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 = *sigp_ctrl;
90 expect = old_val.value;
91 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
93 struct bsca_block *sca = vcpu->kvm->arch.sca;
94 union bsca_sigp_ctrl *sigp_ctrl =
95 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
96 union bsca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
102 expect = old_val.value;
103 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
105 read_unlock(&vcpu->kvm->arch.sca_lock);
108 /* another external call is pending */
111 kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
115 static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
119 if (!kvm_s390_use_sca_entries())
121 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
122 read_lock(&vcpu->kvm->arch.sca_lock);
123 if (vcpu->kvm->arch.use_esca) {
124 struct esca_block *sca = vcpu->kvm->arch.sca;
125 union esca_sigp_ctrl *sigp_ctrl =
126 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
127 union esca_sigp_ctrl old = *sigp_ctrl;
130 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
132 struct bsca_block *sca = vcpu->kvm->arch.sca;
133 union bsca_sigp_ctrl *sigp_ctrl =
134 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
135 union bsca_sigp_ctrl old = *sigp_ctrl;
138 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
140 read_unlock(&vcpu->kvm->arch.sca_lock);
141 WARN_ON(rc != expect); /* cannot clear? */
144 int psw_extint_disabled(struct kvm_vcpu *vcpu)
146 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
149 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
151 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
154 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
156 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
159 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
161 return psw_extint_disabled(vcpu) &&
162 psw_ioint_disabled(vcpu) &&
163 psw_mchk_disabled(vcpu);
166 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
168 if (psw_extint_disabled(vcpu) ||
169 !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
171 if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
172 /* No timer interrupts when single stepping */
177 static int ckc_irq_pending(struct kvm_vcpu *vcpu)
179 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
180 const u64 ckc = vcpu->arch.sie_block->ckc;
182 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
183 if ((s64)ckc >= (s64)now)
185 } else if (ckc >= now) {
188 return ckc_interrupts_enabled(vcpu);
191 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
193 return !psw_extint_disabled(vcpu) &&
194 (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK);
197 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
199 if (!cpu_timer_interrupts_enabled(vcpu))
201 return kvm_s390_get_cpu_timer(vcpu) >> 63;
204 static uint64_t isc_to_isc_bits(int isc)
206 return (0x80 >> isc) << 24;
209 static inline u32 isc_to_int_word(u8 isc)
211 return ((u32)isc << 27) | 0x80000000;
214 static inline u8 int_word_to_isc(u32 int_word)
216 return (int_word & 0x38000000) >> 27;
220 * To use atomic bitmap functions, we have to provide a bitmap address
221 * that is u64 aligned. However, the ipm might be u32 aligned.
222 * Therefore, we logically start the bitmap at the very beginning of the
223 * struct and fixup the bit number.
225 #define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE)
228 * gisa_set_iam - change the GISA interruption alert mask
230 * @gisa: gisa to operate on
231 * @iam: new IAM value to use
233 * Change the IAM atomically with the next alert address and the IPM
234 * of the GISA if the GISA is not part of the GIB alert list. All three
235 * fields are located in the first long word of the GISA.
237 * Returns: 0 on success
238 * -EBUSY in case the gisa is part of the alert list
240 static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam)
245 word = READ_ONCE(gisa->u64.word[0]);
246 if ((u64)gisa != word >> 32)
248 _word = (word & ~0xffUL) | iam;
249 } while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
255 * gisa_clear_ipm - clear the GISA interruption pending mask
257 * @gisa: gisa to operate on
259 * Clear the IPM atomically with the next alert address and the IAM
260 * of the GISA unconditionally. All three fields are located in the
261 * first long word of the GISA.
263 static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa)
268 word = READ_ONCE(gisa->u64.word[0]);
269 _word = word & ~(0xffUL << 24);
270 } while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
274 * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM
276 * @gi: gisa interrupt struct to work on
278 * Atomically restores the interruption alert mask if none of the
279 * relevant ISCs are pending and return the IPM.
