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