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 pr_fmt(fmt) "kvm-s390: " fmt 11 12 #include <linux/cpufeature.h> 13 #include <linux/interrupt.h> 14 #include <linux/kvm_host.h> 15 #include <linux/hrtimer.h> 16 #include <linux/export.h> 17 #include <linux/mmu_context.h> 18 #include <linux/nospec.h> 19 #include <linux/signal.h> 20 #include <linux/slab.h> 21 #include <linux/bitmap.h> 22 #include <linux/vmalloc.h> 23 #include <asm/access-regs.h> 24 #include <asm/asm-offsets.h> 25 #include <asm/dis.h> 26 #include <linux/uaccess.h> 27 #include <asm/sclp.h> 28 #include <asm/isc.h> 29 #include <asm/nmi.h> 30 #include <asm/airq.h> 31 #include <asm/tpi.h> 32 #include "s390.h" 33 #include "gaccess.h" 34 #include "trace-s390.h" 35 #include "pci.h" 36 #include "gmap.h" 37 38 #define PFAULT_INIT 0x0600 39 #define PFAULT_DONE 0x0680 40 #define VIRTIO_PARAM 0x0d00 41 42 static struct kvm_s390_gib *gib; 43 44 /* handle external calls via sigp interpretation facility */ 45 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id) 46 { 47 struct esca_block *sca = vcpu->kvm->arch.sca; 48 union esca_sigp_ctrl sigp_ctrl; 49 50 if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND)) 51 return 0; 52 if (kvm_is_ucontrol(vcpu->kvm)) 53 return 0; 54 55 BUG_ON(!kvm_s390_use_sca_entries()); 56 57 sigp_ctrl = sca->cpu[vcpu->vcpu_id].sigp_ctrl; 58 if (src_id) 59 *src_id = sigp_ctrl.scn; 60 61 return sigp_ctrl.c; 62 } 63 64 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id) 65 { 66 union esca_sigp_ctrl old_val, new_val = {.scn = src_id, .c = 1}; 67 struct esca_block *sca = vcpu->kvm->arch.sca; 68 union esca_sigp_ctrl *sigp_ctrl; 69 int expect, rc; 70 71 BUG_ON(!kvm_s390_use_sca_entries()); 72 if (kvm_is_ucontrol(vcpu->kvm)) 73 return -EINVAL; 74 75 sigp_ctrl = &sca->cpu[vcpu->vcpu_id].sigp_ctrl; 76 old_val = READ_ONCE(*sigp_ctrl); 77 old_val.c = 0; 78 79 expect = old_val.value; 80 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value); 81 82 if (rc != expect) { 83 /* another external call is pending */ 84 return -EBUSY; 85 } 86 kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND); 87 return 0; 88 } 89 90 static void sca_clear_ext_call(struct kvm_vcpu *vcpu) 91 { 92 struct esca_block *sca = vcpu->kvm->arch.sca; 93 union esca_sigp_ctrl *sigp_ctrl; 94 95 if (!kvm_s390_use_sca_entries() || !vcpu->arch.initialized || kvm_is_ucontrol(vcpu->kvm)) 96 return; 97 98 /* Initialize after the above check, to prevent going out of bounds */ 99 sigp_ctrl = &sca->cpu[vcpu->vcpu_id].sigp_ctrl; 100 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND); 101 102 WRITE_ONCE(sigp_ctrl->value, 0); 103 } 104 105 int psw_extint_disabled(struct kvm_vcpu *vcpu) 106 { 107 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT); 108 } 109 110 static int psw_ioint_disabled(struct kvm_vcpu *vcpu) 111 { 112 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO); 113 } 114 115 static int psw_mchk_disabled(struct kvm_vcpu *vcpu) 116 { 117 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK); 118 } 119 120 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu) 121 { 122 return psw_extint_disabled(vcpu) && 123 psw_ioint_disabled(vcpu) && 124 psw_mchk_disabled(vcpu); 125 } 126 127 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu) 128 { 129 if (psw_extint_disabled(vcpu) || 130 !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK)) 131 return 0; 132 if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu)) 133 /* No timer interrupts when single stepping */ 134 return 0; 135 return 1; 136 } 137 138 static int ckc_irq_pending(struct kvm_vcpu *vcpu) 139 { 140 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm); 141 const u64 ckc = vcpu->arch.sie_block->ckc; 142 143 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) { 144 if ((s64)ckc >= (s64)now) 145 return 0; 146 } else if (ckc >= now) { 147 return 0; 148 } 149 return ckc_interrupts_enabled(vcpu); 150 } 151 152 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu) 153 { 154 return !psw_extint_disabled(vcpu) && 155 (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK); 156 } 157 158 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu) 159 { 160 if (!cpu_timer_interrupts_enabled(vcpu)) 161 return 0; 162 return kvm_s390_get_cpu_timer(vcpu) >> 63; 163 } 164 165 static uint64_t isc_to_isc_bits(int isc) 166 { 167 return (0x80 >> isc) << 24; 168 } 169 170 static inline u32 isc_to_int_word(u8 isc) 171 { 172 return ((u32)isc << 27) | 0x80000000; 173 } 174 175 static inline u8 int_word_to_isc(u32 int_word) 176 { 177 return (int_word & 0x38000000) >> 27; 178 } 179 180 /* 181 * To use atomic bitmap functions, we have to provide a bitmap address 182 * that is u64 aligned. However, the ipm might be u32 aligned. 183 * Therefore, we logically start the bitmap at the very beginning of the 184 * struct and fixup the bit number. 185 */ 186 #define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE) 187 188 /** 189 * gisa_set_iam - change the GISA interruption alert mask 190 * 191 * @gisa: gisa to operate on 192 * @iam: new IAM value to use 193 * 194 * Change the IAM atomically with the next alert address and the IPM 195 * of the GISA if the GISA is not part of the GIB alert list. All three 196 * fields are located in the first long word of the GISA. 197 * 198 * Returns: 0 on success 199 * -EBUSY in case the gisa is part of the alert list 200 */ 201 static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam) 202 { 203 u64 word, _word; 204 205 word = READ_ONCE(gisa->u64.word[0]); 206 do { 207 if ((u64)gisa != word >> 32) 208 return -EBUSY; 209 _word = (word & ~0xffUL) | iam; 210 } while (!try_cmpxchg(&gisa->u64.word[0], &word, _word)); 211 212 return 0; 213 } 214 215 /** 216 * gisa_clear_ipm - clear the GISA interruption pending mask 217 * 218 * @gisa: gisa to operate on 219 * 220 * Clear the IPM atomically with the next alert address and the IAM 221 * of the GISA unconditionally. All three fields are located in the 222 * first long word of the GISA. 223 */ 224 static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa) 225 { 226 u64 word, _word; 227 228 word = READ_ONCE(gisa->u64.word[0]); 229 do { 230 _word = word & ~(0xffUL << 24); 231 } while (!try_cmpxchg(&gisa->u64.word[0], &word, _word)); 232 } 233 234 /** 235 * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM 236 * 237 * @gi: gisa interrupt struct to work on 238 * 239 * Atomically restores the interruption alert mask if none of the 240 * relevant ISCs are pending and return the IPM. 241 * 242 * Returns: the relevant pending ISCs 243 */ 244 static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi) 245 { 246 u8 pending_mask, alert_mask; 247 u64 word, _word; 248 249 word = READ_ONCE(gi->origin->u64.word[0]); 250 do { 251 alert_mask = READ_ONCE(gi->alert.mask); 252 pending_mask = (u8)(word >> 24) & alert_mask; 253 if (pending_mask) 254 return pending_mask; 255 _word = (word & ~0xffUL) | alert_mask; 256 } while (!try_cmpxchg(&gi->origin->u64.word[0], &word, _word)); 257 258 return 0; 259 } 260 261 static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc) 262 { 263 set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa); 264 } 265 266 static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa) 267 { 268 return READ_ONCE(gisa->ipm); 269 } 270 271 static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc) 272 { 273 return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa); 274 } 275 276 static inline int gisa_test_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc) 277 { 278 return test_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *)gisa); 279 } 280 281 static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu) 282 { 283 unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs | 284 vcpu->arch.local_int.pending_irqs; 285 286 pending &= ~vcpu->kvm->arch.float_int.masked_irqs; 287 return pending; 288 } 289 290 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu) 291 { 292 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int; 293 unsigned long pending_mask; 294 295 pending_mask = pending_irqs_no_gisa(vcpu); 296 if (gi->origin) 297 pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7; 298 return pending_mask; 299 } 300 301 static inline int isc_to_irq_type(unsigned long isc) 302 { 303 return IRQ_PEND_IO_ISC_0 - isc; 304 } 305 306 static inline int irq_type_to_isc(unsigned long irq_type) 307 { 308 return IRQ_PEND_IO_ISC_0 - irq_type; 309 } 310 311 static unsigned long disable_iscs(struct kvm_vcpu *vcpu, 312 unsigned long active_mask) 313 { 314 int i; 315 316 for (i = 0; i <= MAX_ISC; i++) 317 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i))) 318 active_mask &= ~(1UL << (isc_to_irq_type(i))); 319 320 return active_mask; 321 } 322 323 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu) 324 { 325 unsigned long active_mask; 326 327 active_mask = pending_irqs(vcpu); 328 if (!active_mask) 329 return 0; 330 331 if (psw_extint_disabled(vcpu)) 332 active_mask &= ~IRQ_PEND_EXT_MASK; 333 if (psw_ioint_disabled(vcpu)) 334 active_mask &= ~IRQ_PEND_IO_MASK; 335 else 336 active_mask = disable_iscs(vcpu, active_mask); 337 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK)) 338 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask); 339 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK)) 340 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask); 341 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK)) 342 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask); 343 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK)) 344 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask); 345 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) { 346 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask); 347 __clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask); 348 } 349 if (psw_mchk_disabled(vcpu)) 350 active_mask &= ~IRQ_PEND_MCHK_MASK; 351 /* PV guest cpus can have a single interruption injected at a time. */ 352 if (kvm_s390_pv_cpu_get_handle(vcpu) && 353 vcpu->arch.sie_block->iictl != IICTL_CODE_NONE) 354 active_mask &= ~(IRQ_PEND_EXT_II_MASK | 355 IRQ_PEND_IO_MASK | 356 IRQ_PEND_MCHK_MASK); 357 /* 358 * Check both floating and local interrupt's cr14 because 359 * bit IRQ_PEND_MCHK_REP could be set in both cases. 360 */ 361 if (!(vcpu->arch.sie_block->gcr[14] & 362 (vcpu->kvm->arch.float_int.mchk.cr14 | 363 vcpu->arch.local_int.irq.mchk.cr14))) 364 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask); 365 366 /* 367 * STOP irqs will never be actively delivered. They are triggered via 368 * intercept requests and cleared when the stop intercept is performed. 369 */ 370 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask); 371 372 return active_mask; 373 } 374 375 static void __set_cpu_idle(struct kvm_vcpu *vcpu) 376 { 377 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT); 378 set_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask); 379 } 380 381 static void __unset_cpu_idle(struct kvm_vcpu *vcpu) 382 { 383 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT); 384 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask); 385 } 386 387 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu) 388 { 389 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT | 390 CPUSTAT_STOP_INT); 391 vcpu->arch.sie_block->lctl = 0x0000; 392 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT); 393 394 if (guestdbg_enabled(vcpu)) { 395 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 | 396 LCTL_CR10 | LCTL_CR11); 397 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT); 398 } 399 } 400 401 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu) 402 { 403 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK)) 404 return; 405 if (psw_ioint_disabled(vcpu)) 406 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT); 407 else 408 vcpu->arch.sie_block->lctl |= LCTL_CR6; 409 } 410 411 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu) 412 { 413 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK)) 414 return; 415 if (psw_extint_disabled(vcpu)) 416 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 417 else 418 vcpu->arch.sie_block->lctl |= LCTL_CR0; 419 } 420 421 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu) 422 { 423 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK)) 424 return; 425 if (psw_mchk_disabled(vcpu)) 426 vcpu->arch.sie_block->ictl |= ICTL_LPSW; 427 else 428 vcpu->arch.sie_block->lctl |= LCTL_CR14; 429 } 430 431 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu) 432 { 433 if (kvm_s390_is_stop_irq_pending(vcpu)) 434 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT); 435 } 436 437 /* Set interception request for non-deliverable interrupts */ 438 static void set_intercept_indicators(struct kvm_vcpu *vcpu) 439 { 440 set_intercept_indicators_io(vcpu); 441 set_intercept_indicators_ext(vcpu); 442 set_intercept_indicators_mchk(vcpu); 443 set_intercept_indicators_stop(vcpu); 444 } 445 446 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu) 447 { 448 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 449 int rc = 0; 450 451 vcpu->stat.deliver_cputm++; 452 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER, 453 0, 0); 454 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 455 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT; 456 vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER; 457 } else { 458 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER, 459 (u16 *)__LC_EXT_INT_CODE); 460 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR); 461 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 462 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 463 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 464 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 465 } 466 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 467 return rc ? -EFAULT : 0; 468 } 469 470 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu) 471 { 472 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 473 int rc = 0; 474 475 vcpu->stat.deliver_ckc++; 476 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP, 477 0, 0); 478 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 479 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT; 480 vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP; 481 } else { 482 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP, 483 (u16 __user *)__LC_EXT_INT_CODE); 484 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR); 485 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 486 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 487 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 488 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 489 } 490 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 491 return rc ? -EFAULT : 0; 492 } 493 494 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu) 495 { 496 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 497 struct kvm_s390_ext_info ext; 498 int rc; 499 500 spin_lock(&li->lock); 501 ext = li->irq.ext; 502 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs); 503 li->irq.ext.ext_params2 = 0; 504 spin_unlock(&li->lock); 505 506 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx", 507 ext.ext_params2); 508 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 509 KVM_S390_INT_PFAULT_INIT, 510 0, ext.ext_params2); 511 512 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE); 513 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR); 514 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 515 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 516 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 517 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 518 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2); 519 return rc ? -EFAULT : 0; 520 } 521 522 static int __write_machine_check(struct kvm_vcpu *vcpu, 523 struct kvm_s390_mchk_info *mchk) 524 { 525 unsigned long ext_sa_addr; 526 unsigned long lc; 527 freg_t fprs[NUM_FPRS]; 528 union mci mci; 529 int rc; 530 531 /* 532 * All other possible payload for a machine check (e.g. the register 533 * contents in the save area) will be handled by the ultravisor, as 534 * the hypervisor does not not have the needed information for 535 * protected guests. 