); 1229 } 1230 out: 1231 return rc; 1232 } 1233 1234 /* Check whether an external call is pending (deliverable or not) */ 1235 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu) 1236 { 1237 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1238 1239 if (!sclp.has_sigpif) 1240 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs); 1241 1242 return sca_ext_call_pending(vcpu, NULL); 1243 } 1244 1245 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop) 1246 { 1247 if (deliverable_irqs(vcpu)) 1248 return 1; 1249 1250 if (kvm_cpu_has_pending_timer(vcpu)) 1251 return 1; 1252 1253 /* external call pending and deliverable */ 1254 if (kvm_s390_ext_call_pending(vcpu) && 1255 !psw_extint_disabled(vcpu) && 1256 (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK)) 1257 return 1; 1258 1259 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu)) 1260 return 1; 1261 return 0; 1262 } 1263 1264 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) 1265 { 1266 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu); 1267 } 1268 1269 static u64 __calculate_sltime(struct kvm_vcpu *vcpu) 1270 { 1271 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm); 1272 const u64 ckc = vcpu->arch.sie_block->ckc; 1273 u64 cputm, sltime = 0; 1274 1275 if (ckc_interrupts_enabled(vcpu)) { 1276 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) { 1277 if ((s64)now < (s64)ckc) 1278 sltime = tod_to_ns((s64)ckc - (s64)now); 1279 } else if (now < ckc) { 1280 sltime = tod_to_ns(ckc - now); 1281 } 1282 /* already expired */ 1283 if (!sltime) 1284 return 0; 1285 if (cpu_timer_interrupts_enabled(vcpu)) { 1286 cputm = kvm_s390_get_cpu_timer(vcpu); 1287 /* already expired? */ 1288 if (cputm >> 63) 1289 return 0; 1290 return min(sltime, tod_to_ns(cputm)); 1291 } 1292 } else if (cpu_timer_interrupts_enabled(vcpu)) { 1293 sltime = kvm_s390_get_cpu_timer(vcpu); 1294 /* already expired? */ 1295 if (sltime >> 63) 1296 return 0; 1297 } 1298 return sltime; 1299 } 1300 1301 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu) 1302 { 1303 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int; 1304 u64 sltime; 1305 1306 vcpu->stat.exit_wait_state++; 1307 1308 /* fast path */ 1309 if (kvm_arch_vcpu_runnable(vcpu)) 1310 return 0; 1311 1312 if (psw_interrupts_disabled(vcpu)) { 1313 VCPU_EVENT(vcpu, 3, "%s", "disabled wait"); 1314 return -EOPNOTSUPP; /* disabled wait */ 1315 } 1316 1317 if (gi->origin && 1318 (gisa_get_ipm_or_restore_iam(gi) & 1319 vcpu->arch.sie_block->gcr[6] >> 24)) 1320 return 0; 1321 1322 if (!ckc_interrupts_enabled(vcpu) && 1323 !cpu_timer_interrupts_enabled(vcpu)) { 1324 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer"); 1325 __set_cpu_idle(vcpu); 1326 goto no_timer; 1327 } 1328 1329 sltime = __calculate_sltime(vcpu); 1330 if (!sltime) 1331 return 0; 1332 1333 __set_cpu_idle(vcpu); 1334 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL); 1335 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime); 1336 no_timer: 1337 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); 1338 kvm_vcpu_block(vcpu); 1339 __unset_cpu_idle(vcpu); 1340 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 1341 1342 hrtimer_cancel(&vcpu->arch.ckc_timer); 1343 return 0; 1344 } 1345 1346 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu) 1347 { 1348 vcpu->valid_wakeup = true; 1349 kvm_vcpu_wake_up(vcpu); 1350 1351 /* 1352 * The VCPU might not be sleeping but rather executing VSIE. Let's 1353 * kick it, so it leaves the SIE to process the request. 1354 */ 1355 kvm_s390_vsie_kick(vcpu); 1356 } 1357 1358 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer) 1359 { 1360 struct kvm_vcpu *vcpu; 1361 u64 sltime; 1362 1363 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer); 1364 sltime = __calculate_sltime(vcpu); 1365 1366 /* 1367 * If the monotonic clock runs faster than the tod clock we might be 1368 * woken up too early and have to go back to sleep to avoid deadlocks. 1369 */ 1370 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime))) 1371 return HRTIMER_RESTART; 1372 kvm_s390_vcpu_wakeup(vcpu); 1373 return HRTIMER_NORESTART; 1374 } 1375 1376 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu) 1377 { 1378 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1379 1380 spin_lock(&li->lock); 1381 li->pending_irqs = 0; 1382 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS); 1383 memset(&li->irq, 0, sizeof(li->irq)); 1384 spin_unlock(&li->lock); 1385 1386 sca_clear_ext_call(vcpu); 1387 } 1388 1389 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu) 1390 { 1391 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1392 int rc = 0; 1393 unsigned long irq_type; 1394 unsigned long irqs; 1395 1396 __reset_intercept_indicators(vcpu); 1397 1398 /* pending ckc conditions might have been invalidated */ 1399 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 1400 if (ckc_irq_pending(vcpu)) 1401 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 1402 1403 /* pending cpu timer conditions might have been invalidated */ 1404 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 1405 if (cpu_timer_irq_pending(vcpu)) 1406 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 1407 1408 while ((irqs = deliverable_irqs(vcpu)) && !rc) { 1409 /* bits are in the reverse order of interrupt priority */ 1410 irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT); 1411 switch (irq_type) { 1412 case IRQ_PEND_IO_ISC_0: 1413 case IRQ_PEND_IO_ISC_1: 1414 case IRQ_PEND_IO_ISC_2: 1415 case IRQ_PEND_IO_ISC_3: 1416 case IRQ_PEND_IO_ISC_4: 1417 case IRQ_PEND_IO_ISC_5: 1418 case IRQ_PEND_IO_ISC_6: 1419 case IRQ_PEND_IO_ISC_7: 1420 rc = __deliver_io(vcpu, irq_type); 1421 break; 1422 case IRQ_PEND_MCHK_EX: 1423 case IRQ_PEND_MCHK_REP: 1424 rc = __deliver_machine_check(vcpu); 1425 break; 1426 case IRQ_PEND_PROG: 1427 rc = __deliver_prog(vcpu); 1428 break; 1429 case IRQ_PEND_EXT_EMERGENCY: 1430 rc = __deliver_emergency_signal(vcpu); 1431 break; 1432 case IRQ_PEND_EXT_EXTERNAL: 1433 rc = __deliver_external_call(vcpu); 1434 break; 1435 case IRQ_PEND_EXT_CLOCK_COMP: 1436 rc = __deliver_ckc(vcpu); 1437 break; 1438 case IRQ_PEND_EXT_CPU_TIMER: 1439 rc = __deliver_cpu_timer(vcpu); 1440 break; 1441 case IRQ_PEND_RESTART: 1442 rc = __deliver_restart(vcpu); 1443 break; 1444 case IRQ_PEND_SET_PREFIX: 1445 rc = __deliver_set_prefix(vcpu); 1446 break; 1447 case IRQ_PEND_PFAULT_INIT: 1448 rc = __deliver_pfault_init(vcpu); 1449 break; 1450 case IRQ_PEND_EXT_SERVICE: 1451 rc = __deliver_service(vcpu); 1452 break; 1453 case IRQ_PEND_EXT_SERVICE_EV: 1454 rc = __deliver_service_ev(vcpu); 1455 break; 1456 case IRQ_PEND_PFAULT_DONE: 1457 rc = __deliver_pfault_done(vcpu); 1458 break; 1459 case IRQ_PEND_VIRTIO: 1460 rc = __deliver_virtio(vcpu); 1461 break; 1462 default: 1463 WARN_ONCE(1, "Unknown pending irq type %ld", irq_type); 1464 clear_bit(irq_type, &li->pending_irqs); 1465 } 1466 } 1467 1468 set_intercept_indicators(vcpu); 1469 1470 return rc; 1471 } 1472 1473 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1474 { 1475 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1476 1477 vcpu->stat.inject_program++; 1478 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code); 1479 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT, 1480 irq->u.pgm.code, 0); 1481 1482 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) { 1483 /* auto detection if no valid ILC was given */ 1484 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK; 1485 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu); 1486 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID; 1487 } 1488 1489 if (irq->u.pgm.code == PGM_PER) { 1490 li->irq.pgm.code |= PGM_PER; 1491 li->irq.pgm.flags = irq->u.pgm.flags; 1492 /* only modify PER related information */ 1493 li->irq.pgm.per_address = irq->u.pgm.per_address; 1494 li->irq.pgm.per_code = irq->u.pgm.per_code; 1495 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid; 1496 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id; 1497 } else if (!