281 * Returns: the relevant pending ISCs
283 static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
285 u8 pending_mask, alert_mask;
289 word = READ_ONCE(gi->origin->u64.word[0]);
290 alert_mask = READ_ONCE(gi->alert.mask);
291 pending_mask = (u8)(word >> 24) & alert_mask;
294 _word = (word & ~0xffUL) | alert_mask;
295 } while (cmpxchg(&gi->origin->u64.word[0], word, _word) != word);
300 static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
302 return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa;
305 static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
307 set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
310 static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
312 return READ_ONCE(gisa->ipm);
315 static inline void gisa_clear_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
317 clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
320 static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
322 return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
325 static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
327 unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs |
328 vcpu->arch.local_int.pending_irqs;
330 pending &= ~vcpu->kvm->arch.float_int.masked_irqs;
334 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
336 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
337 unsigned long pending_mask;
339 pending_mask = pending_irqs_no_gisa(vcpu);
341 pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
345 static inline int isc_to_irq_type(unsigned long isc)
347 return IRQ_PEND_IO_ISC_0 - isc;
350 static inline int irq_type_to_isc(unsigned long irq_type)
352 return IRQ_PEND_IO_ISC_0 - irq_type;
355 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
356 unsigned long active_mask)
360 for (i = 0; i <= MAX_ISC; i++)
361 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
362 active_mask &= ~(1UL << (isc_to_irq_type(i)));
367 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
369 unsigned long active_mask;
371 active_mask = pending_irqs(vcpu);
375 if (psw_extint_disabled(vcpu))
376 active_mask &= ~IRQ_PEND_EXT_MASK;
377 if (psw_ioint_disabled(vcpu))
378 active_mask &= ~IRQ_PEND_IO_MASK;
380 active_mask = disable_iscs(vcpu, active_mask);
381 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
382 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
383 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
384 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
385 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
386 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
387 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
388 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
389 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) {
390 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
391 __clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask);
393 if (psw_mchk_disabled(vcpu))
394 active_mask &= ~IRQ_PEND_MCHK_MASK;
395 /* PV guest cpus can have a single interruption injected at a time. */
396 if (kvm_s390_pv_cpu_get_handle(vcpu) &&
397 vcpu->arch.sie_block->iictl != IICTL_CODE_NONE)
398 active_mask &= ~(IRQ_PEND_EXT_II_MASK |
402 * Check both floating and local interrupt's cr14 because
403 * bit IRQ_PEND_MCHK_REP could be set in both cases.
405 if (!(vcpu->arch.sie_block->gcr[14] &
406 (vcpu->kvm->arch.float_int.mchk.cr14 |
407 vcpu->arch.local_int.irq.mchk.cr14)))
408 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
411 * STOP irqs will never be actively delivered. They are triggered via
412 * intercept requests and cleared when the stop intercept is performed.
414 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
419 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
421 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
422 set_bit(kvm_vcpu_get_idx(vcpu), vcpu->kvm->arch.idle_mask);
425 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
427 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
428 clear_bit(kvm_vcpu_get_idx(vcpu), vcpu->kvm->arch.idle_mask);
431 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
433 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
435 vcpu->arch.sie_block->lctl = 0x0000;
436 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
438 if (guestdbg_enabled(vcpu)) {
439 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
440 LCTL_CR10 | LCTL_CR11);
441 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
445 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
447 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
449 if (psw_ioint_disabled(vcpu))
450 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
452 vcpu->arch.sie_block->lctl |= LCTL_CR6;
455 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
457 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
459 if (psw_extint_disabled(vcpu))
460 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
462 vcpu->arch.sie_block->lctl |= LCTL_CR0;
465 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
467 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
469 if (psw_mchk_disabled(vcpu))
470 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
472 vcpu->arch.sie_block->lctl |= LCTL_CR14;
475 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
477 if (kvm_s390_is_stop_irq_pending(vcpu))
478 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
481 /* Set interception request for non-deliverable interrupts */
482 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
484 set_intercept_indicators_io(vcpu);
485 set_intercept_indicators_ext(vcpu);
486 set_intercept_indicators_mchk(vcpu);
487 set_intercept_indicators_stop(vcpu);
490 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
492 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
495 vcpu->stat.deliver_cputm++;
496 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
498 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
499 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
500 vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER;
502 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
503 (u16 *)__LC_EXT_INT_CODE);
504 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
505 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
