1 // SPDX-License-Identifier: GPL-2.0
3 * handling kvm guest interrupts
5 * Copyright IBM Corp. 2008, 2020
7 * Author(s): Carsten Otte <cotte@de.ibm.com>
10 #define KMSG_COMPONENT "kvm-s390"
11 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
13 #include <linux/interrupt.h>
14 #include <linux/kvm_host.h>
15 #include <linux/hrtimer.h>
16 #include <linux/mmu_context.h>
17 #include <linux/nospec.h>
18 #include <linux/signal.h>
19 #include <linux/slab.h>
20 #include <linux/bitmap.h>
21 #include <linux/vmalloc.h>
22 #include <asm/asm-offsets.h>
24 #include <linux/uaccess.h>
28 #include <asm/switch_to.h>
33 #include "trace-s390.h"
35 #define PFAULT_INIT 0x0600
36 #define PFAULT_DONE 0x0680
37 #define VIRTIO_PARAM 0x0d00
39 static struct kvm_s390_gib *gib;
41 /* handle external calls via sigp interpretation facility */
42 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
46 if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND))
49 BUG_ON(!kvm_s390_use_sca_entries());
50 read_lock(&vcpu->kvm->arch.sca_lock);
51 if (vcpu->kvm->arch.use_esca) {
52 struct esca_block *sca = vcpu->kvm->arch.sca;
53 union esca_sigp_ctrl sigp_ctrl =
54 sca->cpu[vcpu->vcpu_id].sigp_ctrl;
59 struct bsca_block *sca = vcpu->kvm->arch.sca;
60 union bsca_sigp_ctrl sigp_ctrl =
61 sca->cpu[vcpu->vcpu_id].sigp_ctrl;
66 read_unlock(&vcpu->kvm->arch.sca_lock);
74 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
78 BUG_ON(!kvm_s390_use_sca_entries());
79 read_lock(&vcpu->kvm->arch.sca_lock);
80 if (vcpu->kvm->arch.use_esca) {
81 struct esca_block *sca = vcpu->kvm->arch.sca;
82 union esca_sigp_ctrl *sigp_ctrl =
83 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
84 union esca_sigp_ctrl new_val = {0}, old_val;
86 old_val = READ_ONCE(*sigp_ctrl);
91 expect = old_val.value;
92 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
94 struct bsca_block *sca = vcpu->kvm->arch.sca;
95 union bsca_sigp_ctrl *sigp_ctrl =
96 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
97 union bsca_sigp_ctrl new_val = {0}, old_val;
99 old_val = READ_ONCE(*sigp_ctrl);
100 new_val.scn = src_id;
104 expect = old_val.value;
105 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
107 read_unlock(&vcpu->kvm->arch.sca_lock);
110 /* another external call is pending */
113 kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
117 static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
121 if (!kvm_s390_use_sca_entries())
123 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
124 read_lock(&vcpu->kvm->arch.sca_lock);
125 if (vcpu->kvm->arch.use_esca) {
126 struct esca_block *sca = vcpu->kvm->arch.sca;
127 union esca_sigp_ctrl *sigp_ctrl =
128 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
129 union esca_sigp_ctrl old;
131 old = READ_ONCE(*sigp_ctrl);
133 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
135 struct bsca_block *sca = vcpu->kvm->arch.sca;
136 union bsca_sigp_ctrl *sigp_ctrl =
137 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
138 union bsca_sigp_ctrl old;
140 old = READ_ONCE(*sigp_ctrl);
142 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
144 read_unlock(&vcpu->kvm->arch.sca_lock);
145 WARN_ON(rc != expect); /* cannot clear? */
148 int psw_extint_disabled(struct kvm_vcpu *vcpu)
150 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
153 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
155 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
158 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
160 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
163 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
165 return psw_extint_disabled(vcpu) &&
166 psw_ioint_disabled(vcpu) &&
167 psw_mchk_disabled(vcpu);
170 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
172 if (psw_extint_disabled(vcpu) ||
173 !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
175 if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
176 /* No timer interrupts when single stepping */
181 static int ckc_irq_pending(struct kvm_vcpu *vcpu)
183 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
184 const u64 ckc = vcpu->arch.sie_block->ckc;
186 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
187 if ((s64)ckc >= (s64)now)
189 } else if (ckc >= now) {
192 return ckc_interrupts_enabled(vcpu);
195 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
197 return !psw_extint_disabled(vcpu) &&
198 (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK);
201 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
203 if (!cpu_timer_interrupts_enabled(vcpu))
205 return kvm_s390_get_cpu_timer(vcpu) >> 63;
208 static uint64_t isc_to_isc_bits(int isc)
210 return (0x80 >> isc) << 24;
213 static inline u32 isc_to_int_word(u8 isc)
215 return ((u32)isc << 27) | 0x80000000;
218 static inline u8 int_word_to_isc(u32 int_word)
220 return (int_word & 0x38000000) >> 27;
224 * To use atomic bitmap functions, we have to provide a bitmap address
225 * that is u64 aligned. However, the ipm might be u32 aligned.
226 * Therefore, we logically start the bitmap at the very beginning of the
227 * struct and fixup the bit number.
229 #define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE)
232 * gisa_set_iam - change the GISA interruption alert mask
234 * @gisa: gisa to operate on
235 * @iam: new IAM value to use
237 * Change the IAM atomically with the next alert address and the IPM
238 * of the GISA if the GISA is not part of the GIB alert list. All three
239 * fields are located in the first long word of the GISA.
241 * Returns: 0 on success
242 * -EBUSY in case the gisa is part of the alert list
244 static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam)
249 word = READ_ONCE(gisa->u64.word[0]);
250 if ((u64)gisa != word >> 32)
252 _word = (word & ~0xffUL) | iam;
253 } while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
259 * gisa_clear_ipm - clear the GISA interruption pending mask
261 * @gisa: gisa to operate on
263 * Clear the IPM atomically with the next alert address and the IAM
264 * of the GISA unconditionally. All three fields are located in the
265 * first long word of the GISA.
267 static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa)
272 word = READ_ONCE(gisa->u64.word[0]);
273 _word = word & ~(0xffUL << 24);
274 } while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
278 * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM
280 * @gi: gisa interrupt struct to work on
282 * Atomically restores the interruption alert mask if none of the
283 * relevant ISCs are pending and return the IPM.
285 * Returns: the relevant pending ISCs
287 static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
289 u8 pending_mask, alert_mask;
293 word = READ_ONCE(gi->origin->u64.word[0]);
294 alert_mask = READ_ONCE(gi->alert.mask);
295 pending_mask = (u8)(word >> 24) & alert_mask;
298 _word = (word & ~0xffUL) | alert_mask;
299 } while (cmpxchg(&gi->origin->u64.word[0], word, _word) != word);
304 static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
306 return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa;
309 static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
311 set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
314 static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
316 return READ_ONCE(gisa->ipm);
319 static inline void gisa_clear_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
321 clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
324 static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
326 return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
329 static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
331 unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs |
332 vcpu->arch.local_int.pending_irqs;
334 pending &= ~vcpu->kvm->arch.float_int.masked_irqs;
338 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
340 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
341 unsigned long pending_mask;
343 pending_mask = pending_irqs_no_gisa(vcpu);
345 pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
349 static inline int isc_to_irq_type(unsigned long isc)
351 return IRQ_PEND_IO_ISC_0 - isc;
354 static inline int irq_type_to_isc(unsigned long irq_type)
356 return IRQ_PEND_IO_ISC_0 - irq_type;
359 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
360 unsigned long active_mask)
364 for (i = 0; i <= MAX_ISC; i++)
365 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
366 active_mask &= ~(1UL << (isc_to_irq_type(i)));
371 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
373 unsigned long active_mask;
375 active_mask = pending_irqs(vcpu);
379 if (psw_extint_disabled(vcpu))
380 active_mask &= ~IRQ_PEND_EXT_MASK;
381 if (psw_ioint_disabled(vcpu))
382 active_mask &= ~IRQ_PEND_IO_MASK;
384 active_mask = disable_iscs(vcpu, active_mask);
385 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
386 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
387 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
388 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
389 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
390 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
391 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
392 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
393 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) {
394 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
395 __clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask);
397 if (psw_mchk_disabled(vcpu))
398 active_mask &= ~IRQ_PEND_MCHK_MASK;
399 /* PV guest cpus can have a single interruption injected at a time. */
400 if (kvm_s390_pv_cpu_get_handle(vcpu) &&
401 vcpu->arch.sie_block->iictl != IICTL_CODE_NONE)
402 active_mask &= ~(IRQ_PEND_EXT_II_MASK |
406 * Check both floating and local interrupt's cr14 because
407 * bit IRQ_PEND_MCHK_REP could be set in both cases.
409 if (!(vcpu->arch.sie_block->gcr[14] &
410 (vcpu->kvm->arch.float_int.mchk.cr14 |
411 vcpu->arch.local_int.irq.mchk.cr14)))
412 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
415 * STOP irqs will never be actively delivered. They are triggered via
416 * intercept requests and cleared when the stop intercept is performed.