536 */ 537 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 538 vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK; 539 vcpu->arch.sie_block->mcic = mchk->mcic; 540 vcpu->arch.sie_block->faddr = mchk->failing_storage_address; 541 vcpu->arch.sie_block->edc = mchk->ext_damage_code; 542 return 0; 543 } 544 545 mci.val = mchk->mcic; 546 /* take care of lazy register loading */ 547 kvm_s390_fpu_store(vcpu->run); 548 save_access_regs(vcpu->run->s.regs.acrs); 549 if (cpu_has_gs() && vcpu->arch.gs_enabled) 550 save_gs_cb(current->thread.gs_cb); 551 552 /* Extended save area */ 553 rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr, 554 sizeof(unsigned long)); 555 /* Only bits 0 through 63-LC are used for address formation */ 556 lc = ext_sa_addr & MCESA_LC_MASK; 557 if (test_kvm_facility(vcpu->kvm, 133)) { 558 switch (lc) { 559 case 0: 560 case 10: 561 ext_sa_addr &= ~0x3ffUL; 562 break; 563 case 11: 564 ext_sa_addr &= ~0x7ffUL; 565 break; 566 case 12: 567 ext_sa_addr &= ~0xfffUL; 568 break; 569 default: 570 ext_sa_addr = 0; 571 break; 572 } 573 } else { 574 ext_sa_addr &= ~0x3ffUL; 575 } 576 577 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) { 578 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs, 579 512)) 580 mci.vr = 0; 581 } else { 582 mci.vr = 0; 583 } 584 if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133) 585 && (lc == 11 || lc == 12)) { 586 if (write_guest_abs(vcpu, ext_sa_addr + 1024, 587 &vcpu->run->s.regs.gscb, 32)) 588 mci.gs = 0; 589 } else { 590 mci.gs = 0; 591 } 592 593 /* General interruption information */ 594 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID); 595 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW, 596 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 597 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW, 598 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 599 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE); 600 601 /* Register-save areas */ 602 if (cpu_has_vx()) { 603 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs); 604 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128); 605 } else { 606 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, 607 vcpu->run->s.regs.fprs, 128); 608 } 609 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA, 610 vcpu->run->s.regs.gprs, 128); 611 rc |= put_guest_lc(vcpu, vcpu->run->s.regs.fpc, 612 (u32 __user *) __LC_FP_CREG_SAVE_AREA); 613 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr, 614 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA); 615 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu), 616 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA); 617 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8, 618 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA); 619 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA, 620 &vcpu->run->s.regs.acrs, 64); 621 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA, 622 &vcpu->arch.sie_block->gcr, 128); 623 624 /* Extended interruption information */ 625 rc |= put_guest_lc(vcpu, mchk->ext_damage_code, 626 (u32 __user *) __LC_EXT_DAMAGE_CODE); 627 rc |= put_guest_lc(vcpu, mchk->failing_storage_address, 628 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR); 629 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout, 630 sizeof(mchk->fixed_logout)); 631 return rc ? -EFAULT : 0; 632 } 633 634 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu) 635 { 636 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; 637 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 638 struct kvm_s390_mchk_info mchk = {}; 639 int deliver = 0; 640 int rc = 0; 641 unsigned long flags; 642 643 spin_lock_irqsave(&fi->lock, flags); 644 spin_lock(&li->lock); 645 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) || 646 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) { 647 /* 648 * If there was an exigent machine check pending, then any 649 * repressible machine checks that might have been pending 650 * are indicated along with it, so always clear bits for 651 * repressible and exigent interrupts 652 */ 653 mchk = li->irq.mchk; 654 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs); 655 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs); 656 memset(&li->irq.mchk, 0, sizeof(mchk)); 657 deliver = 1; 658 } 659 /* 660 * We indicate floating repressible conditions along with 661 * other pending conditions. Channel Report Pending and Channel 662 * Subsystem damage are the only two and are indicated by 663 * bits in mcic and masked in cr14. 664 */ 665 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) { 666 mchk.mcic |= fi->mchk.mcic; 667 mchk.cr14 |= fi->mchk.cr14; 668 memset(&fi->mchk, 0, sizeof(mchk)); 669 deliver = 1; 670 } 671 spin_unlock(&li->lock); 672 spin_unlock_irqrestore(&fi->lock, flags); 673 674 if (deliver) { 675 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx", 676 mchk.mcic); 677 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 678 KVM_S390_MCHK, 679 mchk.cr14, mchk.mcic); 680 vcpu->stat.deliver_machine_check++; 681 rc = __write_machine_check(vcpu, &mchk); 682 } 683 return rc; 684 } 685 686 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu) 687 { 688 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 689 int rc = 0; 690 691 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart"); 692 vcpu->stat.deliver_restart_signal++; 693 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0); 694 695 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 696 vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART; 697 } else { 698 rc = write_guest_lc(vcpu, 699 offsetof(struct lowcore, restart_old_psw), 700 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 701 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw), 702 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 703 } 704 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs); 705 return rc ? -EFAULT : 0; 706 } 707 708 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu) 709 { 710 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 711 struct kvm_s390_prefix_info prefix; 712 713 spin_lock(&li->lock); 714 prefix = li->irq.prefix; 715 li->irq.prefix.address = 0; 716 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs); 717 spin_unlock(&li->lock); 718 719 vcpu->stat.deliver_prefix_signal++; 720 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 721 KVM_S390_SIGP_SET_PREFIX, 722 prefix.address, 0); 723 724 kvm_s390_set_prefix(vcpu, prefix.address); 725 return 0; 726 } 727 728 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu) 729 { 730 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 731 int rc; 732 int cpu_addr; 733 734 spin_lock(&li->lock); 735 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS); 736 clear_bit(cpu_addr, li->sigp_emerg_pending); 737 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS)) 738 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs); 739 spin_unlock(&li->lock); 740 741 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg"); 742 vcpu->stat.deliver_emergency_signal++; 743 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY, 744 cpu_addr, 0); 745 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 746 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT; 747 vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG; 748 vcpu->arch.sie_block->extcpuaddr = cpu_addr; 749 return 0; 750 } 751 752 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG, 753 (u16 *)__LC_EXT_INT_CODE); 754 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR); 755 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 756 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 757 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 758 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 759 return rc ? -EFAULT : 0; 760 } 761 762 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu) 763 { 764 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 765 struct kvm_s390_extcall_info extcall; 766 int rc; 767 768 spin_lock(&li->lock); 769 extcall = li->irq.extcall; 770 li->irq.extcall.code = 0; 771 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs); 772 spin_unlock(&li->lock); 773 774 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call"); 775 vcpu->stat.deliver_external_call++; 776 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 777 KVM_S390_INT_EXTERNAL_CALL, 778 extcall.code, 0); 779 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 780 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT; 781 vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL; 782 vcpu->arch.sie_block->extcpuaddr = extcall.code; 783 return 0; 784 } 785 786 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL, 787 (u16 *)__LC_EXT_INT_CODE); 788 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR); 789 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 790 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 791 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw, 792 sizeof(psw_t)); 793 return rc ? -EFAULT : 0; 794 } 795 796 static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code) 797 { 798 switch (code) { 799 case PGM_SPECIFICATION: 800 vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION; 801 break; 802 case PGM_OPERAND: 803 vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND; 804 break; 805 default: 806 return -EINVAL; 807 } 808 return 0; 809 } 810 811 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu) 812 { 813 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 814 struct kvm_s390_pgm_info pgm_info; 815 int rc = 0, nullifying = false; 816 u16 ilen; 817 818 spin_lock(&li->lock); 819 pgm_info = li->irq.pgm; 820 clear_bit(IRQ_PEND_PROG, &li->pending_irqs); 821 memset(&li->irq.pgm, 0, sizeof(pgm_info)); 822 spin_unlock(&li->lock); 823 824 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK; 825 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d", 826 pgm_info.code, ilen); 827 vcpu->stat.deliver_program++; 828 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT, 829 pgm_info.code, 0); 830 831 /* PER is handled by the ultravisor */ 832 if (kvm_s390_pv_cpu_is_protected(vcpu)) 833 return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER); 834 835 switch (pgm_info.code & ~PGM_PER) { 836 case PGM_AFX_TRANSLATION: 837 case PGM_ASX_TRANSLATION: 838 case PGM_EX_TRANSLATION: 839 case PGM_LFX_TRANSLATION: 840 case PGM_LSTE_SEQUENCE: 841 case PGM_LSX_TRANSLATION: 842 case PGM_LX_TRANSLATION: 843 case PGM_PRIMARY_AUTHORITY: 844 case PGM_SECONDARY_AUTHORITY: 845 nullifying = true; 846 fallthrough; 847 case PGM_SPACE_SWITCH: 848 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code, 849 (u64 *)__LC_TRANS_EXC_CODE); 850 break; 851 case PGM_ALEN_TRANSLATION: 852 case PGM_ALE_SEQUENCE: 853 case PGM_ASTE_INSTANCE: 854 case PGM_ASTE_SEQUENCE: 855 case PGM_ASTE_VALIDITY: 856 case PGM_EXTENDED_AUTHORITY: 857 rc = put_guest_lc(vcpu, pgm_info.exc_access_id, 858 (u8 *)__LC_EXC_ACCESS_ID); 859 nullifying = true; 860 break; 861 case PGM_ASCE_TYPE: 862 case PGM_PAGE_TRANSLATION: 863 case PGM_REGION_FIRST_TRANS: 864 case PGM_REGION_SECOND_TRANS: 865 case PGM_REGION_THIRD_TRANS: 866 case PGM_SEGMENT_TRANSLATION: 867 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code, 868 (u64 *)__LC_TRANS_EXC_CODE); 869 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id, 870 (u8 *)__LC_EXC_ACCESS_ID); 871 rc |= put_guest_lc(vcpu, pgm_info.op_access_id, 872 (u8 *)__LC_OP_ACCESS_ID); 873 nullifying = true; 874 break; 875 case PGM_MONITOR: 876 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr, 877 (u16 *)__LC_MON_CLASS_NR); 878 rc |= put_guest_lc(vcpu, pgm_info.mon_code, 879 (u64 *)__LC_MON_CODE); 880 break; 881 case PGM_VECTOR_PROCESSING: 882 case PGM_DATA: 883 rc = put_guest_lc(vcpu, pgm_info.data_exc_code, 884 (u32 *)__LC_DATA_EXC_CODE); 885 break; 886 case PGM_PROTECTION: 887 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code, 888 (u64 *)__LC_TRANS_EXC_CODE); 889 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id, 890 (u8 *)__LC_EXC_ACCESS_ID); 891 break; 892 case PGM_STACK_FULL: 893 case PGM_STACK_EMPTY: 894 case PGM_STACK_SPECIFICATION: 895 case PGM_STACK_TYPE: 896 case PGM_STACK_OPERATION: 897 case PGM_TRACE_TABEL: 898 case PGM_CRYPTO_OPERATION: 899 nullifying = true; 900 break; 901 } 902 903 if (pgm_info.code & PGM_PER) { 904 rc |= put_guest_lc(vcpu, pgm_info.per_code, 905 (u8 *) __LC_PER_CODE); 906 rc |= put_guest_lc(vcpu, pgm_info.per_atmid, 907 (u8 *)__LC_PER_ATMID); 908 rc |= put_guest_lc(vcpu, pgm_info.per_address, 909 (u64 *) __LC_PER_ADDRESS); 910 rc |= put_guest_lc(vcpu, pgm_info.per_access_id, 911 (u8 *) __LC_PER_ACCESS_ID); 912 } 913 914 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND)) 915 kvm_s390_rewind_psw(vcpu, ilen); 916 917 /* bit 1+2 of the target are the ilc, so we can directly use ilen */ 918 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC); 919 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea, 920 (u64 *) __LC_PGM_LAST_BREAK); 921 rc |= put_guest_lc(vcpu, pgm_info.code, (u16 *)__LC_PGM_CODE); 922 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW, 923 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 924 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW, 925 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 926 return rc ? -EFAULT : 0; 927 } 928 929 #define SCCB_MASK 0xFFFFFFF8 930 #define SCCB_EVENT_PENDING 0x3 931 932 static int write_sclp(struct kvm_vcpu *vcpu, u32 parm) 933 { 934 int rc; 935 936 if (kvm_s390_pv_cpu_get_handle(vcpu)) { 937 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT; 938 vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG; 939 vcpu->arch.sie_block->eiparams = parm; 940 return 0; 941 } 942 943 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE); 944 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR); 945 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 946 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 947 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 948 &vcpu->arch.sie_block->gpsw, sizeof(psw_t)); 949 rc |= put_guest_lc(vcpu, parm, 950 (u32 *)__LC_EXT_PARAMS); 951 952 return rc ? -EFAULT : 0; 953 } 954 955 static int __must_check __deliver_service(struct kvm_vcpu *vcpu) 956 { 957 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; 958 struct kvm_s390_ext_info ext; 959 unsigned long flags; 960 961 spin_lock_irqsave(&fi->lock, flags); 962 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) || 963 !