(irq->u.pgm.code & PGM_PER)) { 1498 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) | 1499 irq->u.pgm.code; 1500 li->irq.pgm.flags = irq->u.pgm.flags; 1501 /* only modify non-PER information */ 1502 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code; 1503 li->irq.pgm.mon_code = irq->u.pgm.mon_code; 1504 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code; 1505 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr; 1506 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id; 1507 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id; 1508 } else { 1509 li->irq.pgm = irq->u.pgm; 1510 } 1511 set_bit(IRQ_PEND_PROG, &li->pending_irqs); 1512 return 0; 1513 } 1514 1515 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1516 { 1517 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1518 1519 vcpu->stat.inject_pfault_init++; 1520 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx", 1521 irq->u.ext.ext_params2); 1522 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT, 1523 irq->u.ext.ext_params, 1524 irq->u.ext.ext_params2); 1525 1526 li->irq.ext = irq->u.ext; 1527 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs); 1528 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1529 return 0; 1530 } 1531 1532 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1533 { 1534 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1535 struct kvm_s390_extcall_info *extcall = &li->irq.extcall; 1536 uint16_t src_id = irq->u.extcall.code; 1537 1538 vcpu->stat.inject_external_call++; 1539 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u", 1540 src_id); 1541 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL, 1542 src_id, 0); 1543 1544 /* sending vcpu invalid */ 1545 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL) 1546 return -EINVAL; 1547 1548 if (sclp.has_sigpif && !kvm_s390_pv_cpu_get_handle(vcpu)) 1549 return sca_inject_ext_call(vcpu, src_id); 1550 1551 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs)) 1552 return -EBUSY; 1553 *extcall = irq->u.extcall; 1554 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1555 return 0; 1556 } 1557 1558 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1559 { 1560 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1561 struct kvm_s390_prefix_info *prefix = &li->irq.prefix; 1562 1563 vcpu->stat.inject_set_prefix++; 1564 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x", 1565 irq->u.prefix.address); 1566 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX, 1567 irq->u.prefix.address, 0); 1568 1569 if (!is_vcpu_stopped(vcpu)) 1570 return -EBUSY; 1571 1572 *prefix = irq->u.prefix; 1573 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs); 1574 return 0; 1575 } 1576 1577 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS) 1578 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1579 { 1580 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1581 struct kvm_s390_stop_info *stop = &li->irq.stop; 1582 int rc = 0; 1583 1584 vcpu->stat.inject_stop_signal++; 1585 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0); 1586 1587 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS) 1588 return -EINVAL; 1589 1590 if (is_vcpu_stopped(vcpu)) { 1591 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS) 1592 rc = kvm_s390_store_status_unloaded(vcpu, 1593 KVM_S390_STORE_STATUS_NOADDR); 1594 return rc; 1595 } 1596 1597 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs)) 1598 return -EBUSY; 1599 stop->flags = irq->u.stop.flags; 1600 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT); 1601 return 0; 1602 } 1603 1604 static int __inject_sigp_restart(struct kvm_vcpu *vcpu) 1605 { 1606 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1607 1608 vcpu->stat.inject_restart++; 1609 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int"); 1610 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0); 1611 1612 set_bit(IRQ_PEND_RESTART, &li->pending_irqs); 1613 return 0; 1614 } 1615 1616 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu, 1617 struct kvm_s390_irq *irq) 1618 { 1619 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1620 1621 vcpu->stat.inject_emergency_signal++; 1622 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u", 1623 irq->u.emerg.code); 1624 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY, 1625 irq->u.emerg.code, 0); 1626 1627 /* sending vcpu invalid */ 1628 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL) 1629 return -EINVAL; 1630 1631 set_bit(irq->u.emerg.code, li->sigp_emerg_pending); 1632 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs); 1633 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1634 return 0; 1635 } 1636 1637 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1638 { 1639 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1640 struct kvm_s390_mchk_info *mchk = &li->irq.mchk; 1641 1642 vcpu->stat.inject_mchk++; 1643 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx", 1644 irq->u.mchk.mcic); 1645 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0, 1646 irq->u.mchk.mcic); 1647 1648 /* 1649 * Because repressible machine checks can be indicated along with 1650 * exigent machine checks (PoP, Chapter 11, Interruption action) 1651 * we need to combine cr14, mcic and external damage code. 1652 * Failing storage address and the logout area should not be or'ed 1653 * together, we just indicate the last occurrence of the corresponding 1654 * machine check 1655 */ 1656 mchk->cr14 |= irq->u.mchk.cr14; 1657 mchk->mcic |= irq->u.mchk.mcic; 1658 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code; 1659 mchk->failing_storage_address = irq->u.mchk.failing_storage_address; 1660 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout, 1661 sizeof(mchk->fixed_logout)); 1662 if (mchk->mcic & MCHK_EX_MASK) 1663 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs); 1664 else if (mchk->mcic & MCHK_REP_MASK) 1665 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs); 1666 return 0; 1667 } 1668 1669 static int __inject_ckc(struct kvm_vcpu *vcpu) 1670 { 1671 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1672 1673 vcpu->stat.inject_ckc++; 1674 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external"); 1675 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP, 1676 0, 0); 1677 1678 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 1679 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1680 return 0; 1681 } 1682 1683 static int __inject_cpu_timer(struct kvm_vcpu *vcpu) 1684 { 1685 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1686 1687 vcpu->stat.inject_cputm++; 1688 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external"); 1689 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER, 1690 0, 0); 1691 1692 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 1693 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1694 return 0; 1695 } 1696 1697 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm, 1698 int isc, u32 schid) 1699 { 1700 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1701 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc]; 1702 struct kvm_s390_interrupt_info *iter; 1703 u16 id = (schid & 0xffff0000U) >> 16; 1704 u16 nr = schid & 0x0000ffffU; 1705 1706 spin_lock(&fi->lock); 1707 list_for_each_entry(iter, isc_list, list) { 1708 if (schid && (id != iter->io.subchannel_id || 1709 nr != iter->io.subchannel_nr)) 1710 continue; 