506 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
507 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
508 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
510 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
511 return rc ? -EFAULT : 0;
514 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
516 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
519 vcpu->stat.deliver_ckc++;
520 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
522 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
523 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
524 vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP;
526 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
527 (u16 __user *)__LC_EXT_INT_CODE);
528 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
529 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
530 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
531 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
532 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
534 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
535 return rc ? -EFAULT : 0;
538 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
540 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
541 struct kvm_s390_ext_info ext;
544 spin_lock(&li->lock);
546 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
547 li->irq.ext.ext_params2 = 0;
548 spin_unlock(&li->lock);
550 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
552 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
553 KVM_S390_INT_PFAULT_INIT,
556 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
557 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
558 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
559 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
560 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
561 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
562 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
563 return rc ? -EFAULT : 0;
566 static int __write_machine_check(struct kvm_vcpu *vcpu,
567 struct kvm_s390_mchk_info *mchk)
569 unsigned long ext_sa_addr;
571 freg_t fprs[NUM_FPRS];
576 * All other possible payload for a machine check (e.g. the register
577 * contents in the save area) will be handled by the ultravisor, as
578 * the hypervisor does not not have the needed information for
581 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
582 vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK;
583 vcpu->arch.sie_block->mcic = mchk->mcic;
584 vcpu->arch.sie_block->faddr = mchk->failing_storage_address;
585 vcpu->arch.sie_block->edc = mchk->ext_damage_code;
589 mci.val = mchk->mcic;
590 /* take care of lazy register loading */
592 save_access_regs(vcpu->run->s.regs.acrs);
593 if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
594 save_gs_cb(current->thread.gs_cb);
596 /* Extended save area */
597 rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
598 sizeof(unsigned long));
599 /* Only bits 0 through 63-LC are used for address formation */
600 lc = ext_sa_addr & MCESA_LC_MASK;
601 if (test_kvm_facility(vcpu->kvm, 133)) {
605 ext_sa_addr &= ~0x3ffUL;
608 ext_sa_addr &= ~0x7ffUL;
611 ext_sa_addr &= ~0xfffUL;
618 ext_sa_addr &= ~0x3ffUL;
621 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
622 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
628 if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
629 && (lc == 11 || lc == 12)) {
630 if (write_guest_abs(vcpu, ext_sa_addr + 1024,
631 &vcpu->run->s.regs.gscb, 32))
637 /* General interruption information */
638 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
639 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
640 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
641 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
642 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
643 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
645 /* Register-save areas */
646 if (MACHINE_HAS_VX) {
647 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
648 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
650 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
651 vcpu->run->s.regs.fprs, 128);
653 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
654 vcpu->run->s.regs.gprs, 128);
655 rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
656 (u32 __user *) __LC_FP_CREG_SAVE_AREA);
657 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
658 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
659 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
660 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
661 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
662 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
663 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
664 &vcpu->run->s.regs.acrs, 64);
665 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
666 &vcpu->arch.sie_block->gcr, 128);
668 /* Extended interruption information */
669 rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
670 (u32 __user *) __LC_EXT_DAMAGE_CODE);
671 rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
672 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
673 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
674 sizeof(mchk->fixed_logout));
675 return rc ? -EFAULT : 0;
678 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
680 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
681 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
682 struct kvm_s390_mchk_info mchk = {};
686 spin_lock(&fi->lock);
687 spin_lock(&li->lock);
688 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
689 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
691 * If there was an exigent machine check pending, then any
692 * repressible machine checks that might have been pending
693 * are indicated along with it, so always clear bits for
694 * repressible and exigent interrupts
697 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
698 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
699 memset(&li->irq.mchk, 0, sizeof(mchk));
703 * We indicate floating repressible conditions along with
704 * other pending conditions. Channel Report Pending and Channel
705 * Subsystem damage are the only two and and are indicated by
706 * bits in mcic and masked in cr14.