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) { 964 spin_unlock_irqrestore(&fi->lock, flags); 965 return 0; 966 } 967 ext = fi->srv_signal; 968 memset(&fi->srv_signal, 0, sizeof(ext)); 969 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs); 970 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs); 971 if (kvm_s390_pv_cpu_is_protected(vcpu)) 972 set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs); 973 spin_unlock_irqrestore(&fi->lock, flags); 974 975 if (!ext.ext_params) 976 return 0; 977 978 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x", 979 ext.ext_params); 980 vcpu->stat.deliver_service_signal++; 981 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE, 982 ext.ext_params, 0); 983 984 return write_sclp(vcpu, ext.ext_params); 985 } 986 987 static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu) 988 { 989 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; 990 struct kvm_s390_ext_info ext; 991 unsigned long flags; 992 993 spin_lock_irqsave(&fi->lock, flags); 994 if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) { 995 spin_unlock_irqrestore(&fi->lock, flags); 996 return 0; 997 } 998 ext = fi->srv_signal; 999 /* only clear the event bits */ 1000 fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING; 1001 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs); 1002 spin_unlock_irqrestore(&fi->lock, flags); 1003 1004 VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event"); 1005 vcpu->stat.deliver_service_signal++; 1006 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE, 1007 ext.ext_params, 0); 1008 1009 return write_sclp(vcpu, ext.ext_params & SCCB_EVENT_PENDING); 1010 } 1011 1012 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu) 1013 { 1014 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; 1015 struct kvm_s390_interrupt_info *inti; 1016 int rc = 0; 1017 unsigned long flags; 1018 1019 spin_lock_irqsave(&fi->lock, flags); 1020 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT], 1021 struct kvm_s390_interrupt_info, 1022 list); 1023 if (inti) { 1024 list_del(&inti->list); 1025 fi->counters[FIRQ_CNTR_PFAULT] -= 1; 1026 } 1027 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT])) 1028 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs); 1029 spin_unlock_irqrestore(&fi->lock, flags); 1030 1031 if (inti) { 1032 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 1033 KVM_S390_INT_PFAULT_DONE, 0, 1034 inti->ext.ext_params2); 1035 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx", 1036 inti->ext.ext_params2); 1037 1038 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, 1039 (u16 *)__LC_EXT_INT_CODE); 1040 rc |= put_guest_lc(vcpu, PFAULT_DONE, 1041 (u16 *)__LC_EXT_CPU_ADDR); 1042 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 1043 &vcpu->arch.sie_block->gpsw, 1044 sizeof(psw_t)); 1045 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 1046 &vcpu->arch.sie_block->gpsw, 1047 sizeof(psw_t)); 1048 rc |= put_guest_lc(vcpu, inti->ext.ext_params2, 1049 (u64 *)__LC_EXT_PARAMS2); 1050 kfree(inti); 1051 } 1052 return rc ? -EFAULT : 0; 1053 } 1054 1055 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu) 1056 { 1057 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; 1058 struct kvm_s390_interrupt_info *inti; 1059 int rc = 0; 1060 unsigned long flags; 1061 1062 spin_lock_irqsave(&fi->lock, flags); 1063 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO], 1064 struct kvm_s390_interrupt_info, 1065 list); 1066 if (inti) { 1067 VCPU_EVENT(vcpu, 4, 1068 "deliver: virtio parm: 0x%x,parm64: 0x%llx", 1069 inti->ext.ext_params, inti->ext.ext_params2); 1070 vcpu->stat.deliver_virtio++; 1071 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 1072 inti->type, 1073 inti->ext.ext_params, 1074 inti->ext.ext_params2); 1075 list_del(&inti->list); 1076 fi->counters[FIRQ_CNTR_VIRTIO] -= 1; 1077 } 1078 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO])) 1079 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs); 1080 spin_unlock_irqrestore(&fi->lock, flags); 1081 1082 if (inti) { 1083 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, 1084 (u16 *)__LC_EXT_INT_CODE); 1085 rc |= put_guest_lc(vcpu, VIRTIO_PARAM, 1086 (u16 *)__LC_EXT_CPU_ADDR); 1087 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW, 1088 &vcpu->arch.sie_block->gpsw, 1089 sizeof(psw_t)); 1090 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, 1091 &vcpu->arch.sie_block->gpsw, 1092 sizeof(psw_t)); 1093 rc |= put_guest_lc(vcpu, inti->ext.ext_params, 1094 (u32 *)__LC_EXT_PARAMS); 1095 rc |= put_guest_lc(vcpu, inti->ext.ext_params2, 1096 (u64 *)__LC_EXT_PARAMS2); 1097 kfree(inti); 1098 } 1099 return rc ? -EFAULT : 0; 1100 } 1101 1102 static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io) 1103 { 1104 int rc; 1105 1106 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 1107 vcpu->arch.sie_block->iictl = IICTL_CODE_IO; 1108 vcpu->arch.sie_block->subchannel_id = io->subchannel_id; 1109 vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr; 1110 vcpu->arch.sie_block->io_int_parm = io->io_int_parm; 1111 vcpu->arch.sie_block->io_int_word = io->io_int_word; 1112 return 0; 1113 } 1114 1115 rc = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID); 1116 rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR); 1117 rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM); 1118 rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD); 1119 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW, 1120 &vcpu->arch.sie_block->gpsw, 1121 sizeof(psw_t)); 1122 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW, 1123 &vcpu->arch.sie_block->gpsw, 1124 sizeof(psw_t)); 1125 return rc ? -EFAULT : 0; 1126 } 1127 1128 static int __must_check __deliver_io(struct kvm_vcpu *vcpu, 1129 unsigned long irq_type) 1130 { 1131 struct list_head *isc_list; 1132 struct kvm_s390_float_interrupt *fi; 1133 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int; 1134 struct kvm_s390_interrupt_info *inti = NULL; 1135 struct kvm_s390_io_info io; 1136 u32 isc; 1137 int rc = 0; 1138 unsigned long flags; 1139 1140 fi = &vcpu->kvm->arch.float_int; 1141 1142 spin_lock_irqsave(&fi->lock, flags); 1143 isc = irq_type_to_isc(irq_type); 1144 isc_list = &fi->lists[isc]; 1145 inti = list_first_entry_or_null(isc_list, 1146 struct kvm_s390_interrupt_info, 1147 list); 1148 if (inti) { 1149 if (inti->type & KVM_S390_INT_IO_AI_MASK) 1150 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)"); 1151 else 1152 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x", 1153 inti->io.subchannel_id >> 8, 1154 inti->io.subchannel_id >> 1 & 0x3, 1155 inti->io.subchannel_nr); 1156 1157 vcpu->stat.deliver_io++; 1158 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 1159 inti->type, 1160 ((__u32)inti->io.subchannel_id << 16) | 1161 inti->io.subchannel_nr, 1162 ((__u64)inti->io.io_int_parm << 32) | 1163 inti->io.io_int_word); 1164 list_del(&inti->list); 1165 fi->counters[FIRQ_CNTR_IO] -= 1; 1166 } 1167 if (list_empty(isc_list)) 1168 clear_bit(irq_type, &fi->pending_irqs); 1169 spin_unlock_irqrestore(&fi->lock, flags); 1170 1171 if (inti) { 1172 rc = __do_deliver_io(vcpu, &(inti->io)); 1173 kfree(inti); 1174 goto out; 1175 } 1176 1177 if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) { 1178 /* 1179 * in case an adapter interrupt was not delivered 1180 * in SIE context KVM will handle the delivery 1181 */ 1182 VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc); 1183 memset(&io, 0, sizeof(io)); 1184 io.io_int_word = isc_to_int_word(isc); 1185 vcpu->stat.deliver_io++; 1186 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, 1187 KVM_S390_INT_IO(1, 0, 0, 0), 1188 ((__u32)io.subchannel_id << 16) | 1189 io.subchannel_nr, 1190 ((__u64)io.io_int_parm << 32) | 1191 io.io_int_word); 1192 rc = __do_deliver_io(vcpu, &io); 1193 } 1194 out: 1195 return rc; 1196 } 1197 1198 /* Check whether an external call is pending (deliverable or not) */ 1199 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu) 1200 { 1201 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1202 1203 if (!kvm_s390_use_sca_entries()) 1204 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs); 1205 1206 return sca_ext_call_pending(vcpu, NULL); 1207 } 1208 1209 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop) 1210 { 1211 if (deliverable_irqs(vcpu)) 1212 return 1; 1213 1214 if (kvm_cpu_has_pending_timer(vcpu)) 1215 return 1; 1216 1217 /* external call pending and deliverable */ 1218 if (kvm_s390_ext_call_pending(vcpu) && 1219 !psw_extint_disabled(vcpu) && 1220 (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK)) 1221 return 1; 1222 1223 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu)) 1224 return 1; 1225 return 0; 1226 } 1227 1228 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) 1229 { 1230 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu); 1231 } 1232 1233 static u64 __calculate_sltime(struct kvm_vcpu *vcpu) 1234 { 1235 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm); 1236 const u64 ckc = vcpu->arch.sie_block->ckc; 1237 u64 cputm, sltime = 0; 1238 1239 if (ckc_interrupts_enabled(vcpu)) { 1240 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) { 1241 if ((s64)now < (s64)ckc) 1242 sltime = tod_to_ns((s64)ckc - (s64)now); 1243 } else if (now < ckc) { 1244 sltime = tod_to_ns(ckc - now); 1245 } 1246 /* already expired */ 1247 if (!sltime) 1248 return 0; 1249 if (cpu_timer_interrupts_enabled(vcpu)) { 1250 cputm = kvm_s390_get_cpu_timer(vcpu); 1251 /* already expired? */ 1252 if (cputm >> 63) 1253 return 0; 1254 return min_t(u64, sltime, tod_to_ns(cputm)); 1255 } 1256 } else if (cpu_timer_interrupts_enabled(vcpu)) { 1257 sltime = kvm_s390_get_cpu_timer(vcpu); 1258 /* already expired? */ 1259 if (sltime >> 63) 1260 return 0; 1261 } 1262 return sltime; 1263 } 1264 1265 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu) 1266 { 1267 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int; 1268 u64 sltime; 1269 1270 vcpu->stat.exit_wait_state++; 1271 1272 /* fast path */ 1273 if (kvm_arch_vcpu_runnable(vcpu)) 1274 return 0; 1275 1276 if (psw_interrupts_disabled(vcpu)) { 1277 VCPU_EVENT(vcpu, 3, "%s", "disabled wait"); 1278 return -EOPNOTSUPP; /* disabled wait */ 1279 } 1280 1281 if (gi->origin && 1282 (gisa_get_ipm_or_restore_iam(gi) & 1283 vcpu->arch.sie_block->gcr[6] >> 24)) 1284 return 0; 1285 1286 if (!ckc_interrupts_enabled(vcpu) && 1287 !cpu_timer_interrupts_enabled(vcpu)) { 1288 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer"); 1289 __set_cpu_idle(vcpu); 1290 goto no_timer; 1291 } 1292 1293 sltime = __calculate_sltime(vcpu); 1294 if (!sltime) 1295 return 0; 1296 1297 __set_cpu_idle(vcpu); 1298 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL); 1299 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime); 1300 no_timer: 1301 kvm_vcpu_srcu_read_unlock(vcpu); 1302 vcpu->kvm->arch.float_int.last_sleep_cpu = vcpu->vcpu_idx; 1303 kvm_vcpu_halt(vcpu); 1304 vcpu->valid_wakeup = false; 1305 __unset_cpu_idle(vcpu); 1306 kvm_vcpu_srcu_read_lock(vcpu); 1307 1308 hrtimer_cancel(&vcpu->arch.ckc_timer); 1309 return 0; 1310 } 1311 1312 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu) 1313 { 1314 vcpu->valid_wakeup = true; 1315 kvm_vcpu_wake_up(vcpu); 1316 1317 /* 1318 * The VCPU might not be sleeping but rather executing VSIE. Let's 1319 * kick it, so it leaves the SIE to process the request. 1320 */ 1321 kvm_s390_vsie_kick(vcpu); 1322 } 1323 1324 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer) 1325 { 1326 struct kvm_vcpu *vcpu; 1327 u64 sltime; 1328 1329 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer); 1330 sltime = __calculate_sltime(vcpu); 1331 1332 /* 1333 * If the monotonic clock runs faster than the tod clock we might be 1334 * woken up too early and have to go back to sleep to avoid deadlocks. 