1711 /* found an appropriate entry */ 1712 list_del_init(&iter->list); 1713 fi->counters[FIRQ_CNTR_IO] -= 1; 1714 if (list_empty(isc_list)) 1715 clear_bit(isc_to_irq_type(isc), &fi->pending_irqs); 1716 spin_unlock(&fi->lock); 1717 return iter; 1718 } 1719 spin_unlock(&fi->lock); 1720 return NULL; 1721 } 1722 1723 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm, 1724 u64 isc_mask, u32 schid) 1725 { 1726 struct kvm_s390_interrupt_info *inti = NULL; 1727 int isc; 1728 1729 for (isc = 0; isc <= MAX_ISC && !inti; isc++) { 1730 if (isc_mask & isc_to_isc_bits(isc)) 1731 inti = get_io_int(kvm, isc, schid); 1732 } 1733 return inti; 1734 } 1735 1736 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid) 1737 { 1738 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 1739 unsigned long active_mask; 1740 int isc; 1741 1742 if (schid) 1743 goto out; 1744 if (!gi->origin) 1745 goto out; 1746 1747 active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32; 1748 while (active_mask) { 1749 isc = __fls(active_mask) ^ (BITS_PER_LONG - 1); 1750 if (gisa_tac_ipm_gisc(gi->origin, isc)) 1751 return isc; 1752 clear_bit_inv(isc, &active_mask); 1753 } 1754 out: 1755 return -EINVAL; 1756 } 1757 1758 /* 1759 * Dequeue and return an I/O interrupt matching any of the interruption 1760 * subclasses as designated by the isc mask in cr6 and the schid (if != 0). 1761 * Take into account the interrupts pending in the interrupt list and in GISA. 1762 * 1763 * Note that for a guest that does not enable I/O interrupts 1764 * but relies on TPI, a flood of classic interrupts may starve 1765 * out adapter interrupts on the same isc. Linux does not do 1766 * that, and it is possible to work around the issue by configuring 1767 * different iscs for classic and adapter interrupts in the guest, 1768 * but we may want to revisit this in the future. 1769 */ 1770 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm, 1771 u64 isc_mask, u32 schid) 1772 { 1773 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 1774 struct kvm_s390_interrupt_info *inti, *tmp_inti; 1775 int isc; 1776 1777 inti = get_top_io_int(kvm, isc_mask, schid); 1778 1779 isc = get_top_gisa_isc(kvm, isc_mask, schid); 1780 if (isc < 0) 1781 /* no AI in GISA */ 1782 goto out; 1783 1784 if (!inti) 1785 /* AI in GISA but no classical IO int */ 1786 goto gisa_out; 1787 1788 /* both types of interrupts present */ 1789 if (int_word_to_isc(inti->io.io_int_word) <= isc) { 1790 /* classical IO int with higher priority */ 1791 gisa_set_ipm_gisc(gi->origin, isc); 1792 goto out; 1793 } 1794 gisa_out: 1795 tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL); 1796 if (tmp_inti) { 1797 tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0); 1798 tmp_inti->io.io_int_word = isc_to_int_word(isc); 1799 if (inti) 1800 kvm_s390_reinject_io_int(kvm, inti); 1801 inti = tmp_inti; 1802 } else 1803 gisa_set_ipm_gisc(gi->origin, isc); 1804 out: 1805 return inti; 1806 } 1807 1808 static int __inject_service(struct kvm *kvm, 1809 struct kvm_s390_interrupt_info *inti) 1810 { 1811 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1812 1813 kvm->stat.inject_service_signal++; 1814 spin_lock(&fi->lock); 1815 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING; 1816 1817 /* We always allow events, track them separately from the sccb ints */ 1818 if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING) 1819 set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs); 1820 1821 /* 1822 * Early versions of the QEMU s390 bios will inject several 1823 * service interrupts after another without handling a 1824 * condition code indicating busy. 1825 * We will silently ignore those superfluous sccb values. 1826 * A future version of QEMU will take care of serialization 1827 * of servc requests 1828 */ 1829 if (fi->srv_signal.ext_params & SCCB_MASK) 1830 goto out; 1831 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK; 1832 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs); 1833 out: 1834 spin_unlock(&fi->lock); 1835 kfree(inti); 1836 return 0; 1837 } 1838 1839 static int __inject_virtio(struct kvm *kvm, 1840 struct kvm_s390_interrupt_info *inti) 1841 { 1842 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1843 1844 kvm->stat.inject_virtio++; 1845 spin_lock(&fi->lock); 1846 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) { 1847 spin_unlock(&fi->lock); 1848 return -EBUSY; 1849 } 1850 fi->counters[FIRQ_CNTR_VIRTIO] += 1; 1851 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]); 1852 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs); 1853 spin_unlock(&fi->lock); 1854 return 0; 1855 } 1856 1857 static int __inject_pfault_done(struct kvm *kvm, 1858 struct kvm_s390_interrupt_info *inti) 1859 { 1860 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1861 1862 kvm->stat.inject_pfault_done++; 1863 spin_lock(&fi->lock); 1864 if (fi->counters[FIRQ_CNTR_PFAULT] >= 1865 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) { 1866 spin_unlock(&fi->lock); 1867 return -EBUSY; 1868 } 1869 fi->counters[FIRQ_CNTR_PFAULT] += 1; 1870 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]); 1871 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs); 1872 spin_unlock(&fi->lock); 1873 return 0; 1874 } 1875 1876 #define CR_PENDING_SUBCLASS 28 1877 static int __inject_float_mchk(struct kvm *kvm, 1878 struct kvm_s390_interrupt_info *inti) 1879 { 1880 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1881 1882 kvm->stat.inject_float_mchk++; 1883 spin_lock(&fi->lock); 1884 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS); 1885 fi->mchk.mcic |= inti->mchk.mcic; 1886 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs); 1887 spin_unlock(&fi->lock); 1888 kfree(inti); 1889 return 0; 1890 } 1891 1892 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti) 1893 { 1894 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 1895 struct kvm_s390_float_interrupt *fi; 1896 struct list_head *list; 1897 int isc; 1898 1899 kvm->stat.inject_io++; 1900 isc = int_word_to_isc(inti->io.io_int_word); 1901 1902 /* 1903 * Do not make use of gisa in protected mode. We do not use the lock 1904 * checking variant as this is just a performance optimization and we 1905 * do not hold the lock here. This is ok as the code will pick 1906 * interrupts from both "lists" for delivery. 1907 */ 1908 if (!kvm_s390_pv_get_handle(kvm) && 1909 gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) { 1910 VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc); 1911 gisa_set_ipm_gisc(gi->origin, isc); 1912 kfree(inti); 1913 return 0; 1914 } 1915 1916 fi = &kvm->arch.float_int; 1917 spin_lock(&fi->lock); 1918 if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) { 1919 spin_unlock(&fi->lock); 1920 return -EBUSY; 1921 } 1922 fi->counters[FIRQ_CNTR_IO] += 1; 1923 1924 if (inti->type & KVM_S390_INT_IO_AI_MASK) 1925 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)"); 1926 else 1927 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x", 1928 inti->io.subchannel_id >> 8, 1929 inti->io.subchannel_id >> 1 & 0x3, 1930 inti->io.subchannel_nr); 1931 list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc]; 1932 list_add_tail(&inti->list, list); 1933 set_bit(isc_to_irq_type(isc), &fi->pending_irqs); 1934 spin_unlock(&fi->lock); 1935 return 0; 1936 } 1937 1938 /* 1939 * Find a destination VCPU for a floating irq and kick it. 1940 */ 1941 static void __floating_irq_kick(struct kvm *kvm, u64 type) 1942 { 1943 struct kvm_vcpu *dst_vcpu; 1944 int sigcpu, online_vcpus, nr_tries = 0; 1945 1946 online_vcpus = atomic_read(&kvm->online_vcpus); 1947 if (!online_vcpus) 1948 return; 1949 1950 /* find idle VCPUs first, then round robin */ 1951 sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus); 1952 if (sigcpu == online_vcpus) { 1953 do { 1954 sigcpu = kvm->arch.float_int.next_rr_cpu++; 1955 kvm->arch.float_int.next_rr_cpu %= online_vcpus; 1956 /* avoid endless loops if all vcpus are stopped */ 1957 if (nr_tries++ >= online_vcpus) 1958 return; 1959 } while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu))); 1960 } 1961 dst_vcpu = kvm_get_vcpu(kvm, sigcpu); 1962 1963 /* make the VCPU drop out of the SIE, or wake it up if sleeping */ 1964 switch (type) { 1965 case KVM_S390_MCHK: 1966 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT); 1967 break; 1968 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 1969 if (!