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_idx, 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_idx, 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_PGM_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_t(u64, 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_halt(vcpu); 1339 vcpu->valid_wakeup = false; 1340 __unset_cpu_idle(vcpu); 1341 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 1342 1343 hrtimer_cancel(&vcpu->arch.ckc_timer); 1344 return 0; 1345 } 1346 1347 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu) 1348 { 1349 vcpu->valid_wakeup = true; 1350 kvm_vcpu_wake_up(vcpu); 1351 1352 /* 1353 * The VCPU might not be sleeping but rather executing VSIE. Let's 1354 * kick it, so it leaves the SIE to process the request. 1355 */ 1356 kvm_s390_vsie_kick(vcpu); 1357 } 1358 1359 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer) 1360 { 1361 struct kvm_vcpu *vcpu; 1362 u64 sltime; 1363 1364 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer); 1365 sltime = __calculate_sltime(vcpu); 1366 1367 /* 1368 * If the monotonic clock runs faster than the tod clock we might be 1369 * woken up too early and have to go back to sleep to avoid deadlocks. 1370 */ 1371 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime))) 1372 return HRTIMER_RESTART; 1373 kvm_s390_vcpu_wakeup(vcpu); 1374 return HRTIMER_NORESTART; 1375 } 1376 1377 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu) 1378 { 1379 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1380 1381 spin_lock(&li->lock); 1382 li->pending_irqs = 0; 1383 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS); 1384 memset(&li->irq, 0, sizeof(li->irq)); 1385 spin_unlock(&li->lock); 1386 1387 sca_clear_ext_call(vcpu); 1388 } 1389 1390 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu) 1391 { 1392 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1393 int rc = 0; 1394 unsigned long irq_type; 1395 unsigned long irqs; 1396 1397 __reset_intercept_indicators(vcpu); 1398 1399 /* pending ckc conditions might have been invalidated */ 1400 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 1401 if (ckc_irq_pending(vcpu)) 1402 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 1403 1404 /* pending cpu timer conditions might have been invalidated */ 1405 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 1406 if (cpu_timer_irq_pending(vcpu)) 1407 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 1408 1409 while ((irqs = deliverable_irqs(vcpu)) && !rc) { 1410 /* bits are in the reverse order of interrupt priority */ 1411 irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT); 1412 switch (irq_type) { 1413 case IRQ_PEND_IO_ISC_0: 1414 case IRQ_PEND_IO_ISC_1: 1415 case IRQ_PEND_IO_ISC_2: 1416 case IRQ_PEND_IO_ISC_3: 1417 case IRQ_PEND_IO_ISC_4: 1418 case IRQ_PEND_IO_ISC_5: 1419 case IRQ_PEND_IO_ISC_6: 1420 case IRQ_PEND_IO_ISC_7: 1421 rc = __deliver_io(vcpu, irq_type); 1422 break; 1423 case IRQ_PEND_MCHK_EX: 1424 case IRQ_PEND_MCHK_REP: 1425 rc = __deliver_machine_check(vcpu); 1426 break; 1427 case IRQ_PEND_PROG: 1428 rc = __deliver_prog(vcpu); 1429 break; 1430 case IRQ_PEND_EXT_EMERGENCY: 1431 rc = __deliver_emergency_signal(vcpu); 1432 break; 1433 case IRQ_PEND_EXT_EXTERNAL: 1434 rc = __deliver_external_call(vcpu); 1435 break; 1436 case IRQ_PEND_EXT_CLOCK_COMP: 1437 rc = __deliver_ckc(vcpu); 1438 break; 1439 case IRQ_PEND_EXT_CPU_TIMER: 1440 rc = __deliver_cpu_timer(vcpu); 1441 break; 1442 case IRQ_PEND_RESTART: 1443 rc = __deliver_restart(vcpu); 1444 break; 1445 case IRQ_PEND_SET_PREFIX: 1446 rc = __deliver_set_prefix(vcpu); 1447 break; 1448 case IRQ_PEND_PFAULT_INIT: 1449 rc = __deliver_pfault_init(vcpu); 1450 break; 1451 case IRQ_PEND_EXT_SERVICE: 1452 rc = __deliver_service(vcpu); 1453 break; 1454 case IRQ_PEND_EXT_SERVICE_EV: 1455 rc = __deliver_service_ev(vcpu); 1456 break; 1457 case IRQ_PEND_PFAULT_DONE: 1458 rc = __deliver_pfault_done(vcpu); 1459 break; 1460 case IRQ_PEND_VIRTIO: 1461 rc = __deliver_virtio(vcpu); 1462 break; 1463 default: 1464 WARN_ONCE(1, "Unknown pending irq type %ld", irq_type); 1465 clear_bit(irq_type, &li->pending_irqs); 1466 } 1467 } 1468 1469 set_intercept_indicators(vcpu); 1470 1471 return rc; 1472 } 1473 1474 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1475 { 1476 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1477 1478 vcpu->stat.inject_program++; 1479 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code); 1480 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT, 1481 irq->u.pgm.code, 0); 1482 1483 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) { 1484 /* auto detection if no valid ILC was given */ 1485 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK; 1486 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu); 1487 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID; 1488 } 1489 1490 if (irq->u.pgm.code == PGM_PER) { 1491 li->irq.pgm.code |= PGM_PER; 1492 li->irq.pgm.flags = irq->u.pgm.flags; 1493 /* only modify PER related information */ 1494 li->irq.pgm.per_address = irq->u.pgm.per_address; 1495 li->irq.pgm.per_code = irq->u.pgm.per_code; 1496 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid; 1497 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id; 1498 } else if (!(irq->u.pgm.code & PGM_PER)) { 1499 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) | 1500 irq->u.pgm.code; 1501 li->irq.pgm.flags = irq->u.pgm.flags; 1502 /* only modify non-PER information */ 1503 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code; 1504 li->irq.pgm.mon_code = irq->u.pgm.mon_code; 1505 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code; 1506 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr; 1507 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id; 1508 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id; 1509 } else { 1510 li->irq.pgm = irq->u.pgm; 1511 } 1512 set_bit(IRQ_PEND_PROG, &li->pending_irqs); 1513 return 0; 1514 } 1515 1516 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1517 { 1518 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1519 1520 vcpu->stat.inject_pfault_init++; 1521 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx", 1522 irq->u.ext.ext_params2); 1523 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT, 1524 irq->u.ext.ext_params, 1525 irq->u.ext.ext_params2); 1526 1527 li->irq.ext = irq->u.ext; 1528 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs); 1529 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1530 return 0; 1531 } 1532 1533 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1534 { 1535 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1536 struct kvm_s390_extcall_info *extcall = &li->irq.extcall; 1537 uint16_t src_id = irq->u.extcall.code; 1538 1539 vcpu->stat.inject_external_call++; 