418 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
423 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
425 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
426 set_bit(kvm_vcpu_get_idx(vcpu), vcpu->kvm->arch.idle_mask);
429 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
431 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
432 clear_bit(kvm_vcpu_get_idx(vcpu), vcpu->kvm->arch.idle_mask);
435 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
437 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
439 vcpu->arch.sie_block->lctl = 0x0000;
440 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
442 if (guestdbg_enabled(vcpu)) {
443 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
444 LCTL_CR10 | LCTL_CR11);
445 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
449 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
451 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
453 if (psw_ioint_disabled(vcpu))
454 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
456 vcpu->arch.sie_block->lctl |= LCTL_CR6;
459 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
461 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
463 if (psw_extint_disabled(vcpu))
464 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
466 vcpu->arch.sie_block->lctl |= LCTL_CR0;
469 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
471 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
473 if (psw_mchk_disabled(vcpu))
474 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
476 vcpu->arch.sie_block->lctl |= LCTL_CR14;
479 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
481 if (kvm_s390_is_stop_irq_pending(vcpu))
482 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
485 /* Set interception request for non-deliverable interrupts */
486 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
488 set_intercept_indicators_io(vcpu);
489 set_intercept_indicators_ext(vcpu);
490 set_intercept_indicators_mchk(vcpu);
491 set_intercept_indicators_stop(vcpu);
494 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
496 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
499 vcpu->stat.deliver_cputm++;
500 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
502 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
503 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
504 vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER;
506 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
507 (u16 *)__LC_EXT_INT_CODE);
508 rc |= put_guest_lc(vcpu, 0, (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));
514 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
515 return rc ? -EFAULT : 0;
518 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
520 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
523 vcpu->stat.deliver_ckc++;
524 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
526 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
527 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
528 vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP;
530 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
531 (u16 __user *)__LC_EXT_INT_CODE);
532 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
533 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
534 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
535 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
536 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
538 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
539 return rc ? -EFAULT : 0;
542 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
544 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
545 struct kvm_s390_ext_info ext;
548 spin_lock(&li->lock);
550 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
551 li->irq.ext.ext_params2 = 0;
552 spin_unlock(&li->lock);
554 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
556 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
557 KVM_S390_INT_PFAULT_INIT,
560 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
561 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
562 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
563 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
564 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
565 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
566 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
567 return rc ? -EFAULT : 0;
570 static int __write_machine_check(struct kvm_vcpu *vcpu,
571 struct kvm_s390_mchk_info *mchk)
573 unsigned long ext_sa_addr;
575 freg_t fprs[NUM_FPRS];
580 * All other possible payload for a machine check (e.g. the register
581 * contents in the save area) will be handled by the ultravisor, as
582 * the hypervisor does not not have the needed information for
585 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
586 vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK;
587 vcpu->arch.sie_block->mcic = mchk->mcic;
588 vcpu->arch.sie_block->faddr = mchk->failing_storage_address;
589 vcpu->arch.sie_block->edc = mchk->ext_damage_code;
593 mci.val = mchk->mcic;
594 /* take care of lazy register loading */
596 save_access_regs(vcpu->run->s.regs.acrs);
597 if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
598 save_gs_cb(current->thread.gs_cb);
600 /* Extended save area */
601 rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
602 sizeof(unsigned long));
603 /* Only bits 0 through 63-LC are used for address formation */
604 lc = ext_sa_addr & MCESA_LC_MASK;
605 if (test_kvm_facility(vcpu->kvm, 133)) {
609 ext_sa_addr &= ~0x3ffUL;
612 ext_sa_addr &= ~0x7ffUL;
615 ext_sa_addr &= ~0xfffUL;
622 ext_sa_addr &= ~0x3ffUL;
625 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
626 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
632 if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
633 && (lc == 11 || lc == 12)) {
634 if (write_guest_abs(vcpu, ext_sa_addr + 1024,
635 &vcpu->run->s.regs.gscb, 32))
641 /* General interruption information */
642 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
643 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
644 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
645 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
646 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
647 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
649 /* Register-save areas */
650 if (MACHINE_HAS_VX) {
651 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
652 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
654 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
655 vcpu->run->s.regs.fprs, 128);
657 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
658 vcpu->run->s.regs.gprs, 128);
659 rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
660 (u32 __user *) __LC_FP_CREG_SAVE_AREA);
661 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
662 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
663 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
664 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
665 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
666 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
667 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
668 &vcpu->run->s.regs.acrs, 64);
669 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
670 &vcpu->arch.sie_block->gcr, 128);
672 /* Extended interruption information */
673 rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
674 (u32 __user *) __LC_EXT_DAMAGE_CODE);
675 rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
676 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
677 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
678 sizeof(mchk->fixed_logout));
679 return rc ? -EFAULT : 0;
682 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
684 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
685 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
686 struct kvm_s390_mchk_info mchk = {};
690 spin_lock(&fi->lock);
691 spin_lock(&li->lock);
692 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
693 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
695 * If there was an exigent machine check pending, then any
696 * repressible machine checks that might have been pending
697 * are indicated along with it, so always clear bits for
698 * repressible and exigent interrupts
701 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
702 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
703 memset(&li->irq.mchk, 0, sizeof(mchk));
707 * We indicate floating repressible conditions along with
708 * other pending conditions. Channel Report Pending and Channel
709 * Subsystem damage are the only two and and are indicated by
710 * bits in mcic and masked in cr14.
712 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
713 mchk.mcic |= fi->mchk.mcic;
714 mchk.cr14 |= fi->mchk.cr14;
715 memset(&fi->mchk, 0, sizeof(mchk));
718 spin_unlock(&li->lock);
719 spin_unlock(&fi->lock);
722 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
724 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
726 mchk.cr14, mchk.mcic);
727 vcpu->stat.deliver_machine_check++;
728 rc = __write_machine_check(vcpu, &mchk);
733 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
735 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
738 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
739 vcpu->stat.deliver_restart_signal++;
740 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
742 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
743 vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART;
745 rc = write_guest_lc(vcpu,
746 offsetof(struct lowcore, restart_old_psw),
747 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
748 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
749 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
751 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
752 return rc ? -EFAULT : 0;
755 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
757 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
758 struct kvm_s390_prefix_info prefix;
760 spin_lock(&li->lock);
761 prefix = li->irq.prefix;
762 li->irq.prefix.address = 0;
763 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
764 spin_unlock(&li->lock);
766 vcpu->stat.deliver_prefix_signal++;
767 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
768 KVM_S390_SIGP_SET_PREFIX,
771 kvm_s390_set_prefix(vcpu, prefix.address);
775 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
777 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
781 spin_lock(&li->lock);
782 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
783 clear_bit(cpu_addr, li->sigp_emerg_pending);
784 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
785 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
786 spin_unlock(&li->lock);
788 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
789 vcpu->stat.deliver_emergency_signal++;
790 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
792 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
793 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
794 vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG;
795 vcpu->arch.sie_block->extcpuaddr = cpu_addr;
799 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
800 (u16 *)__LC_EXT_INT_CODE);
801 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
802 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
803 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
804 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
805 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
806 return rc ? -EFAULT : 0;
809 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
811 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
812 struct kvm_s390_extcall_info extcall;
815 spin_lock(&li->lock);
816 extcall = li->irq.extcall;
817 li->irq.extcall.code = 0;
818 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
819 spin_unlock(&li->lock);
821 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
822 vcpu->stat.deliver_external_call++;
823 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
824 KVM_S390_INT_EXTERNAL_CALL,
826 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
827 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
828 vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL;
829 vcpu->arch.sie_block->extcpuaddr = extcall.code;
833 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
834 (u16 *)__LC_EXT_INT_CODE);
835 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
836 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
837 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
838 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
840 return rc ? -EFAULT : 0;
843 static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code)
846 case PGM_SPECIFICATION:
847 vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION;
850 vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND;
858 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
860 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
861 struct kvm_s390_pgm_info pgm_info;
862 int rc = 0, nullifying = false;
865 spin_lock(&li->lock);
866 pgm_info = li->irq.pgm;
867 clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
868 memset(&li->irq.pgm, 0, sizeof(pgm_info));
869 spin_unlock(&li->lock);
871 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
872 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