(type & KVM_S390_INT_IO_AI_MASK && 1970 kvm->arch.gisa_int.origin) || 1971 kvm_s390_pv_cpu_get_handle(dst_vcpu)) 1972 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT); 1973 break; 1974 default: 1975 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT); 1976 break; 1977 } 1978 kvm_s390_vcpu_wakeup(dst_vcpu); 1979 } 1980 1981 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti) 1982 { 1983 u64 type = READ_ONCE(inti->type); 1984 int rc; 1985 1986 switch (type) { 1987 case KVM_S390_MCHK: 1988 rc = __inject_float_mchk(kvm, inti); 1989 break; 1990 case KVM_S390_INT_VIRTIO: 1991 rc = __inject_virtio(kvm, inti); 1992 break; 1993 case KVM_S390_INT_SERVICE: 1994 rc = __inject_service(kvm, inti); 1995 break; 1996 case KVM_S390_INT_PFAULT_DONE: 1997 rc = __inject_pfault_done(kvm, inti); 1998 break; 1999 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2000 rc = __inject_io(kvm, inti); 2001 break; 2002 default: 2003 rc = -EINVAL; 2004 } 2005 if (rc) 2006 return rc; 2007 2008 __floating_irq_kick(kvm, type); 2009 return 0; 2010 } 2011 2012 int kvm_s390_inject_vm(struct kvm *kvm, 2013 struct kvm_s390_interrupt *s390int) 2014 { 2015 struct kvm_s390_interrupt_info *inti; 2016 int rc; 2017 2018 inti = kzalloc(sizeof(*inti), GFP_KERNEL); 2019 if (!inti) 2020 return -ENOMEM; 2021 2022 inti->type = s390int->type; 2023 switch (inti->type) { 2024 case KVM_S390_INT_VIRTIO: 2025 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx", 2026 s390int->parm, s390int->parm64); 2027 inti->ext.ext_params = s390int->parm; 2028 inti->ext.ext_params2 = s390int->parm64; 2029 break; 2030 case KVM_S390_INT_SERVICE: 2031 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm); 2032 inti->ext.ext_params = s390int->parm; 2033 break; 2034 case KVM_S390_INT_PFAULT_DONE: 2035 inti->ext.ext_params2 = s390int->parm64; 2036 break; 2037 case KVM_S390_MCHK: 2038 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx", 2039 s390int->parm64); 2040 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */ 2041 inti->mchk.mcic = s390int->parm64; 2042 break; 2043 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2044 inti->io.subchannel_id = s390int->parm >> 16; 2045 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu; 2046 inti->io.io_int_parm = s390int->parm64 >> 32; 2047 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull; 2048 break; 2049 default: 2050 kfree(inti); 2051 return -EINVAL; 2052 } 2053 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64, 2054 2); 2055 2056 rc = __inject_vm(kvm, inti); 2057 if (rc) 2058 kfree(inti); 2059 return rc; 2060 } 2061 2062 int kvm_s390_reinject_io_int(struct kvm *kvm, 2063 struct kvm_s390_interrupt_info *inti) 2064 { 2065 return __inject_vm(kvm, inti); 2066 } 2067 2068 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int, 2069 struct kvm_s390_irq *irq) 2070 { 2071 irq->type = s390int->type; 2072 switch (irq->type) { 2073 case KVM_S390_PROGRAM_INT: 2074 if (s390int->parm & 0xffff0000) 2075 return -EINVAL; 2076 irq->u.pgm.code = s390int->parm; 2077 break; 2078 case KVM_S390_SIGP_SET_PREFIX: 2079 irq->u.prefix.address = s390int->parm; 2080 break; 2081 case KVM_S390_SIGP_STOP: 2082 irq->u.stop.flags = s390int->parm; 2083 break; 2084 case KVM_S390_INT_EXTERNAL_CALL: 2085 if (s390int->parm & 0xffff0000) 2086 return -EINVAL; 2087 irq->u.extcall.code = s390int->parm; 2088 break; 2089 case KVM_S390_INT_EMERGENCY: 2090 if (s390int->parm & 0xffff0000) 2091 return -EINVAL; 2092 irq->u.emerg.code = s390int->parm; 2093 break; 2094 case KVM_S390_MCHK: 2095 irq->u.mchk.mcic = s390int->parm64; 2096 break; 2097 case KVM_S390_INT_PFAULT_INIT: 2098 irq->u.ext.ext_params = s390int->parm; 2099 irq->u.ext.ext_params2 = s390int->parm64; 2100 break; 2101 case KVM_S390_RESTART: 2102 case KVM_S390_INT_CLOCK_COMP: 2103 case KVM_S390_INT_CPU_TIMER: 2104 break; 2105 default: 2106 return -EINVAL; 2107 } 2108 return 0; 2109 } 2110 2111 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu) 2112 { 2113 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2114 2115 return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs); 2116 } 2117 2118 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu) 2119 { 2120 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2121 2122 spin_lock(&li->lock); 2123 li->irq.stop.flags = 0; 2124 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs); 2125 spin_unlock(&li->lock); 2126 } 2127 2128 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 2129 { 2130 int rc; 2131 2132 switch (irq->type) { 2133 case KVM_S390_PROGRAM_INT: 2134 rc = __inject_prog(vcpu, irq); 2135 break; 2136 case KVM_S390_SIGP_SET_PREFIX: 2137 rc = __inject_set_prefix(vcpu, irq); 2138 break; 2139 case KVM_S390_SIGP_STOP: 2140 rc = __inject_sigp_stop(vcpu, irq); 2141 break; 2142 case KVM_S390_RESTART: 2143 rc = __inject_sigp_restart(vcpu); 2144 break; 2145 case KVM_S390_INT_CLOCK_COMP: 2146 rc = __inject_ckc(vcpu); 2147 break; 2148 case KVM_S390_INT_CPU_TIMER: 2149 rc = __inject_cpu_timer(vcpu); 2150 break; 2151 case KVM_S390_INT_EXTERNAL_CALL: 2152 rc = __inject_extcall(vcpu, irq); 2153 break; 2154 case KVM_S390_INT_EMERGENCY: 2155 rc = __inject_sigp_emergency(vcpu, irq); 2156 break; 2157 case KVM_S390_MCHK: 2158 rc = __inject_mchk(vcpu, irq); 2159 break; 2160 case KVM_S390_INT_PFAULT_INIT: 2161 rc = __inject_pfault_init(vcpu, irq); 2162 break; 2163 case KVM_S390_INT_VIRTIO: 2164 case KVM_S390_INT_SERVICE: 2165 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2166 default: 2167 rc = -EINVAL; 2168 } 2169 2170 return rc; 2171 } 2172 2173 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 2174 { 2175 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2176 int rc; 2177 2178 spin_lock(&li->lock); 2179 rc = do_inject_vcpu(vcpu, irq); 2180 spin_unlock(&li->lock); 2181 if (!rc) 2182 kvm_s390_vcpu_wakeup(vcpu); 2183 return rc; 2184 } 2185 2186 static inline void clear_irq_list(struct list_head *_list) 2187 { 2188 struct kvm_s390_interrupt_info *inti, *n; 2189 2190 list_for_each_entry_safe(inti, n, _list, list) { 2191 list_del(&inti->list); 2192 kfree(inti); 2193 } 2194 } 2195 2196 static void inti_to_irq(struct kvm_s390_interrupt_info *inti, 2197 struct kvm_s390_irq *irq) 2198 { 2199 irq->type = inti->type; 2200 switch (inti->type) { 2201 case KVM_S390_INT_PFAULT_INIT: 2202 case KVM_S390_INT_PFAULT_DONE: 2203 case KVM_S390_INT_VIRTIO: 2204 irq->u.ext = inti->ext; 2205 break; 2206 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2207 irq->u.io = inti->io; 2208 break; 2209 } 2210 } 2211 2212 void kvm_s390_clear_float_irqs(struct kvm *kvm) 2213 { 2214 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2215 int i; 2216 2217 mutex_lock(&kvm->lock); 2218 if (!kvm_s390_pv_is_protected(kvm)) 2219 fi->masked_irqs = 0; 2220 mutex_unlock(&kvm->lock); 2221 spin_lock(&fi->lock); 2222 fi->pending_irqs = 0; 2223 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal)); 2224 memset(&fi->mchk, 0, sizeof(fi->mchk)); 2225 for (i = 0; i < FIRQ_LIST_COUNT; i++) 2226 