708 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
709 mchk.mcic |= fi->mchk.mcic;
710 mchk.cr14 |= fi->mchk.cr14;
711 memset(&fi->mchk, 0, sizeof(mchk));
714 spin_unlock(&li->lock);
715 spin_unlock(&fi->lock);
718 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
720 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
722 mchk.cr14, mchk.mcic);
723 vcpu->stat.deliver_machine_check++;
724 rc = __write_machine_check(vcpu, &mchk);
729 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
731 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
734 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
735 vcpu->stat.deliver_restart_signal++;
736 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
738 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
739 vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART;
741 rc = write_guest_lc(vcpu,
742 offsetof(struct lowcore, restart_old_psw),
743 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
744 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
745 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
747 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
748 return rc ? -EFAULT : 0;
751 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
753 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
754 struct kvm_s390_prefix_info prefix;
756 spin_lock(&li->lock);
757 prefix = li->irq.prefix;
758 li->irq.prefix.address = 0;
759 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
760 spin_unlock(&li->lock);
762 vcpu->stat.deliver_prefix_signal++;
763 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
764 KVM_S390_SIGP_SET_PREFIX,
767 kvm_s390_set_prefix(vcpu, prefix.address);
771 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
773 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
777 spin_lock(&li->lock);
778 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
779 clear_bit(cpu_addr, li->sigp_emerg_pending);
780 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
781 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
782 spin_unlock(&li->lock);
784 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
785 vcpu->stat.deliver_emergency_signal++;
786 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
788 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
789 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
790 vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG;
791 vcpu->arch.sie_block->extcpuaddr = cpu_addr;
795 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
796 (u16 *)__LC_EXT_INT_CODE);
797 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
798 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
799 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
800 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
801 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
802 return rc ? -EFAULT : 0;
805 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
807 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
808 struct kvm_s390_extcall_info extcall;
811 spin_lock(&li->lock);
812 extcall = li->irq.extcall;
813 li->irq.extcall.code = 0;
814 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
815 spin_unlock(&li->lock);
817 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
818 vcpu->stat.deliver_external_call++;
819 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
820 KVM_S390_INT_EXTERNAL_CALL,
822 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
823 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
824 vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL;
825 vcpu->arch.sie_block->extcpuaddr = extcall.code;
829 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
830 (u16 *)__LC_EXT_INT_CODE);
831 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
832 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
833 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
834 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
836 return rc ? -EFAULT : 0;
839 static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code)
842 case PGM_SPECIFICATION:
843 vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION;
846 vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND;
854 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
856 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
857 struct kvm_s390_pgm_info pgm_info;
858 int rc = 0, nullifying = false;
861 spin_lock(&li->lock);
862 pgm_info = li->irq.pgm;
863 clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
864 memset(&li->irq.pgm, 0, sizeof(pgm_info));
865 spin_unlock(&li->lock);
867 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
868 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
869 pgm_info.code, ilen);
870 vcpu->stat.deliver_program++;
871 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
874 /* PER is handled by the ultravisor */
875 if (kvm_s390_pv_cpu_is_protected(vcpu))
876 return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER);
878 switch (pgm_info.code & ~PGM_PER) {
879 case PGM_AFX_TRANSLATION:
880 case PGM_ASX_TRANSLATION:
881 case PGM_EX_TRANSLATION:
882 case PGM_LFX_TRANSLATION:
883 case PGM_LSTE_SEQUENCE:
884 case PGM_LSX_TRANSLATION:
885 case PGM_LX_TRANSLATION:
886 case PGM_PRIMARY_AUTHORITY:
887 case PGM_SECONDARY_AUTHORITY:
890 case PGM_SPACE_SWITCH:
891 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
892 (u64 *)__LC_TRANS_EXC_CODE);
894 case PGM_ALEN_TRANSLATION:
895 case PGM_ALE_SEQUENCE:
896 case PGM_ASTE_INSTANCE:
897 case PGM_ASTE_SEQUENCE:
898 case PGM_ASTE_VALIDITY:
899 case PGM_EXTENDED_AUTHORITY:
900 rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
901 (u8 *)__LC_EXC_ACCESS_ID);
905 case PGM_PAGE_TRANSLATION:
906 case PGM_REGION_FIRST_TRANS:
907 case PGM_REGION_SECOND_TRANS:
908 case PGM_REGION_THIRD_TRANS:
909 case PGM_SEGMENT_TRANSLATION:
910 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
911 (u64 *)__LC_TRANS_EXC_CODE);
912 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
913 (u8 *)__LC_EXC_ACCESS_ID);
914 rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
915 (u8 *)__LC_OP_ACCESS_ID);
919 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
920 (u16 *)__LC_MON_CLASS_NR);
921 rc |= put_guest_lc(vcpu, pgm_info.mon_code,
922 (u64 *)__LC_MON_CODE);
924 case PGM_VECTOR_PROCESSING:
926 rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
927 (u32 *)__LC_DATA_EXC_CODE);
930 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
931 (u64 *)__LC_TRANS_EXC_CODE);
932 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
933 (u8 *)__LC_EXC_ACCESS_ID);
936 case PGM_STACK_EMPTY:
937 case PGM_STACK_SPECIFICATION:
939 case PGM_STACK_OPERATION:
940 case PGM_TRACE_TABEL:
941 case PGM_CRYPTO_OPERATION:
946 if (pgm_info.code & PGM_PER) {
947 rc |= put_guest_lc(vcpu, pgm_info.per_code,
948 (u8 *) __LC_PER_CODE);
949 rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
950 (u8 *)__LC_PER_ATMID);
951 rc |= put_guest_lc(vcpu, pgm_info.per_address,
952 (u64 *) __LC_PER_ADDRESS);
953 rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
954 (u8 *) __LC_PER_ACCESS_ID);
957 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
958 kvm_s390_rewind_psw(vcpu, ilen);
960 /* bit 1+2 of the target are the ilc, so we can directly use ilen */
961 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
962 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
963 (u64 *) __LC_LAST_BREAK);
964 rc |= put_guest_lc(vcpu, pgm_info.code,
965 (u16 *)__LC_PGM_INT_CODE);
966 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
967 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
968 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
969 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
970 return rc ? -EFAULT : 0;
973 #define SCCB_MASK 0xFFFFFFF8
974 #define SCCB_EVENT_PENDING 0x3
976 static int write_sclp(struct kvm_vcpu *vcpu, u32 parm)
980 if (kvm_s390_pv_cpu_get_handle(vcpu)) {
981 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
982 vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG;
983 vcpu->arch.sie_block->eiparams = parm;
987 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
988 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
989 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
990 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
991 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
992 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
993 rc |= put_guest_lc(vcpu, parm,
994 (u32 *)__LC_EXT_PARAMS);
996 return rc ? -EFAULT : 0;
999 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
1001 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1002 struct kvm_s390_ext_info ext;
1004 spin_lock(&fi->lock);
1005 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) ||
1006 !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
1007 spin_unlock(&fi->lock);
1010 ext = fi->srv_signal;
1011 memset(&fi->srv_signal, 0, sizeof(ext));
1012 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1013 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1014 if (kvm_s390_pv_cpu_is_protected(vcpu))
1015 set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs);
1016 spin_unlock(&fi->lock);
1018 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
1020 vcpu->stat.deliver_service_signal++;
1021 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1024 return write_sclp(vcpu, ext.ext_params);
1027 static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)
1029 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1030 struct kvm_s390_ext_info ext;
1032 spin_lock(&fi->lock);
1033 if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) {
1034 spin_unlock(&fi->lock);
1037 ext = fi->srv_signal;
1038 /* only clear the event bit */
1039 fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING;
1040 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1041 spin_unlock(&fi->lock);
1043 VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event");
1044 vcpu->stat.deliver_service_signal++;
1045 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1048 return write_sclp(vcpu, SCCB_EVENT_PENDING);
1051 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
1053 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1054 struct kvm_s390_interrupt_info *inti;
1057 spin_lock(&fi->lock);
1058 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
1059 struct kvm_s390_interrupt_info,
1062 list_del(&inti->list);
1063 fi->counters[FIRQ_CNTR_PFAULT] -= 1;
1065 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
1066 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1067 spin_unlock(&fi->lock);
1070 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1071 KVM_S390_INT_PFAULT_DONE, 0,
1072 inti->ext.ext_params2);
1073 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
1074 inti->ext.ext_params2);
1076 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1077 (u16 *)__LC_EXT_INT_CODE);
1078 rc |= put_guest_lc(vcpu, PFAULT_DONE,
1079 (u16 *)__LC_EXT_CPU_ADDR);
1080 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1081 &vcpu->arch.sie_block->gpsw,
1083 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1084 &vcpu->arch.sie_block->gpsw,