1540 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u", 1541 src_id); 1542 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL, 1543 src_id, 0); 1544 1545 /* sending vcpu invalid */ 1546 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL) 1547 return -EINVAL; 1548 1549 if (sclp.has_sigpif && !kvm_s390_pv_cpu_get_handle(vcpu)) 1550 return sca_inject_ext_call(vcpu, src_id); 1551 1552 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs)) 1553 return -EBUSY; 1554 *extcall = irq->u.extcall; 1555 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1556 return 0; 1557 } 1558 1559 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1560 { 1561 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1562 struct kvm_s390_prefix_info *prefix = &li->irq.prefix; 1563 1564 vcpu->stat.inject_set_prefix++; 1565 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x", 1566 irq->u.prefix.address); 1567 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX, 1568 irq->u.prefix.address, 0); 1569 1570 if (!is_vcpu_stopped(vcpu)) 1571 return -EBUSY; 1572 1573 *prefix = irq->u.prefix; 1574 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs); 1575 return 0; 1576 } 1577 1578 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS) 1579 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1580 { 1581 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1582 struct kvm_s390_stop_info *stop = &li->irq.stop; 1583 int rc = 0; 1584 1585 vcpu->stat.inject_stop_signal++; 1586 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0); 1587 1588 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS) 1589 return -EINVAL; 1590 1591 if (is_vcpu_stopped(vcpu)) { 1592 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS) 1593 rc = kvm_s390_store_status_unloaded(vcpu, 1594 KVM_S390_STORE_STATUS_NOADDR); 1595 return rc; 1596 } 1597 1598 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs)) 1599 return -EBUSY; 1600 stop->flags = irq->u.stop.flags; 1601 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT); 1602 return 0; 1603 } 1604 1605 static int __inject_sigp_restart(struct kvm_vcpu *vcpu) 1606 { 1607 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1608 1609 vcpu->stat.inject_restart++; 1610 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int"); 1611 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0); 1612 1613 set_bit(IRQ_PEND_RESTART, &li->pending_irqs); 1614 return 0; 1615 } 1616 1617 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu, 1618 struct kvm_s390_irq *irq) 1619 { 1620 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1621 1622 vcpu->stat.inject_emergency_signal++; 1623 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u", 1624 irq->u.emerg.code); 1625 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY, 1626 irq->u.emerg.code, 0); 1627 1628 /* sending vcpu invalid */ 1629 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL) 1630 return -EINVAL; 1631 1632 set_bit(irq->u.emerg.code, li->sigp_emerg_pending); 1633 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs); 1634 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1635 return 0; 1636 } 1637 1638 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1639 { 1640 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1641 struct kvm_s390_mchk_info *mchk = &li->irq.mchk; 1642 1643 vcpu->stat.inject_mchk++; 1644 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx", 1645 irq->u.mchk.mcic); 1646 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0, 1647 irq->u.mchk.mcic); 1648 1649 /* 1650 * Because repressible machine checks can be indicated along with 1651 * exigent machine checks (PoP, Chapter 11, Interruption action) 1652 * we need to combine cr14, mcic and external damage code. 1653 * Failing storage address and the logout area should not be or'ed 1654 * together, we just indicate the last occurrence of the corresponding 1655 * machine check 1656 */ 1657 mchk->cr14 |= irq->u.mchk.cr14; 1658 mchk->mcic |= irq->u.mchk.mcic; 1659 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code; 1660 mchk->failing_storage_address = irq->u.mchk.failing_storage_address; 1661 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout, 1662 sizeof(mchk->fixed_logout)); 1663 if (mchk->mcic & MCHK_EX_MASK) 1664 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs); 1665 else if (mchk->mcic & MCHK_REP_MASK) 1666 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs); 1667 return 0; 1668 } 1669 1670 static int __inject_ckc(struct kvm_vcpu *vcpu) 1671 { 1672 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1673 1674 vcpu->stat.inject_ckc++; 1675 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external"); 1676 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP, 1677 0, 0); 1678 1679 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 1680 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1681 return 0; 1682 } 1683 1684 static int __inject_cpu_timer(struct kvm_vcpu *vcpu) 1685 { 1686 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1687 1688 vcpu->stat.inject_cputm++; 1689 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external"); 1690 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER, 1691 0, 0); 1692 1693 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 1694 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1695 return 0; 1696 } 1697 1698 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm, 1699 int isc, u32 schid) 1700 { 1701 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1702 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc]; 1703 struct kvm_s390_interrupt_info *iter; 1704 u16 id = (schid & 0xffff0000U) >> 16; 1705 u16 nr = schid & 0x0000ffffU; 1706 1707 spin_lock(&fi->lock); 1708 list_for_each_entry(iter, isc_list, list) { 1709 if (schid && (id != iter->io.subchannel_id || 1710 nr != iter->io.subchannel_nr)) 1711 continue; 1712 /* found an appropriate entry */ 1713 list_del_init(&iter->list); 1714 fi->counters[FIRQ_CNTR_IO] -= 1; 1715 if (list_empty(isc_list)) 1716 clear_bit(isc_to_irq_type(isc), &fi->pending_irqs); 1717 spin_unlock(&fi->lock); 1718 return iter; 1719 } 1720 spin_unlock(&fi->lock); 1721 return NULL; 1722 } 1723 1724 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm, 1725 u64 isc_mask, u32 schid) 1726 { 1727 struct kvm_s390_interrupt_info *inti = NULL; 1728 int isc; 1729 1730 for (isc = 0; isc <= MAX_ISC && !inti; isc++) { 1731 if (isc_mask & isc_to_isc_bits(isc)) 1732 inti = get_io_int(kvm, isc, schid); 1733 } 1734 return inti; 1735 } 1736 1737 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid) 1738 { 1739 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 1740 unsigned long active_mask; 1741 int isc; 1742 1743 if (schid) 1744 goto out; 1745 if (!gi->origin) 1746 goto out; 1747 1748 active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32; 1749 while (active_mask) { 1750 isc = __fls(active_mask) ^ (BITS_PER_LONG - 1); 1751 if (gisa_tac_ipm_gisc(gi->origin, isc)) 1752 return isc; 1753 clear_bit_inv(isc, &active_mask); 1754 } 1755 out: 1756 return -EINVAL; 1757 } 1758 1759 /* 1760 * Dequeue and return an I/O interrupt matching any of the interruption 1761 * subclasses as designated by the isc mask in cr6 and the schid (if != 0). 