clear_irq_list(&fi->lists[i]); 2227 for (i = 0; i < FIRQ_MAX_COUNT; i++) 2228 fi->counters[i] = 0; 2229 spin_unlock(&fi->lock); 2230 kvm_s390_gisa_clear(kvm); 2231 }; 2232 2233 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len) 2234 { 2235 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 2236 struct kvm_s390_interrupt_info *inti; 2237 struct kvm_s390_float_interrupt *fi; 2238 struct kvm_s390_irq *buf; 2239 struct kvm_s390_irq *irq; 2240 int max_irqs; 2241 int ret = 0; 2242 int n = 0; 2243 int i; 2244 2245 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0) 2246 return -EINVAL; 2247 2248 /* 2249 * We are already using -ENOMEM to signal 2250 * userspace it may retry with a bigger buffer, 2251 * so we need to use something else for this case 2252 */ 2253 buf = vzalloc(len); 2254 if (!buf) 2255 return -ENOBUFS; 2256 2257 max_irqs = len / sizeof(struct kvm_s390_irq); 2258 2259 if (gi->origin && gisa_get_ipm(gi->origin)) { 2260 for (i = 0; i <= MAX_ISC; i++) { 2261 if (n == max_irqs) { 2262 /* signal userspace to try again */ 2263 ret = -ENOMEM; 2264 goto out_nolock; 2265 } 2266 if (gisa_tac_ipm_gisc(gi->origin, i)) { 2267 irq = (struct kvm_s390_irq *) &buf[n]; 2268 irq->type = KVM_S390_INT_IO(1, 0, 0, 0); 2269 irq->u.io.io_int_word = isc_to_int_word(i); 2270 n++; 2271 } 2272 } 2273 } 2274 fi = &kvm->arch.float_int; 2275 spin_lock(&fi->lock); 2276 for (i = 0; i < FIRQ_LIST_COUNT; i++) { 2277 list_for_each_entry(inti, &fi->lists[i], list) { 2278 if (n == max_irqs) { 2279 /* signal userspace to try again */ 2280 ret = -ENOMEM; 2281 goto out; 2282 } 2283 inti_to_irq(inti, &buf[n]); 2284 n++; 2285 } 2286 } 2287 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) || 2288 test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) { 2289 if (n == max_irqs) { 2290 /* signal userspace to try again */ 2291 ret = -ENOMEM; 2292 goto out; 2293 } 2294 irq = (struct kvm_s390_irq *) &buf[n]; 2295 irq->type = KVM_S390_INT_SERVICE; 2296 irq->u.ext = fi->srv_signal; 2297 n++; 2298 } 2299 if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) { 2300 if (n == max_irqs) { 2301 /* signal userspace to try again */ 2302 ret = -ENOMEM; 2303 goto out; 2304 } 2305 irq = (struct kvm_s390_irq *) &buf[n]; 2306 irq->type = KVM_S390_MCHK; 2307 irq->u.mchk = fi->mchk; 2308 n++; 2309 } 2310 2311 out: 2312 spin_unlock(&fi->lock); 2313 out_nolock: 2314 if (!ret && n > 0) { 2315 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n)) 2316 ret = -EFAULT; 2317 } 2318 vfree(buf); 2319 2320 return ret < 0 ? ret : n; 2321 } 2322 2323 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr) 2324 { 2325 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2326 struct kvm_s390_ais_all ais; 2327 2328 if (attr->attr < sizeof(ais)) 2329 return -EINVAL; 2330 2331 if (!test_kvm_facility(kvm, 72)) 2332 return -EOPNOTSUPP; 2333 2334 mutex_lock(&fi->ais_lock); 2335 ais.simm = fi->simm; 2336 ais.nimm = fi->nimm; 2337 mutex_unlock(&fi->ais_lock); 2338 2339 if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais))) 2340 return -EFAULT; 2341 2342 return 0; 2343 } 2344 2345 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr) 2346 { 2347 int r; 2348 2349 switch (attr->group) { 2350 case KVM_DEV_FLIC_GET_ALL_IRQS: 2351 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr, 2352 attr->attr); 2353 break; 2354 case KVM_DEV_FLIC_AISM_ALL: 2355 r = flic_ais_mode_get_all(dev->kvm, attr); 2356 break; 2357 default: 2358 r = -EINVAL; 2359 } 2360 2361 return r; 2362 } 2363 2364 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti, 2365 u64 addr) 2366 { 2367 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr; 2368 void *target = NULL; 2369 void __user *source; 2370 u64 size; 2371 2372 if (get_user(inti->type, (u64 __user *)addr)) 2373 return -EFAULT; 2374 2375 switch (inti->type) { 2376 case KVM_S390_INT_PFAULT_INIT: 2377 case KVM_S390_INT_PFAULT_DONE: 2378 case KVM_S390_INT_VIRTIO: 2379 case KVM_S390_INT_SERVICE: 2380 target = (void *) &inti->ext; 2381 source = &uptr->u.ext; 2382 size = sizeof(inti->ext); 2383 break; 2384 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2385 target = (void *) &inti->io; 2386 source = &uptr->u.io; 2387 size = sizeof(inti->io); 2388 break; 2389 case KVM_S390_MCHK: 2390 target = (void *) &inti->mchk; 2391 source = &uptr->u.mchk; 2392 size = sizeof(inti->mchk); 2393 break; 2394 default: 2395 return -EINVAL; 2396 } 2397 2398 if (copy_from_user(target, source, size)) 2399 return -EFAULT; 2400 2401 return 0; 2402 } 2403 2404 static int enqueue_floating_irq(struct kvm_device *dev, 2405 struct kvm_device_attr *attr) 2406 { 2407 struct kvm_s390_interrupt_info *inti = NULL; 2408 int r = 0; 2409 int len = attr->attr; 2410 2411 if (len % sizeof(struct kvm_s390_irq) != 0) 2412 return -EINVAL; 2413 else if (len > KVM_S390_FLIC_MAX_BUFFER) 2414 return -EINVAL; 2415 2416 while (len >= sizeof(struct kvm_s390_irq)) { 2417 inti = kzalloc(sizeof(*inti), GFP_KERNEL); 2418 if (!inti) 2419 return -ENOMEM; 2420 2421 r = copy_irq_from_user(inti, attr->addr); 2422 if (r) { 2423 kfree(inti); 2424 return r; 2425 } 2426 r = __inject_vm(dev->kvm, inti); 2427 if (r) { 2428 kfree(inti); 2429 return r; 2430 } 2431 len -= sizeof(struct kvm_s390_irq); 2432 attr->addr += sizeof(struct kvm_s390_irq); 2433 } 2434 2435 return r; 2436 } 2437 2438 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id) 2439 { 2440 if (id >= MAX_S390_IO_ADAPTERS) 2441 return NULL; 2442 id = array_index_nospec(id, MAX_S390_IO_ADAPTERS); 2443 return kvm->arch.adapters[id]; 2444 } 2445 2446 static int register_io_adapter(struct kvm_device *dev, 2447 struct kvm_device_attr *attr) 2448 { 2449 struct s390_io_adapter *adapter; 2450 struct kvm_s390_io_adapter adapter_info; 2451 2452 if (copy_from_user(&adapter_info, 2453 (void __user *)attr->addr, sizeof(adapter_info))) 2454 return -EFAULT; 2455 2456 if (adapter_info.id >= MAX_S390_IO_ADAPTERS) 2457 return -EINVAL; 2458 2459 adapter_info.id = array_index_nospec(adapter_info.id, 2460 MAX_S390_IO_ADAPTERS); 2461 2462 if (dev->kvm->arch.adapters[adapter_info.id] != NULL) 2463 return -EINVAL; 2464 2465 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL); 2466 if (!adapter) 2467 return -ENOMEM; 2468 2469 adapter->id = adapter_info.id; 2470 adapter->isc = adapter_info.isc; 2471 adapter->maskable = adapter_info.maskable; 2472 adapter->masked = false; 2473 adapter->swap = adapter_info.swap; 2474 adapter->suppressible = (adapter_info.flags) & 2475 KVM_S390_ADAPTER_SUPPRESSIBLE; 2476 dev->kvm->arch.adapters[adapter->id] = adapter; 2477 2478 return 0; 2479 } 2480 2481 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked) 2482 { 2483 int ret; 2484 struct s390_io_adapter *adapter = get_io_adapter(kvm, id); 2485 2486 if (!adapter || !adapter->maskable) 2487 return -EINVAL; 2488 ret = adapter->masked; 2489 adapter->masked = masked; 2490 return ret; 2491 } 2492 2493 void kvm_s390_destroy_adapters(struct kvm *kvm) 2494 { 2495 int i; 2496 2497 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) 2498 kfree(kvm->arch.adapters[i]); 2499 } 2500 2501 static int modify_io_adapter(struct kvm_device *dev, 2502 struct kvm_device_attr *attr) 2503 { 2504 struct kvm_s390_io_adapter_req req; 2505 struct s390_io_adapter *adapter; 2506 int ret; 2507 2508 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req))) 2509 return -EFAULT; 2510 2511 adapter = get_io_adapter(dev->kvm, req.id); 2512 if (!adapter) 2513 return -EINVAL; 2514 switch (req.type) { 2515 case KVM_S390_IO_ADAPTER_MASK: 2516 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask); 2517 if (ret > 0) 2518 ret = 0; 2519 break; 2520 /* 2521 * The following operations are no longer needed and therefore no-ops. 2522 * The gpa to hva translation is done when an IRQ route is set up. The 2523 * set_irq code uses get_user_pages_remote() to do the actual write. 