1086 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1087 (u64 *)__LC_EXT_PARAMS2);
1090 return rc ? -EFAULT : 0;
1093 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
1095 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1096 struct kvm_s390_interrupt_info *inti;
1099 spin_lock(&fi->lock);
1100 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
1101 struct kvm_s390_interrupt_info,
1105 "deliver: virtio parm: 0x%x,parm64: 0x%llx",
1106 inti->ext.ext_params, inti->ext.ext_params2);
1107 vcpu->stat.deliver_virtio++;
1108 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1110 inti->ext.ext_params,
1111 inti->ext.ext_params2);
1112 list_del(&inti->list);
1113 fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1115 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1116 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1117 spin_unlock(&fi->lock);
1120 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1121 (u16 *)__LC_EXT_INT_CODE);
1122 rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1123 (u16 *)__LC_EXT_CPU_ADDR);
1124 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1125 &vcpu->arch.sie_block->gpsw,
1127 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1128 &vcpu->arch.sie_block->gpsw,
1130 rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1131 (u32 *)__LC_EXT_PARAMS);
1132 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1133 (u64 *)__LC_EXT_PARAMS2);
1136 return rc ? -EFAULT : 0;
1139 static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1143 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
1144 vcpu->arch.sie_block->iictl = IICTL_CODE_IO;
1145 vcpu->arch.sie_block->subchannel_id = io->subchannel_id;
1146 vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr;
1147 vcpu->arch.sie_block->io_int_parm = io->io_int_parm;
1148 vcpu->arch.sie_block->io_int_word = io->io_int_word;
1152 rc = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1153 rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1154 rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1155 rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1156 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1157 &vcpu->arch.sie_block->gpsw,
1159 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1160 &vcpu->arch.sie_block->gpsw,
1162 return rc ? -EFAULT : 0;
1165 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1166 unsigned long irq_type)
1168 struct list_head *isc_list;
1169 struct kvm_s390_float_interrupt *fi;
1170 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1171 struct kvm_s390_interrupt_info *inti = NULL;
1172 struct kvm_s390_io_info io;
1176 fi = &vcpu->kvm->arch.float_int;
1178 spin_lock(&fi->lock);
1179 isc = irq_type_to_isc(irq_type);
1180 isc_list = &fi->lists[isc];
1181 inti = list_first_entry_or_null(isc_list,
1182 struct kvm_s390_interrupt_info,
1185 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1186 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1188 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1189 inti->io.subchannel_id >> 8,
1190 inti->io.subchannel_id >> 1 & 0x3,
1191 inti->io.subchannel_nr);
1193 vcpu->stat.deliver_io++;
1194 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1196 ((__u32)inti->io.subchannel_id << 16) |
1197 inti->io.subchannel_nr,
1198 ((__u64)inti->io.io_int_parm << 32) |
1199 inti->io.io_int_word);
1200 list_del(&inti->list);
1201 fi->counters[FIRQ_CNTR_IO] -= 1;
1203 if (list_empty(isc_list))
1204 clear_bit(irq_type, &fi->pending_irqs);
1205 spin_unlock(&fi->lock);
1208 rc = __do_deliver_io(vcpu, &(inti->io));
1213 if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1215 * in case an adapter interrupt was not delivered
1216 * in SIE context KVM will handle the delivery
1218 VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1219 memset(&io, 0, sizeof(io));
1220 io.io_int_word = isc_to_int_word(isc);
1221 vcpu->stat.deliver_io++;
1222 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1223 KVM_S390_INT_IO(1, 0, 0, 0),
1224 ((__u32)io.subchannel_id << 16) |
1226 ((__u64)io.io_int_parm << 32) |
1228 rc = __do_deliver_io(vcpu, &io);
1234 /* Check whether an external call is pending (deliverable or not) */
1235 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1237 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1239 if (!sclp.has_sigpif)
1240 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1242 return sca_ext_call_pending(vcpu, NULL);
1245 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1247 if (deliverable_irqs(vcpu))
1250 if (kvm_cpu_has_pending_timer(vcpu))
1253 /* external call pending and deliverable */
1254 if (kvm_s390_ext_call_pending(vcpu) &&
1255 !psw_extint_disabled(vcpu) &&
1256 (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1259 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1264 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1266 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1269 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1271 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1272 const u64 ckc = vcpu->arch.sie_block->ckc;
1273 u64 cputm, sltime = 0;
1275 if (ckc_interrupts_enabled(vcpu)) {
1276 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1277 if ((s64)now < (s64)ckc)
1278 sltime = tod_to_ns((s64)ckc - (s64)now);
1279 } else if (now < ckc) {
1280 sltime = tod_to_ns(ckc - now);
1282 /* already expired */
1285 if (cpu_timer_interrupts_enabled(vcpu)) {
1286 cputm = kvm_s390_get_cpu_timer(vcpu);
1287 /* already expired? */