873 pgm_info.code, ilen);
874 vcpu->stat.deliver_program++;
875 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
878 /* PER is handled by the ultravisor */
879 if (kvm_s390_pv_cpu_is_protected(vcpu))
880 return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER);
882 switch (pgm_info.code & ~PGM_PER) {
883 case PGM_AFX_TRANSLATION:
884 case PGM_ASX_TRANSLATION:
885 case PGM_EX_TRANSLATION:
886 case PGM_LFX_TRANSLATION:
887 case PGM_LSTE_SEQUENCE:
888 case PGM_LSX_TRANSLATION:
889 case PGM_LX_TRANSLATION:
890 case PGM_PRIMARY_AUTHORITY:
891 case PGM_SECONDARY_AUTHORITY:
894 case PGM_SPACE_SWITCH:
895 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
896 (u64 *)__LC_TRANS_EXC_CODE);
898 case PGM_ALEN_TRANSLATION:
899 case PGM_ALE_SEQUENCE:
900 case PGM_ASTE_INSTANCE:
901 case PGM_ASTE_SEQUENCE:
902 case PGM_ASTE_VALIDITY:
903 case PGM_EXTENDED_AUTHORITY:
904 rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
905 (u8 *)__LC_EXC_ACCESS_ID);
909 case PGM_PAGE_TRANSLATION:
910 case PGM_REGION_FIRST_TRANS:
911 case PGM_REGION_SECOND_TRANS:
912 case PGM_REGION_THIRD_TRANS:
913 case PGM_SEGMENT_TRANSLATION:
914 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
915 (u64 *)__LC_TRANS_EXC_CODE);
916 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
917 (u8 *)__LC_EXC_ACCESS_ID);
918 rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
919 (u8 *)__LC_OP_ACCESS_ID);
923 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
924 (u16 *)__LC_MON_CLASS_NR);
925 rc |= put_guest_lc(vcpu, pgm_info.mon_code,
926 (u64 *)__LC_MON_CODE);
928 case PGM_VECTOR_PROCESSING:
930 rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
931 (u32 *)__LC_DATA_EXC_CODE);
934 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
935 (u64 *)__LC_TRANS_EXC_CODE);
936 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
937 (u8 *)__LC_EXC_ACCESS_ID);
940 case PGM_STACK_EMPTY:
941 case PGM_STACK_SPECIFICATION:
943 case PGM_STACK_OPERATION:
944 case PGM_TRACE_TABEL:
945 case PGM_CRYPTO_OPERATION:
950 if (pgm_info.code & PGM_PER) {
951 rc |= put_guest_lc(vcpu, pgm_info.per_code,
952 (u8 *) __LC_PER_CODE);
953 rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
954 (u8 *)__LC_PER_ATMID);
955 rc |= put_guest_lc(vcpu, pgm_info.per_address,
956 (u64 *) __LC_PER_ADDRESS);
957 rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
958 (u8 *) __LC_PER_ACCESS_ID);
961 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
962 kvm_s390_rewind_psw(vcpu, ilen);
964 /* bit 1+2 of the target are the ilc, so we can directly use ilen */
965 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
966 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
967 (u64 *) __LC_LAST_BREAK);
968 rc |= put_guest_lc(vcpu, pgm_info.code,
969 (u16 *)__LC_PGM_INT_CODE);
970 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
971 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
972 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
973 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
974 return rc ? -EFAULT : 0;
977 #define SCCB_MASK 0xFFFFFFF8
978 #define SCCB_EVENT_PENDING 0x3
980 static int write_sclp(struct kvm_vcpu *vcpu, u32 parm)
984 if (kvm_s390_pv_cpu_get_handle(vcpu)) {
985 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
986 vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG;
987 vcpu->arch.sie_block->eiparams = parm;
991 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
992 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
993 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
994 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
995 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
996 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
997 rc |= put_guest_lc(vcpu, parm,
998 (u32 *)__LC_EXT_PARAMS);
1000 return rc ? -EFAULT : 0;
1003 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
1005 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1006 struct kvm_s390_ext_info ext;
1008 spin_lock(&fi->lock);
1009 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) ||
1010 !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
1011 spin_unlock(&fi->lock);
1014 ext = fi->srv_signal;
1015 memset(&fi->srv_signal, 0, sizeof(ext));
1016 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1017 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1018 if (kvm_s390_pv_cpu_is_protected(vcpu))
1019 set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs);
1020 spin_unlock(&fi->lock);
1022 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
1024 vcpu->stat.deliver_service_signal++;
1025 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1028 return write_sclp(vcpu, ext.ext_params);
1031 static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)
1033 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1034 struct kvm_s390_ext_info ext;
1036 spin_lock(&fi->lock);
1037 if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) {
1038 spin_unlock(&fi->lock);
1041 ext = fi->srv_signal;
1042 /* only clear the event bit */
1043 fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING;
1044 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1045 spin_unlock(&fi->lock);
1047 VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event");
1048 vcpu->stat.deliver_service_signal++;
1049 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1052 return write_sclp(vcpu, SCCB_EVENT_PENDING);
1055 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
1057 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1058 struct kvm_s390_interrupt_info *inti;
1061 spin_lock(&fi->lock);
1062 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
1063 struct kvm_s390_interrupt_info,
1066 list_del(&inti->list);
1067 fi->counters[FIRQ_CNTR_PFAULT] -= 1;
1069 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
1070 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1071 spin_unlock(&fi->lock);
1074 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1075 KVM_S390_INT_PFAULT_DONE, 0,
1076 inti->ext.ext_params2);
1077 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
1078 inti->ext.ext_params2);
1080 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1081 (u16 *)__LC_EXT_INT_CODE);
1082 rc |= put_guest_lc(vcpu, PFAULT_DONE,
1083 (u16 *)__LC_EXT_CPU_ADDR);
1084 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1085 &vcpu->arch.sie_block->gpsw,
1087 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1088 &vcpu->arch.sie_block->gpsw,
1090 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1091 (u64 *)__LC_EXT_PARAMS2);
1094 return rc ? -EFAULT : 0;
1097 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
1099 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1100 struct kvm_s390_interrupt_info *inti;
1103 spin_lock(&fi->lock);
1104 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
1105 struct kvm_s390_interrupt_info,
1109 "deliver: virtio parm: 0x%x,parm64: 0x%llx",
1110 inti->ext.ext_params, inti->ext.ext_params2);
1111 vcpu->stat.deliver_virtio++;
1112 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1114 inti->ext.ext_params,
1115 inti->ext.ext_params2);
1116 list_del(&inti->list);
1117 fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1119 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1120 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1121 spin_unlock(&fi->lock);
1124 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1125 (u16 *)__LC_EXT_INT_CODE);
1126 rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1127 (u16 *)__LC_EXT_CPU_ADDR);
1128 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1129 &vcpu->arch.sie_block->gpsw,
1131 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1132 &vcpu->arch.sie_block->gpsw,
1134 rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1135 (u32 *)__LC_EXT_PARAMS);
1136 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1137 (u64 *)__LC_EXT_PARAMS2);
1140 return rc ? -EFAULT : 0;
1143 static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1147 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
1148 vcpu->arch.sie_block->iictl = IICTL_CODE_IO;
1149 vcpu->arch.sie_block->subchannel_id = io->subchannel_id;
1150 vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr;
1151 vcpu->arch.sie_block->io_int_parm = io->io_int_parm;
1152 vcpu->arch.sie_block->io_int_word = io->io_int_word;
1156 rc = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1157 rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1158 rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1159 rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1160 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1161 &vcpu->arch.sie_block->gpsw,
1163 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1164 &vcpu->arch.sie_block->gpsw,
1166 return rc ? -EFAULT : 0;
1169 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1170 unsigned long irq_type)
1172 struct list_head *isc_list;
1173 struct kvm_s390_float_interrupt *fi;
1174 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1175 struct kvm_s390_interrupt_info *inti = NULL;
1176 struct kvm_s390_io_info io;
1180 fi = &vcpu->kvm->arch.float_int;
1182 spin_lock(&fi->lock);
1183 isc = irq_type_to_isc(irq_type);
1184 isc_list = &fi->lists[isc];
1185 inti = list_first_entry_or_null(isc_list,
1186 struct kvm_s390_interrupt_info,
1189 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1190 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1192 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1193 inti->io.subchannel_id >> 8,
1194 inti->io.subchannel_id >> 1 & 0x3,
1195 inti->io.subchannel_nr);
1197 vcpu->stat.deliver_io++;
1198 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1200 ((__u32)inti->io.subchannel_id << 16) |
1201 inti->io.subchannel_nr,
1202 ((__u64)inti->io.io_int_parm << 32) |
1203 inti->io.io_int_word);
1204 list_del(&inti->list);
1205 fi->counters[FIRQ_CNTR_IO] -= 1;
1207 if (list_empty(isc_list))
1208 clear_bit(irq_type, &fi->pending_irqs);
1209 spin_unlock(&fi->lock);
1212 rc = __do_deliver_io(vcpu, &(inti->io));
1217 if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1219 * in case an adapter interrupt was not delivered
1220 * in SIE context KVM will handle the delivery
1222 VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1223 memset(&io, 0, sizeof(io));
1224 io.io_int_word = isc_to_int_word(isc);
1225 vcpu->stat.deliver_io++;
1226 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1227 KVM_S390_INT_IO(1, 0, 0, 0),
1228 ((__u32)io.subchannel_id << 16) |
1230 ((__u64)io.io_int_parm << 32) |
1232 rc = __do_deliver_io(vcpu, &io);
1238 /* Check whether an external call is pending (deliverable or not) */
1239 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1241 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1243 if (!sclp.has_sigpif)
1244 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1246 return sca_ext_call_pending(vcpu, NULL);
1249 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1251 if (deliverable_irqs(vcpu))
1254 if (kvm_cpu_has_pending_timer(vcpu))
1257 /* external call pending and deliverable */
1258 if (kvm_s390_ext_call_pending(vcpu) &&
1259 !psw_extint_disabled(vcpu) &&
1260 (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1263 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1268 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1270 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1273 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1275 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1276 const u64 ckc = vcpu->arch.sie_block->ckc;
1277 u64 cputm, sltime = 0;
1279 if (ckc_interrupts_enabled(vcpu)) {
1280 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1281 if ((s64)now < (s64)ckc)
1282 sltime = tod_to_ns((s64)ckc - (s64)now);
1283 } else if (now < ckc) {
1284 sltime = tod_to_ns(ckc - now);
1286 /* already expired */
1289 if (cpu_timer_interrupts_enabled(vcpu)) {
1290 cputm = kvm_s390_get_cpu_timer(vcpu);
1291 /* already expired? */
1294 return min(sltime, tod_to_ns(cputm));
1296 } else if (cpu_timer_interrupts_enabled(vcpu)) {
1297 sltime = kvm_s390_get_cpu_timer(vcpu);
1298 /* already expired? */
1305 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1307 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1310 vcpu->stat.exit_wait_state++;
1313 if (kvm_arch_vcpu_runnable(vcpu))
1316 if (psw_interrupts_disabled(vcpu)) {
1317 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1318 return -EOPNOTSUPP; /* disabled wait */
1322 (gisa_get_ipm_or_restore_iam(gi) &
1323 vcpu->arch.sie_block->gcr[6] >> 24))
1326 if (!ckc_interrupts_enabled(vcpu) &&
1327 !cpu_timer_interrupts_enabled(vcpu)) {
1328 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1329 __set_cpu_idle(vcpu);
1333 sltime = __calculate_sltime(vcpu);
1337 __set_cpu_idle(vcpu);
1338 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1339 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1341 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1342 kvm_vcpu_block(vcpu);
1343 __unset_cpu_idle(vcpu);
1344 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1346 hrtimer_cancel(&vcpu->arch.ckc_timer);
1350 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1352 vcpu->valid_wakeup = true;
1353 kvm_vcpu_wake_up(vcpu);
1356 * The VCPU might not be sleeping but rather executing VSIE. Let's
1357 * kick it, so it leaves the SIE to process the request.