1762 * Take into account the interrupts pending in the interrupt list and in GISA. 1763 * 1764 * Note that for a guest that does not enable I/O interrupts 1765 * but relies on TPI, a flood of classic interrupts may starve 1766 * out adapter interrupts on the same isc. Linux does not do 1767 * that, and it is possible to work around the issue by configuring 1768 * different iscs for classic and adapter interrupts in the guest, 1769 * but we may want to revisit this in the future. 1770 */ 1771 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm, 1772 u64 isc_mask, u32 schid) 1773 { 1774 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 1775 struct kvm_s390_interrupt_info *inti, *tmp_inti; 1776 int isc; 1777 1778 inti = get_top_io_int(kvm, isc_mask, schid); 1779 1780 isc = get_top_gisa_isc(kvm, isc_mask, schid); 1781 if (isc < 0) 1782 /* no AI in GISA */ 1783 goto out; 1784 1785 if (!inti) 1786 /* AI in GISA but no classical IO int */ 1787 goto gisa_out; 1788 1789 /* both types of interrupts present */ 1790 if (int_word_to_isc(inti->io.io_int_word) <= isc) { 1791 /* classical IO int with higher priority */ 1792 gisa_set_ipm_gisc(gi->origin, isc); 1793 goto out; 1794 } 1795 gisa_out: 1796 tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT); 1797 if (tmp_inti) { 1798 tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0); 1799 tmp_inti->io.io_int_word = isc_to_int_word(isc); 1800 if (inti) 1801 kvm_s390_reinject_io_int(kvm, inti); 1802 inti = tmp_inti; 1803 } else 1804 gisa_set_ipm_gisc(gi->origin, isc); 1805 out: 1806 return inti; 1807 } 1808 1809 static int __inject_service(struct kvm *kvm, 1810 struct kvm_s390_interrupt_info *inti) 1811 { 1812 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1813 1814 kvm->stat.inject_service_signal++; 1815 spin_lock(&fi->lock); 1816 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING; 1817 1818 /* We always allow events, track them separately from the sccb ints */ 1819 if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING) 1820 set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs); 1821 1822 /* 1823 * Early versions of the QEMU s390 bios will inject several 1824 * service interrupts after another without handling a 1825 * condition code indicating busy. 1826 * We will silently ignore those superfluous sccb values. 1827 * A future version of QEMU will take care of serialization 1828 * of servc requests 1829 */ 1830 if (fi->srv_signal.ext_params & SCCB_MASK) 1831 goto out; 1832 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK; 1833 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs); 1834 out: 1835 spin_unlock(&fi->lock); 1836 kfree(inti); 1837 return 0; 1838 } 1839 1840 static int __inject_virtio(struct kvm *kvm, 1841 struct kvm_s390_interrupt_info *inti) 1842 { 1843 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1844 1845 kvm->stat.inject_virtio++; 1846 spin_lock(&fi->lock); 1847 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) { 1848 spin_unlock(&fi->lock); 1849 return -EBUSY; 1850 } 1851 fi->counters[FIRQ_CNTR_VIRTIO] += 1; 1852 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]); 1853 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs); 1854 spin_unlock(&fi->lock); 1855 return 0; 1856 } 1857 1858 static int __inject_pfault_done(struct kvm *kvm, 1859 struct kvm_s390_interrupt_info *inti) 1860 { 1861 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1862 1863 kvm->stat.inject_pfault_done++; 1864 spin_lock(&fi->lock); 1865 if (fi->counters[FIRQ_CNTR_PFAULT] >= 1866 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) { 1867 spin_unlock(&fi->lock); 1868 return -EBUSY; 1869 } 1870 fi->counters[FIRQ_CNTR_PFAULT] += 1; 1871 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]); 1872 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs); 1873 spin_unlock(&fi->lock); 1874 return 0; 1875 } 1876 1877 #define CR_PENDING_SUBCLASS 28 1878 static int __inject_float_mchk(struct kvm *kvm, 1879 struct kvm_s390_interrupt_info *inti) 1880 { 1881 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1882 1883 kvm->stat.inject_float_mchk++; 1884 spin_lock(&fi->lock); 1885 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS); 1886 fi->mchk.mcic |= inti->mchk.mcic; 1887 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs); 1888 spin_unlock(&fi->lock); 1889 kfree(inti); 1890 return 0; 1891 } 1892 1893 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti) 1894 { 1895 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 1896 struct kvm_s390_float_interrupt *fi; 1897 struct list_head *list; 1898 int isc; 1899 1900 kvm->stat.inject_io++; 1901 isc = int_word_to_isc(inti->io.io_int_word); 1902 1903 /* 1904 * We do not use the lock checking variant as this is just a 1905 * performance optimization and we do not hold the lock here. 1906 * This is ok as the code will pick interrupts from both "lists" 1907 * for delivery. 1908 */ 1909 if (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_ACCOUNT); 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 int kvm_s390_is_restart_irq_pending(struct kvm_vcpu *vcpu) 2119 { 2120 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2121 2122 return test_bit(IRQ_PEND_RESTART, &li->pending_irqs); 2123 } 2124 2125 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu) 2126 { 2127 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2128 2129 spin_lock(&li->lock); 2130 li->irq.stop.flags = 0; 2131 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs); 2132 spin_unlock(&li->lock); 2133 } 2134 2135 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 2136 { 2137 int rc; 2138 2139 switch (irq->type) { 2140 case KVM_S390_PROGRAM_INT: 2141 rc = __inject_prog(vcpu, irq); 2142 break; 2143 case KVM_S390_SIGP_SET_PREFIX: 2144 rc = __inject_set_prefix(vcpu, irq); 2145 break; 2146 case KVM_S390_SIGP_STOP: 2147 rc = __inject_sigp_stop(vcpu, irq); 2148 break; 2149 case KVM_S390_RESTART: 2150 rc = __inject_sigp_restart(vcpu); 2151 break; 2152 case KVM_S390_INT_CLOCK_COMP: 2153 rc = __inject_ckc(vcpu); 2154 break; 2155 case KVM_S390_INT_CPU_TIMER: 2156 rc = __inject_cpu_timer(vcpu); 2157 break; 2158 case KVM_S390_INT_EXTERNAL_CALL: 2159 rc = __inject_extcall(vcpu, irq); 2160 break; 2161 case KVM_S390_INT_EMERGENCY: 2162 rc = __inject_sigp_emergency(vcpu, irq); 2163 break; 2164 case