1335 */ 1336 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime))) 1337 return HRTIMER_RESTART; 1338 kvm_s390_vcpu_wakeup(vcpu); 1339 return HRTIMER_NORESTART; 1340 } 1341 1342 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu) 1343 { 1344 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1345 1346 spin_lock(&li->lock); 1347 li->pending_irqs = 0; 1348 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS); 1349 memset(&li->irq, 0, sizeof(li->irq)); 1350 spin_unlock(&li->lock); 1351 1352 sca_clear_ext_call(vcpu); 1353 } 1354 1355 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu) 1356 { 1357 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1358 int rc = 0; 1359 bool delivered = false; 1360 unsigned long irq_type; 1361 unsigned long irqs; 1362 1363 __reset_intercept_indicators(vcpu); 1364 1365 /* pending ckc conditions might have been invalidated */ 1366 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 1367 if (ckc_irq_pending(vcpu)) 1368 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 1369 1370 /* pending cpu timer conditions might have been invalidated */ 1371 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 1372 if (cpu_timer_irq_pending(vcpu)) 1373 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 1374 1375 while ((irqs = deliverable_irqs(vcpu)) && !rc) { 1376 /* bits are in the reverse order of interrupt priority */ 1377 irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT); 1378 switch (irq_type) { 1379 case IRQ_PEND_IO_ISC_0: 1380 case IRQ_PEND_IO_ISC_1: 1381 case IRQ_PEND_IO_ISC_2: 1382 case IRQ_PEND_IO_ISC_3: 1383 case IRQ_PEND_IO_ISC_4: 1384 case IRQ_PEND_IO_ISC_5: 1385 case IRQ_PEND_IO_ISC_6: 1386 case IRQ_PEND_IO_ISC_7: 1387 rc = __deliver_io(vcpu, irq_type); 1388 break; 1389 case IRQ_PEND_MCHK_EX: 1390 case IRQ_PEND_MCHK_REP: 1391 rc = __deliver_machine_check(vcpu); 1392 break; 1393 case IRQ_PEND_PROG: 1394 rc = __deliver_prog(vcpu); 1395 break; 1396 case IRQ_PEND_EXT_EMERGENCY: 1397 rc = __deliver_emergency_signal(vcpu); 1398 break; 1399 case IRQ_PEND_EXT_EXTERNAL: 1400 rc = __deliver_external_call(vcpu); 1401 break; 1402 case IRQ_PEND_EXT_CLOCK_COMP: 1403 rc = __deliver_ckc(vcpu); 1404 break; 1405 case IRQ_PEND_EXT_CPU_TIMER: 1406 rc = __deliver_cpu_timer(vcpu); 1407 break; 1408 case IRQ_PEND_RESTART: 1409 rc = __deliver_restart(vcpu); 1410 break; 1411 case IRQ_PEND_SET_PREFIX: 1412 rc = __deliver_set_prefix(vcpu); 1413 break; 1414 case IRQ_PEND_PFAULT_INIT: 1415 rc = __deliver_pfault_init(vcpu); 1416 break; 1417 case IRQ_PEND_EXT_SERVICE: 1418 rc = __deliver_service(vcpu); 1419 break; 1420 case IRQ_PEND_EXT_SERVICE_EV: 1421 rc = __deliver_service_ev(vcpu); 1422 break; 1423 case IRQ_PEND_PFAULT_DONE: 1424 rc = __deliver_pfault_done(vcpu); 1425 break; 1426 case IRQ_PEND_VIRTIO: 1427 rc = __deliver_virtio(vcpu); 1428 break; 1429 default: 1430 WARN_ONCE(1, "Unknown pending irq type %ld", irq_type); 1431 clear_bit(irq_type, &li->pending_irqs); 1432 } 1433 delivered |= !rc; 1434 } 1435 1436 /* 1437 * We delivered at least one interrupt and modified the PC. Force a 1438 * singlestep event now. 1439 */ 1440 if (delivered && guestdbg_sstep_enabled(vcpu)) { 1441 struct kvm_debug_exit_arch *debug_exit = &vcpu->run->debug.arch; 1442 1443 debug_exit->addr = vcpu->arch.sie_block->gpsw.addr; 1444 debug_exit->type = KVM_SINGLESTEP; 1445 vcpu->guest_debug |= KVM_GUESTDBG_EXIT_PENDING; 1446 } 1447 1448 set_intercept_indicators(vcpu); 1449 1450 return rc; 1451 } 1452 1453 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1454 { 1455 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1456 1457 vcpu->stat.inject_program++; 1458 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code); 1459 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT, 1460 irq->u.pgm.code, 0); 1461 1462 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) { 1463 /* auto detection if no valid ILC was given */ 1464 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK; 1465 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu); 1466 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID; 1467 } 1468 1469 if (irq->u.pgm.code == PGM_PER) { 1470 li->irq.pgm.code |= PGM_PER; 1471 li->irq.pgm.flags = irq->u.pgm.flags; 1472 /* only modify PER related information */ 1473 li->irq.pgm.per_address = irq->u.pgm.per_address; 1474 li->irq.pgm.per_code = irq->u.pgm.per_code; 1475 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid; 1476 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id; 1477 } else if (!(irq->u.pgm.code & PGM_PER)) { 1478 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) | 1479 irq->u.pgm.code; 1480 li->irq.pgm.flags = irq->u.pgm.flags; 1481 /* only modify non-PER information */ 1482 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code; 1483 li->irq.pgm.mon_code = irq->u.pgm.mon_code; 1484 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code; 1485 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr; 1486 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id; 1487 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id; 1488 } else { 1489 li->irq.pgm = irq->u.pgm; 1490 } 1491 set_bit(IRQ_PEND_PROG, &li->pending_irqs); 1492 return 0; 1493 } 1494 1495 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1496 { 1497 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1498 1499 vcpu->stat.inject_pfault_init++; 1500 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx", 1501 irq->u.ext.ext_params2); 1502 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT, 1503 irq->u.ext.ext_params, 1504 irq->u.ext.ext_params2); 1505 1506 li->irq.ext = irq->u.ext; 1507 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs); 1508 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1509 return 0; 1510 } 1511 1512 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1513 { 1514 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1515 struct kvm_s390_extcall_info *extcall = &li->irq.extcall; 1516 uint16_t src_id = irq->u.extcall.code; 1517 1518 vcpu->stat.inject_external_call++; 1519 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u", 1520 src_id); 1521 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL, 1522 src_id, 0); 1523 1524 /* sending vcpu invalid */ 1525 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL) 1526 return -EINVAL; 1527 1528 if (kvm_s390_use_sca_entries() && !kvm_s390_pv_cpu_get_handle(vcpu)) 1529 return sca_inject_ext_call(vcpu, src_id); 1530 1531 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs)) 1532 return -EBUSY; 1533 *extcall = irq->u.extcall; 1534 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1535 return 0; 1536 } 1537 1538 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1539 { 1540 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1541 struct kvm_s390_prefix_info *prefix = &li->irq.prefix; 1542 1543 vcpu->stat.inject_set_prefix++; 1544 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x", 1545 irq->u.prefix.address); 1546 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX, 1547 irq->u.prefix.address, 0); 1548 1549 if (!is_vcpu_stopped(vcpu)) 1550 return -EBUSY; 1551 1552 *prefix = irq->u.prefix; 1553 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs); 1554 return 0; 1555 } 1556 1557 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS) 1558 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq, bool *storestatus) 1559 { 1560 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1561 struct kvm_s390_stop_info *stop = &li->irq.stop; 1562 1563 vcpu->stat.inject_stop_signal++; 1564 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0); 1565 1566 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS) 1567 return -EINVAL; 1568 if (is_vcpu_stopped(vcpu)) { 1569 if (!(irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)) 1570 return 0; 1571 *storestatus = true; 1572 return -EWOULDBLOCK; 1573 } 1574 1575 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs)) 1576 return -EBUSY; 1577 stop->flags = irq->u.stop.flags; 1578 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT); 1579 return 0; 1580 } 1581 1582 static int __inject_sigp_restart(struct kvm_vcpu *vcpu) 1583 { 1584 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1585 1586 vcpu->stat.inject_restart++; 1587 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int"); 1588 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0); 1589 1590 set_bit(IRQ_PEND_RESTART, &li->pending_irqs); 1591 return 0; 1592 } 1593 1594 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu, 1595 struct kvm_s390_irq *irq) 1596 { 1597 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1598 1599 vcpu->stat.inject_emergency_signal++; 1600 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u", 1601 irq->u.emerg.code); 1602 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY, 1603 irq->u.emerg.code, 0); 1604 1605 /* sending vcpu invalid */ 1606 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL) 1607 return -EINVAL; 1608 1609 set_bit(irq->u.emerg.code, li->sigp_emerg_pending); 1610 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs); 1611 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1612 return 0; 1613 } 1614 1615 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 1616 { 1617 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1618 struct kvm_s390_mchk_info *mchk = &li->irq.mchk; 1619 1620 vcpu->stat.inject_mchk++; 1621 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx", 1622 irq->u.mchk.mcic); 1623 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0, 1624 irq->u.mchk.mcic); 1625 1626 /* 1627 * Because repressible machine checks can be indicated along with 1628 * exigent machine checks (PoP, Chapter 11, Interruption action) 1629 * we need to combine cr14, mcic and external damage code. 1630 * Failing storage address and the logout area should not be or'ed 1631 * together, we just indicate the last occurrence of the corresponding 1632 * machine check 1633 */ 1634 mchk->cr14 |= irq->u.mchk.cr14; 1635 mchk->mcic |= irq->u.mchk.mcic; 1636 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code; 1637 mchk->failing_storage_address = irq->u.mchk.failing_storage_address; 1638 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout, 1639 sizeof(mchk->fixed_logout)); 1640 if (mchk->mcic & MCHK_EX_MASK) 1641 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs); 1642 else if (mchk->mcic & MCHK_REP_MASK) 1643 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs); 1644 return 0; 1645 } 1646 1647 static int __inject_ckc(struct kvm_vcpu *vcpu) 1648 { 1649 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1650 1651 vcpu->stat.inject_ckc++; 1652 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external"); 1653 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP, 1654 0, 0); 1655 1656 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs); 1657 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1658 return 0; 1659 } 1660 1661 static int __inject_cpu_timer(struct kvm_vcpu *vcpu) 1662 { 1663 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 1664 1665 vcpu->stat.inject_cputm++; 1666 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external"); 1667 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER, 1668 0, 0); 1669 1670 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs); 1671 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1672 return 0; 1673 } 1674 1675 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm, 1676 int isc, u32 schid) 1677 { 1678 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1679 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc]; 1680 struct kvm_s390_interrupt_info *iter; 1681 u16 id = (schid & 0xffff0000U) >> 16; 1682 u16 nr = schid & 0x0000ffffU; 1683 unsigned long flags; 1684 1685 spin_lock_irqsave(&fi->lock, flags); 1686 list_for_each_entry(iter, isc_list, list) { 1687 if (schid && (id != iter->io.subchannel_id || 1688 nr != iter->io.subchannel_nr)) 1689 continue; 1690 /* found an appropriate entry */ 1691 list_del_init(&iter->list); 1692 fi->counters[FIRQ_CNTR_IO] -= 1; 1693 if (list_empty(isc_list)) 1694 clear_bit(isc_to_irq_type(isc), &fi->pending_irqs); 1695 spin_unlock_irqrestore(&fi->lock, flags); 1696 return iter; 1697 } 1698 spin_unlock_irqrestore(&fi->lock, flags); 1699 return NULL; 1700 } 1701 1702 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm, 1703 u64 isc_mask, u32 schid) 1704 { 1705 struct kvm_s390_interrupt_info *inti = NULL; 1706 int isc; 1707 1708 for (isc = 0; isc <= MAX_ISC && !inti; isc++) { 1709 if (isc_mask & isc_to_isc_bits(isc)) 1710 inti = get_io_int(kvm, isc, schid); 1711 } 1712 return inti; 1713 } 1714 1715 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid) 1716 { 1717 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 1718 unsigned long active_mask; 1719 int isc; 1720 1721 if (schid) 1722 goto out; 1723 if (!gi->origin) 1724 goto out; 1725 1726 active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32; 1727 while (active_mask) { 1728 isc = __fls(active_mask) ^ (BITS_PER_LONG - 1); 1729 if (gisa_tac_ipm_gisc(gi->origin, isc)) 1730 return isc; 1731 clear_bit_inv(isc, &active_mask); 1732 } 1733 out: 1734 return -EINVAL; 1735 } 1736 1737 /* 1738 * Dequeue and return an I/O interrupt matching any of the interruption 1739 * subclasses as designated by the isc mask in cr6 and the schid (if != 0). 1740 * Take into account the interrupts pending in the interrupt list and in GISA. 1741 * 1742 * Note that for a guest that does not enable I/O interrupts 1743 * but relies on TPI, a flood of classic interrupts may starve 1744 * out adapter interrupts on the same isc. Linux does not do 1745 * that, and it is possible to work around the issue by configuring 1746 * different iscs for classic and adapter interrupts in the guest, 1747 * but we may want to revisit this in the future. 1748 */ 1749 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm, 1750 u64 isc_mask, u32 schid) 1751 { 1752 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 1753 struct kvm_s390_interrupt_info *inti, *tmp_inti; 1754 int isc; 1755 1756 inti = get_top_io_int(kvm, isc_mask, schid); 1757 1758 isc = get_top_gisa_isc(kvm, isc_mask, schid); 1759 if (isc < 0) 1760 /* no AI in GISA */ 1761 goto out; 1762 1763 if (!inti) 1764 /* AI in GISA but no classical IO int */ 1765 goto gisa_out; 1766 1767 /* both types of interrupts present */ 1768 if (int_word_to_isc(inti->io.io_int_word) <= isc) { 1769 /* classical IO int with higher priority */ 1770 gisa_set_ipm_gisc(gi->origin, isc); 1771 goto out; 1772 } 1773 gisa_out: 1774 tmp_inti = kzalloc_obj(*inti, GFP_KERNEL_ACCOUNT); 1775 if (tmp_inti) { 1776 tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0); 1777 tmp_inti->io.io_int_word = isc_to_int_word(isc); 1778 if (inti) 1779 kvm_s390_reinject_io_int(kvm, inti); 1780 inti = tmp_inti; 1781 } else 1782 gisa_set_ipm_gisc(gi->origin, isc); 1783 out: 1784 return inti; 1785 } 1786 1787 static int __inject_service(struct kvm *kvm, 1788 struct kvm_s390_interrupt_info *inti) 1789 { 1790 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1791 unsigned long flags; 1792 1793 kvm->stat.inject_service_signal++; 1794 spin_lock_irqsave(&fi->lock, flags); 1795 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING; 1796 1797 /* We always allow events, track them separately from the sccb ints */ 1798 if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING) 1799 set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs); 1800 1801 /* 1802 * Early versions of the QEMU s390 bios will inject several 1803 * service interrupts after another without handling a 1804 * condition code indicating busy. 1805 * We will silently ignore those superfluous sccb values. 