1359 kvm_s390_vsie_kick(vcpu);
1362 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1364 struct kvm_vcpu *vcpu;
1367 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1368 sltime = __calculate_sltime(vcpu);
1371 * If the monotonic clock runs faster than the tod clock we might be
1372 * woken up too early and have to go back to sleep to avoid deadlocks.
1374 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1375 return HRTIMER_RESTART;
1376 kvm_s390_vcpu_wakeup(vcpu);
1377 return HRTIMER_NORESTART;
1380 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1382 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1384 spin_lock(&li->lock);
1385 li->pending_irqs = 0;
1386 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1387 memset(&li->irq, 0, sizeof(li->irq));
1388 spin_unlock(&li->lock);
1390 sca_clear_ext_call(vcpu);
1393 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1395 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1397 unsigned long irq_type;
1400 __reset_intercept_indicators(vcpu);
1402 /* pending ckc conditions might have been invalidated */
1403 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1404 if (ckc_irq_pending(vcpu))
1405 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1407 /* pending cpu timer conditions might have been invalidated */
1408 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1409 if (cpu_timer_irq_pending(vcpu))
1410 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1412 while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1413 /* bits are in the reverse order of interrupt priority */
1414 irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1416 case IRQ_PEND_IO_ISC_0:
1417 case IRQ_PEND_IO_ISC_1:
1418 case IRQ_PEND_IO_ISC_2:
1419 case IRQ_PEND_IO_ISC_3:
1420 case IRQ_PEND_IO_ISC_4:
1421 case IRQ_PEND_IO_ISC_5:
1422 case IRQ_PEND_IO_ISC_6:
1423 case IRQ_PEND_IO_ISC_7:
1424 rc = __deliver_io(vcpu, irq_type);
1426 case IRQ_PEND_MCHK_EX:
1427 case IRQ_PEND_MCHK_REP:
1428 rc = __deliver_machine_check(vcpu);
1431 rc = __deliver_prog(vcpu);
1433 case IRQ_PEND_EXT_EMERGENCY:
1434 rc = __deliver_emergency_signal(vcpu);
1436 case IRQ_PEND_EXT_EXTERNAL:
1437 rc = __deliver_external_call(vcpu);
1439 case IRQ_PEND_EXT_CLOCK_COMP:
1440 rc = __deliver_ckc(vcpu);
1442 case IRQ_PEND_EXT_CPU_TIMER:
1443 rc = __deliver_cpu_timer(vcpu);
1445 case IRQ_PEND_RESTART:
1446 rc = __deliver_restart(vcpu);
1448 case IRQ_PEND_SET_PREFIX:
1449 rc = __deliver_set_prefix(vcpu);
1451 case IRQ_PEND_PFAULT_INIT:
1452 rc = __deliver_pfault_init(vcpu);
1454 case IRQ_PEND_EXT_SERVICE:
1455 rc = __deliver_service(vcpu);
1457 case IRQ_PEND_EXT_SERVICE_EV:
1458 rc = __deliver_service_ev(vcpu);
1460 case IRQ_PEND_PFAULT_DONE:
1461 rc = __deliver_pfault_done(vcpu);
1463 case IRQ_PEND_VIRTIO:
1464 rc = __deliver_virtio(vcpu);
1467 WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1468 clear_bit(irq_type, &li->pending_irqs);
1472 set_intercept_indicators(vcpu);
1477 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1479 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1481 vcpu->stat.inject_program++;
1482 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1483 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1484 irq->u.pgm.code, 0);
1486 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1487 /* auto detection if no valid ILC was given */
1488 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1489 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1490 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1493 if (irq->u.pgm.code == PGM_PER) {
1494 li->irq.pgm.code |= PGM_PER;
1495 li->irq.pgm.flags = irq->u.pgm.flags;
1496 /* only modify PER related information */
1497 li->irq.pgm.per_address = irq->u.pgm.per_address;
1498 li->irq.pgm.per_code = irq->u.pgm.per_code;
1499 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1500 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1501 } else if (!(irq->u.pgm.code & PGM_PER)) {
1502 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1504 li->irq.pgm.flags = irq->u.pgm.flags;
1505 /* only modify non-PER information */
1506 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1507 li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1508 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1509 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1510 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1511 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1513 li->irq.pgm = irq->u.pgm;
1515 set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1519 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1521 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1523 vcpu->stat.inject_pfault_init++;
1524 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1525 irq->u.ext.ext_params2);
1526 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1527 irq->u.ext.ext_params,
1528 irq->u.ext.ext_params2);
1530 li->irq.ext = irq->u.ext;
1531 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1532 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1536 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1538 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1539 struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1540 uint16_t src_id = irq->u.extcall.code;
1542 vcpu->stat.inject_external_call++;
1543 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1545 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1548 /* sending vcpu invalid */
1549 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1552 if (sclp.has_sigpif && !kvm_s390_pv_cpu_get_handle(vcpu))
1553 return sca_inject_ext_call(vcpu, src_id);
1555 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1557 *extcall = irq->u.extcall;
1558 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1562 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1564 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1565 struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1567 vcpu->stat.inject_set_prefix++;
1568 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1569 irq->u.prefix.address);
1570 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1571 irq->u.prefix.address, 0);
1573 if (!is_vcpu_stopped(vcpu))
1576 *prefix = irq->u.prefix;
1577 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1581 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1582 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1584 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1585 struct kvm_s390_stop_info *stop = &li->irq.stop;
1588 vcpu->stat.inject_stop_signal++;
1589 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1591 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1594 if (is_vcpu_stopped(vcpu)) {
1595 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1596 rc = kvm_s390_store_status_unloaded(vcpu,
1597 KVM_S390_STORE_STATUS_NOADDR);
1601 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1603 stop->flags = irq->u.stop.flags;
1604 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1608 static int __inject_sigp_restart(struct kvm_vcpu *vcpu)
1610 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1612 vcpu->stat.inject_restart++;
1613 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1614 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1616 set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1620 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1621 struct kvm_s390_irq *irq)
1623 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1625 vcpu->stat.inject_emergency_signal++;
1626 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1628 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1629 irq->u.emerg.code, 0);
1631 /* sending vcpu invalid */
1632 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1635 set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1636 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1637 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1641 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1643 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1644 struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1646 vcpu->stat.inject_mchk++;
1647 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1649 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1653 * Because repressible machine checks can be indicated along with
1654 * exigent machine checks (PoP, Chapter 11, Interruption action)
1655 * we need to combine cr14, mcic and external damage code.
1656 * Failing storage address and the logout area should not be or'ed
1657 * together, we just indicate the last occurrence of the corresponding
1660 mchk->cr14 |= irq->u.mchk.cr14;
1661 mchk->mcic |= irq->u.mchk.mcic;
1662 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1663 mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1664 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1665 sizeof(mchk->fixed_logout));
1666 if (mchk->mcic & MCHK_EX_MASK)
1667 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1668 else if (mchk->mcic & MCHK_REP_MASK)
1669 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
1673 static int __inject_ckc(struct kvm_vcpu *vcpu)
1675 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1677 vcpu->stat.inject_ckc++;
1678 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1679 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1682 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1683 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1687 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1689 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1691 vcpu->stat.inject_cputm++;
1692 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1693 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1696 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1697 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1701 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1704 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1705 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1706 struct kvm_s390_interrupt_info *iter;
1707 u16 id = (schid & 0xffff0000U) >> 16;
1708 u16 nr = schid & 0x0000ffffU;
1710 spin_lock(&fi->lock);
1711 list_for_each_entry(iter, isc_list, list) {
1712 if (schid && (id != iter->io.subchannel_id ||
1713 nr != iter->io.subchannel_nr))
1715 /* found an appropriate entry */
1716 list_del_init(&iter->list);
1717 fi->counters[FIRQ_CNTR_IO] -= 1;
1718 if (list_empty(isc_list))
1719 clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1720 spin_unlock(&fi->lock);
1723 spin_unlock(&fi->lock);
1727 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1728 u64 isc_mask, u32 schid)
1730 struct kvm_s390_interrupt_info *inti = NULL;
1733 for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1734 if (isc_mask & isc_to_isc_bits(isc))
1735 inti = get_io_int(kvm, isc, schid);
1740 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1742 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1743 unsigned long active_mask;
1751 active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1752 while (active_mask) {
1753 isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1754 if (gisa_tac_ipm_gisc(gi->origin, isc))
1756 clear_bit_inv(isc, &active_mask);
1763 * Dequeue and return an I/O interrupt matching any of the interruption
1764 * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1765 * Take into account the interrupts pending in the interrupt list and in GISA.