1290 return min_t(u64, sltime, tod_to_ns(cputm));
1292 } else if (cpu_timer_interrupts_enabled(vcpu)) {
1293 sltime = kvm_s390_get_cpu_timer(vcpu);
1294 /* already expired? */
1301 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1303 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1306 vcpu->stat.exit_wait_state++;
1309 if (kvm_arch_vcpu_runnable(vcpu))
1312 if (psw_interrupts_disabled(vcpu)) {
1313 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1314 return -EOPNOTSUPP; /* disabled wait */
1318 (gisa_get_ipm_or_restore_iam(gi) &
1319 vcpu->arch.sie_block->gcr[6] >> 24))
1322 if (!ckc_interrupts_enabled(vcpu) &&
1323 !cpu_timer_interrupts_enabled(vcpu)) {
1324 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1325 __set_cpu_idle(vcpu);
1329 sltime = __calculate_sltime(vcpu);
1333 __set_cpu_idle(vcpu);
1334 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1335 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1337 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1338 kvm_vcpu_block(vcpu);
1339 __unset_cpu_idle(vcpu);
1340 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1342 hrtimer_cancel(&vcpu->arch.ckc_timer);
1346 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1348 vcpu->valid_wakeup = true;
1349 kvm_vcpu_wake_up(vcpu);
1352 * The VCPU might not be sleeping but rather executing VSIE. Let's
1353 * kick it, so it leaves the SIE to process the request.
1355 kvm_s390_vsie_kick(vcpu);
1358 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1360 struct kvm_vcpu *vcpu;
1363 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1364 sltime = __calculate_sltime(vcpu);
1367 * If the monotonic clock runs faster than the tod clock we might be
1368 * woken up too early and have to go back to sleep to avoid deadlocks.
1370 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1371 return HRTIMER_RESTART;
1372 kvm_s390_vcpu_wakeup(vcpu);
1373 return HRTIMER_NORESTART;
1376 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1378 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1380 spin_lock(&li->lock);
1381 li->pending_irqs = 0;
1382 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1383 memset(&li->irq, 0, sizeof(li->irq));
1384 spin_unlock(&li->lock);
1386 sca_clear_ext_call(vcpu);
1389 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1391 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1393 unsigned long irq_type;
1396 __reset_intercept_indicators(vcpu);
1398 /* pending ckc conditions might have been invalidated */
1399 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1400 if (ckc_irq_pending(vcpu))
1401 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1403 /* pending cpu timer conditions might have been invalidated */
1404 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1405 if (cpu_timer_irq_pending(vcpu))
1406 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1408 while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1409 /* bits are in the reverse order of interrupt priority */
1410 irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1412 case IRQ_PEND_IO_ISC_0:
1413 case IRQ_PEND_IO_ISC_1:
1414 case IRQ_PEND_IO_ISC_2:
1415 case IRQ_PEND_IO_ISC_3:
1416 case IRQ_PEND_IO_ISC_4:
1417 case IRQ_PEND_IO_ISC_5:
1418 case IRQ_PEND_IO_ISC_6:
1419 case IRQ_PEND_IO_ISC_7:
1420 rc = __deliver_io(vcpu, irq_type);
1422 case IRQ_PEND_MCHK_EX:
1423 case IRQ_PEND_MCHK_REP:
1424 rc = __deliver_machine_check(vcpu);
1427 rc = __deliver_prog(vcpu);
1429 case IRQ_PEND_EXT_EMERGENCY:
1430 rc = __deliver_emergency_signal(vcpu);
1432 case IRQ_PEND_EXT_EXTERNAL:
1433 rc = __deliver_external_call(vcpu);
1435 case IRQ_PEND_EXT_CLOCK_COMP:
1436 rc = __deliver_ckc(vcpu);
1438 case IRQ_PEND_EXT_CPU_TIMER:
1439 rc = __deliver_cpu_timer(vcpu);
1441 case IRQ_PEND_RESTART:
1442 rc = __deliver_restart(vcpu);
1444 case IRQ_PEND_SET_PREFIX:
1445 rc = __deliver_set_prefix(vcpu);
1447 case IRQ_PEND_PFAULT_INIT:
1448 rc = __deliver_pfault_init(vcpu);
1450 case IRQ_PEND_EXT_SERVICE:
1451 rc = __deliver_service(vcpu);
1453 case IRQ_PEND_EXT_SERVICE_EV:
1454 rc = __deliver_service_ev(vcpu);
1456 case IRQ_PEND_PFAULT_DONE:
1457 rc = __deliver_pfault_done(vcpu);
1459 case IRQ_PEND_VIRTIO:
1460 rc = __deliver_virtio(vcpu);
1463 WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1464 clear_bit(irq_type, &li->pending_irqs);
1468 set_intercept_indicators(vcpu);
1473 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1475 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1477 vcpu->stat.inject_program++;
1478 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1479 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1480 irq->u.pgm.code, 0);
1482 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1483 /* auto detection if no valid ILC was given */
1484 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1485 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1486 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1489 if (irq->u.pgm.code == PGM_PER) {
1490 li->irq.pgm.code |= PGM_PER;
1491 li->irq.pgm.flags = irq->u.pgm.flags;
1492 /* only modify PER related information */
1493 li->irq.pgm.per_address = irq->u.pgm.per_address;
1494 li->irq.pgm.per_code = irq->u.pgm.per_code;
1495 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1496 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1497 } else if (!(irq->u.pgm.code & PGM_PER)) {
1498 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1500 li->irq.pgm.flags = irq->u.pgm.flags;
1501 /* only modify non-PER information */