KVM_S390_MCHK: 2165 rc = __inject_mchk(vcpu, irq); 2166 break; 2167 case KVM_S390_INT_PFAULT_INIT: 2168 rc = __inject_pfault_init(vcpu, irq); 2169 break; 2170 case KVM_S390_INT_VIRTIO: 2171 case KVM_S390_INT_SERVICE: 2172 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2173 default: 2174 rc = -EINVAL; 2175 } 2176 2177 return rc; 2178 } 2179 2180 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 2181 { 2182 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2183 int rc; 2184 2185 spin_lock(&li->lock); 2186 rc = do_inject_vcpu(vcpu, irq); 2187 spin_unlock(&li->lock); 2188 if (!rc) 2189 kvm_s390_vcpu_wakeup(vcpu); 2190 return rc; 2191 } 2192 2193 static inline void clear_irq_list(struct list_head *_list) 2194 { 2195 struct kvm_s390_interrupt_info *inti, *n; 2196 2197 list_for_each_entry_safe(inti, n, _list, list) { 2198 list_del(&inti->list); 2199 kfree(inti); 2200 } 2201 } 2202 2203 static void inti_to_irq(struct kvm_s390_interrupt_info *inti, 2204 struct kvm_s390_irq *irq) 2205 { 2206 irq->type = inti->type; 2207 switch (inti->type) { 2208 case KVM_S390_INT_PFAULT_INIT: 2209 case KVM_S390_INT_PFAULT_DONE: 2210 case KVM_S390_INT_VIRTIO: 2211 irq->u.ext = inti->ext; 2212 break; 2213 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2214 irq->u.io = inti->io; 2215 break; 2216 } 2217 } 2218 2219 void kvm_s390_clear_float_irqs(struct kvm *kvm) 2220 { 2221 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2222 int i; 2223 2224 mutex_lock(&kvm->lock); 2225 if (!kvm_s390_pv_is_protected(kvm)) 2226 fi->masked_irqs = 0; 2227 mutex_unlock(&kvm->lock); 2228 spin_lock(&fi->lock); 2229 fi->pending_irqs = 0; 2230 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal)); 2231 memset(&fi->mchk, 0, sizeof(fi->mchk)); 2232 for (i = 0; i < FIRQ_LIST_COUNT; i++) 2233 clear_irq_list(&fi->lists[i]); 2234 for (i = 0; i < FIRQ_MAX_COUNT; i++) 2235 fi->counters[i] = 0; 2236 spin_unlock(&fi->lock); 2237 kvm_s390_gisa_clear(kvm); 2238 }; 2239 2240 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len) 2241 { 2242 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 2243 struct kvm_s390_interrupt_info *inti; 2244 struct kvm_s390_float_interrupt *fi; 2245 struct kvm_s390_irq *buf; 2246 struct kvm_s390_irq *irq; 2247 int max_irqs; 2248 int ret = 0; 2249 int n = 0; 2250 int i; 2251 2252 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0) 2253 return -EINVAL; 2254 2255 /* 2256 * We are already using -ENOMEM to signal 2257 * userspace it may retry with a bigger buffer, 2258 * so we need to use something else for this case 2259 */ 2260 buf = vzalloc(len); 2261 if (!buf) 2262 return -ENOBUFS; 2263 2264 max_irqs = len / sizeof(struct kvm_s390_irq); 2265 2266 if (gi->origin && gisa_get_ipm(gi->origin)) { 2267 for (i = 0; i <= MAX_ISC; i++) { 2268 if (n == max_irqs) { 2269 /* signal userspace to try again */ 2270 ret = -ENOMEM; 2271 goto out_nolock; 2272 } 2273 if (gisa_tac_ipm_gisc(gi->origin, i)) { 2274 irq = (struct kvm_s390_irq *) &buf[n]; 2275 irq->type = KVM_S390_INT_IO(1, 0, 0, 0); 2276 irq->u.io.io_int_word = isc_to_int_word(i); 2277 n++; 2278 } 2279 } 2280 } 2281 fi = &kvm->arch.float_int; 2282 spin_lock(&fi->lock); 2283 for (i = 0; i < FIRQ_LIST_COUNT; i++) { 2284 list_for_each_entry(inti, &fi->lists[i], list) { 2285 if (n == max_irqs) { 2286 /* signal userspace to try again */ 2287 ret = -ENOMEM; 2288 goto out; 2289 } 2290 inti_to_irq(inti, &buf[n]); 2291 n++; 2292 } 2293 } 2294 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) || 2295 test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) { 2296 if (n == max_irqs) { 2297 /* signal userspace to try again */ 2298 ret = -ENOMEM; 2299 goto out; 2300 } 2301 irq = (struct kvm_s390_irq *) &buf[n]; 2302 irq->type = KVM_S390_INT_SERVICE; 2303 irq->u.ext = fi->srv_signal; 2304 n++; 2305 } 2306 if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) { 2307 if (n == max_irqs) { 2308 /* signal userspace to try again */ 2309 ret = -ENOMEM; 2310 goto out; 2311 } 2312 irq = (struct kvm_s390_irq *) &buf[n]; 2313 irq->type = KVM_S390_MCHK; 2314 irq->u.mchk = fi->mchk; 2315 n++; 2316 } 2317 2318 out: 2319 spin_unlock(&fi->lock); 2320 out_nolock: 2321 if (!ret && n > 0) { 2322 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n)) 2323 ret = -EFAULT; 2324 } 2325 vfree(buf); 2326 2327 return ret < 0 ? ret : n; 2328 } 2329 2330 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr) 2331 { 2332 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2333 struct kvm_s390_ais_all ais; 2334 2335 if (attr->attr < sizeof(ais)) 2336 return -EINVAL; 2337 2338 if (!test_kvm_facility(kvm, 72)) 2339 return -EOPNOTSUPP; 2340 2341 mutex_lock(&fi->ais_lock); 2342 ais.simm = fi->simm; 2343 ais.nimm = fi->nimm; 2344 mutex_unlock(&fi->ais_lock); 2345 2346 if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais))) 2347 return -EFAULT; 2348 2349 return 0; 2350 } 2351 2352 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr) 2353 { 2354 int r; 2355 2356 switch (attr->group) { 2357 case KVM_DEV_FLIC_GET_ALL_IRQS: 2358 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr, 2359 attr->attr); 2360 break; 2361 case KVM_DEV_FLIC_AISM_ALL: 2362 r = flic_ais_mode_get_all(dev->kvm, attr); 2363 break; 2364 default: 2365 r = -EINVAL; 2366 } 2367 2368 return r; 2369 } 2370 2371 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti, 2372 u64 addr) 2373 { 2374 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr; 2375 void *target = NULL; 2376 void __user *source; 2377 u64 size; 2378 2379 if (get_user(inti->type, (u64 __user *)addr)) 2380 return -EFAULT; 2381 2382 switch (inti->type) { 2383 case KVM_S390_INT_PFAULT_INIT: 2384 case KVM_S390_INT_PFAULT_DONE: 2385 case KVM_S390_INT_VIRTIO: 2386 case KVM_S390_INT_SERVICE: 2387 target = (void *) &inti->ext; 2388 source = &uptr->u.ext; 2389 size = sizeof(inti->ext); 2390 break; 2391 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2392 target = (void *) &inti->io; 2393 source = &uptr->u.io; 2394 size = sizeof(inti->io); 2395 break; 2396 case KVM_S390_MCHK: 2397 target = (void *) &inti->mchk; 2398 source = &uptr->u.mchk; 2399 size = sizeof(inti->mchk); 2400 break; 2401 default: 2402 return -EINVAL; 2403 } 2404 2405 if (copy_from_user(target, source, size)) 2406 return -EFAULT; 2407 2408 return 0; 2409 } 2410 2411 static int enqueue_floating_irq(struct kvm_device *dev, 2412 struct kvm_device_attr *attr) 2413 { 2414 struct kvm_s390_interrupt_info *inti = NULL; 2415 int r = 0; 2416 int len = attr->attr; 2417 2418 if (len % sizeof(struct kvm_s390_irq) != 0) 2419 return -EINVAL; 2420 else if (len > KVM_S390_FLIC_MAX_BUFFER) 2421 return -EINVAL; 2422 2423 while (len >= sizeof(struct kvm_s390_irq)) { 2424 inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT); 2425 if (!inti) 2426 return -ENOMEM; 2427 2428 r = copy_irq_from_user(inti, attr->addr); 2429 if (r) { 2430 kfree(inti); 2431 return r; 2432 } 2433 r = __inject_vm(dev->kvm, inti); 2434 if (r) { 2435 kfree(inti); 2436 return r; 2437 } 2438 len -= sizeof(struct kvm_s390_irq); 2439 attr->addr += sizeof(struct kvm_s390_irq); 2440 } 2441 2442 return r; 2443 } 2444 2445 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id) 2446 { 2447 if (id >= MAX_S390_IO_ADAPTERS) 2448 return NULL; 2449 id = array_index_nospec(id, MAX_S390_IO_ADAPTERS); 2450 return kvm->arch.adapters[id]; 2451 } 2452 2453 static int register_io_adapter(struct kvm_device *dev, 2454 struct kvm_device_attr *attr) 2455 { 2456 struct s390_io_adapter *adapter; 2457 struct kvm_s390_io_adapter adapter_info; 2458 2459 if (copy_from_user(&adapter_info, 2460 (void __user *)attr->addr, sizeof(adapter_info))) 2461 return -EFAULT; 2462 2463 if (adapter_info.id >= MAX_S390_IO_ADAPTERS) 2464 return -EINVAL; 2465 2466 adapter_info.id = array_index_nospec(adapter_info.id, 2467 MAX_S390_IO_ADAPTERS); 2468 2469 if (dev->kvm->arch.adapters[adapter_info.id] != NULL) 2470 return -EINVAL; 2471 2472 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL_ACCOUNT); 2473 if (!adapter) 2474 return -ENOMEM; 2475 2476 adapter->id = adapter_info.id; 2477 adapter->isc = adapter_info.isc; 2478 adapter->maskable = adapter_info.maskable; 2479 adapter->masked = false; 2480 adapter->swap = adapter_info.swap; 2481 adapter->suppressible = (adapter_info.flags) & 2482 KVM_S390_ADAPTER_SUPPRESSIBLE; 2483 dev->kvm->arch.adapters[adapter->id] = adapter; 2484 2485 return 0; 2486 } 2487 2488 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked) 2489 { 2490 int ret; 2491 struct s390_io_adapter *adapter = get_io_adapter(kvm, id); 2492 2493 if (!adapter || !adapter->maskable) 2494 return -EINVAL; 2495 ret = adapter->masked; 2496 adapter->masked = masked; 2497 return ret; 2498 } 2499 2500 void kvm_s390_destroy_adapters(struct kvm *kvm) 2501 { 2502 int i; 2503 2504 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) 2505 kfree(kvm->arch.adapters[i]); 2506 } 2507 2508 static int modify_io_adapter(struct kvm_device *dev, 2509 struct kvm_device_attr *attr) 2510 { 2511 struct kvm_s390_io_adapter_req req; 2512 struct s390_io_adapter *adapter; 2513 int ret; 2514 2515 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req))) 2516 return -EFAULT; 2517 2518 adapter = get_io_adapter(dev->kvm, req.id); 2519 if (!adapter) 2520 return -EINVAL; 2521 switch (req.type) { 2522 case KVM_S390_IO_ADAPTER_MASK: 2523 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask); 2524 if (ret > 0) 2525 ret = 0; 2526 break; 2527 /* 2528 * The following operations are no longer needed and therefore no-ops. 2529 * The gpa to hva translation is done when an IRQ route is set up. The 2530 * set_irq code uses get_user_pages_remote() to do the actual write. 2531 */ 2532 case KVM_S390_IO_ADAPTER_MAP: 2533 case KVM_S390_IO_ADAPTER_UNMAP: 2534 ret = 0; 2535 break; 2536 default: 2537 ret = -EINVAL; 2538 } 2539 2540 return ret; 2541 } 2542 2543 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr) 2544 2545 { 2546 const u64 isc_mask = 0xffUL << 24; /* all iscs set */ 2547 u32 schid; 2548 2549 if (attr->flags) 2550 return -EINVAL; 2551 if (attr->attr != sizeof(schid)) 2552 return -EINVAL; 2553 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid))) 2554 return -EFAULT; 2555 if (!schid) 2556 return -EINVAL; 2557 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid)); 2558 /* 2559 * If userspace is conforming to the architecture, we can have at most 2560 * one pending I/O interrupt per subchannel, so this is effectively a 2561 * clear all. 2562 */ 2563 return 0; 2564 } 2565 2566 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr) 2567 { 2568 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2569 struct kvm_s390_ais_req req; 2570 int ret = 0; 2571 2572 if (!test_kvm_facility(kvm, 72)) 2573 return -EOPNOTSUPP; 2574 2575 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req))) 2576 return -EFAULT; 2577 2578 if (req.isc > MAX_ISC) 2579 return -EINVAL; 2580 2581 trace_kvm_s390_modify_ais_mode(req.isc, 2582 (fi->simm & AIS_MODE_MASK(req.isc)) ? 2583 (fi->nimm & AIS_MODE_MASK(req.isc)) ? 2584 2 : KVM_S390_AIS_MODE_SINGLE : 2585 KVM_S390_AIS_MODE_ALL, req.mode); 2586 2587 mutex_lock(&fi->ais_lock); 2588 switch (req.mode) { 2589 case KVM_S390_AIS_MODE_ALL: 2590 fi->simm &= ~AIS_MODE_MASK(req.isc); 2591 fi->nimm &= ~AIS_MODE_MASK(req.isc); 2592 break; 2593 case KVM_S390_AIS_MODE_SINGLE: 2594 fi->simm |= AIS_MODE_MASK(req.isc); 2595 fi->nimm &= ~AIS_MODE_MASK(req.isc); 2596 break; 2597 default: 2598 ret = -EINVAL; 2599 } 2600 mutex_unlock(&fi->ais_lock); 2601 2602 return ret; 2603 } 2604 2605 static int kvm_s390_inject_airq(struct kvm *kvm, 2606 struct s390_io_adapter *adapter) 2607 { 2608 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2609 struct kvm_s390_interrupt s390int = { 2610 .type = KVM_S390_INT_IO(1, 0, 0, 0), 2611 .parm = 0, 2612 .parm64 = isc_to_int_word(adapter->isc), 2613 }; 2614 int ret = 0; 2615 2616 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible) 2617 return kvm_s390_inject_vm(kvm, &s390int); 2618 2619 mutex_lock(&fi->ais_lock); 2620 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) { 2621 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc); 2622 goto out; 2623 } 2624 2625 ret = kvm_s390_inject_vm(kvm, &s390int); 2626 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) { 2627 fi->nimm |= AIS_MODE_MASK(adapter->isc); 2628 trace_kvm_s390_modify_ais_mode(adapter->isc, 2629 KVM_S390_AIS_MODE_SINGLE, 2); 2630 } 2631 out: 2632 mutex_unlock(&fi->ais_lock); 2633 return ret; 2634 } 2635 2636 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr) 2637 { 2638 unsigned int id = attr->attr; 2639 struct s390_io_adapter *adapter = get_io_adapter(kvm, id); 2640 2641 if (!adapter) 2642 return -EINVAL; 2643 2644 return kvm_s390_inject_airq(kvm, adapter); 2645 } 2646 2647 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr) 2648 { 2649 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2650 struct kvm_s390_ais_all ais; 2651 2652 if (!test_kvm_facility(kvm, 72)) 2653 return -EOPNOTSUPP; 2654 2655 if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais))) 2656 return -EFAULT; 2657 2658 mutex_lock(&fi->ais_lock); 2659 fi->simm = ais.simm; 2660 fi->nimm = ais.nimm; 2661 mutex_unlock(&fi->ais_lock); 2662 2663 return 0; 2664 } 2665 2666 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr) 2667 { 2668 int r = 0; 2669 unsigned long i; 2670 struct kvm_vcpu *vcpu; 2671 2672 switch (attr->group) { 2673 case KVM_DEV_FLIC_ENQUEUE: 2674 r = enqueue_floating_irq(dev, attr); 2675 break; 2676 case KVM_DEV_FLIC_CLEAR_IRQS: 2677 kvm_s390_clear_float_irqs(dev->kvm); 2678 break; 2679 case KVM_DEV_FLIC_APF_ENABLE: 2680 dev->kvm->arch.gmap->pfault_enabled = 1; 2681 break; 2682 case KVM_DEV_FLIC_APF_DISABLE_WAIT: 2683 dev->kvm->arch.gmap->pfault_enabled = 0; 2684 /* 2685 * Make sure no async faults are in transition when 2686 * clearing the queues. So we don't need to worry 2687 * about late coming workers. 