2524 */ 2525 case KVM_S390_IO_ADAPTER_MAP: 2526 case KVM_S390_IO_ADAPTER_UNMAP: 2527 ret = 0; 2528 break; 2529 default: 2530 ret = -EINVAL; 2531 } 2532 2533 return ret; 2534 } 2535 2536 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr) 2537 2538 { 2539 const u64 isc_mask = 0xffUL << 24; /* all iscs set */ 2540 u32 schid; 2541 2542 if (attr->flags) 2543 return -EINVAL; 2544 if (attr->attr != sizeof(schid)) 2545 return -EINVAL; 2546 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid))) 2547 return -EFAULT; 2548 if (!schid) 2549 return -EINVAL; 2550 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid)); 2551 /* 2552 * If userspace is conforming to the architecture, we can have at most 2553 * one pending I/O interrupt per subchannel, so this is effectively a 2554 * clear all. 2555 */ 2556 return 0; 2557 } 2558 2559 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr) 2560 { 2561 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2562 struct kvm_s390_ais_req req; 2563 int ret = 0; 2564 2565 if (!test_kvm_facility(kvm, 72)) 2566 return -EOPNOTSUPP; 2567 2568 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req))) 2569 return -EFAULT; 2570 2571 if (req.isc > MAX_ISC) 2572 return -EINVAL; 2573 2574 trace_kvm_s390_modify_ais_mode(req.isc, 2575 (fi->simm & AIS_MODE_MASK(req.isc)) ? 2576 (fi->nimm & AIS_MODE_MASK(req.isc)) ? 2577 2 : KVM_S390_AIS_MODE_SINGLE : 2578 KVM_S390_AIS_MODE_ALL, req.mode); 2579 2580 mutex_lock(&fi->ais_lock); 2581 switch (req.mode) { 2582 case KVM_S390_AIS_MODE_ALL: 2583 fi->simm &= ~AIS_MODE_MASK(req.isc); 2584 fi->nimm &= ~AIS_MODE_MASK(req.isc); 2585 break; 2586 case KVM_S390_AIS_MODE_SINGLE: 2587 fi->simm |= AIS_MODE_MASK(req.isc); 2588 fi->nimm &= ~AIS_MODE_MASK(req.isc); 2589 break; 2590 default: 2591 ret = -EINVAL; 2592 } 2593 mutex_unlock(&fi->ais_lock); 2594 2595 return ret; 2596 } 2597 2598 static int kvm_s390_inject_airq(struct kvm *kvm, 2599 struct s390_io_adapter *adapter) 2600 { 2601 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2602 struct kvm_s390_interrupt s390int = { 2603 .type = KVM_S390_INT_IO(1, 0, 0, 0), 2604 .parm = 0, 2605 .parm64 = isc_to_int_word(adapter->isc), 2606 }; 2607 int ret = 0; 2608 2609 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible) 2610 return kvm_s390_inject_vm(kvm, &s390int); 2611 2612 mutex_lock(&fi->ais_lock); 2613 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) { 2614 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc); 2615 goto out; 2616 } 2617 2618 ret = kvm_s390_inject_vm(kvm, &s390int); 2619 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) { 2620 fi->nimm |= AIS_MODE_MASK(adapter->isc); 2621 trace_kvm_s390_modify_ais_mode(adapter->isc, 2622 KVM_S390_AIS_MODE_SINGLE, 2); 2623 } 2624 out: 2625 mutex_unlock(&fi->ais_lock); 2626 return ret; 2627 } 2628 2629 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr) 2630 { 2631 unsigned int id = attr->attr; 2632 struct s390_io_adapter *adapter = get_io_adapter(kvm, id); 2633 2634 if (!adapter) 2635 return -EINVAL; 2636 2637 return kvm_s390_inject_airq(kvm, adapter); 2638 } 2639 2640 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr) 2641 { 2642 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2643 struct kvm_s390_ais_all ais; 2644 2645 if (!test_kvm_facility(kvm, 72)) 2646 return -EOPNOTSUPP; 2647 2648 if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais))) 2649 return -EFAULT; 2650 2651 mutex_lock(&fi->ais_lock); 2652 fi->simm = ais.simm; 2653 fi->nimm = ais.nimm; 2654 mutex_unlock(&fi->ais_lock); 2655 2656 return 0; 2657 } 2658 2659 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr) 2660 { 2661 int r = 0; 2662 unsigned int i; 2663 struct kvm_vcpu *vcpu; 2664 2665 switch (attr->group) { 2666 case KVM_DEV_FLIC_ENQUEUE: 2667 r = enqueue_floating_irq(dev, attr); 2668 break; 2669 case KVM_DEV_FLIC_CLEAR_IRQS: 2670 kvm_s390_clear_float_irqs(dev->kvm); 2671 break; 2672 case KVM_DEV_FLIC_APF_ENABLE: 2673 dev->kvm->arch.gmap->pfault_enabled = 1; 2674 break; 2675 case KVM_DEV_FLIC_APF_DISABLE_WAIT: 2676 dev->kvm->arch.gmap->pfault_enabled = 0; 2677 /* 2678 * Make sure no async faults are in transition when 2679 * clearing the queues. So we don't need to worry 2680 * about late coming workers. 2681 */ 2682 synchronize_srcu(&dev->kvm->srcu); 2683 kvm_for_each_vcpu(i, vcpu, dev->kvm) 2684 kvm_clear_async_pf_completion_queue(vcpu); 2685 break; 2686 case KVM_DEV_FLIC_ADAPTER_REGISTER: 2687 r = register_io_adapter(dev, attr); 2688 break; 2689 case KVM_DEV_FLIC_ADAPTER_MODIFY: 2690 r = modify_io_adapter(dev, attr); 2691 break; 2692 case KVM_DEV_FLIC_CLEAR_IO_IRQ: 2693 r = clear_io_irq(dev->kvm, attr); 2694 break; 2695 case KVM_DEV_FLIC_AISM: 2696 r = modify_ais_mode(dev->kvm, attr); 2697 break; 2698 case KVM_DEV_FLIC_AIRQ_INJECT: 2699 r = flic_inject_airq(dev->kvm, attr); 2700 break; 2701 case KVM_DEV_FLIC_AISM_ALL: 2702 r = flic_ais_mode_set_all(dev->kvm, attr); 2703 break; 2704 default: 2705 r = -EINVAL; 2706 } 2707 2708 return r; 2709 } 2710 2711 static int flic_has_attr(struct kvm_device *dev, 2712 struct kvm_device_attr *attr) 2713 { 2714 switch (attr->group) { 2715 case KVM_DEV_FLIC_GET_ALL_IRQS: 2716 case KVM_DEV_FLIC_ENQUEUE: 2717 case KVM_DEV_FLIC_CLEAR_IRQS: 2718 case KVM_DEV_FLIC_APF_ENABLE: 2719 case KVM_DEV_FLIC_APF_DISABLE_WAIT: 2720 case KVM_DEV_FLIC_ADAPTER_REGISTER: 2721 case KVM_DEV_FLIC_ADAPTER_MODIFY: 2722 case KVM_DEV_FLIC_CLEAR_IO_IRQ: 2723 case KVM_DEV_FLIC_AISM: 2724 case KVM_DEV_FLIC_AIRQ_INJECT: 2725 case KVM_DEV_FLIC_AISM_ALL: 2726 return 0; 2727 } 2728 return -ENXIO; 2729 } 2730 2731 static int flic_create(struct kvm_device *dev, u32 type) 2732 { 2733 if (!dev) 2734 return -EINVAL; 2735 if (dev->kvm->arch.flic) 2736 return -EINVAL; 2737 dev->kvm->arch.flic = dev; 2738 return 0; 2739 } 2740 2741 static void flic_destroy(struct kvm_device *dev) 2742 { 2743 dev->kvm->arch.flic = NULL; 2744 kfree(dev); 2745 } 2746 2747 /* s390 floating irq controller (flic) */ 2748 struct kvm_device_ops kvm_flic_ops = { 2749 .name = "kvm-flic", 2750 .get_attr = flic_get_attr, 2751 .set_attr = flic_set_attr, 2752 .has_attr = flic_has_attr, 2753 .create = flic_create, 2754 .destroy = flic_destroy, 2755 }; 2756 2757 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap) 2758 { 2759 unsigned long bit; 2760 2761 bit = bit_nr + (addr % PAGE_SIZE) * 8; 2762 2763 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit; 2764 } 2765 2766 static struct page *get_map_page(struct kvm *kvm, u64 uaddr) 2767 { 2768 struct page *page = NULL; 2769 2770 mmap_read_lock(kvm->mm); 2771 get_user_pages_remote(NULL, kvm->mm, uaddr, 1, FOLL_WRITE, 2772 &page, NULL, NULL); 2773 mmap_read_unlock(kvm->mm); 2774 return page; 2775 } 2776 2777 static int adapter_indicators_set(struct kvm *kvm, 2778 struct s390_io_adapter *adapter, 2779 struct kvm_s390_adapter_int *adapter_int) 2780 { 2781 unsigned long bit; 2782 int summary_set, idx; 2783 struct page *ind_page, *summary_page; 2784 void *map; 2785 2786 ind_page = get_map_page(kvm, adapter_int->ind_addr); 2787 if (!ind_page) 2788 return -1; 2789 summary_page = get_map_page(kvm, adapter_int->summary_addr); 2790 if (!summary_page) { 2791 put_page(ind_page); 2792 return -1; 2793 } 2794 2795 idx = srcu_read_lock(&kvm->srcu); 2796 map = page_address(ind_page); 2797 bit = get_ind_bit(adapter_int->ind_addr, 2798 adapter_int->ind_offset, adapter->swap); 2799 set_bit(bit, map); 2800 mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT); 2801 set_page_dirty_lock(ind_page); 2802 map = page_address(summary_page); 2803 bit = get_ind_bit(adapter_int->summary_addr, 2804 adapter_int->summary_offset, adapter->swap); 2805 summary_set = test_and_set_bit(bit, map); 2806 mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT); 2807 set_page_dirty_lock(summary_page); 2808 srcu_read_unlock(&kvm->srcu, idx); 2809 2810 put_page(ind_page); 2811 put_page(summary_page); 2812 return summary_set ? 