1767 * Note that for a guest that does not enable I/O interrupts
1768 * but relies on TPI, a flood of classic interrupts may starve
1769 * out adapter interrupts on the same isc. Linux does not do
1770 * that, and it is possible to work around the issue by configuring
1771 * different iscs for classic and adapter interrupts in the guest,
1772 * but we may want to revisit this in the future.
1774 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1775 u64 isc_mask, u32 schid)
1777 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1778 struct kvm_s390_interrupt_info *inti, *tmp_inti;
1781 inti = get_top_io_int(kvm, isc_mask, schid);
1783 isc = get_top_gisa_isc(kvm, isc_mask, schid);
1789 /* AI in GISA but no classical IO int */
1792 /* both types of interrupts present */
1793 if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1794 /* classical IO int with higher priority */
1795 gisa_set_ipm_gisc(gi->origin, isc);
1799 tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1801 tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1802 tmp_inti->io.io_int_word = isc_to_int_word(isc);
1804 kvm_s390_reinject_io_int(kvm, inti);
1807 gisa_set_ipm_gisc(gi->origin, isc);
1812 static int __inject_service(struct kvm *kvm,
1813 struct kvm_s390_interrupt_info *inti)
1815 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1817 kvm->stat.inject_service_signal++;
1818 spin_lock(&fi->lock);
1819 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1821 /* We always allow events, track them separately from the sccb ints */
1822 if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING)
1823 set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1826 * Early versions of the QEMU s390 bios will inject several
1827 * service interrupts after another without handling a
1828 * condition code indicating busy.
1829 * We will silently ignore those superfluous sccb values.
1830 * A future version of QEMU will take care of serialization
1833 if (fi->srv_signal.ext_params & SCCB_MASK)
1835 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1836 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1838 spin_unlock(&fi->lock);
1843 static int __inject_virtio(struct kvm *kvm,
1844 struct kvm_s390_interrupt_info *inti)
1846 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1848 kvm->stat.inject_virtio++;
1849 spin_lock(&fi->lock);
1850 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1851 spin_unlock(&fi->lock);
1854 fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1855 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1856 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1857 spin_unlock(&fi->lock);
1861 static int __inject_pfault_done(struct kvm *kvm,
1862 struct kvm_s390_interrupt_info *inti)
1864 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1866 kvm->stat.inject_pfault_done++;
1867 spin_lock(&fi->lock);
1868 if (fi->counters[FIRQ_CNTR_PFAULT] >=
1869 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1870 spin_unlock(&fi->lock);
1873 fi->counters[FIRQ_CNTR_PFAULT] += 1;
1874 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1875 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1876 spin_unlock(&fi->lock);
1880 #define CR_PENDING_SUBCLASS 28
1881 static int __inject_float_mchk(struct kvm *kvm,
1882 struct kvm_s390_interrupt_info *inti)
1884 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1886 kvm->stat.inject_float_mchk++;
1887 spin_lock(&fi->lock);
1888 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1889 fi->mchk.mcic |= inti->mchk.mcic;
1890 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1891 spin_unlock(&fi->lock);
1896 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1898 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1899 struct kvm_s390_float_interrupt *fi;
1900 struct list_head *list;
1903 kvm->stat.inject_io++;
1904 isc = int_word_to_isc(inti->io.io_int_word);
1907 * Do not make use of gisa in protected mode. We do not use the lock
1908 * checking variant as this is just a performance optimization and we
1909 * do not hold the lock here. This is ok as the code will pick
1910 * interrupts from both "lists" for delivery.
1912 if (!kvm_s390_pv_get_handle(kvm) &&
1913 gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1914 VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1915 gisa_set_ipm_gisc(gi->origin, isc);
1920 fi = &kvm->arch.float_int;
1921 spin_lock(&fi->lock);
1922 if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1923 spin_unlock(&fi->lock);
1926 fi->counters[FIRQ_CNTR_IO] += 1;
1928 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1929 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1931 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1932 inti->io.subchannel_id >> 8,
1933 inti->io.subchannel_id >> 1 & 0x3,
1934 inti->io.subchannel_nr);
1935 list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1936 list_add_tail(&inti->list, list);
1937 set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1938 spin_unlock(&fi->lock);
1943 * Find a destination VCPU for a floating irq and kick it.
1945 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1947 struct kvm_vcpu *dst_vcpu;
1948 int sigcpu, online_vcpus, nr_tries = 0;
1950 online_vcpus = atomic_read(&kvm->online_vcpus);
1954 /* find idle VCPUs first, then round robin */
1955 sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1956 if (sigcpu == online_vcpus) {
1958 sigcpu = kvm->arch.float_int.next_rr_cpu++;
1959 kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1960 /* avoid endless loops if all vcpus are stopped */
1961 if (nr_tries++ >= online_vcpus)
1963 } while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1965 dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1967 /* make the VCPU drop out of the SIE, or wake it up if sleeping */
1970 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1972 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1973 if (!(type & KVM_S390_INT_IO_AI_MASK &&
1974 kvm->arch.gisa_int.origin) ||
1975 kvm_s390_pv_cpu_get_handle(dst_vcpu))
1976 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1979 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1982 kvm_s390_vcpu_wakeup(dst_vcpu);
1985 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1987 u64 type = READ_ONCE(inti->type);
1992 rc = __inject_float_mchk(kvm, inti);
1994 case KVM_S390_INT_VIRTIO:
1995 rc = __inject_virtio(kvm, inti);
1997 case KVM_S390_INT_SERVICE:
1998 rc = __inject_service(kvm, inti);
2000 case KVM_S390_INT_PFAULT_DONE:
2001 rc = __inject_pfault_done(kvm, inti);
2003 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2004 rc = __inject_io(kvm, inti);
2012 __floating_irq_kick(kvm, type);
2016 int kvm_s390_inject_vm(struct kvm *kvm,
2017 struct kvm_s390_interrupt *s390int)
2019 struct kvm_s390_interrupt_info *inti;
2022 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
2026 inti->type = s390int->type;
2027 switch (inti->type) {
2028 case KVM_S390_INT_VIRTIO:
2029 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
2030 s390int->parm, s390int->parm64);
2031 inti->ext.ext_params = s390int->parm;
2032 inti->ext.ext_params2 = s390int->parm64;
2034 case KVM_S390_INT_SERVICE:
2035 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
2036 inti->ext.ext_params = s390int->parm;
2038 case KVM_S390_INT_PFAULT_DONE:
2039 inti->ext.ext_params2 = s390int->parm64;
2042 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
2044 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
2045 inti->mchk.mcic = s390int->parm64;
2047 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2048 inti->io.subchannel_id = s390int->parm >> 16;
2049 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
2050 inti->io.io_int_parm = s390int->parm64 >> 32;
2051 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
2057 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
2060 rc = __inject_vm(kvm, inti);
2066 int kvm_s390_reinject_io_int(struct kvm *kvm,
2067 struct kvm_s390_interrupt_info *inti)
2069 return __inject_vm(kvm, inti);
2072 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
2073 struct kvm_s390_irq *irq)
2075 irq->type = s390int->type;
2076 switch (irq->type) {
2077 case KVM_S390_PROGRAM_INT:
2078 if (s390int->parm & 0xffff0000)
2080 irq->u.pgm.code = s390int->parm;
2082 case KVM_S390_SIGP_SET_PREFIX:
2083 irq->u.prefix.address = s390int->parm;
2085 case KVM_S390_SIGP_STOP:
2086 irq->u.stop.flags = s390int->parm;
2088 case KVM_S390_INT_EXTERNAL_CALL:
2089 if (s390int->parm & 0xffff0000)
2091 irq->u.extcall.code = s390int->parm;
2093 case KVM_S390_INT_EMERGENCY:
2094 if (s390int->parm & 0xffff0000)
2096 irq->u.emerg.code = s390int->parm;
2099 irq->u.mchk.mcic = s390int->parm64;
2101 case KVM_S390_INT_PFAULT_INIT:
2102 irq->u.ext.ext_params = s390int->parm;
2103 irq->u.ext.ext_params2 = s390int->parm64;
2105 case KVM_S390_RESTART:
2106 case KVM_S390_INT_CLOCK_COMP:
2107 case KVM_S390_INT_CPU_TIMER:
2115 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
2117 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2119 return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2122 int kvm_s390_is_restart_irq_pending(struct kvm_vcpu *vcpu)
2124 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2126 return test_bit(IRQ_PEND_RESTART, &li->pending_irqs);