2688 */ 2689 synchronize_srcu(&dev->kvm->srcu); 2690 kvm_for_each_vcpu(i, vcpu, dev->kvm) 2691 kvm_clear_async_pf_completion_queue(vcpu); 2692 break; 2693 case KVM_DEV_FLIC_ADAPTER_REGISTER: 2694 r = register_io_adapter(dev, attr); 2695 break; 2696 case KVM_DEV_FLIC_ADAPTER_MODIFY: 2697 r = modify_io_adapter(dev, attr); 2698 break; 2699 case KVM_DEV_FLIC_CLEAR_IO_IRQ: 2700 r = clear_io_irq(dev->kvm, attr); 2701 break; 2702 case KVM_DEV_FLIC_AISM: 2703 r = modify_ais_mode(dev->kvm, attr); 2704 break; 2705 case KVM_DEV_FLIC_AIRQ_INJECT: 2706 r = flic_inject_airq(dev->kvm, attr); 2707 break; 2708 case KVM_DEV_FLIC_AISM_ALL: 2709 r = flic_ais_mode_set_all(dev->kvm, attr); 2710 break; 2711 default: 2712 r = -EINVAL; 2713 } 2714 2715 return r; 2716 } 2717 2718 static int flic_has_attr(struct kvm_device *dev, 2719 struct kvm_device_attr *attr) 2720 { 2721 switch (attr->group) { 2722 case KVM_DEV_FLIC_GET_ALL_IRQS: 2723 case KVM_DEV_FLIC_ENQUEUE: 2724 case KVM_DEV_FLIC_CLEAR_IRQS: 2725 case KVM_DEV_FLIC_APF_ENABLE: 2726 case KVM_DEV_FLIC_APF_DISABLE_WAIT: 2727 case KVM_DEV_FLIC_ADAPTER_REGISTER: 2728 case KVM_DEV_FLIC_ADAPTER_MODIFY: 2729 case KVM_DEV_FLIC_CLEAR_IO_IRQ: 2730 case KVM_DEV_FLIC_AISM: 2731 case KVM_DEV_FLIC_AIRQ_INJECT: 2732 case KVM_DEV_FLIC_AISM_ALL: 2733 return 0; 2734 } 2735 return -ENXIO; 2736 } 2737 2738 static int flic_create(struct kvm_device *dev, u32 type) 2739 { 2740 if (!dev) 2741 return -EINVAL; 2742 if (dev->kvm->arch.flic) 2743 return -EINVAL; 2744 dev->kvm->arch.flic = dev; 2745 return 0; 2746 } 2747 2748 static void flic_destroy(struct kvm_device *dev) 2749 { 2750 dev->kvm->arch.flic = NULL; 2751 kfree(dev); 2752 } 2753 2754 /* s390 floating irq controller (flic) */ 2755 struct kvm_device_ops kvm_flic_ops = { 2756 .name = "kvm-flic", 2757 .get_attr = flic_get_attr, 2758 .set_attr = flic_set_attr, 2759 .has_attr = flic_has_attr, 2760 .create = flic_create, 2761 .destroy = flic_destroy, 2762 }; 2763 2764 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap) 2765 { 2766 unsigned long bit; 2767 2768 bit = bit_nr + (addr % PAGE_SIZE) * 8; 2769 2770 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit; 2771 } 2772 2773 static struct page *get_map_page(struct kvm *kvm, u64 uaddr) 2774 { 2775 struct page *page = NULL; 2776 2777 mmap_read_lock(kvm->mm); 2778 get_user_pages_remote(kvm->mm, uaddr, 1, FOLL_WRITE, 2779 &page, NULL, NULL); 2780 mmap_read_unlock(kvm->mm); 2781 return page; 2782 } 2783 2784 static int adapter_indicators_set(struct kvm *kvm, 2785 struct s390_io_adapter *adapter, 2786 struct kvm_s390_adapter_int *adapter_int) 2787 { 2788 unsigned long bit; 2789 int summary_set, idx; 2790 struct page *ind_page, *summary_page; 2791 void *map; 2792 2793 ind_page = get_map_page(kvm, adapter_int->ind_addr); 2794 if (!ind_page) 2795 return -1; 2796 summary_page = get_map_page(kvm, adapter_int->summary_addr); 2797 if (!summary_page) { 2798 put_page(ind_page); 2799 return -1; 2800 } 2801 2802 idx = srcu_read_lock(&kvm->srcu); 2803 map = page_address(ind_page); 2804 bit = get_ind_bit(adapter_int->ind_addr, 2805 adapter_int->ind_offset, adapter->swap); 2806 set_bit(bit, map); 2807 mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT); 2808 set_page_dirty_lock(ind_page); 2809 map = page_address(summary_page); 2810 bit = get_ind_bit(adapter_int->summary_addr, 2811 adapter_int->summary_offset, adapter->swap); 2812 summary_set = test_and_set_bit(bit, map); 2813 mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT); 2814 set_page_dirty_lock(summary_page); 2815 srcu_read_unlock(&kvm->srcu, idx); 2816 2817 put_page(ind_page); 2818 put_page(summary_page); 2819 return summary_set ? 0 : 1; 2820 } 2821 2822 /* 2823 * < 0 - not injected due to error 2824 * = 0 - coalesced, summary indicator already active 2825 * > 0 - injected interrupt 2826 */ 2827 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e, 2828 struct kvm *kvm, int irq_source_id, int level, 2829 bool line_status) 2830 { 2831 int ret; 2832 struct s390_io_adapter *adapter; 2833 2834 /* We're only interested in the 0->1 transition. */ 2835 if (!level) 2836 return 0; 2837 adapter = get_io_adapter(kvm, e->adapter.adapter_id); 2838 if (!adapter) 2839 return -1; 2840 ret = adapter_indicators_set(kvm, adapter, &e->adapter); 2841 if ((ret > 0) && !adapter->masked) { 2842 ret = kvm_s390_inject_airq(kvm, adapter); 2843 if (ret == 0) 2844 ret = 1; 2845 } 2846 return ret; 2847 } 2848 2849 /* 2850 * Inject the machine check to the guest. 2851 */ 2852 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu, 2853 struct mcck_volatile_info *mcck_info) 2854 { 2855 struct kvm_s390_interrupt_info inti; 2856 struct kvm_s390_irq irq; 2857 struct kvm_s390_mchk_info *mchk; 2858 union mci mci; 2859 __u64 cr14 = 0; /* upper bits are not used */ 2860 int rc; 2861 2862 mci.val = mcck_info->mcic; 2863 if (mci.sr) 2864 cr14 |= CR14_RECOVERY_SUBMASK; 2865 if (mci.dg) 2866 cr14 |= CR14_DEGRADATION_SUBMASK; 2867 if (mci.w) 2868 cr14 |= CR14_WARNING_SUBMASK; 2869 2870 mchk = mci.ck ? &inti.mchk : &irq.u.mchk; 2871 mchk->cr14 = cr14; 2872 mchk->mcic = mcck_info->mcic; 2873 mchk->ext_damage_code = mcck_info->ext_damage_code; 2874 mchk->failing_storage_address = mcck_info->failing_storage_address; 2875 if (mci.ck) { 2876 /* Inject the floating machine check */ 2877 inti.type = KVM_S390_MCHK; 2878 rc = __inject_vm(vcpu->kvm, &inti); 2879 } else { 2880 /* Inject the machine check to specified vcpu */ 2881 irq.type = KVM_S390_MCHK; 2882 rc = kvm_s390_inject_vcpu(vcpu, &irq); 2883 } 2884 WARN_ON_ONCE(rc); 2885 } 2886 2887 int kvm_set_routing_entry(struct kvm *kvm, 2888 struct kvm_kernel_irq_routing_entry *e, 2889 const struct kvm_irq_routing_entry *ue) 2890 { 2891 u64 uaddr; 2892 2893 switch (ue->type) { 2894 /* we store the userspace addresses instead of the guest addresses */ 2895 case KVM_IRQ_ROUTING_S390_ADAPTER: 2896 e->set = set_adapter_int; 2897 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.summary_addr); 2898 if (uaddr == -EFAULT) 2899 return -EFAULT; 2900 e->adapter.summary_addr = uaddr; 2901 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.ind_addr); 2902 if (uaddr == -EFAULT) 2903 return -EFAULT; 2904 e->adapter.ind_addr = uaddr; 2905 e->adapter.summary_offset = ue->u.adapter.summary_offset; 2906 e->adapter.ind_offset = ue->u.adapter.ind_offset; 2907 e->adapter.adapter_id = ue->u.adapter.adapter_id; 2908 return 0; 2909 default: 2910 return -EINVAL; 2911 } 2912 } 2913 2914 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, 2915 int irq_source_id, int level, bool line_status) 2916 { 2917 return -EINVAL; 2918 } 2919 2920 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len) 2921 { 2922 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2923 struct kvm_s390_irq *buf; 2924 int r = 0; 2925 int n; 2926 2927 buf = vmalloc(len); 2928 if (!buf) 2929 return -ENOMEM; 2930 2931 if (copy_from_user((void *) buf, irqstate, len)) { 2932 r = -EFAULT; 2933 goto out_free; 2934 } 2935 2936 /* 2937 * Don't allow setting the interrupt state 2938 * when there are already interrupts pending 2939 */ 2940 spin_lock(&li->lock); 2941 if (li->pending_irqs) { 2942 r = -EBUSY; 2943 goto out_unlock; 2944 } 2945 2946 for (n = 0; n < len / sizeof(*buf); n++) { 2947 r = do_inject_vcpu(vcpu, &buf[n]); 2948 if (r) 2949 break; 2950 } 2951 2952 out_unlock: 2953 spin_unlock(&li->lock); 2954 out_free: 2955 vfree(buf); 2956 2957 return r; 2958 } 2959 2960 static void store_local_irq(struct kvm_s390_local_interrupt *li, 2961 struct kvm_s390_irq *irq, 2962 unsigned long irq_type) 2963 { 2964 switch (irq_type) { 2965 case IRQ_PEND_MCHK_EX: 2966 case IRQ_PEND_MCHK_REP: 2967 irq->type = KVM_S390_MCHK; 2968 irq->u.mchk = li->irq.mchk; 2969 break; 2970 case IRQ_PEND_PROG: 2971 irq->type = KVM_S390_PROGRAM_INT; 2972 irq->u.pgm = li->irq.pgm; 2973 break; 2974 case IRQ_PEND_PFAULT_INIT: 2975 irq->type = KVM_S390_INT_PFAULT_INIT; 2976 irq->u.ext = li->irq.ext; 2977 break; 2978 case IRQ_PEND_EXT_EXTERNAL: 2979 irq->type = KVM_S390_INT_EXTERNAL_CALL; 2980 irq->u.extcall = li->irq.extcall; 2981 break; 2982 case IRQ_PEND_EXT_CLOCK_COMP: 2983 irq->type = KVM_S390_INT_CLOCK_COMP; 2984 break; 2985 case IRQ_PEND_EXT_CPU_TIMER: 2986 irq->type = KVM_S390_INT_CPU_TIMER; 2987 break; 2988 case IRQ_PEND_SIGP_STOP: 2989 irq->type = KVM_S390_SIGP_STOP; 2990 irq->u.stop = li->irq.stop; 2991 break; 2992 case IRQ_PEND_RESTART: 2993 irq->type = KVM_S390_RESTART; 2994 break; 2995 case IRQ_PEND_SET_PREFIX: 2996 irq->type = KVM_S390_SIGP_SET_PREFIX; 2997 irq->u.prefix = li->irq.prefix; 2998 break; 2999 } 3000 } 3001 3002 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len) 3003 { 3004 int scn; 3005 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS); 3006 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 3007 unsigned long pending_irqs; 3008 struct kvm_s390_irq irq; 3009 unsigned long irq_type; 3010 int cpuaddr; 3011 int n = 0; 3012 3013 spin_lock(&li->lock); 3014 pending_irqs = li->pending_irqs; 3015 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending, 3016 sizeof(sigp_emerg_pending)); 3017 spin_unlock(&li->lock); 3018 3019 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) { 3020 memset(&irq, 0, sizeof(irq)); 3021 if (irq_type == IRQ_PEND_EXT_EMERGENCY) 3022 continue; 3023 if (n + sizeof(irq) > len) 3024 return -ENOBUFS; 3025 store_local_irq(&vcpu->arch.local_int, &irq, irq_type); 3026 if (copy_to_user(&buf[n], &irq, sizeof(irq))) 3027 return -EFAULT; 3028 n += sizeof(irq); 3029 } 3030 3031 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) { 3032 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) { 3033 memset(&irq, 0, sizeof(irq)); 3034 if (n + sizeof(irq) > len) 3035 return -ENOBUFS; 3036 irq.type = KVM_S390_INT_EMERGENCY; 3037 irq.u.emerg.code = cpuaddr; 3038 if (copy_to_user(&buf[n], &irq, sizeof(irq))) 3039 return -EFAULT; 3040 n += sizeof(irq); 3041 } 3042 } 3043 3044 if (sca_ext_call_pending(vcpu, &scn)) { 3045 if (n + sizeof(irq) > len) 3046 return -ENOBUFS; 3047 memset(&irq, 0, sizeof(irq)); 3048 irq.type = KVM_S390_INT_EXTERNAL_CALL; 3049 irq.u.extcall.code = scn; 3050 if (copy_to_user(&buf[n], &irq, sizeof(irq))) 3051 return -EFAULT; 3052 n += sizeof(irq); 3053 } 3054 3055 return n; 3056 } 3057 3058 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask) 3059 { 3060 int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus); 3061 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3062 struct kvm_vcpu *vcpu; 3063 u8 vcpu_isc_mask; 3064 3065 for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) { 3066 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 3067 if (psw_ioint_disabled(vcpu)) 3068 continue; 3069 vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24); 3070 if (deliverable_mask & vcpu_isc_mask) { 3071 /* lately kicked but not yet running */ 3072 if (test_and_set_bit(vcpu_idx, gi->kicked_mask)) 3073 return; 3074 kvm_s390_vcpu_wakeup(vcpu); 3075 return; 3076 } 3077 } 3078 } 3079 3080 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer) 3081 { 3082 struct kvm_s390_gisa_interrupt *gi = 3083 container_of(timer, struct kvm_s390_gisa_interrupt, timer); 3084 struct kvm *kvm = 3085 container_of(gi->origin, struct sie_page2, gisa)->kvm; 3086 u8 pending_mask; 3087 3088 pending_mask = gisa_get_ipm_or_restore_iam(gi); 3089 if (pending_mask) { 3090 __airqs_kick_single_vcpu(kvm, pending_mask); 3091 hrtimer_forward_now(timer, ns_to_ktime(gi->expires)); 3092 return HRTIMER_RESTART; 3093 } 3094 3095 return HRTIMER_NORESTART; 3096 } 3097 3098 #define NULL_GISA_ADDR 0x00000000UL 3099 #define NONE_GISA_ADDR 0x00000001UL 3100 #define GISA_ADDR_MASK 0xfffff000UL 3101 3102 static void process_gib_alert_list(void) 3103 { 3104 struct kvm_s390_gisa_interrupt *gi; 3105 struct kvm_s390_gisa *gisa; 3106 struct kvm *kvm; 3107 u32 final, origin = 0UL; 3108 3109 do { 3110 /* 3111 * If the NONE_GISA_ADDR is still stored in the alert list 3112 * origin, we will leave the outer loop. No further GISA has 3113 * been added to the alert list by millicode while processing 3114 * the current alert list. 3115 */ 3116 final = (origin & NONE_GISA_ADDR); 3117 /* 3118 * Cut off the alert list and store the NONE_GISA_ADDR in the 3119 * alert list origin to avoid further GAL interruptions. 3120 * A new alert list can be build up by millicode in parallel 3121 * for guests not in the yet cut-off alert list. When in the 3122 * final loop, store the NULL_GISA_ADDR instead. This will re- 3123 * enable GAL interruptions on the host again. 