0 : 1; 2813 } 2814 2815 /* 2816 * < 0 - not injected due to error 2817 * = 0 - coalesced, summary indicator already active 2818 * > 0 - injected interrupt 2819 */ 2820 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e, 2821 struct kvm *kvm, int irq_source_id, int level, 2822 bool line_status) 2823 { 2824 int ret; 2825 struct s390_io_adapter *adapter; 2826 2827 /* We're only interested in the 0->1 transition. */ 2828 if (!level) 2829 return 0; 2830 adapter = get_io_adapter(kvm, e->adapter.adapter_id); 2831 if (!adapter) 2832 return -1; 2833 ret = adapter_indicators_set(kvm, adapter, &e->adapter); 2834 if ((ret > 0) && !adapter->masked) { 2835 ret = kvm_s390_inject_airq(kvm, adapter); 2836 if (ret == 0) 2837 ret = 1; 2838 } 2839 return ret; 2840 } 2841 2842 /* 2843 * Inject the machine check to the guest. 2844 */ 2845 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu, 2846 struct mcck_volatile_info *mcck_info) 2847 { 2848 struct kvm_s390_interrupt_info inti; 2849 struct kvm_s390_irq irq; 2850 struct kvm_s390_mchk_info *mchk; 2851 union mci mci; 2852 __u64 cr14 = 0; /* upper bits are not used */ 2853 int rc; 2854 2855 mci.val = mcck_info->mcic; 2856 if (mci.sr) 2857 cr14 |= CR14_RECOVERY_SUBMASK; 2858 if (mci.dg) 2859 cr14 |= CR14_DEGRADATION_SUBMASK; 2860 if (mci.w) 2861 cr14 |= CR14_WARNING_SUBMASK; 2862 2863 mchk = mci.ck ? &inti.mchk : &irq.u.mchk; 2864 mchk->cr14 = cr14; 2865 mchk->mcic = mcck_info->mcic; 2866 mchk->ext_damage_code = mcck_info->ext_damage_code; 2867 mchk->failing_storage_address = mcck_info->failing_storage_address; 2868 if (mci.ck) { 2869 /* Inject the floating machine check */ 2870 inti.type = KVM_S390_MCHK; 2871 rc = __inject_vm(vcpu->kvm, &inti); 2872 } else { 2873 /* Inject the machine check to specified vcpu */ 2874 irq.type = KVM_S390_MCHK; 2875 rc = kvm_s390_inject_vcpu(vcpu, &irq); 2876 } 2877 WARN_ON_ONCE(rc); 2878 } 2879 2880 int kvm_set_routing_entry(struct kvm *kvm, 2881 struct kvm_kernel_irq_routing_entry *e, 2882 const struct kvm_irq_routing_entry *ue) 2883 { 2884 u64 uaddr; 2885 2886 switch (ue->type) { 2887 /* we store the userspace addresses instead of the guest addresses */ 2888 case KVM_IRQ_ROUTING_S390_ADAPTER: 2889 e->set = set_adapter_int; 2890 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.summary_addr); 2891 if (uaddr == -EFAULT) 2892 return -EFAULT; 2893 e->adapter.summary_addr = uaddr; 2894 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.ind_addr); 2895 if (uaddr == -EFAULT) 2896 return -EFAULT; 2897 e->adapter.ind_addr = uaddr; 2898 e->adapter.summary_offset = ue->u.adapter.summary_offset; 2899 e->adapter.ind_offset = ue->u.adapter.ind_offset; 2900 e->adapter.adapter_id = ue->u.adapter.adapter_id; 2901 return 0; 2902 default: 2903 return -EINVAL; 2904 } 2905 } 2906 2907 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, 2908 int irq_source_id, int level, bool line_status) 2909 { 2910 return -EINVAL; 2911 } 2912 2913 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len) 2914 { 2915 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2916 struct kvm_s390_irq *buf; 2917 int r = 0; 2918 int n; 2919 2920 buf = vmalloc(len); 2921 if (!buf) 2922 return -ENOMEM; 2923 2924 if (copy_from_user((void *) buf, irqstate, len)) { 2925 r = -EFAULT; 2926 goto out_free; 2927 } 2928 2929 /* 2930 * Don't allow setting the interrupt state 2931 * when there are already interrupts pending 2932 */ 2933 spin_lock(&li->lock); 2934 if (li->pending_irqs) { 2935 r = -EBUSY; 2936 goto out_unlock; 2937 } 2938 2939 for (n = 0; n < len / sizeof(*buf); n++) { 2940 r = do_inject_vcpu(vcpu, &buf[n]); 2941 if (r) 2942 break; 2943 } 2944 2945 out_unlock: 2946 spin_unlock(&li->lock); 2947 out_free: 2948 vfree(buf); 2949 2950 return r; 2951 } 2952 2953 static void store_local_irq(struct kvm_s390_local_interrupt *li, 2954 struct kvm_s390_irq *irq, 2955 unsigned long irq_type) 2956 { 2957 switch (irq_type) { 2958 case IRQ_PEND_MCHK_EX: 2959 case IRQ_PEND_MCHK_REP: 2960 irq->type = KVM_S390_MCHK; 2961 irq->u.mchk = li->irq.mchk; 2962 break; 2963 case IRQ_PEND_PROG: 2964 irq->type = KVM_S390_PROGRAM_INT; 2965 irq->u.pgm = li->irq.pgm; 2966 break; 2967 case IRQ_PEND_PFAULT_INIT: 2968 irq->type = KVM_S390_INT_PFAULT_INIT; 2969 irq->u.ext = li->irq.ext; 2970 break; 2971 case IRQ_PEND_EXT_EXTERNAL: 2972 irq->type = KVM_S390_INT_EXTERNAL_CALL; 2973 irq->u.extcall = li->irq.extcall; 2974 break; 2975 case IRQ_PEND_EXT_CLOCK_COMP: 2976 irq->type = KVM_S390_INT_CLOCK_COMP; 2977 break; 2978 case IRQ_PEND_EXT_CPU_TIMER: 2979 irq->type = KVM_S390_INT_CPU_TIMER; 2980 break; 2981 case IRQ_PEND_SIGP_STOP: 2982 irq->type = KVM_S390_SIGP_STOP; 2983 irq->u.stop = li->irq.stop; 2984 break; 2985 case IRQ_PEND_RESTART: 2986 irq->type = KVM_S390_RESTART; 2987 break; 2988 case IRQ_PEND_SET_PREFIX: 2989 irq->type = KVM_S390_SIGP_SET_PREFIX; 2990 irq->u.prefix = li->irq.prefix; 2991 break; 2992 } 2993 } 2994 2995 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len) 2996 { 2997 int scn; 2998 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS); 2999 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 3000 unsigned long pending_irqs; 3001 struct kvm_s390_irq irq; 3002 unsigned long irq_type; 3003 int cpuaddr; 3004 int n = 0; 3005 3006 spin_lock(&li->lock); 3007 pending_irqs = li->pending_irqs; 3008 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending, 3009 sizeof(sigp_emerg_pending)); 3010 spin_unlock(&li->lock); 3011 3012 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) { 3013 memset(&irq, 0, sizeof(irq)); 3014 if (irq_type == IRQ_PEND_EXT_EMERGENCY) 3015 continue; 3016 if (n + sizeof(irq) > len) 3017 return -ENOBUFS; 3018 store_local_irq(&vcpu->arch.local_int, &irq, irq_type); 3019 if (copy_to_user(&buf[n], &irq, sizeof(irq))) 3020 return -EFAULT; 3021 n += sizeof(irq); 3022 } 3023 3024 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) { 3025 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) { 3026 memset(&irq, 0, sizeof(irq)); 3027 if (n + sizeof(irq) > len) 3028 return -ENOBUFS; 3029 irq.type = KVM_S390_INT_EMERGENCY; 3030 irq.u.emerg.code = cpuaddr; 3031 if (copy_to_user(&buf[n], &irq, sizeof(irq))) 3032 return -EFAULT; 3033 n += sizeof(irq); 3034 } 3035 } 3036 3037 if (sca_ext_call_pending(vcpu, &scn)) { 3038 if (n + sizeof(irq) > len) 3039 return -ENOBUFS; 3040 memset(&irq, 0, sizeof(irq)); 3041 irq.type = KVM_S390_INT_EXTERNAL_CALL; 3042 irq.u.extcall.code = scn; 3043 if (copy_to_user(&buf[n], &irq, sizeof(irq))) 3044 return -EFAULT; 3045 n += sizeof(irq); 3046 } 3047 3048 return n; 3049 } 3050 3051 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask) 3052 { 3053 int vcpu_id, online_vcpus = atomic_read(&kvm->online_vcpus); 3054 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3055 struct kvm_vcpu *vcpu; 3056 3057 for_each_set_bit(vcpu_id, kvm->arch.idle_mask, online_vcpus) { 3058 vcpu = kvm_get_vcpu(kvm, vcpu_id); 3059 if (psw_ioint_disabled(vcpu)) 3060 continue; 3061 deliverable_mask &= (u8)(vcpu->arch.sie_block->gcr[6] >> 24); 3062 if (deliverable_mask) { 3063 /* lately kicked but not yet running */ 3064 if (test_and_set_bit(vcpu_id, gi->kicked_mask)) 3065 return; 3066 kvm_s390_vcpu_wakeup(vcpu); 3067 return; 3068 } 3069 } 3070 } 3071 3072 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer) 3073 { 3074 struct