1806 * A future version of QEMU will take care of serialization 1807 * of servc requests 1808 */ 1809 if (fi->srv_signal.ext_params & SCCB_MASK) 1810 goto out; 1811 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK; 1812 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs); 1813 out: 1814 spin_unlock_irqrestore(&fi->lock, flags); 1815 kfree(inti); 1816 return 0; 1817 } 1818 1819 static int __inject_virtio(struct kvm *kvm, 1820 struct kvm_s390_interrupt_info *inti) 1821 { 1822 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1823 unsigned long flags; 1824 1825 kvm->stat.inject_virtio++; 1826 spin_lock_irqsave(&fi->lock, flags); 1827 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) { 1828 spin_unlock_irqrestore(&fi->lock, flags); 1829 return -EBUSY; 1830 } 1831 fi->counters[FIRQ_CNTR_VIRTIO] += 1; 1832 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]); 1833 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs); 1834 spin_unlock_irqrestore(&fi->lock, flags); 1835 return 0; 1836 } 1837 1838 static int __inject_pfault_done(struct kvm *kvm, 1839 struct kvm_s390_interrupt_info *inti) 1840 { 1841 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1842 unsigned long flags; 1843 1844 kvm->stat.inject_pfault_done++; 1845 spin_lock_irqsave(&fi->lock, flags); 1846 if (fi->counters[FIRQ_CNTR_PFAULT] >= 1847 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) { 1848 spin_unlock_irqrestore(&fi->lock, flags); 1849 return -EBUSY; 1850 } 1851 fi->counters[FIRQ_CNTR_PFAULT] += 1; 1852 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]); 1853 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs); 1854 spin_unlock_irqrestore(&fi->lock, flags); 1855 return 0; 1856 } 1857 1858 #define CR_PENDING_SUBCLASS 28 1859 static int __inject_float_mchk(struct kvm *kvm, 1860 struct kvm_s390_interrupt_info *inti) 1861 { 1862 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 1863 unsigned long flags; 1864 1865 kvm->stat.inject_float_mchk++; 1866 spin_lock_irqsave(&fi->lock, flags); 1867 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS); 1868 fi->mchk.mcic |= inti->mchk.mcic; 1869 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs); 1870 spin_unlock_irqrestore(&fi->lock, flags); 1871 kfree(inti); 1872 return 0; 1873 } 1874 1875 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti) 1876 { 1877 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 1878 struct kvm_s390_float_interrupt *fi; 1879 struct list_head *list; 1880 int isc; 1881 unsigned long flags; 1882 1883 kvm->stat.inject_io++; 1884 isc = int_word_to_isc(inti->io.io_int_word); 1885 1886 /* 1887 * We do not use the lock checking variant as this is just a 1888 * performance optimization and we do not hold the lock here. 1889 * This is ok as the code will pick interrupts from both "lists" 1890 * for delivery. 1891 */ 1892 if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) { 1893 VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc); 1894 gisa_set_ipm_gisc(gi->origin, isc); 1895 kfree(inti); 1896 return 0; 1897 } 1898 1899 fi = &kvm->arch.float_int; 1900 spin_lock_irqsave(&fi->lock, flags); 1901 if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) { 1902 spin_unlock_irqrestore(&fi->lock, flags); 1903 return -EBUSY; 1904 } 1905 fi->counters[FIRQ_CNTR_IO] += 1; 1906 1907 if (inti->type & KVM_S390_INT_IO_AI_MASK) 1908 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)"); 1909 else 1910 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x", 1911 inti->io.subchannel_id >> 8, 1912 inti->io.subchannel_id >> 1 & 0x3, 1913 inti->io.subchannel_nr); 1914 list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc]; 1915 list_add_tail(&inti->list, list); 1916 set_bit(isc_to_irq_type(isc), &fi->pending_irqs); 1917 spin_unlock_irqrestore(&fi->lock, flags); 1918 return 0; 1919 } 1920 1921 static u64 inti_to_irq_pend_mask(u64 type, int isc) 1922 { 1923 switch (type) { 1924 case KVM_S390_MCHK: 1925 /* Only repressible machine checks are floating */ 1926 return BIT(IRQ_PEND_MCHK_REP); 1927 case KVM_S390_INT_VIRTIO: 1928 return BIT(IRQ_PEND_VIRTIO); 1929 case KVM_S390_INT_SERVICE: 1930 return BIT(IRQ_PEND_EXT_SERVICE) | 1931 BIT(IRQ_PEND_EXT_SERVICE_EV); 1932 case KVM_S390_INT_PFAULT_DONE: 1933 return BIT(IRQ_PEND_PFAULT_DONE); 1934 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 1935 return BIT(isc_to_irq_type(isc)); 1936 default: 1937 return 0; 1938 } 1939 } 1940 1941 /* 1942 * Setup intervention masks to catch running vcpus that hopefully open 1943 * their masks soonish and kick sleeping vcpus to motivate them to 1944 * take IRQs. 1945 */ 1946 static void vcpu_intervention_kick(struct kvm_vcpu *vcpu, u64 type) 1947 { 1948 /* make the VCPU drop out of the SIE, or wake it up if sleeping */ 1949 switch (type) { 1950 case KVM_S390_MCHK: 1951 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT); 1952 break; 1953 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 1954 if (!(type & KVM_S390_INT_IO_AI_MASK && 1955 vcpu->kvm->arch.gisa_int.origin) || 1956 kvm_s390_pv_cpu_get_handle(vcpu)) 1957 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT); 1958 break; 1959 default: 1960 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT); 1961 break; 1962 } 1963 kvm_s390_vcpu_wakeup(vcpu); 1964 } 1965 1966 static void kick_cpu_irq(struct kvm *kvm, u64 type, u64 parm) 1967 { 1968 struct kvm_vcpu *dst_vcpu; 1969 int sigcpu, online_vcpus, nr_tries = 0; 1970 u64 irq_pend_mask; 1971 unsigned long i; 1972 1973 online_vcpus = atomic_read(&kvm->online_vcpus); 1974 if (!online_vcpus) 1975 return; 1976 1977 irq_pend_mask = inti_to_irq_pend_mask(type, parm); 1978 for (sigcpu = kvm->arch.float_int.last_sleep_cpu; ; sigcpu++) { 1979 sigcpu %= online_vcpus; 1980 dst_vcpu = kvm_get_vcpu(kvm, sigcpu); 1981 if (!is_vcpu_stopped(dst_vcpu) && 1982 deliverable_irqs(dst_vcpu) & irq_pend_mask) 1983 break; 1984 /* avoid endless loops if all vcpus are stopped */ 1985 if (nr_tries++ >= online_vcpus) { 1986 dst_vcpu = NULL; 1987 break; 1988 } 1989 } 1990 1991 /* Nobody was enabled, time to wake all of them */ 1992 if (!dst_vcpu) { 1993 kvm_for_each_vcpu(i, dst_vcpu, kvm) 1994 vcpu_intervention_kick(dst_vcpu, type); 1995 return; 1996 } 1997 1998 vcpu_intervention_kick(dst_vcpu, type); 1999 } 2000 2001 void kvm_s390_pv_sclp_kick(struct kvm_vcpu *vcpu) 2002 { 2003 /* 2004 * The cpu that called sclp likely will also take the IRQ, no 2005 * need to kick anyone. 2006 */ 2007 if (deliverable_irqs(vcpu) & BIT(IRQ_PEND_EXT_SERVICE)) 2008 return; 2009 2010 /* 2011 * For the other cases we might have sleeping cpus with open 2012 * masks. Time to find and kick them. 2013 */ 2014 kick_cpu_irq(vcpu->kvm, KVM_S390_INT_SERVICE, -1); 2015 } 2016 2017 /* 2018 * Find a destination VCPU for a floating irq and kick it. 2019 */ 2020 static void __floating_irq_kick(struct kvm *kvm, u64 type, u64 parm) 2021 { 2022 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2023 2024 /* 2025 * No need to kick on non-ev service IRQs for PV VMs, we're 2026 * not allowed to inject anyway. We need to wait for the sclp 2027 * instruction notification AFTER re-entry of the vcpu that 2028 * handled the instruction intercept. 2029 */ 2030 if (type == KVM_S390_INT_SERVICE && !(parm & SCCB_EVENT_PENDING) && 2031 test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs)) 2032 return; 2033 2034 kick_cpu_irq(kvm, type, parm); 2035 } 2036 2037 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti) 2038 { 2039 u64 type = READ_ONCE(inti->type); 2040 u64 parm = -1; 2041 int rc; 2042 2043 switch (type) { 2044 case KVM_S390_MCHK: 2045 rc = __inject_float_mchk(kvm, inti); 2046 break; 2047 case KVM_S390_INT_VIRTIO: 2048 rc = __inject_virtio(kvm, inti); 2049 break; 2050 case KVM_S390_INT_SERVICE: 2051 parm = inti->ext.ext_params & SCCB_EVENT_PENDING; 2052 rc = __inject_service(kvm, inti); 2053 break; 2054 case KVM_S390_INT_PFAULT_DONE: 2055 rc = __inject_pfault_done(kvm, inti); 2056 break; 2057 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2058 /* Grab isc here since __inject_io() might free inti */ 2059 parm = int_word_to_isc(inti->io.io_int_word); 2060 rc = __inject_io(kvm, inti); 2061 break; 2062 default: 2063 rc = -EINVAL; 2064 } 2065 if (rc) 2066 return rc; 2067 2068 __floating_irq_kick(kvm, type, parm); 2069 return 0; 2070 } 2071 2072 int kvm_s390_inject_vm(struct kvm *kvm, 2073 struct kvm_s390_interrupt *s390int, struct kvm_s390_interrupt_info *inti) 2074 { 2075 int rc; 2076 2077 inti->type = s390int->type; 2078 switch (inti->type) { 2079 case KVM_S390_INT_VIRTIO: 2080 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx", 2081 s390int->parm, s390int->parm64); 2082 inti->ext.ext_params = s390int->parm; 2083 inti->ext.ext_params2 = s390int->parm64; 2084 break; 2085 case KVM_S390_INT_SERVICE: 2086 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm); 2087 inti->ext.ext_params = s390int->parm; 2088 break; 2089 case KVM_S390_INT_PFAULT_DONE: 2090 inti->ext.ext_params2 = s390int->parm64; 2091 break; 2092 case KVM_S390_MCHK: 2093 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx", 2094 s390int->parm64); 2095 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */ 2096 inti->mchk.mcic = s390int->parm64; 2097 break; 2098 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2099 inti->io.subchannel_id = s390int->parm >> 16; 2100 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu; 2101 inti->io.io_int_parm = s390int->parm64 >> 32; 2102 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull; 2103 break; 2104 default: 2105 return -EINVAL; 2106 } 2107 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64, 2108 2); 2109 2110 rc = __inject_vm(kvm, inti); 2111 2112 return rc; 2113 } 2114 2115 int kvm_s390_reinject_io_int(struct kvm *kvm, 2116 struct kvm_s390_interrupt_info *inti) 2117 { 2118 return __inject_vm(kvm, inti); 2119 } 2120 2121 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int, 2122 struct kvm_s390_irq *irq) 2123 { 2124 irq->type = s390int->type; 2125 switch (irq->type) { 2126 case KVM_S390_PROGRAM_INT: 2127 if (s390int->parm & 0xffff0000) 2128 return -EINVAL; 2129 irq->u.pgm.code = s390int->parm; 2130 break; 2131 case KVM_S390_SIGP_SET_PREFIX: 2132 irq->u.prefix.address = s390int->parm; 2133 break; 2134 case KVM_S390_SIGP_STOP: 2135 irq->u.stop.flags = s390int->parm; 2136 break; 2137 case KVM_S390_INT_EXTERNAL_CALL: 2138 if (s390int->parm & 0xffff0000) 2139 return -EINVAL; 2140 irq->u.extcall.code = s390int->parm; 2141 break; 2142 case KVM_S390_INT_EMERGENCY: 2143 if (s390int->parm & 0xffff0000) 2144 return -EINVAL; 2145 irq->u.emerg.code = s390int->parm; 2146 break; 2147 case KVM_S390_MCHK: 2148 irq->u.mchk.mcic = s390int->parm64; 2149 break; 2150 case KVM_S390_INT_PFAULT_INIT: 2151 irq->u.ext.ext_params = s390int->parm; 2152 irq->u.ext.ext_params2 = s390int->parm64; 2153 break; 2154 case KVM_S390_RESTART: 2155 case KVM_S390_INT_CLOCK_COMP: 2156 case KVM_S390_INT_CPU_TIMER: 2157 break; 2158 default: 2159 return -EINVAL; 2160 } 2161 return 0; 2162 } 2163 2164 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu) 2165 { 2166 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2167 2168 return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs); 2169 } 2170 2171 int kvm_s390_is_restart_irq_pending(struct kvm_vcpu *vcpu) 2172 { 2173 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2174 2175 return test_bit(IRQ_PEND_RESTART, &li->pending_irqs); 2176 } 2177 2178 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu) 2179 { 2180 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2181 2182 spin_lock(&li->lock); 2183 li->irq.stop.flags = 0; 2184 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs); 2185 spin_unlock(&li->lock); 2186 } 2187 2188 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq, bool *storestatus) 2189 { 2190 int rc; 2191 2192 switch (irq->type) { 2193 case KVM_S390_PROGRAM_INT: 2194 rc = __inject_prog(vcpu, irq); 2195 break; 2196 case KVM_S390_SIGP_SET_PREFIX: 2197 rc = __inject_set_prefix(vcpu, irq); 2198 break; 2199 case KVM_S390_SIGP_STOP: 2200 rc = __inject_sigp_stop(vcpu, irq, storestatus); 2201 break; 2202 case KVM_S390_RESTART: 2203 rc = __inject_sigp_restart(vcpu); 2204 break; 2205 case KVM_S390_INT_CLOCK_COMP: 2206 rc = __inject_ckc(vcpu); 2207 break; 2208 case KVM_S390_INT_CPU_TIMER: 2209 rc = __inject_cpu_timer(vcpu); 2210 break; 2211 case KVM_S390_INT_EXTERNAL_CALL: 2212 rc = __inject_extcall(vcpu, irq); 2213 break; 2214 case KVM_S390_INT_EMERGENCY: 2215 rc = __inject_sigp_emergency(vcpu, irq); 2216 break; 2217 case KVM_S390_MCHK: 2218 rc = __inject_mchk(vcpu, irq); 2219 break; 2220 case KVM_S390_INT_PFAULT_INIT: 2221 rc = __inject_pfault_init(vcpu, irq); 2222 break; 2223 case KVM_S390_INT_VIRTIO: 2224 case KVM_S390_INT_SERVICE: 2225 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2226 default: 2227 rc = -EINVAL; 2228 } 2229 2230 return rc; 2231 } 2232 2233 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq) 2234 { 2235 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 2236 bool storestatus = false; 2237 int rc; 2238 2239 spin_lock(&li->lock); 2240 rc = do_inject_vcpu(vcpu, irq, &storestatus); 2241 spin_unlock(&li->lock); 2242 2243 if (rc == -EWOULDBLOCK && storestatus) 2244 rc = kvm_s390_store_status_unloaded(vcpu, KVM_S390_STORE_STATUS_NOADDR); 2245 2246 if (!rc) 2247 kvm_s390_vcpu_wakeup(vcpu); 2248 return rc; 2249 } 2250 2251 static inline void clear_irq_list(struct list_head *_list) 2252 { 2253 struct kvm_s390_interrupt_info *inti, *n; 2254 2255 list_for_each_entry_safe(inti, n, _list, list) { 2256 list_del(&inti->list); 2257 kfree(inti); 2258 } 2259 } 2260 2261 static void inti_to_irq(struct kvm_s390_interrupt_info *inti, 2262 struct kvm_s390_irq *irq) 2263 { 2264 irq->type = inti->type; 2265 switch (inti->type) { 2266 case KVM_S390_INT_PFAULT_INIT: 2267 case KVM_S390_INT_PFAULT_DONE: 2268 case KVM_S390_INT_VIRTIO: 2269 irq->u.ext = inti->ext; 2270 break; 2271 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2272 irq->u.io = inti->io; 2273 break; 2274 } 2275 } 2276 2277 void kvm_s390_clear_float_irqs(struct kvm *kvm) 2278 { 2279 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2280 int i; 2281 unsigned long flags; 2282 2283 mutex_lock(&kvm->lock); 2284 if (!kvm_s390_pv_is_protected(kvm)) 2285 fi->masked_irqs = 0; 2286 mutex_unlock(&kvm->lock); 2287 spin_lock_irqsave(&fi->lock, flags); 2288 fi->pending_irqs = 0; 2289 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal)); 2290 memset(&fi->mchk, 0, sizeof(fi->mchk)); 2291 for (i = 0; i < FIRQ_LIST_COUNT; i++) 2292 clear_irq_list(&fi->lists[i]); 2293 for (i = 0; i < FIRQ_MAX_COUNT; i++) 2294 fi->counters[i] = 0; 2295 spin_unlock_irqrestore(&fi->lock, flags); 2296 kvm_s390_gisa_clear(kvm); 2297 }; 2298 2299 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len) 2300 { 2301 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 2302 struct kvm_s390_interrupt_info *inti; 2303 struct kvm_s390_float_interrupt *fi; 2304 struct kvm_s390_irq *buf; 2305 struct kvm_s390_irq *irq; 2306 int max_irqs; 2307 int ret = 0; 2308 int n = 0; 2309 int i; 2310 unsigned long flags; 2311 2312 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0) 2313 return -EINVAL; 2314 2315 /* 2316 * We are already using -ENOMEM to signal 2317 * userspace it may retry with a bigger buffer, 2318 * so we need to use something else for this case 2319 */ 2320 buf = vzalloc(len); 2321 if (!buf) 2322 return -ENOBUFS; 2323 2324 max_irqs = len / sizeof(struct kvm_s390_irq); 2325 2326 if (gi->origin && gisa_get_ipm(gi->origin)) { 2327 for (i = 0; i <= MAX_ISC; i++) { 2328 if (n == max_irqs) { 2329 /* signal userspace to try again */ 2330 ret = -ENOMEM; 2331 goto out_nolock; 2332 } 2333 if (gisa_test_ipm_gisc(gi->origin, i)) { 2334 irq = (struct kvm_s390_irq *) &buf[n]; 2335 irq->type = KVM_S390_INT_IO(1, 0, 0, 0); 2336 irq->u.io.io_int_word = isc_to_int_word(i); 2337 n++; 2338 } 2339 } 2340 } 2341 fi = &kvm->arch.float_int; 2342 spin_lock_irqsave(&fi->lock, flags); 2343 for (i = 0; i < FIRQ_LIST_COUNT; i++) { 2344 