3124 */ 3125 origin = xchg(&gib->alert_list_origin, 3126 (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR); 3127 /* 3128 * Loop through the just cut-off alert list and start the 3129 * gisa timers to kick idle vcpus to consume the pending 3130 * interruptions asap. 3131 */ 3132 while (origin & GISA_ADDR_MASK) { 3133 gisa = (struct kvm_s390_gisa *)(u64)origin; 3134 origin = gisa->next_alert; 3135 gisa->next_alert = (u32)(u64)gisa; 3136 kvm = container_of(gisa, struct sie_page2, gisa)->kvm; 3137 gi = &kvm->arch.gisa_int; 3138 if (hrtimer_active(&gi->timer)) 3139 hrtimer_cancel(&gi->timer); 3140 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL); 3141 } 3142 } while (!final); 3143 3144 } 3145 3146 void kvm_s390_gisa_clear(struct kvm *kvm) 3147 { 3148 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3149 3150 if (!gi->origin) 3151 return; 3152 gisa_clear_ipm(gi->origin); 3153 VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin); 3154 } 3155 3156 void kvm_s390_gisa_init(struct kvm *kvm) 3157 { 3158 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3159 3160 if (!css_general_characteristics.aiv) 3161 return; 3162 gi->origin = &kvm->arch.sie_page2->gisa; 3163 gi->alert.mask = 0; 3164 spin_lock_init(&gi->alert.ref_lock); 3165 gi->expires = 50 * 1000; /* 50 usec */ 3166 hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 3167 gi->timer.function = gisa_vcpu_kicker; 3168 memset(gi->origin, 0, sizeof(struct kvm_s390_gisa)); 3169 gi->origin->next_alert = (u32)(u64)gi->origin; 3170 VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin); 3171 } 3172 3173 void kvm_s390_gisa_enable(struct kvm *kvm) 3174 { 3175 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3176 struct kvm_vcpu *vcpu; 3177 unsigned long i; 3178 u32 gisa_desc; 3179 3180 if (gi->origin) 3181 return; 3182 kvm_s390_gisa_init(kvm); 3183 gisa_desc = kvm_s390_get_gisa_desc(kvm); 3184 if (!gisa_desc) 3185 return; 3186 kvm_for_each_vcpu(i, vcpu, kvm) { 3187 mutex_lock(&vcpu->mutex); 3188 vcpu->arch.sie_block->gd = gisa_desc; 3189 vcpu->arch.sie_block->eca |= ECA_AIV; 3190 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u", 3191 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id); 3192 mutex_unlock(&vcpu->mutex); 3193 } 3194 } 3195 3196 void kvm_s390_gisa_destroy(struct kvm *kvm) 3197 { 3198 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3199 struct kvm_s390_gisa *gisa = gi->origin; 3200 3201 if (!gi->origin) 3202 return; 3203 if (gi->alert.mask) 3204 KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x", 3205 kvm, gi->alert.mask); 3206 while (gisa_in_alert_list(gi->origin)) 3207 cpu_relax(); 3208 hrtimer_cancel(&gi->timer); 3209 gi->origin = NULL; 3210 VM_EVENT(kvm, 3, "gisa 0x%pK destroyed", gisa); 3211 } 3212 3213 void kvm_s390_gisa_disable(struct kvm *kvm) 3214 { 3215 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3216 struct kvm_vcpu *vcpu; 3217 unsigned long i; 3218 3219 if (!gi->origin) 3220 return; 3221 kvm_for_each_vcpu(i, vcpu, kvm) { 3222 mutex_lock(&vcpu->mutex); 3223 vcpu->arch.sie_block->eca &= ~ECA_AIV; 3224 vcpu->arch.sie_block->gd = 0U; 3225 mutex_unlock(&vcpu->mutex); 3226 VCPU_EVENT(vcpu, 3, "AIV disabled for cpu %03u", vcpu->vcpu_id); 3227 } 3228 kvm_s390_gisa_destroy(kvm); 3229 } 3230 3231 /** 3232 * kvm_s390_gisc_register - register a guest ISC 3233 * 3234 * @kvm: the kernel vm to work with 3235 * @gisc: the guest interruption sub class to register 3236 * 3237 * The function extends the vm specific alert mask to use. 3238 * The effective IAM mask in the GISA is updated as well 3239 * in case the GISA is not part of the GIB alert list. 3240 * It will be updated latest when the IAM gets restored 3241 * by gisa_get_ipm_or_restore_iam(). 3242 * 3243 * Returns: the nonspecific ISC (NISC) the gib alert mechanism 3244 * has registered with the channel subsystem. 3245 * -ENODEV in case the vm uses no GISA 3246 * -ERANGE in case the guest ISC is invalid 3247 */ 3248 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc) 3249 { 3250 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3251 3252 if (!gi->origin) 3253 return -ENODEV; 3254 if (gisc > MAX_ISC) 3255 return -ERANGE; 3256 3257 spin_lock(&gi->alert.ref_lock); 3258 gi->alert.ref_count[gisc]++; 3259 if (gi->alert.ref_count[gisc] == 1) { 3260 gi->alert.mask |= 0x80 >> gisc; 3261 gisa_set_iam(gi->origin, gi->alert.mask); 3262 } 3263 spin_unlock(&gi->alert.ref_lock); 3264 3265 return gib->nisc; 3266 } 3267 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register); 3268 3269 /** 3270 * kvm_s390_gisc_unregister - unregister a guest ISC 3271 * 3272 * @kvm: the kernel vm to work with 3273 * @gisc: the guest interruption sub class to register 3274 * 3275 * The function reduces the vm specific alert mask to use. 3276 * The effective IAM mask in the GISA is updated as well 3277 * in case the GISA is not part of the GIB alert list. 3278 * It will be updated latest when the IAM gets restored 3279 * by gisa_get_ipm_or_restore_iam(). 3280 * 3281 * Returns: the nonspecific ISC (NISC) the gib alert mechanism 3282 * has registered with the channel subsystem. 3283 * -ENODEV in case the vm uses no GISA 3284 * -ERANGE in case the guest ISC is invalid 3285 * -EINVAL in case the guest ISC is not registered 3286 */ 3287 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc) 3288 { 3289 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3290 int rc = 0; 3291 3292 if (!gi->origin) 3293 return -ENODEV; 3294 if (gisc > MAX_ISC) 3295 return -ERANGE; 3296 3297 spin_lock(&gi->alert.ref_lock); 3298 if (gi->alert.ref_count[gisc] == 0) { 3299 rc = -EINVAL; 3300 goto out; 3301 } 3302 gi->alert.ref_count[gisc]--; 3303 if (gi->alert.ref_count[gisc] == 0) { 3304 gi->alert.mask &= ~(0x80 >> gisc); 3305 gisa_set_iam(gi->origin, gi->alert.mask); 3306 } 3307 out: 3308 spin_unlock(&gi->alert.ref_lock); 3309 3310 return rc; 3311 } 3312 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister); 3313 3314 static void gib_alert_irq_handler(struct airq_struct *airq, bool floating) 3315 { 3316 inc_irq_stat(IRQIO_GAL); 3317 process_gib_alert_list(); 3318 } 3319 3320 static struct airq_struct gib_alert_irq = { 3321 .handler = gib_alert_irq_handler, 3322 .lsi_ptr = &gib_alert_irq.lsi_mask, 3323 }; 3324 3325 void kvm_s390_gib_destroy(void) 3326 { 3327 if (!gib) 3328 return; 3329 chsc_sgib(0); 3330 unregister_adapter_interrupt(&gib_alert_irq); 3331 free_page((unsigned long)gib); 3332 gib = NULL; 3333 } 3334 3335 int kvm_s390_gib_init(u8 nisc) 3336 { 3337 int rc = 0; 3338 3339 if (!css_general_characteristics.aiv) { 3340 KVM_EVENT(3, "%s", "gib not initialized, no AIV facility"); 3341 goto out; 3342 } 3343 3344 gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA); 3345 if (!gib) { 3346 rc = -ENOMEM; 3347 goto out; 3348 } 3349 3350 gib_alert_irq.isc = nisc; 3351 if (register_adapter_interrupt(&gib_alert_irq)) { 3352 pr_err("Registering the GIB alert interruption handler failed\n"); 3353 rc = -EIO; 3354 goto out_free_gib; 3355 } 3356 3357 gib->nisc = nisc; 3358 if (chsc_sgib((u32)(u64)gib)) { 3359 pr_err("Associating the GIB with the AIV facility failed\n"); 3360 free_page((unsigned long)gib); 3361 gib = NULL; 3362 rc = -EIO; 3363 goto out_unreg_gal; 3364 } 3365 3366 KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc); 3367 goto out; 3368 3369 out_unreg_gal: 3370 unregister_adapter_interrupt(&gib_alert_irq); 3371 out_free_gib: 3372 free_page((unsigned long)gib); 3373 gib = NULL; 3374 out: 3375 return rc; 3376 } 3377