2129 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
2131 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2133 spin_lock(&li->lock);
2134 li->irq.stop.flags = 0;
2135 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2136 spin_unlock(&li->lock);
2139 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2143 switch (irq->type) {
2144 case KVM_S390_PROGRAM_INT:
2145 rc = __inject_prog(vcpu, irq);
2147 case KVM_S390_SIGP_SET_PREFIX:
2148 rc = __inject_set_prefix(vcpu, irq);
2150 case KVM_S390_SIGP_STOP:
2151 rc = __inject_sigp_stop(vcpu, irq);
2153 case KVM_S390_RESTART:
2154 rc = __inject_sigp_restart(vcpu);
2156 case KVM_S390_INT_CLOCK_COMP:
2157 rc = __inject_ckc(vcpu);
2159 case KVM_S390_INT_CPU_TIMER:
2160 rc = __inject_cpu_timer(vcpu);
2162 case KVM_S390_INT_EXTERNAL_CALL:
2163 rc = __inject_extcall(vcpu, irq);
2165 case KVM_S390_INT_EMERGENCY:
2166 rc = __inject_sigp_emergency(vcpu, irq);
2169 rc = __inject_mchk(vcpu, irq);
2171 case KVM_S390_INT_PFAULT_INIT:
2172 rc = __inject_pfault_init(vcpu, irq);
2174 case KVM_S390_INT_VIRTIO:
2175 case KVM_S390_INT_SERVICE:
2176 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2184 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2186 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2189 spin_lock(&li->lock);
2190 rc = do_inject_vcpu(vcpu, irq);
2191 spin_unlock(&li->lock);
2193 kvm_s390_vcpu_wakeup(vcpu);
2197 static inline void clear_irq_list(struct list_head *_list)
2199 struct kvm_s390_interrupt_info *inti, *n;
2201 list_for_each_entry_safe(inti, n, _list, list) {
2202 list_del(&inti->list);
2207 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2208 struct kvm_s390_irq *irq)
2210 irq->type = inti->type;
2211 switch (inti->type) {
2212 case KVM_S390_INT_PFAULT_INIT:
2213 case KVM_S390_INT_PFAULT_DONE:
2214 case KVM_S390_INT_VIRTIO:
2215 irq->u.ext = inti->ext;
2217 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2218 irq->u.io = inti->io;
2223 void kvm_s390_clear_float_irqs(struct kvm *kvm)
2225 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2228 mutex_lock(&kvm->lock);
2229 if (!kvm_s390_pv_is_protected(kvm))
2230 fi->masked_irqs = 0;
2231 mutex_unlock(&kvm->lock);
2232 spin_lock(&fi->lock);
2233 fi->pending_irqs = 0;
2234 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2235 memset(&fi->mchk, 0, sizeof(fi->mchk));
2236 for (i = 0; i < FIRQ_LIST_COUNT; i++)
2237 clear_irq_list(&fi->lists[i]);
2238 for (i = 0; i < FIRQ_MAX_COUNT; i++)
2239 fi->counters[i] = 0;
2240 spin_unlock(&fi->lock);
2241 kvm_s390_gisa_clear(kvm);
2244 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2246 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2247 struct kvm_s390_interrupt_info *inti;
2248 struct kvm_s390_float_interrupt *fi;
2249 struct kvm_s390_irq *buf;
2250 struct kvm_s390_irq *irq;
2256 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2260 * We are already using -ENOMEM to signal
2261 * userspace it may retry with a bigger buffer,
2262 * so we need to use something else for this case
2268 max_irqs = len / sizeof(struct kvm_s390_irq);
2270 if (gi->origin && gisa_get_ipm(gi->origin)) {
2271 for (i = 0; i <= MAX_ISC; i++) {
2272 if (n == max_irqs) {
2273 /* signal userspace to try again */
2277 if (gisa_tac_ipm_gisc(gi->origin, i)) {
2278 irq = (struct kvm_s390_irq *) &buf[n];
2279 irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2280 irq->u.io.io_int_word = isc_to_int_word(i);
2285 fi = &kvm->arch.float_int;
2286 spin_lock(&fi->lock);
2287 for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2288 list_for_each_entry(inti, &fi->lists[i], list) {
2289 if (n == max_irqs) {
2290 /* signal userspace to try again */
2294 inti_to_irq(inti, &buf[n]);
2298 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) ||
2299 test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) {
2300 if (n == max_irqs) {
2301 /* signal userspace to try again */
2305 irq = (struct kvm_s390_irq *) &buf[n];
2306 irq->type = KVM_S390_INT_SERVICE;
2307 irq->u.ext = fi->srv_signal;
2310 if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2311 if (n == max_irqs) {
2312 /* signal userspace to try again */
2316 irq = (struct kvm_s390_irq *) &buf[n];
2317 irq->type = KVM_S390_MCHK;
2318 irq->u.mchk = fi->mchk;
2323 spin_unlock(&fi->lock);
2325 if (!ret && n > 0) {
2326 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2331 return ret < 0 ? ret : n;
2334 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2336 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2337 struct kvm_s390_ais_all ais;
2339 if (attr->attr < sizeof(ais))
2342 if (!test_kvm_facility(kvm, 72))
2345 mutex_lock(&fi->ais_lock);
2346 ais.simm = fi->simm;
2347 ais.nimm = fi->nimm;
2348 mutex_unlock(&fi->ais_lock);
2350 if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2356 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2360 switch (attr->group) {
2361 case KVM_DEV_FLIC_GET_ALL_IRQS:
2362 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2365 case KVM_DEV_FLIC_AISM_ALL:
2366 r = flic_ais_mode_get_all(dev->kvm, attr);
2375 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2378 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2379 void *target = NULL;
2380 void __user *source;
2383 if (get_user(inti->type, (u64 __user *)addr))
2386 switch (inti->type) {
2387 case KVM_S390_INT_PFAULT_INIT:
2388 case KVM_S390_INT_PFAULT_DONE:
2389 case KVM_S390_INT_VIRTIO:
2390 case KVM_S390_INT_SERVICE:
2391 target = (void *) &inti->ext;
2392 source = &uptr->u.ext;
2393 size = sizeof(inti->ext);
2395 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2396 target = (void *) &inti->io;
2397 source = &uptr->u.io;
2398 size = sizeof(inti->io);
2401 target = (void *) &inti->mchk;
2402 source = &uptr->u.mchk;
2403 size = sizeof(inti->mchk);
2409 if (copy_from_user(target, source, size))
2415 static int enqueue_floating_irq(struct kvm_device *dev,
2416 struct kvm_device_attr *attr)
2418 struct kvm_s390_interrupt_info *inti = NULL;
2420 int len = attr->attr;
2422 if (len % sizeof(struct kvm_s390_irq) != 0)
2424 else if (len > KVM_S390_FLIC_MAX_BUFFER)
2427 while (len >= sizeof(struct kvm_s390_irq)) {
2428 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
2432 r = copy_irq_from_user(inti, attr->addr);
2437 r = __inject_vm(dev->kvm, inti);
2442 len -= sizeof(struct kvm_s390_irq);
2443 attr->addr += sizeof(struct kvm_s390_irq);
2449 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2451 if (id >= MAX_S390_IO_ADAPTERS)
2453 id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2454 return kvm->arch.adapters[id];
2457 static int register_io_adapter(struct kvm_device *dev,
2458 struct kvm_device_attr *attr)
2460 struct s390_io_adapter *adapter;
2461 struct kvm_s390_io_adapter adapter_info;
2463 if (copy_from_user(&adapter_info,
2464 (void __user *)attr->addr, sizeof(adapter_info)))
2467 if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2470 adapter_info.id = array_index_nospec(adapter_info.id,
2471 MAX_S390_IO_ADAPTERS);
2473 if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2476 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2480 adapter->id = adapter_info.id;
2481 adapter->isc = adapter_info.isc;
2482 adapter->maskable = adapter_info.maskable;
2483 adapter->masked = false;
2484 adapter->swap = adapter_info.swap;
2485 adapter->suppressible = (adapter_info.flags) &
2486 KVM_S390_ADAPTER_SUPPRESSIBLE;
2487 dev->kvm->arch.adapters[adapter->id] = adapter;
2492 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2495 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2497 if (!adapter || !adapter->maskable)
2499 ret = adapter->masked;
2500 adapter->masked = masked;
2504 void kvm_s390_destroy_adapters(struct kvm *kvm)
2508 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++)
2509 kfree(kvm->arch.adapters[i]);
2512 static int modify_io_adapter(struct kvm_device *dev,
2513 struct kvm_device_attr *attr)
2515 struct kvm_s390_io_adapter_req req;
2516 struct s390_io_adapter *adapter;
2519 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2522 adapter = get_io_adapter(dev->kvm, req.id);
2526 case KVM_S390_IO_ADAPTER_MASK:
2527 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2532 * The following operations are no longer needed and therefore no-ops.
2533 * The gpa to hva translation is done when an IRQ route is set up. The
2534 * set_irq code uses get_user_pages_remote() to do the actual write.
2536 case KVM_S390_IO_ADAPTER_MAP:
2537 case KVM_S390_IO_ADAPTER_UNMAP:
2547 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2550 const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2555 if (attr->attr != sizeof(schid))
2557 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2561 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2563 * If userspace is conforming to the architecture, we can have at most
2564 * one pending I/O interrupt per subchannel, so this is effectively a
2570 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2572 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2573 struct kvm_s390_ais_req req;
2576 if (!test_kvm_facility(kvm, 72))
2579 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2582 if (req.isc > MAX_ISC)
2585 trace_kvm_s390_modify_ais_mode(req.isc,
2586 (fi->simm & AIS_MODE_MASK(req.isc)) ?