kvm_s390_gisa_interrupt *gi = 3075 container_of(timer, struct kvm_s390_gisa_interrupt, timer); 3076 struct kvm *kvm = 3077 container_of(gi->origin, struct sie_page2, gisa)->kvm; 3078 u8 pending_mask; 3079 3080 pending_mask = gisa_get_ipm_or_restore_iam(gi); 3081 if (pending_mask) { 3082 __airqs_kick_single_vcpu(kvm, pending_mask); 3083 hrtimer_forward_now(timer, ns_to_ktime(gi->expires)); 3084 return HRTIMER_RESTART; 3085 } 3086 3087 return HRTIMER_NORESTART; 3088 } 3089 3090 #define NULL_GISA_ADDR 0x00000000UL 3091 #define NONE_GISA_ADDR 0x00000001UL 3092 #define GISA_ADDR_MASK 0xfffff000UL 3093 3094 static void process_gib_alert_list(void) 3095 { 3096 struct kvm_s390_gisa_interrupt *gi; 3097 struct kvm_s390_gisa *gisa; 3098 struct kvm *kvm; 3099 u32 final, origin = 0UL; 3100 3101 do { 3102 /* 3103 * If the NONE_GISA_ADDR is still stored in the alert list 3104 * origin, we will leave the outer loop. No further GISA has 3105 * been added to the alert list by millicode while processing 3106 * the current alert list. 3107 */ 3108 final = (origin & NONE_GISA_ADDR); 3109 /* 3110 * Cut off the alert list and store the NONE_GISA_ADDR in the 3111 * alert list origin to avoid further GAL interruptions. 3112 * A new alert list can be build up by millicode in parallel 3113 * for guests not in the yet cut-off alert list. When in the 3114 * final loop, store the NULL_GISA_ADDR instead. This will re- 3115 * enable GAL interruptions on the host again. 3116 */ 3117 origin = xchg(&gib->alert_list_origin, 3118 (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR); 3119 /* 3120 * Loop through the just cut-off alert list and start the 3121 * gisa timers to kick idle vcpus to consume the pending 3122 * interruptions asap. 3123 */ 3124 while (origin & GISA_ADDR_MASK) { 3125 gisa = (struct kvm_s390_gisa *)(u64)origin; 3126 origin = gisa->next_alert; 3127 gisa->next_alert = (u32)(u64)gisa; 3128 kvm = container_of(gisa, struct sie_page2, gisa)->kvm; 3129 gi = &kvm->arch.gisa_int; 3130 if (hrtimer_active(&gi->timer)) 3131 hrtimer_cancel(&gi->timer); 3132 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL); 3133 } 3134 } while (!final); 3135 3136 } 3137 3138 void kvm_s390_gisa_clear(struct kvm *kvm) 3139 { 3140 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3141 3142 if (!gi->origin) 3143 return; 3144 gisa_clear_ipm(gi->origin); 3145 VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin); 3146 } 3147 3148 void kvm_s390_gisa_init(struct kvm *kvm) 3149 { 3150 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3151 3152 if (!css_general_characteristics.aiv) 3153 return; 3154 gi->origin = &kvm->arch.sie_page2->gisa; 3155 gi->alert.mask = 0; 3156 spin_lock_init(&gi->alert.ref_lock); 3157 gi->expires = 50 * 1000; /* 50 usec */ 3158 hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 3159 gi->timer.function = gisa_vcpu_kicker; 3160 memset(gi->origin, 0, sizeof(struct kvm_s390_gisa)); 3161 gi->origin->next_alert = (u32)(u64)gi->origin; 3162 VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin); 3163 } 3164 3165 void kvm_s390_gisa_destroy(struct kvm *kvm) 3166 { 3167 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3168 3169 if (!gi->origin) 3170 return; 3171 if (gi->alert.mask) 3172 KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x", 3173 kvm, gi->alert.mask); 3174 while (gisa_in_alert_list(gi->origin)) 3175 cpu_relax(); 3176 hrtimer_cancel(&gi->timer); 3177 gi->origin = NULL; 3178 } 3179 3180 /** 3181 * kvm_s390_gisc_register - register a guest ISC 3182 * 3183 * @kvm: the kernel vm to work with 3184 * @gisc: the guest interruption sub class to register 3185 * 3186 * The function extends the vm specific alert mask to use. 3187 * The effective IAM mask in the GISA is updated as well 3188 * in case the GISA is not part of the GIB alert list. 3189 * It will be updated latest when the IAM gets restored 3190 * by gisa_get_ipm_or_restore_iam(). 3191 * 3192 * Returns: the nonspecific ISC (NISC) the gib alert mechanism 3193 * has registered with the channel subsystem. 3194 * -ENODEV in case the vm uses no GISA 3195 * -ERANGE in case the guest ISC is invalid 3196 */ 3197 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc) 3198 { 3199 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3200 3201 if (!gi->origin) 3202 return -ENODEV; 3203 if (gisc > MAX_ISC) 3204 return -ERANGE; 3205 3206 spin_lock(&gi->alert.ref_lock); 3207 gi->alert.ref_count[gisc]++; 3208 if (gi->alert.ref_count[gisc] == 1) { 3209 gi->alert.mask |= 0x80 >> gisc; 3210 gisa_set_iam(gi->origin, gi->alert.mask); 3211 } 3212 spin_unlock(&gi->alert.ref_lock); 3213 3214 return gib->nisc; 3215 } 3216 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register); 3217 3218 /** 3219 * kvm_s390_gisc_unregister - unregister a guest ISC 3220 * 3221 * @kvm: the kernel vm to work with 3222 * @gisc: the guest interruption sub class to register 3223 * 3224 * The function reduces the vm specific alert mask to use. 3225 * The effective IAM mask in the GISA is updated as well 3226 * in case the GISA is not part of the GIB alert list. 3227 * It will be updated latest when the IAM gets restored 3228 * by gisa_get_ipm_or_restore_iam(). 3229 * 3230 * Returns: the nonspecific ISC (NISC) the gib alert mechanism 3231 * has registered with the channel subsystem. 3232 * -ENODEV in case the vm uses no GISA 3233 * -ERANGE in case the guest ISC is invalid 3234 * -EINVAL in case the guest ISC is not registered 3235 */ 3236 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc) 3237 { 3238 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3239 int rc = 0; 3240 3241 if (!gi->origin) 3242 return -ENODEV; 3243 if (gisc > MAX_ISC) 3244 return -ERANGE; 3245 3246 spin_lock(&gi->alert.ref_lock); 3247 if (gi->alert.ref_count[gisc] == 0) { 3248 rc = -EINVAL; 3249 goto out; 3250 } 3251 gi->alert.ref_count[gisc]--; 3252 if (gi->alert.ref_count[gisc] == 0) { 3253 gi->alert.mask &= ~(0x80 >> gisc); 3254 gisa_set_iam(gi->origin, gi->alert.mask); 3255 } 3256 out: 3257 spin_unlock(&gi->alert.ref_lock); 3258 3259 return rc; 3260 } 3261 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister); 3262 3263 static void gib_alert_irq_handler(struct airq_struct *airq, bool floating) 3264 { 3265 inc_irq_stat(IRQIO_GAL); 3266 process_gib_alert_list(); 3267 } 3268 3269 static struct airq_struct gib_alert_irq = { 3270 .handler = gib_alert_irq_handler, 3271 .lsi_ptr = &gib_alert_irq.lsi_mask, 3272 }; 3273 3274 void kvm_s390_gib_destroy(void) 3275 { 3276 if (!gib) 3277 return; 3278 chsc_sgib(0); 3279 unregister_adapter_interrupt(&gib_alert_irq); 3280 free_page((unsigned long)gib); 3281 gib = NULL; 3282 } 3283 3284 int kvm_s390_gib_init(u8 nisc) 3285 { 3286 int rc = 0; 3287 3288 if (!css_general_characteristics.aiv) { 3289 KVM_EVENT(3, "%s", "gib not initialized, no AIV facility"); 3290 goto out; 3291 } 3292 3293 gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL | GFP_DMA); 3294 if (!gib) { 3295 rc = -ENOMEM; 3296 goto out; 3297 } 3298 3299 gib_alert_irq.isc = nisc; 3300 if (register_adapter_interrupt(&gib_alert_irq)) { 3301 pr_err("Registering the GIB alert interruption handler failed\n"); 3302 rc = -EIO; 3303 goto out_free_gib; 3304 } 3305 3306 gib->nisc = nisc; 3307 if (chsc_sgib((u32)(u64)gib)) { 3308 pr_err("Associating the GIB with the AIV facility failed\n"); 3309 free_page((unsigned long)gib); 3310 gib = NULL; 3311 rc = -EIO; 3312 goto out_unreg_gal; 3313 } 3314 3315 KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc); 3316 goto out; 3317 3318 out_unreg_gal: 3319 unregister_adapter_interrupt(&gib_alert_irq); 3320 out_free_gib: 3321 free_page((unsigned long)gib); 3322 gib = NULL; 3323 out: 3324 return rc; 3325 } 3326