list_for_each_entry(inti, &fi->lists[i], list) { 2345 if (n == max_irqs) { 2346 /* signal userspace to try again */ 2347 ret = -ENOMEM; 2348 goto out; 2349 } 2350 inti_to_irq(inti, &buf[n]); 2351 n++; 2352 } 2353 } 2354 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) || 2355 test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) { 2356 if (n == max_irqs) { 2357 /* signal userspace to try again */ 2358 ret = -ENOMEM; 2359 goto out; 2360 } 2361 irq = (struct kvm_s390_irq *) &buf[n]; 2362 irq->type = KVM_S390_INT_SERVICE; 2363 irq->u.ext = fi->srv_signal; 2364 n++; 2365 } 2366 if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) { 2367 if (n == max_irqs) { 2368 /* signal userspace to try again */ 2369 ret = -ENOMEM; 2370 goto out; 2371 } 2372 irq = (struct kvm_s390_irq *) &buf[n]; 2373 irq->type = KVM_S390_MCHK; 2374 irq->u.mchk = fi->mchk; 2375 n++; 2376 } 2377 2378 out: 2379 spin_unlock_irqrestore(&fi->lock, flags); 2380 out_nolock: 2381 if (!ret && n > 0) { 2382 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n)) 2383 ret = -EFAULT; 2384 } 2385 vfree(buf); 2386 2387 return ret < 0 ? ret : n; 2388 } 2389 2390 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr) 2391 { 2392 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2393 struct kvm_s390_ais_all ais; 2394 unsigned long flags; 2395 2396 if (attr->attr < sizeof(ais)) 2397 return -EINVAL; 2398 2399 if (!test_kvm_facility(kvm, 72)) 2400 return -EOPNOTSUPP; 2401 2402 spin_lock_irqsave(&fi->ais_lock, flags); 2403 ais.simm = fi->simm; 2404 ais.nimm = fi->nimm; 2405 spin_unlock_irqrestore(&fi->ais_lock, flags); 2406 2407 if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais))) 2408 return -EFAULT; 2409 2410 return 0; 2411 } 2412 2413 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr) 2414 { 2415 int r; 2416 2417 switch (attr->group) { 2418 case KVM_DEV_FLIC_GET_ALL_IRQS: 2419 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr, 2420 attr->attr); 2421 break; 2422 case KVM_DEV_FLIC_AISM_ALL: 2423 r = flic_ais_mode_get_all(dev->kvm, attr); 2424 break; 2425 default: 2426 r = -EINVAL; 2427 } 2428 2429 return r; 2430 } 2431 2432 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti, 2433 u64 addr) 2434 { 2435 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr; 2436 void *target = NULL; 2437 void __user *source; 2438 u64 size; 2439 2440 if (get_user(inti->type, (u64 __user *)addr)) 2441 return -EFAULT; 2442 2443 switch (inti->type) { 2444 case KVM_S390_INT_PFAULT_INIT: 2445 case KVM_S390_INT_PFAULT_DONE: 2446 case KVM_S390_INT_VIRTIO: 2447 case KVM_S390_INT_SERVICE: 2448 target = (void *) &inti->ext; 2449 source = &uptr->u.ext; 2450 size = sizeof(inti->ext); 2451 break; 2452 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX: 2453 target = (void *) &inti->io; 2454 source = &uptr->u.io; 2455 size = sizeof(inti->io); 2456 break; 2457 case KVM_S390_MCHK: 2458 target = (void *) &inti->mchk; 2459 source = &uptr->u.mchk; 2460 size = sizeof(inti->mchk); 2461 break; 2462 default: 2463 return -EINVAL; 2464 } 2465 2466 if (copy_from_user(target, source, size)) 2467 return -EFAULT; 2468 2469 return 0; 2470 } 2471 2472 static int enqueue_floating_irq(struct kvm_device *dev, 2473 struct kvm_device_attr *attr) 2474 { 2475 struct kvm_s390_interrupt_info *inti = NULL; 2476 int r = 0; 2477 int len = attr->attr; 2478 2479 if (len % sizeof(struct kvm_s390_irq) != 0) 2480 return -EINVAL; 2481 else if (len > KVM_S390_FLIC_MAX_BUFFER) 2482 return -EINVAL; 2483 2484 while (len >= sizeof(struct kvm_s390_irq)) { 2485 inti = kzalloc_obj(*inti, GFP_KERNEL_ACCOUNT); 2486 if (!inti) 2487 return -ENOMEM; 2488 2489 r = copy_irq_from_user(inti, attr->addr); 2490 if (r) { 2491 kfree(inti); 2492 return r; 2493 } 2494 r = __inject_vm(dev->kvm, inti); 2495 if (r) { 2496 kfree(inti); 2497 return r; 2498 } 2499 len -= sizeof(struct kvm_s390_irq); 2500 attr->addr += sizeof(struct kvm_s390_irq); 2501 } 2502 2503 return r; 2504 } 2505 2506 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id) 2507 { 2508 if (id >= MAX_S390_IO_ADAPTERS) 2509 return NULL; 2510 id = array_index_nospec(id, MAX_S390_IO_ADAPTERS); 2511 return kvm->arch.adapters[id]; 2512 } 2513 2514 static int register_io_adapter(struct kvm_device *dev, 2515 struct kvm_device_attr *attr) 2516 { 2517 struct s390_io_adapter *adapter; 2518 struct kvm_s390_io_adapter adapter_info; 2519 int rc = 0; 2520 2521 mutex_lock(&dev->kvm->lock); 2522 if (copy_from_user(&adapter_info, 2523 (void __user *)attr->addr, sizeof(adapter_info))) { 2524 rc = -EFAULT; 2525 goto out; 2526 } 2527 if (adapter_info.id >= MAX_S390_IO_ADAPTERS) { 2528 rc = -EINVAL; 2529 goto out; 2530 } 2531 adapter_info.id = array_index_nospec(adapter_info.id, 2532 MAX_S390_IO_ADAPTERS); 2533 2534 if (dev->kvm->arch.adapters[adapter_info.id] != NULL) { 2535 rc = -EINVAL; 2536 goto out; 2537 } 2538 adapter = kzalloc_obj(*adapter, GFP_KERNEL_ACCOUNT); 2539 if (!adapter) { 2540 rc = -ENOMEM; 2541 goto out; 2542 } 2543 2544 INIT_LIST_HEAD(&adapter->maps); 2545 spin_lock_init(&adapter->maps_lock); 2546 adapter->nr_maps = 0; 2547 adapter->id = adapter_info.id; 2548 adapter->isc = adapter_info.isc; 2549 adapter->maskable = adapter_info.maskable; 2550 adapter->masked = false; 2551 adapter->swap = adapter_info.swap; 2552 adapter->suppressible = adapter_info.flags & 2553 KVM_S390_ADAPTER_SUPPRESSIBLE; 2554 dev->kvm->arch.adapters[adapter->id] = adapter; 2555 2556 out: 2557 mutex_unlock(&dev->kvm->lock); 2558 return rc; 2559 } 2560 2561 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked) 2562 { 2563 int ret; 2564 struct s390_io_adapter *adapter = get_io_adapter(kvm, id); 2565 2566 if (!adapter || !adapter->maskable) 2567 return -EINVAL; 2568 ret = adapter->masked; 2569 adapter->masked = masked; 2570 return ret; 2571 } 2572 2573 static struct page *pin_map_page(struct kvm *kvm, u64 uaddr, 2574 unsigned int gup_flags) 2575 { 2576 struct mm_struct *mm = kvm->mm; 2577 struct page *page = NULL; 2578 int locked = 1; 2579 2580 if (mmget_not_zero(mm)) { 2581 mmap_read_lock(mm); 2582 pin_user_pages_remote(mm, uaddr, 1, FOLL_WRITE | gup_flags, 2583 &page, &locked); 2584 if (locked) 2585 mmap_read_unlock(mm); 2586 mmput(mm); 2587 } 2588 2589 return page; 2590 } 2591 2592 static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr) 2593 { 2594 struct s390_io_adapter *adapter = get_io_adapter(kvm, id); 2595 struct s390_map_info *map; 2596 unsigned long flags; 2597 __u64 host_addr; 2598 int ret, idx; 2599 2600 if (!adapter || !addr) 2601 return -EINVAL; 2602 2603 map = kzalloc_obj(*map, GFP_KERNEL_ACCOUNT); 2604 if (!map) 2605 return -ENOMEM; 2606 2607 INIT_LIST_HEAD(&map->list); 2608 idx = srcu_read_lock(&kvm->srcu); 2609 host_addr = gpa_to_hva(kvm, addr); 2610 if (kvm_is_error_hva(host_addr)) { 2611 srcu_read_unlock(&kvm->srcu, idx); 2612 ret = -EFAULT; 2613 goto out; 2614 } 2615 srcu_read_unlock(&kvm->srcu, idx); 2616 map->guest_addr = addr; 2617 map->addr = host_addr; 2618 map->page = pin_map_page(kvm, host_addr, FOLL_LONGTERM); 2619 if (!map->page) { 2620 /* 2621 * Long-term pinning may fail for memory types such as file-backed 2622 * memory. Verify that short-term pinning succeeds so that the 2623 * non-atomic irqfd path can handle interrupt injection. 2624 */ 2625 map->page = pin_map_page(kvm, host_addr, 0); 2626 if (!map->page) { 2627 ret = -EINVAL; 2628 goto out; 2629 } 2630 unpin_user_page(map->page); 2631 map->page = NULL; 2632 map->pinned = false; 2633 /* Add an entry to preserve MAP/UNMAP symmetry. */ 2634 } else { 2635 map->pinned = true; 2636 } 2637 spin_lock_irqsave(&adapter->maps_lock, flags); 2638 if (adapter->nr_maps < MAX_S390_ADAPTER_MAPS) { 2639 list_add_tail(&map->list, &adapter->maps); 2640 adapter->nr_maps++; 2641 ret = 0; 2642 } else { 2643 ret = -EINVAL; 2644 } 2645 spin_unlock_irqrestore(&adapter->maps_lock, flags); 2646 if (ret && map->page) 2647 unpin_user_page(map->page); 2648 out: 2649 if (ret) 2650 kfree(map); 2651 return ret; 2652 } 2653 2654 static int kvm_s390_adapter_unmap(struct kvm *kvm, unsigned int id, __u64 addr) 2655 { 2656 struct s390_io_adapter *adapter = get_io_adapter(kvm, id); 2657 struct s390_map_info *map, *tmp, *map_to_free; 2658 struct page *map_page_to_put = NULL; 2659 u64 map_addr_to_mark = 0; 2660 bool map_pinned = false; 2661 unsigned long flags; 2662 int found = 0, idx; 2663 2664 if (!adapter || !addr) 2665 return -EINVAL; 2666 2667 spin_lock_irqsave(&adapter->maps_lock, flags); 2668 list_for_each_entry_safe(map, tmp, &adapter->maps, list) { 2669 if (map->guest_addr == addr) { 2670 found = 1; 2671 adapter->nr_maps--; 2672 list_del(&map->list); 2673 map_page_to_put = map->page; 2674 map_addr_to_mark = map->guest_addr; 2675 map_pinned = map->pinned; 2676 map_to_free = map; 2677 break; 2678 } 2679 } 2680 spin_unlock_irqrestore(&adapter->maps_lock, flags); 2681 2682 if (found) { 2683 kfree(map_to_free); 2684 if (map_pinned) { 2685 /* 2686 * Only long-term pinned pages need to be marked dirty 2687 * and released. Fallback entries exist only for 2688 * MAP/UNMAP symmetry. 2689 */ 2690 idx = srcu_read_lock(&kvm->srcu); 2691 mark_page_dirty(kvm, map_addr_to_mark >> PAGE_SHIFT); 2692 set_page_dirty_lock(map_page_to_put); 2693 srcu_read_unlock(&kvm->srcu, idx); 2694 unpin_user_page(map_page_to_put); 2695 } 2696 } 2697 2698 return found ? 0 : -ENOENT; 2699 } 2700 2701 void kvm_s390_unmap_all_adapters(struct kvm *kvm) 2702 { 2703 struct s390_map_info *map, *tmp; 2704 unsigned long flags; 2705 int i, idx; 2706 2707 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) { 2708 struct s390_io_adapter *adapter = kvm->arch.adapters[i]; 2709 LIST_HEAD(local_list); 2710 2711 if (!adapter) 2712 continue; 2713 2714 spin_lock_irqsave(&adapter->maps_lock, flags); 2715 list_splice_init(&adapter->maps, &local_list); 2716 adapter->nr_maps = 0; 2717 spin_unlock_irqrestore(&adapter->maps_lock, flags); 2718 2719 list_for_each_entry_safe(map, tmp, &local_list, list) { 2720 list_del(&map->list); 2721 if (map->pinned) { 2722 idx = srcu_read_lock(&kvm->srcu); 2723 mark_page_dirty(kvm, map->guest_addr >> PAGE_SHIFT); 2724 set_page_dirty_lock(map->page); 2725 srcu_read_unlock(&kvm->srcu, idx); 2726 unpin_user_page(map->page); 2727 } 2728 kfree(map); 2729 } 2730 } 2731 } 2732 2733 void kvm_s390_destroy_adapters(struct kvm *kvm) 2734 { 2735 int i; 2736 2737 kvm_s390_unmap_all_adapters(kvm); 2738 2739 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) { 2740 kfree(kvm->arch.adapters[i]); 2741 kvm->arch.adapters[i] = NULL; 2742 } 2743 } 2744 2745 static int modify_io_adapter(struct kvm_device *dev, 2746 struct kvm_device_attr *attr) 2747 { 2748 struct kvm_s390_io_adapter_req req; 2749 struct s390_io_adapter *adapter; 2750 int ret; 2751 2752 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req))) 2753 return -EFAULT; 2754 2755 adapter = get_io_adapter(dev->kvm, req.id); 2756 if (!adapter) 2757 return -EINVAL; 2758 switch (req.type) { 2759 case KVM_S390_IO_ADAPTER_MASK: 2760 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask); 2761 if (ret > 0) 2762 ret = 0; 2763 break; 2764 case KVM_S390_IO_ADAPTER_MAP: 2765 case KVM_S390_IO_ADAPTER_UNMAP: 2766 /* If in Secure Execution mode do not long term pin. */ 2767 mutex_lock(&dev->kvm->lock); 2768 if (kvm_s390_pv_is_protected(dev->kvm)) { 2769 mutex_unlock(&dev->kvm->lock); 2770 return 0; 2771 } 2772 if (req.type == KVM_S390_IO_ADAPTER_MAP) { 2773 dev->kvm->stat.io_390_adapter_map++; 2774 ret = kvm_s390_adapter_map(dev->kvm, req.id, req.addr); 2775 } else { 2776 dev->kvm->stat.io_390_adapter_unmap++; 2777 ret = kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr); 2778 } 2779 mutex_unlock(&dev->kvm->lock); 2780 break; 2781 default: 2782 ret = -EINVAL; 2783 } 2784 2785 return ret; 2786 } 2787 2788 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr) 2789 2790 { 2791 const u64 isc_mask = 0xffUL << 24; /* all iscs set */ 2792 u32 schid; 2793 2794 if (attr->flags) 2795 return -EINVAL; 2796 if (attr->attr != sizeof(schid)) 2797 return -EINVAL; 2798 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid))) 2799 return -EFAULT; 2800 if (!schid) 2801 return -EINVAL; 2802 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid)); 2803 /* 2804 * If userspace is conforming to the architecture, we can have at most 2805 * one pending I/O interrupt per subchannel, so this is effectively a 2806 * clear all. 2807 */ 2808 return 0; 2809 } 2810 2811 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr) 2812 { 2813 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2814 struct kvm_s390_ais_req req; 2815 int ret = 0; 2816 unsigned long flags; 2817 2818 if (!test_kvm_facility(kvm, 72)) 2819 return -EOPNOTSUPP; 2820 2821 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req))) 2822 return -EFAULT; 2823 2824 if (req.isc > MAX_ISC) 2825 return -EINVAL; 2826 2827 trace_kvm_s390_modify_ais_mode(req.isc, 2828 (fi->simm & AIS_MODE_MASK(req.isc)) ? 2829 (fi->nimm & AIS_MODE_MASK(req.isc)) ? 2830 2 : KVM_S390_AIS_MODE_SINGLE : 2831 KVM_S390_AIS_MODE_ALL, req.mode); 2832 2833 spin_lock_irqsave(&fi->ais_lock, flags); 2834 switch (req.mode) { 2835 case KVM_S390_AIS_MODE_ALL: 2836 fi->simm &= ~AIS_MODE_MASK(req.isc); 2837 fi->nimm &= ~AIS_MODE_MASK(req.isc); 2838 break; 2839 case KVM_S390_AIS_MODE_SINGLE: 2840 fi->simm |= AIS_MODE_MASK(req.isc); 2841 fi->nimm &= ~AIS_MODE_MASK(req.isc); 2842 break; 2843 default: 2844 ret = -EINVAL; 2845 } 2846 spin_unlock_irqrestore(&fi->ais_lock, flags); 2847 2848 return ret; 2849 } 2850 2851 static int kvm_s390_inject_airq(struct kvm *kvm, 2852 struct s390_io_adapter *adapter) 2853 { 2854 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2855 struct kvm_s390_interrupt s390int = { 2856 .type = KVM_S390_INT_IO(1, 0, 0, 0), 2857 .parm = 0, 2858 .parm64 = isc_to_int_word(adapter->isc), 2859 }; 2860 struct kvm_s390_interrupt_info *inti; 2861 unsigned long flags; 2862 2863 int ret = 0; 2864 2865 inti = kzalloc_obj(*inti, GFP_KERNEL_ACCOUNT); 2866 if (!inti) 2867 return -ENOMEM; 2868 2869 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible) { 2870 ret = kvm_s390_inject_vm(kvm, &s390int, inti); 2871 if (ret) 2872 kfree(inti); 2873 return ret; 2874 } 2875 2876 spin_lock_irqsave(&fi->ais_lock, flags); 2877 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) { 2878 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc); 2879 spin_unlock_irqrestore(&fi->ais_lock, flags); 2880 kfree(inti); 2881 return ret; 2882 } 2883 2884 ret = kvm_s390_inject_vm(kvm, &s390int, inti); 2885 2886 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) { 2887 fi->nimm |= AIS_MODE_MASK(adapter->isc); 2888 trace_kvm_s390_modify_ais_mode(adapter->isc, 2889 KVM_S390_AIS_MODE_SINGLE, 2); 2890 } 2891 2892 spin_unlock_irqrestore(&fi->ais_lock, flags); 2893 if (ret) 2894 kfree(inti); 2895 return ret; 2896 } 2897 2898 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr) 2899 { 2900 unsigned int id = attr->attr; 2901 struct s390_io_adapter *adapter = get_io_adapter(kvm, id); 2902 2903 kvm->stat.io_flic_inject_airq++; 2904 2905 if (!adapter) 2906 return -EINVAL; 2907 2908 return kvm_s390_inject_airq(kvm, adapter); 2909 } 2910 2911 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr) 2912 { 2913 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 2914 struct kvm_s390_ais_all ais; 2915 unsigned long flags; 2916 2917 if (!test_kvm_facility(kvm, 72)) 2918 return -EOPNOTSUPP; 2919 2920 if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais))) 2921 return -EFAULT; 2922 2923 spin_lock_irqsave(&fi->ais_lock, flags); 2924 fi->simm = ais.simm; 2925 fi->nimm = ais.nimm; 2926 spin_unlock_irqrestore(&fi->ais_lock, flags); 2927 2928 return 0; 2929 } 2930 2931 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr) 2932 { 2933 int r = 0; 2934 unsigned long i; 2935 struct kvm_vcpu *vcpu; 2936 2937 switch (attr->group) { 2938 case KVM_DEV_FLIC_ENQUEUE: 2939 r = enqueue_floating_irq(dev, attr); 2940 break; 2941 case KVM_DEV_FLIC_CLEAR_IRQS: 2942 kvm_s390_clear_float_irqs(dev->kvm); 2943 break; 2944 case KVM_DEV_FLIC_APF_ENABLE: 2945 if (kvm_is_ucontrol(dev->kvm)) 2946 return -EINVAL; 2947 set_bit(GMAP_FLAG_PFAULT_ENABLED, &dev->kvm->arch.gmap->flags); 2948 break; 2949 case KVM_DEV_FLIC_APF_DISABLE_WAIT: 2950 if (kvm_is_ucontrol(dev->kvm)) 2951 return -EINVAL; 2952 clear_bit(GMAP_FLAG_PFAULT_ENABLED, &dev->kvm->arch.gmap->flags); 2953 /* 2954 * Make sure no async faults are in transition when 2955 * clearing the queues. So we don't need to worry 2956 * about late coming workers. 