2587 (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2588 2 : KVM_S390_AIS_MODE_SINGLE :
2589 KVM_S390_AIS_MODE_ALL, req.mode);
2591 mutex_lock(&fi->ais_lock);
2593 case KVM_S390_AIS_MODE_ALL:
2594 fi->simm &= ~AIS_MODE_MASK(req.isc);
2595 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2597 case KVM_S390_AIS_MODE_SINGLE:
2598 fi->simm |= AIS_MODE_MASK(req.isc);
2599 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2604 mutex_unlock(&fi->ais_lock);
2609 static int kvm_s390_inject_airq(struct kvm *kvm,
2610 struct s390_io_adapter *adapter)
2612 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2613 struct kvm_s390_interrupt s390int = {
2614 .type = KVM_S390_INT_IO(1, 0, 0, 0),
2616 .parm64 = isc_to_int_word(adapter->isc),
2620 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2621 return kvm_s390_inject_vm(kvm, &s390int);
2623 mutex_lock(&fi->ais_lock);
2624 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2625 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2629 ret = kvm_s390_inject_vm(kvm, &s390int);
2630 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2631 fi->nimm |= AIS_MODE_MASK(adapter->isc);
2632 trace_kvm_s390_modify_ais_mode(adapter->isc,
2633 KVM_S390_AIS_MODE_SINGLE, 2);
2636 mutex_unlock(&fi->ais_lock);
2640 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2642 unsigned int id = attr->attr;
2643 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2648 return kvm_s390_inject_airq(kvm, adapter);
2651 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2653 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2654 struct kvm_s390_ais_all ais;
2656 if (!test_kvm_facility(kvm, 72))
2659 if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2662 mutex_lock(&fi->ais_lock);
2663 fi->simm = ais.simm;
2664 fi->nimm = ais.nimm;
2665 mutex_unlock(&fi->ais_lock);
2670 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2674 struct kvm_vcpu *vcpu;
2676 switch (attr->group) {
2677 case KVM_DEV_FLIC_ENQUEUE:
2678 r = enqueue_floating_irq(dev, attr);
2680 case KVM_DEV_FLIC_CLEAR_IRQS:
2681 kvm_s390_clear_float_irqs(dev->kvm);
2683 case KVM_DEV_FLIC_APF_ENABLE:
2684 dev->kvm->arch.gmap->pfault_enabled = 1;
2686 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2687 dev->kvm->arch.gmap->pfault_enabled = 0;
2689 * Make sure no async faults are in transition when
2690 * clearing the queues. So we don't need to worry
2691 * about late coming workers.
2693 synchronize_srcu(&dev->kvm->srcu);
2694 kvm_for_each_vcpu(i, vcpu, dev->kvm)
2695 kvm_clear_async_pf_completion_queue(vcpu);
2697 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2698 r = register_io_adapter(dev, attr);
2700 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2701 r = modify_io_adapter(dev, attr);
2703 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2704 r = clear_io_irq(dev->kvm, attr);
2706 case KVM_DEV_FLIC_AISM:
2707 r = modify_ais_mode(dev->kvm, attr);
2709 case KVM_DEV_FLIC_AIRQ_INJECT:
2710 r = flic_inject_airq(dev->kvm, attr);
2712 case KVM_DEV_FLIC_AISM_ALL:
2713 r = flic_ais_mode_set_all(dev->kvm, attr);
2722 static int flic_has_attr(struct kvm_device *dev,
2723 struct kvm_device_attr *attr)
2725 switch (attr->group) {
2726 case KVM_DEV_FLIC_GET_ALL_IRQS:
2727 case KVM_DEV_FLIC_ENQUEUE:
2728 case KVM_DEV_FLIC_CLEAR_IRQS:
2729 case KVM_DEV_FLIC_APF_ENABLE:
2730 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2731 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2732 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2733 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2734 case KVM_DEV_FLIC_AISM:
2735 case KVM_DEV_FLIC_AIRQ_INJECT:
2736 case KVM_DEV_FLIC_AISM_ALL:
2742 static int flic_create(struct kvm_device *dev, u32 type)
2746 if (dev->kvm->arch.flic)
2748 dev->kvm->arch.flic = dev;
2752 static void flic_destroy(struct kvm_device *dev)
2754 dev->kvm->arch.flic = NULL;
2758 /* s390 floating irq controller (flic) */
2759 struct kvm_device_ops kvm_flic_ops = {
2761 .get_attr = flic_get_attr,
2762 .set_attr = flic_set_attr,
2763 .has_attr = flic_has_attr,
2764 .create = flic_create,
2765 .destroy = flic_destroy,
2768 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2772 bit = bit_nr + (addr % PAGE_SIZE) * 8;
2774 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2777 static struct page *get_map_page(struct kvm *kvm, u64 uaddr)
2779 struct page *page = NULL;
2781 mmap_read_lock(kvm->mm);
2782 get_user_pages_remote(kvm->mm, uaddr, 1, FOLL_WRITE,
2784 mmap_read_unlock(kvm->mm);
2788 static int adapter_indicators_set(struct kvm *kvm,
2789 struct s390_io_adapter *adapter,
2790 struct kvm_s390_adapter_int *adapter_int)
2793 int summary_set, idx;
2794 struct page *ind_page, *summary_page;
2797 ind_page = get_map_page(kvm, adapter_int->ind_addr);
2800 summary_page = get_map_page(kvm, adapter_int->summary_addr);
2801 if (!summary_page) {
2806 idx = srcu_read_lock(&kvm->srcu);
2807 map = page_address(ind_page);
2808 bit = get_ind_bit(adapter_int->ind_addr,
2809 adapter_int->ind_offset, adapter->swap);
2811 mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT);
2812 set_page_dirty_lock(ind_page);
2813 map = page_address(summary_page);
2814 bit = get_ind_bit(adapter_int->summary_addr,
2815 adapter_int->summary_offset, adapter->swap);
2816 summary_set = test_and_set_bit(bit, map);
2817 mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT);
2818 set_page_dirty_lock(summary_page);
2819 srcu_read_unlock(&kvm->srcu, idx);
2822 put_page(summary_page);
2823 return summary_set ? 0 : 1;
2827 * < 0 - not injected due to error
2828 * = 0 - coalesced, summary indicator already active
2829 * > 0 - injected interrupt
2831 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2832 struct kvm *kvm, int irq_source_id, int level,
2836 struct s390_io_adapter *adapter;
2838 /* We're only interested in the 0->1 transition. */
2841 adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2844 ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2845 if ((ret > 0) && !adapter->masked) {
2846 ret = kvm_s390_inject_airq(kvm, adapter);
2854 * Inject the machine check to the guest.