2957 */ 2958 synchronize_srcu(&dev->kvm->srcu); 2959 kvm_for_each_vcpu(i, vcpu, dev->kvm) 2960 kvm_clear_async_pf_completion_queue(vcpu); 2961 break; 2962 case KVM_DEV_FLIC_ADAPTER_REGISTER: 2963 r = register_io_adapter(dev, attr); 2964 break; 2965 case KVM_DEV_FLIC_ADAPTER_MODIFY: 2966 r = modify_io_adapter(dev, attr); 2967 break; 2968 case KVM_DEV_FLIC_CLEAR_IO_IRQ: 2969 r = clear_io_irq(dev->kvm, attr); 2970 break; 2971 case KVM_DEV_FLIC_AISM: 2972 r = modify_ais_mode(dev->kvm, attr); 2973 break; 2974 case KVM_DEV_FLIC_AIRQ_INJECT: 2975 r = flic_inject_airq(dev->kvm, attr); 2976 break; 2977 case KVM_DEV_FLIC_AISM_ALL: 2978 r = flic_ais_mode_set_all(dev->kvm, attr); 2979 break; 2980 default: 2981 r = -EINVAL; 2982 } 2983 2984 return r; 2985 } 2986 2987 static int flic_has_attr(struct kvm_device *dev, 2988 struct kvm_device_attr *attr) 2989 { 2990 switch (attr->group) { 2991 case KVM_DEV_FLIC_GET_ALL_IRQS: 2992 case KVM_DEV_FLIC_ENQUEUE: 2993 case KVM_DEV_FLIC_CLEAR_IRQS: 2994 case KVM_DEV_FLIC_APF_ENABLE: 2995 case KVM_DEV_FLIC_APF_DISABLE_WAIT: 2996 case KVM_DEV_FLIC_ADAPTER_REGISTER: 2997 case KVM_DEV_FLIC_ADAPTER_MODIFY: 2998 case KVM_DEV_FLIC_CLEAR_IO_IRQ: 2999 case KVM_DEV_FLIC_AISM: 3000 case KVM_DEV_FLIC_AIRQ_INJECT: 3001 case KVM_DEV_FLIC_AISM_ALL: 3002 return 0; 3003 } 3004 return -ENXIO; 3005 } 3006 3007 static int flic_create(struct kvm_device *dev, u32 type) 3008 { 3009 if (!dev) 3010 return -EINVAL; 3011 if (dev->kvm->arch.flic) 3012 return -EINVAL; 3013 dev->kvm->arch.flic = dev; 3014 return 0; 3015 } 3016 3017 static void flic_destroy(struct kvm_device *dev) 3018 { 3019 dev->kvm->arch.flic = NULL; 3020 kfree(dev); 3021 } 3022 3023 /* s390 floating irq controller (flic) */ 3024 struct kvm_device_ops kvm_flic_ops = { 3025 .name = "kvm-flic", 3026 .get_attr = flic_get_attr, 3027 .set_attr = flic_set_attr, 3028 .has_attr = flic_has_attr, 3029 .create = flic_create, 3030 .destroy = flic_destroy, 3031 }; 3032 3033 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap) 3034 { 3035 unsigned long bit; 3036 3037 bit = bit_nr + (addr % PAGE_SIZE) * 8; 3038 3039 /* kvm_set_routing_entry() should never allow this to happen */ 3040 WARN_ON_ONCE(bit > (PAGE_SIZE * BITS_PER_BYTE - 1)); 3041 3042 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit; 3043 } 3044 3045 static struct s390_map_info *get_map_info(struct s390_io_adapter *adapter, 3046 u64 addr) 3047 { 3048 struct s390_map_info *map; 3049 3050 if (!adapter) 3051 return NULL; 3052 3053 list_for_each_entry(map, &adapter->maps, list) { 3054 if (map->addr == addr) { 3055 if (!map->pinned) 3056 return NULL; 3057 return map; 3058 } 3059 } 3060 return NULL; 3061 } 3062 3063 static int adapter_indicators_set(struct kvm *kvm, 3064 struct s390_io_adapter *adapter, 3065 struct kvm_s390_adapter_int *adapter_int) 3066 { 3067 struct s390_map_info *ind_info, *summary_info; 3068 struct page *ind_page, *summary_page; 3069 unsigned long bit; 3070 int summary_set; 3071 void *map; 3072 3073 ind_page = NULL; 3074 3075 scoped_guard(spinlock_irqsave, &adapter->maps_lock) { 3076 ind_info = get_map_info(adapter, adapter_int->ind_addr); 3077 if (ind_info) { 3078 map = page_address(ind_info->page); 3079 bit = get_ind_bit(ind_info->addr, adapter_int->ind_offset, adapter->swap); 3080 set_bit(bit, map); 3081 } 3082 } 3083 if (!ind_info) { 3084 ind_page = pin_map_page(kvm, adapter_int->ind_addr, 0); 3085 if (!ind_page) 3086 return -1; 3087 map = page_address(ind_page); 3088 bit = get_ind_bit(adapter_int->ind_addr, 3089 adapter_int->ind_offset, adapter->swap); 3090 set_bit(bit, map); 3091 set_page_dirty_lock(ind_page); 3092 unpin_user_page(ind_page); 3093 } 3094 scoped_guard(srcu, &kvm->srcu) 3095 mark_page_dirty(kvm, gpa_to_gfn(adapter_int->ind_gaddr)); 3096 3097 scoped_guard(spinlock_irqsave, &adapter->maps_lock) { 3098 summary_info = get_map_info(adapter, adapter_int->summary_addr); 3099 if (summary_info) { 3100 map = page_address(summary_info->page); 3101 bit = get_ind_bit(summary_info->addr, adapter_int->summary_offset, 3102 adapter->swap); 3103 summary_set = test_and_set_bit(bit, map); 3104 } 3105 } 3106 if (!summary_info) { 3107 summary_page = pin_map_page(kvm, adapter_int->summary_addr, 0); 3108 if (!summary_page) 3109 return -1; 3110 map = page_address(summary_page); 3111 bit = get_ind_bit(adapter_int->summary_addr, 3112 adapter_int->summary_offset, adapter->swap); 3113 summary_set = test_and_set_bit(bit, map); 3114 set_page_dirty_lock(summary_page); 3115 unpin_user_page(summary_page); 3116 } 3117 scoped_guard(srcu, &kvm->srcu) 3118 mark_page_dirty(kvm, gpa_to_gfn(adapter_int->summary_gaddr)); 3119 3120 return summary_set ? 0 : 1; 3121 } 3122 3123 static int adapter_indicators_set_fast(struct kvm *kvm, 3124 struct s390_io_adapter *adapter, 3125 struct kvm_s390_adapter_int *adapter_int, 3126 int setbit) 3127 { 3128 struct s390_map_info *ind_info, *summary_info; 3129 unsigned long bit; 3130 int summary_set; 3131 void *map; 3132 3133 guard(srcu)(&kvm->srcu); 3134 guard(spinlock)(&adapter->maps_lock); 3135 3136 ind_info = get_map_info(adapter, adapter_int->ind_addr); 3137 if (!ind_info) 3138 return -EWOULDBLOCK; 3139 3140 map = page_address(ind_info->page); 3141 bit = get_ind_bit(ind_info->addr, adapter_int->ind_offset, adapter->swap); 3142 if (setbit) { 3143 set_bit(bit, map); 3144 mark_page_dirty(kvm, gpa_to_gfn(adapter_int->ind_gaddr)); 3145 } 3146 3147 summary_info = get_map_info(adapter, adapter_int->summary_addr); 3148 if (!summary_info) 3149 return -EWOULDBLOCK; 3150 3151 map = page_address(summary_info->page); 3152 bit = get_ind_bit(summary_info->addr, adapter_int->summary_offset, 3153 adapter->swap); 3154 /* If setbit then set summary bit. Else if falling back to the slow path */ 3155 /* with setbit==0 then clear the summary bit so the slow path re-injects */ 3156 if (setbit) 3157 summary_set = test_and_set_bit(bit, map); 3158 else 3159 summary_set = test_and_clear_bit(bit, map); 3160 mark_page_dirty(kvm, gpa_to_gfn(adapter_int->summary_gaddr)); 3161 3162 return summary_set ? 0 : 1; 3163 } 3164 3165 /* 3166 * < 0 - not injected due to error 3167 * = 0 - coalesced, summary indicator already active 3168 * > 0 - injected interrupt 3169 */ 3170 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e, 3171 struct kvm *kvm, int irq_source_id, int level, 3172 bool line_status) 3173 { 3174 int ret; 3175 struct s390_io_adapter *adapter; 3176 3177 kvm->stat.io_set_adapter_int++; 3178 3179 /* We're only interested in the 0->1 transition. */ 3180 if (!level) 3181 return 0; 3182 adapter = get_io_adapter(kvm, e->adapter.adapter_id); 3183 if (!adapter) 3184 return -1; 3185 ret = adapter_indicators_set(kvm, adapter, &e->adapter); 3186 if ((ret > 0) && !adapter->masked) { 3187 ret = kvm_s390_inject_airq(kvm, adapter); 3188 if (ret == 0) 3189 ret = 1; 3190 } 3191 return ret; 3192 } 3193 3194 /* 3195 * Inject the machine check to the guest. 3196 */ 3197 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu, 3198 struct mcck_volatile_info *mcck_info) 3199 { 3200 struct kvm_s390_irq irq = {}; 3201 union mci mci; 3202 __u64 cr14 = 0; /* upper bits are not used */ 3203 int rc; 3204 3205 mci.val = mcck_info->mcic; 3206 3207 /* log machine checks being reinjected on all debugs */ 3208 VCPU_EVENT(vcpu, 2, "guest machine check %lx", mci.val); 3209 KVM_EVENT(2, "guest machine check %lx", mci.val); 3210 pr_info("guest machine check pid %d: %lx", current->pid, mci.val); 3211 3212 if (mci.sr) 3213 cr14 |= CR14_RECOVERY_SUBMASK; 3214 if (mci.dg) 3215 cr14 |= CR14_DEGRADATION_SUBMASK; 3216 if (mci.w) 3217 cr14 |= CR14_WARNING_SUBMASK; 3218 3219 irq.u.mchk.cr14 = cr14; 3220 irq.u.mchk.mcic = mcck_info->mcic; 3221 irq.u.mchk.ext_damage_code = mcck_info->ext_damage_code; 3222 irq.u.mchk.failing_storage_address = mcck_info->failing_storage_address; 3223 3224 /* Inject the machine check to specified vcpu */ 3225 irq.type = KVM_S390_MCHK; 3226 rc = kvm_s390_inject_vcpu(vcpu, &irq); 3227 WARN_ON_ONCE(rc); 3228 } 3229 3230 int kvm_set_routing_entry(struct kvm *kvm, 3231 struct kvm_kernel_irq_routing_entry *e, 3232 const struct kvm_irq_routing_entry *ue) 3233 { 3234 const struct kvm_irq_routing_s390_adapter *adapter; 3235 u64 uaddr_s, uaddr_i; 3236 int idx; 3237 3238 switch (ue->type) { 3239 case KVM_IRQ_ROUTING_S390_ADAPTER: 3240 if (kvm_is_ucontrol(kvm)) 3241 return -EINVAL; 3242 e->set = set_adapter_int; 3243 3244 adapter = &ue->u.adapter; 3245 if (adapter->summary_addr + (adapter->summary_offset / 8) >= 3246 (adapter->summary_addr & PAGE_MASK) + PAGE_SIZE) 3247 return -EINVAL; 3248 if (adapter->ind_addr + (adapter->ind_offset / 8) >= 3249 (adapter->ind_addr & PAGE_MASK) + PAGE_SIZE) 3250 return -EINVAL; 3251 3252 idx = srcu_read_lock(&kvm->srcu); 3253 uaddr_s = gpa_to_hva(kvm, ue->u.adapter.summary_addr); 3254 uaddr_i = gpa_to_hva(kvm, ue->u.adapter.ind_addr); 3255 srcu_read_unlock(&kvm->srcu, idx); 3256 3257 if (kvm_is_error_hva(uaddr_s) || kvm_is_error_hva(uaddr_i)) 3258 return -EFAULT; 3259 e->adapter.summary_addr = uaddr_s; 3260 e->adapter.summary_gaddr = ue->u.adapter.summary_addr; 3261 e->adapter.ind_addr = uaddr_i; 3262 e->adapter.ind_gaddr = ue->u.adapter.ind_addr; 3263 e->adapter.summary_offset = ue->u.adapter.summary_offset; 3264 e->adapter.ind_offset = ue->u.adapter.ind_offset; 3265 e->adapter.adapter_id = ue->u.adapter.adapter_id; 3266 return 0; 3267 default: 3268 return -EINVAL; 3269 } 3270 } 3271 3272 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm, 3273 int irq_source_id, int level, bool line_status) 3274 { 3275 return -EINVAL; 3276 } 3277 3278 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len) 3279 { 3280 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 3281 struct kvm_s390_irq *buf __free(kvfree) = NULL; 3282 bool tmp, storestatus = false; 3283 int r = 0; 3284 int n; 3285 3286 buf = vmalloc(len); 3287 if (!buf) 3288 return -ENOMEM; 3289 3290 if (copy_from_user((void *)buf, irqstate, len)) 3291 return -EFAULT; 3292 3293 scoped_guard(spinlock, &li->lock) { 3294 /* 3295 * Don't allow setting the interrupt state 3296 * when there are already interrupts pending 3297 */ 3298 if (li->pending_irqs) 3299 return -EBUSY; 3300 3301 for (n = 0; n < len / sizeof(*buf); n++) { 3302 tmp = false; 3303 r = do_inject_vcpu(vcpu, &buf[n], &tmp); 3304 if (r == -EWOULDBLOCK && tmp) { 3305 storestatus = true; 3306 r = 0; 3307 } 3308 if (r) 3309 break; 3310 } 3311 } 3312 if (storestatus) { 3313 scoped_guard(srcu, &vcpu->kvm->srcu) 3314 n = kvm_s390_store_status_unloaded(vcpu, KVM_S390_STORE_STATUS_NOADDR); 3315 return r ? r : n; 3316 } 3317 3318 return r; 3319 } 3320 3321 static void store_local_irq(struct kvm_s390_local_interrupt *li, 3322 struct kvm_s390_irq *irq, 3323 unsigned long irq_type) 3324 { 3325 switch (irq_type) { 3326 case IRQ_PEND_MCHK_EX: 3327 case IRQ_PEND_MCHK_REP: 3328 irq->type = KVM_S390_MCHK; 3329 irq->u.mchk = li->irq.mchk; 3330 break; 3331 case IRQ_PEND_PROG: 3332 irq->type = KVM_S390_PROGRAM_INT; 3333 irq->u.pgm = li->irq.pgm; 3334 break; 3335 case IRQ_PEND_PFAULT_INIT: 3336 irq->type = KVM_S390_INT_PFAULT_INIT; 3337 irq->u.ext = li->irq.ext; 3338 break; 3339 case IRQ_PEND_EXT_EXTERNAL: 3340 irq->type = KVM_S390_INT_EXTERNAL_CALL; 3341 irq->u.extcall = li->irq.extcall; 3342 break; 3343 case IRQ_PEND_EXT_CLOCK_COMP: 3344 irq->type = KVM_S390_INT_CLOCK_COMP; 3345 break; 3346 case IRQ_PEND_EXT_CPU_TIMER: 3347 irq->type = KVM_S390_INT_CPU_TIMER; 3348 break; 3349 case IRQ_PEND_SIGP_STOP: 3350 irq->type = KVM_S390_SIGP_STOP; 3351 irq->u.stop = li->irq.stop; 3352 break; 3353 case IRQ_PEND_RESTART: 3354 irq->type = KVM_S390_RESTART; 3355 break; 3356 case IRQ_PEND_SET_PREFIX: 3357 irq->type = KVM_S390_SIGP_SET_PREFIX; 3358 irq->u.prefix = li->irq.prefix; 3359 break; 3360 } 3361 } 3362 3363 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len) 3364 { 3365 int scn; 3366 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS); 3367 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; 3368 unsigned long pending_irqs; 3369 struct kvm_s390_irq irq; 3370 unsigned long irq_type; 3371 int cpuaddr; 3372 int n = 0; 3373 3374 spin_lock(&li->lock); 3375 pending_irqs = li->pending_irqs; 3376 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending, 3377 sizeof(sigp_emerg_pending)); 3378 spin_unlock(&li->lock); 3379 3380 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) { 3381 memset(&irq, 0, sizeof(irq)); 3382 if (irq_type == IRQ_PEND_EXT_EMERGENCY) 3383 continue; 3384 if (n + sizeof(irq) > len) 3385 return -ENOBUFS; 3386 store_local_irq(&vcpu->arch.local_int, &irq, irq_type); 3387 if (copy_to_user(&buf[n], &irq, sizeof(irq))) 3388 return -EFAULT; 3389 n += sizeof(irq); 3390 } 3391 3392 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) { 3393 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) { 3394 memset(&irq, 0, sizeof(irq)); 3395 if (n + sizeof(irq) > len) 3396 return -ENOBUFS; 3397 irq.type = KVM_S390_INT_EMERGENCY; 3398 irq.u.emerg.code = cpuaddr; 3399 if (copy_to_user(&buf[n], &irq, sizeof(irq))) 3400 return -EFAULT; 3401 n += sizeof(irq); 3402 } 3403 } 3404 3405 if (sca_ext_call_pending(vcpu, &scn)) { 3406 if (n + sizeof(irq) > len) 3407 return -ENOBUFS; 3408 memset(&irq, 0, sizeof(irq)); 3409 irq.type = KVM_S390_INT_EXTERNAL_CALL; 3410 irq.u.extcall.code = scn; 3411 if (copy_to_user(&buf[n], &irq, sizeof(irq))) 3412 return -EFAULT; 3413 n += sizeof(irq); 3414 } 3415 3416 return n; 3417 } 3418 3419 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask) 3420 { 3421 int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus); 3422 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3423 struct kvm_vcpu *vcpu; 3424 u8 vcpu_isc_mask; 3425 3426 for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) { 3427 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 3428 if (psw_ioint_disabled(vcpu)) 3429 continue; 3430 vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24); 3431 if (deliverable_mask & vcpu_isc_mask) { 3432 /* lately kicked but not yet running */ 3433 if (test_and_set_bit(vcpu_idx, gi->kicked_mask)) 3434 return; 3435 kvm_s390_vcpu_wakeup(vcpu); 3436 return; 3437 } 3438 } 3439 } 3440 3441 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer) 3442 { 3443 struct kvm_s390_gisa_interrupt *gi = 3444 container_of(timer, struct kvm_s390_gisa_interrupt, timer); 3445 struct kvm *kvm = 3446 container_of(gi->origin, struct sie_page2, gisa)->kvm; 3447 u8 pending_mask; 3448 3449 pending_mask = gisa_get_ipm_or_restore_iam(gi); 3450 if (pending_mask) { 3451 __airqs_kick_single_vcpu(kvm, pending_mask); 3452 hrtimer_forward_now(timer, ns_to_ktime(gi->expires)); 3453 return HRTIMER_RESTART; 3454 } 3455 3456 return HRTIMER_NORESTART; 3457 } 3458 3459 #define NULL_GISA_ADDR 0x00000000UL 3460 #define NONE_GISA_ADDR 0x00000001UL 3461 #define GISA_ADDR_MASK 0xfffff000UL 3462 3463 static void process_gib_alert_list(void) 3464 { 3465 struct kvm_s390_gisa_interrupt *gi; 3466 u32 final, gisa_phys, origin = 0UL; 3467 struct kvm_s390_gisa *gisa; 3468 struct kvm *kvm; 3469 3470 do { 3471 /* 3472 * If the NONE_GISA_ADDR is still stored in the alert list 3473 * origin, we will leave the outer loop. No further GISA has 3474 * been added to the alert list by millicode while processing 3475 * the current alert list. 3476 */ 3477 final = (origin & NONE_GISA_ADDR); 3478 /* 3479 * Cut off the alert list and store the NONE_GISA_ADDR in the 3480 * alert list origin to avoid further GAL interruptions. 