2856 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2857 struct mcck_volatile_info *mcck_info)
2859 struct kvm_s390_interrupt_info inti;
2860 struct kvm_s390_irq irq;
2861 struct kvm_s390_mchk_info *mchk;
2863 __u64 cr14 = 0; /* upper bits are not used */
2866 mci.val = mcck_info->mcic;
2868 cr14 |= CR14_RECOVERY_SUBMASK;
2870 cr14 |= CR14_DEGRADATION_SUBMASK;
2872 cr14 |= CR14_WARNING_SUBMASK;
2874 mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2876 mchk->mcic = mcck_info->mcic;
2877 mchk->ext_damage_code = mcck_info->ext_damage_code;
2878 mchk->failing_storage_address = mcck_info->failing_storage_address;
2880 /* Inject the floating machine check */
2881 inti.type = KVM_S390_MCHK;
2882 rc = __inject_vm(vcpu->kvm, &inti);
2884 /* Inject the machine check to specified vcpu */
2885 irq.type = KVM_S390_MCHK;
2886 rc = kvm_s390_inject_vcpu(vcpu, &irq);
2891 int kvm_set_routing_entry(struct kvm *kvm,
2892 struct kvm_kernel_irq_routing_entry *e,
2893 const struct kvm_irq_routing_entry *ue)
2898 /* we store the userspace addresses instead of the guest addresses */
2899 case KVM_IRQ_ROUTING_S390_ADAPTER:
2900 e->set = set_adapter_int;
2901 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.summary_addr);
2902 if (uaddr == -EFAULT)
2904 e->adapter.summary_addr = uaddr;
2905 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.ind_addr);
2906 if (uaddr == -EFAULT)
2908 e->adapter.ind_addr = uaddr;
2909 e->adapter.summary_offset = ue->u.adapter.summary_offset;
2910 e->adapter.ind_offset = ue->u.adapter.ind_offset;
2911 e->adapter.adapter_id = ue->u.adapter.adapter_id;
2918 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2919 int irq_source_id, int level, bool line_status)
2924 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2926 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2927 struct kvm_s390_irq *buf;
2935 if (copy_from_user((void *) buf, irqstate, len)) {
2941 * Don't allow setting the interrupt state
2942 * when there are already interrupts pending
2944 spin_lock(&li->lock);
2945 if (li->pending_irqs) {
2950 for (n = 0; n < len / sizeof(*buf); n++) {
2951 r = do_inject_vcpu(vcpu, &buf[n]);
2957 spin_unlock(&li->lock);
2964 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2965 struct kvm_s390_irq *irq,
2966 unsigned long irq_type)
2969 case IRQ_PEND_MCHK_EX:
2970 case IRQ_PEND_MCHK_REP:
2971 irq->type = KVM_S390_MCHK;
2972 irq->u.mchk = li->irq.mchk;
2975 irq->type = KVM_S390_PROGRAM_INT;
2976 irq->u.pgm = li->irq.pgm;
2978 case IRQ_PEND_PFAULT_INIT:
2979 irq->type = KVM_S390_INT_PFAULT_INIT;
2980 irq->u.ext = li->irq.ext;
2982 case IRQ_PEND_EXT_EXTERNAL:
2983 irq->type = KVM_S390_INT_EXTERNAL_CALL;
2984 irq->u.extcall = li->irq.extcall;
2986 case IRQ_PEND_EXT_CLOCK_COMP:
2987 irq->type = KVM_S390_INT_CLOCK_COMP;
2989 case IRQ_PEND_EXT_CPU_TIMER:
2990 irq->type = KVM_S390_INT_CPU_TIMER;
2992 case IRQ_PEND_SIGP_STOP:
2993 irq->type = KVM_S390_SIGP_STOP;
2994 irq->u.stop = li->irq.stop;
2996 case IRQ_PEND_RESTART:
2997 irq->type = KVM_S390_RESTART;
2999 case IRQ_PEND_SET_PREFIX:
3000 irq->type = KVM_S390_SIGP_SET_PREFIX;
3001 irq->u.prefix = li->irq.prefix;
3006 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
3009 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
3010 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
3011 unsigned long pending_irqs;
3012 struct kvm_s390_irq irq;
3013 unsigned long irq_type;
3017 spin_lock(&li->lock);
3018 pending_irqs = li->pending_irqs;
3019 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
3020 sizeof(sigp_emerg_pending));
3021 spin_unlock(&li->lock);
3023 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
3024 memset(&irq, 0, sizeof(irq));
3025 if (irq_type == IRQ_PEND_EXT_EMERGENCY)
3027 if (n + sizeof(irq) > len)
3029 store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
3030 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3035 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
3036 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
3037 memset(&irq, 0, sizeof(irq));
3038 if (n + sizeof(irq) > len)
3040 irq.type = KVM_S390_INT_EMERGENCY;
3041 irq.u.emerg.code = cpuaddr;
3042 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3048 if (sca_ext_call_pending(vcpu, &scn)) {
3049 if (n + sizeof(irq) > len)
3051 memset(&irq, 0, sizeof(irq));
3052 irq.type = KVM_S390_INT_EXTERNAL_CALL;
3053 irq.u.extcall.code = scn;
3054 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3062 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
3064 int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus);
3065 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3066 struct kvm_vcpu *vcpu;
3069 for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) {
3070 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
3071 if (psw_ioint_disabled(vcpu))
3073 vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
3074 if (deliverable_mask & vcpu_isc_mask) {
3075 /* lately kicked but not yet running */
3076 if (test_and_set_bit(vcpu_idx, gi->kicked_mask))
3078 kvm_s390_vcpu_wakeup(vcpu);
3084 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3086 struct kvm_s390_gisa_interrupt *gi =
3087 container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3089 container_of(gi->origin, struct sie_page2, gisa)->kvm;
3092 pending_mask = gisa_get_ipm_or_restore_iam(gi);
3094 __airqs_kick_single_vcpu(kvm, pending_mask);
3095 hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3096 return HRTIMER_RESTART;
3099 return HRTIMER_NORESTART;
3102 #define NULL_GISA_ADDR 0x00000000UL
3103 #define NONE_GISA_ADDR 0x00000001UL
3104 #define GISA_ADDR_MASK 0xfffff000UL
3106 static void process_gib_alert_list(void)
3108 struct kvm_s390_gisa_interrupt *gi;
3109 struct kvm_s390_gisa *gisa;
3111 u32 final, origin = 0UL;
3115 * If the NONE_GISA_ADDR is still stored in the alert list
3116 * origin, we will leave the outer loop. No further GISA has
3117 * been added to the alert list by millicode while processing
3118 * the current alert list.
3120 final = (origin & NONE_GISA_ADDR);
3122 * Cut off the alert list and store the NONE_GISA_ADDR in the
3123 * alert list origin to avoid further GAL interruptions.
3124 * A new alert list can be build up by millicode in parallel
3125 * for guests not in the yet cut-off alert list. When in the
3126 * final loop, store the NULL_GISA_ADDR instead. This will re-
3127 * enable GAL interruptions on the host again.
3129 origin = xchg(&gib->alert_list_origin,
3130 (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3132 * Loop through the just cut-off alert list and start the
3133 * gisa timers to kick idle vcpus to consume the pending
3134 * interruptions asap.
3136 while (origin & GISA_ADDR_MASK) {
3137 gisa = (struct kvm_s390_gisa *)(u64)origin;
3138 origin = gisa->next_alert;
3139 gisa->next_alert = (u32)(u64)gisa;
3140 kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3141 gi = &kvm->arch.gisa_int;
3142 if (hrtimer_active(&gi->timer))
3143 hrtimer_cancel(&gi->timer);
3144 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3150 void kvm_s390_gisa_clear(struct kvm *kvm)
3152 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3156 gisa_clear_ipm(gi->origin);
3157 VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3160 void kvm_s390_gisa_init(struct kvm *kvm)
3162 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3164 if (!css_general_characteristics.aiv)
3166 gi->origin = &kvm->arch.sie_page2->gisa;
3168 spin_lock_init(&gi->alert.ref_lock);
3169 gi->expires = 50 * 1000; /* 50 usec */
3170 hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3171 gi->timer.function = gisa_vcpu_kicker;
3172 memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3173 gi->origin->next_alert = (u32)(u64)gi->origin;
3174 VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3177 void kvm_s390_gisa_destroy(struct kvm *kvm)
3179 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3184 KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
3185 kvm, gi->alert.mask);
3186 while (gisa_in_alert_list(gi->origin))
3188 hrtimer_cancel(&gi->timer);
3193 * kvm_s390_gisc_register - register a guest ISC
3195 * @kvm: the kernel vm to work with
3196 * @gisc: the guest interruption sub class to register
3198 * The function extends the vm specific alert mask to use.
3199 * The effective IAM mask in the GISA is updated as well
3200 * in case the GISA is not part of the GIB alert list.
3201 * It will be updated latest when the IAM gets restored
3202 * by gisa_get_ipm_or_restore_iam().
3204 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3205 * has registered with the channel subsystem.
3206 * -ENODEV in case the vm uses no GISA
3207 * -ERANGE in case the guest ISC is invalid
3209 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3211 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3218 spin_lock(&gi->alert.ref_lock);
3219 gi->alert.ref_count[gisc]++;
3220 if (gi->alert.ref_count[gisc] == 1) {
3221 gi->alert.mask |= 0x80 >> gisc;
3222 gisa_set_iam(gi->origin, gi->alert.mask);
3224 spin_unlock(&gi->alert.ref_lock);
3228 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3231 * kvm_s390_gisc_unregister - unregister a guest ISC
3233 * @kvm: the kernel vm to work with
3234 * @gisc: the guest interruption sub class to register
3236 * The function reduces the vm specific alert mask to use.
3237 * The effective IAM mask in the GISA is updated as well
3238 * in case the GISA is not part of the GIB alert list.
3239 * It will be updated latest when the IAM gets restored
3240 * by gisa_get_ipm_or_restore_iam().
3242 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3243 * has registered with the channel subsystem.
3244 * -ENODEV in case the vm uses no GISA
3245 * -ERANGE in case the guest ISC is invalid
3246 * -EINVAL in case the guest ISC is not registered
3248 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3250 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3258 spin_lock(&gi->alert.ref_lock);
3259 if (gi->alert.ref_count[gisc] == 0) {
3263 gi->alert.ref_count[gisc]--;
3264 if (gi->alert.ref_count[gisc] == 0) {
3265 gi->alert.mask &= ~(0x80 >> gisc);
3266 gisa_set_iam(gi->origin, gi->alert.mask);
3269 spin_unlock(&gi->alert.ref_lock);
3273 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3275 static void gib_alert_irq_handler(struct airq_struct *airq, bool floating)
3277 inc_irq_stat(IRQIO_GAL);
3278 process_gib_alert_list();
3281 static struct airq_struct gib_alert_irq = {
3282 .handler = gib_alert_irq_handler,
3283 .lsi_ptr = &gib_alert_irq.lsi_mask,
3286 void kvm_s390_gib_destroy(void)
3291 unregister_adapter_interrupt(&gib_alert_irq);
3292 free_page((unsigned long)gib);
3296 int kvm_s390_gib_init(u8 nisc)
3300 if (!css_general_characteristics.aiv) {
3301 KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3305 gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL | GFP_DMA);
3311 gib_alert_irq.isc = nisc;
3312 if (register_adapter_interrupt(&gib_alert_irq)) {
3313 pr_err("Registering the GIB alert interruption handler failed\n");
3319 if (chsc_sgib((u32)(u64)gib)) {
3320 pr_err("Associating the GIB with the AIV facility failed\n");
3321 free_page((unsigned long)gib);
3327 KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3331 unregister_adapter_interrupt(&gib_alert_irq);
3333 free_page((unsigned long)gib);