3481 * A new alert list can be build up by millicode in parallel 3482 * for guests not in the yet cut-off alert list. When in the 3483 * final loop, store the NULL_GISA_ADDR instead. This will re- 3484 * enable GAL interruptions on the host again. 3485 */ 3486 origin = xchg(&gib->alert_list_origin, 3487 (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR); 3488 /* 3489 * Loop through the just cut-off alert list and start the 3490 * gisa timers to kick idle vcpus to consume the pending 3491 * interruptions asap. 3492 */ 3493 while (origin & GISA_ADDR_MASK) { 3494 gisa_phys = origin; 3495 gisa = phys_to_virt(gisa_phys); 3496 origin = gisa->next_alert; 3497 gisa->next_alert = gisa_phys; 3498 kvm = container_of(gisa, struct sie_page2, gisa)->kvm; 3499 gi = &kvm->arch.gisa_int; 3500 if (hrtimer_active(&gi->timer)) 3501 hrtimer_cancel(&gi->timer); 3502 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL); 3503 } 3504 } while (!final); 3505 3506 } 3507 3508 void kvm_s390_gisa_clear(struct kvm *kvm) 3509 { 3510 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3511 3512 if (!gi->origin) 3513 return; 3514 gisa_clear_ipm(gi->origin); 3515 VM_EVENT(kvm, 3, "gisa 0x%p cleared", gi->origin); 3516 } 3517 3518 void kvm_s390_gisa_init(struct kvm *kvm) 3519 { 3520 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3521 3522 if (!css_general_characteristics.aiv) 3523 return; 3524 gi->origin = &kvm->arch.sie_page2->gisa; 3525 gi->alert.mask = 0; 3526 spin_lock_init(&gi->alert.ref_lock); 3527 gi->expires = 50 * 1000; /* 50 usec */ 3528 hrtimer_setup(&gi->timer, gisa_vcpu_kicker, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 3529 memset(gi->origin, 0, sizeof(struct kvm_s390_gisa)); 3530 gi->origin->next_alert = (u32)virt_to_phys(gi->origin); 3531 VM_EVENT(kvm, 3, "gisa 0x%p initialized", gi->origin); 3532 } 3533 3534 void kvm_s390_gisa_enable(struct kvm *kvm) 3535 { 3536 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3537 struct kvm_vcpu *vcpu; 3538 unsigned long i; 3539 u32 gisa_desc; 3540 3541 if (gi->origin) 3542 return; 3543 kvm_s390_gisa_init(kvm); 3544 gisa_desc = kvm_s390_get_gisa_desc(kvm); 3545 if (!gisa_desc) 3546 return; 3547 kvm_for_each_vcpu(i, vcpu, kvm) { 3548 mutex_lock(&vcpu->mutex); 3549 vcpu->arch.sie_block->gd = gisa_desc; 3550 vcpu->arch.sie_block->eca |= ECA_AIV; 3551 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u", 3552 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id); 3553 mutex_unlock(&vcpu->mutex); 3554 } 3555 } 3556 3557 void kvm_s390_gisa_destroy(struct kvm *kvm) 3558 { 3559 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3560 struct kvm_s390_gisa *gisa = gi->origin; 3561 3562 if (!gi->origin) 3563 return; 3564 WARN(gi->alert.mask != 0x00, 3565 "unexpected non zero alert.mask 0x%02x", 3566 gi->alert.mask); 3567 gi->alert.mask = 0x00; 3568 if (gisa_set_iam(gi->origin, gi->alert.mask)) 3569 process_gib_alert_list(); 3570 hrtimer_cancel(&gi->timer); 3571 gi->origin = NULL; 3572 VM_EVENT(kvm, 3, "gisa 0x%p destroyed", gisa); 3573 } 3574 3575 void kvm_s390_gisa_disable(struct kvm *kvm) 3576 { 3577 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3578 struct kvm_vcpu *vcpu; 3579 unsigned long i; 3580 3581 if (!gi->origin) 3582 return; 3583 kvm_for_each_vcpu(i, vcpu, kvm) { 3584 mutex_lock(&vcpu->mutex); 3585 vcpu->arch.sie_block->eca &= ~ECA_AIV; 3586 vcpu->arch.sie_block->gd = 0U; 3587 mutex_unlock(&vcpu->mutex); 3588 VCPU_EVENT(vcpu, 3, "AIV disabled for cpu %03u", vcpu->vcpu_id); 3589 } 3590 kvm_s390_gisa_destroy(kvm); 3591 } 3592 3593 /** 3594 * kvm_s390_gisc_register - register a guest ISC 3595 * 3596 * @kvm: the kernel vm to work with 3597 * @gisc: the guest interruption sub class to register 3598 * 3599 * The function extends the vm specific alert mask to use. 3600 * The effective IAM mask in the GISA is updated as well 3601 * in case the GISA is not part of the GIB alert list. 3602 * It will be updated latest when the IAM gets restored 3603 * by gisa_get_ipm_or_restore_iam(). 3604 * 3605 * Returns: the nonspecific ISC (NISC) the gib alert mechanism 3606 * has registered with the channel subsystem. 3607 * -ENODEV in case the vm uses no GISA 3608 * -ERANGE in case the guest ISC is invalid 3609 */ 3610 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc) 3611 { 3612 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3613 3614 if (!gi->origin) 3615 return -ENODEV; 3616 if (gisc > MAX_ISC) 3617 return -ERANGE; 3618 3619 spin_lock(&gi->alert.ref_lock); 3620 gi->alert.ref_count[gisc]++; 3621 if (gi->alert.ref_count[gisc] == 1) { 3622 gi->alert.mask |= 0x80 >> gisc; 3623 gisa_set_iam(gi->origin, gi->alert.mask); 3624 } 3625 spin_unlock(&gi->alert.ref_lock); 3626 3627 return gib->nisc; 3628 } 3629 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register); 3630 3631 /** 3632 * kvm_s390_gisc_unregister - unregister a guest ISC 3633 * 3634 * @kvm: the kernel vm to work with 3635 * @gisc: the guest interruption sub class to register 3636 * 3637 * The function reduces the vm specific alert mask to use. 3638 * The effective IAM mask in the GISA is updated as well 3639 * in case the GISA is not part of the GIB alert list. 3640 * It will be updated latest when the IAM gets restored 3641 * by gisa_get_ipm_or_restore_iam(). 3642 * 3643 * Returns: the nonspecific ISC (NISC) the gib alert mechanism 3644 * has registered with the channel subsystem. 3645 * -ENODEV in case the vm uses no GISA 3646 * -ERANGE in case the guest ISC is invalid 3647 * -EINVAL in case the guest ISC is not registered 3648 */ 3649 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc) 3650 { 3651 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int; 3652 int rc = 0; 3653 3654 if (!gi->origin) 3655 return -ENODEV; 3656 if (gisc > MAX_ISC) 3657 return -ERANGE; 3658 3659 spin_lock(&gi->alert.ref_lock); 3660 if (gi->alert.ref_count[gisc] == 0) { 3661 rc = -EINVAL; 3662 goto out; 3663 } 3664 gi->alert.ref_count[gisc]--; 3665 if (gi->alert.ref_count[gisc] == 0) { 3666 gi->alert.mask &= ~(0x80 >> gisc); 3667 gisa_set_iam(gi->origin, gi->alert.mask); 3668 } 3669 out: 3670 spin_unlock(&gi->alert.ref_lock); 3671 3672 return rc; 3673 } 3674 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister); 3675 3676 static void aen_host_forward(unsigned long si) 3677 { 3678 struct kvm_s390_gisa_interrupt *gi; 3679 struct zpci_gaite *gaite; 3680 struct kvm *kvm; 3681 3682 gaite = aift->gait + si; 3683 if (gaite->count == 0) 3684 return; 3685 if (gaite->aisb != 0) 3686 set_bit_inv(gaite->aisbo, phys_to_virt(gaite->aisb)); 3687 3688 kvm = kvm_s390_pci_si_to_kvm(aift, si); 3689 if (!kvm) 3690 return; 3691 gi = &kvm->arch.gisa_int; 3692 3693 if (!(gi->origin->g1.simm & AIS_MODE_MASK(gaite->gisc)) || 3694 !(gi->origin->g1.nimm & AIS_MODE_MASK(gaite->gisc))) { 3695 gisa_set_ipm_gisc(gi->origin, gaite->gisc); 3696 if (hrtimer_active(&gi->timer)) 3697 hrtimer_cancel(&gi->timer); 3698 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL); 3699 kvm->stat.aen_forward++; 3700 } 3701 } 3702 3703 static void aen_process_gait(u8 isc) 3704 { 3705 bool found = false, first = true; 3706 union zpci_sic_iib iib = {{0}}; 3707 unsigned long si, flags; 3708 3709 spin_lock_irqsave(&aift->gait_lock, flags); 3710 3711 if (!aift->gait) { 3712 spin_unlock_irqrestore(&aift->gait_lock, flags); 3713 return; 3714 } 3715 3716 for (si = 0;;) { 3717 /* Scan adapter summary indicator bit vector */ 3718 si = airq_iv_scan(aift->sbv, si, airq_iv_end(aift->sbv)); 3719 if (si == -1UL) { 3720 if (first || found) { 3721 /* Re-enable interrupts. */ 3722 zpci_set_irq_ctrl(SIC_IRQ_MODE_SINGLE, isc, 3723 &iib); 3724 first = found = false; 3725 } else { 3726 /* Interrupts on and all bits processed */ 3727 break; 3728 } 3729 found = false; 3730 si = 0; 3731 /* Scan again after re-enabling interrupts */ 3732 continue; 3733 } 3734 found = true; 3735 aen_host_forward(si); 3736 } 3737 3738 spin_unlock_irqrestore(&aift->gait_lock, flags); 3739 } 3740 3741 static void gib_alert_irq_handler(struct airq_struct *airq, 3742 struct tpi_info *tpi_info) 3743 { 3744 struct tpi_adapter_info *info = (struct tpi_adapter_info *)tpi_info; 3745 3746 inc_irq_stat(IRQIO_GAL); 3747 3748 if ((info->forward || info->error) && 3749 IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) { 3750 aen_process_gait(info->isc); 3751 if (info->aism != 0) 3752 process_gib_alert_list(); 3753 } else { 3754 process_gib_alert_list(); 3755 } 3756 } 3757 3758 static struct airq_struct gib_alert_irq = { 3759 .handler = gib_alert_irq_handler, 3760 }; 3761 3762 void kvm_s390_gib_destroy(void) 3763 { 3764 if (!gib) 3765 return; 3766 if (kvm_s390_pci_interp_allowed() && aift) { 3767 mutex_lock(&aift->aift_lock); 3768 kvm_s390_pci_aen_exit(); 3769 mutex_unlock(&aift->aift_lock); 3770 } 3771 chsc_sgib(0); 3772 unregister_adapter_interrupt(&gib_alert_irq); 3773 kfree(gib); 3774 gib = NULL; 3775 } 3776 3777 int __init kvm_s390_gib_init(u8 nisc) 3778 { 3779 u32 gib_origin; 3780 int rc = 0; 3781 3782 if (!css_general_characteristics.aiv) { 3783 KVM_EVENT(3, "%s", "gib not initialized, no AIV facility"); 3784 goto out; 3785 } 3786 3787 gib = kzalloc(PAGE_SIZE, GFP_KERNEL_ACCOUNT | GFP_DMA); 3788 if (!gib) { 3789 rc = -ENOMEM; 3790 goto out; 3791 } 3792 3793 gib_alert_irq.isc = nisc; 3794 if (register_adapter_interrupt(&gib_alert_irq)) { 3795 pr_err("Registering the GIB alert interruption handler failed\n"); 3796 rc = -EIO; 3797 goto out_free_gib; 3798 } 3799 /* adapter interrupts used for AP (applicable here) don't use the LSI */ 3800 *gib_alert_irq.lsi_ptr = 0xff; 3801 3802 gib->nisc = nisc; 3803 gib_origin = virt_to_phys(gib); 3804 if (chsc_sgib(gib_origin)) { 3805 pr_err("Associating the GIB with the AIV facility failed\n"); 3806 kfree(gib); 3807 gib = NULL; 3808 rc = -EIO; 3809 goto out_unreg_gal; 3810 } 3811 3812 if (kvm_s390_pci_interp_allowed()) { 3813 if (kvm_s390_pci_aen_init(nisc)) { 3814 pr_err("Initializing AEN for PCI failed\n"); 3815 rc = -EIO; 3816 goto out_unreg_gal; 3817 } 3818 } 3819 3820 KVM_EVENT(3, "gib 0x%p (nisc=%d) initialized", gib, gib->nisc); 3821 goto out; 3822 3823 out_unreg_gal: 3824 unregister_adapter_interrupt(&gib_alert_irq); 3825 out_free_gib: 3826 kfree(gib); 3827 gib = NULL; 3828 out: 3829 return rc; 3830 } 3831 3832 /* 3833 * kvm_arch_set_irq_inatomic: fast-path for irqfd injection 3834 */ 3835 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e, 3836 struct kvm *kvm, int irq_source_id, int level, 3837 bool line_status) 3838 { 3839 int ret, setbit; 3840 struct s390_io_adapter *adapter; 3841 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int; 3842 struct kvm_s390_interrupt_info *inti; 3843 struct kvm_s390_interrupt s390int = { 3844 .type = KVM_S390_INT_IO(1, 0, 0, 0), 3845 .parm = 0, 3846 }; 3847 3848 kvm->stat.io_390_inatomic++; 3849 3850 /* We're only interested in the 0->1 transition. */ 3851 if (!level) 3852 return 0; 3853 if (e->type != KVM_IRQ_ROUTING_S390_ADAPTER) 3854 return -EWOULDBLOCK; 3855 3856 adapter = get_io_adapter(kvm, e->adapter.adapter_id); 3857 if (!adapter) 3858 return -EWOULDBLOCK; 3859 3860 s390int.parm64 = isc_to_int_word(adapter->isc); 3861 setbit = 1; 3862 ret = adapter_indicators_set_fast(kvm, adapter, &e->adapter, setbit); 3863 if (ret < 0) 3864 return -EWOULDBLOCK; 3865 if (!ret || adapter->masked) { 3866 kvm->stat.io_390_inatomic_no_inject++; 3867 return 0; 3868 } 3869 3870 inti = kzalloc_obj(*inti, GFP_ATOMIC); 3871 if (!inti) { 3872 setbit = 0; 3873 adapter_indicators_set_fast(kvm, adapter, &e->adapter, setbit); 3874 return -EWOULDBLOCK; 3875 } 3876 3877 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible) { 3878 ret = kvm_s390_inject_vm(kvm, &s390int, inti); 3879 if (ret == 0) { 3880 return ret; 3881 } else { 3882 setbit = 0; 3883 adapter_indicators_set_fast(kvm, adapter, &e->adapter, setbit); 3884 kfree(inti); 3885 return -EWOULDBLOCK; 3886 } 3887 } 3888 3889 spin_lock(&fi->ais_lock); 3890 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) { 3891 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc); 3892 spin_unlock(&fi->ais_lock); 3893 kfree(inti); 3894 kvm->stat.io_390_inatomic_no_inject++; 3895 return 0; 3896 } 3897 3898 ret = kvm_s390_inject_vm(kvm, &s390int, inti); 3899 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) { 3900 fi->nimm |= AIS_MODE_MASK(adapter->isc); 3901 trace_kvm_s390_modify_ais_mode(adapter->isc, 3902 KVM_S390_AIS_MODE_SINGLE, 2); 3903 } else if (ret) { 3904 spin_unlock(&fi->ais_lock); 3905 setbit = 0; 3906 adapter_indicators_set_fast(kvm, adapter, &e->adapter, setbit); 3907 kfree(inti); 3908 return -EWOULDBLOCK; 3909 } 3910 3911 spin_unlock(&fi->ais_lock); 3912 return 0; 3913 } 3914