1502 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1503 li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1504 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1505 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1506 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1507 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1509 li->irq.pgm = irq->u.pgm;
1511 set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1515 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1517 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1519 vcpu->stat.inject_pfault_init++;
1520 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1521 irq->u.ext.ext_params2);
1522 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1523 irq->u.ext.ext_params,
1524 irq->u.ext.ext_params2);
1526 li->irq.ext = irq->u.ext;
1527 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1528 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1532 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1534 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1535 struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1536 uint16_t src_id = irq->u.extcall.code;
1538 vcpu->stat.inject_external_call++;
1539 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1541 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1544 /* sending vcpu invalid */
1545 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1548 if (sclp.has_sigpif && !kvm_s390_pv_cpu_get_handle(vcpu))
1549 return sca_inject_ext_call(vcpu, src_id);
1551 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1553 *extcall = irq->u.extcall;
1554 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1558 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1560 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1561 struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1563 vcpu->stat.inject_set_prefix++;
1564 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1565 irq->u.prefix.address);
1566 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1567 irq->u.prefix.address, 0);
1569 if (!is_vcpu_stopped(vcpu))
1572 *prefix = irq->u.prefix;
1573 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1577 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1578 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1580 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1581 struct kvm_s390_stop_info *stop = &li->irq.stop;
1584 vcpu->stat.inject_stop_signal++;
1585 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1587 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1590 if (is_vcpu_stopped(vcpu)) {
1591 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1592 rc = kvm_s390_store_status_unloaded(vcpu,
1593 KVM_S390_STORE_STATUS_NOADDR);
1597 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1599 stop->flags = irq->u.stop.flags;
1600 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1604 static int __inject_sigp_restart(struct kvm_vcpu *vcpu)
1606 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1608 vcpu->stat.inject_restart++;
1609 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1610 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1612 set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1616 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1617 struct kvm_s390_irq *irq)
1619 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1621 vcpu->stat.inject_emergency_signal++;
1622 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1624 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1625 irq->u.emerg.code, 0);
1627 /* sending vcpu invalid */
1628 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1631 set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1632 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1633 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1637 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1639 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1640 struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1642 vcpu->stat.inject_mchk++;
1643 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1645 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1649 * Because repressible machine checks can be indicated along with
1650 * exigent machine checks (PoP, Chapter 11, Interruption action)
1651 * we need to combine cr14, mcic and external damage code.
1652 * Failing storage address and the logout area should not be or'ed
1653 * together, we just indicate the last occurrence of the corresponding
1656 mchk->cr14 |= irq->u.mchk.cr14;
1657 mchk->mcic |= irq->u.mchk.mcic;
1658 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1659 mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1660 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1661 sizeof(mchk->fixed_logout));
1662 if (mchk->mcic & MCHK_EX_MASK)
1663 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1664 else if (mchk->mcic & MCHK_REP_MASK)
1665 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
1669 static int __inject_ckc(struct kvm_vcpu *vcpu)
1671 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1673 vcpu->stat.inject_ckc++;
1674 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1675 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1678 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1679 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1683 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1685 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1687 vcpu->stat.inject_cputm++;
1688 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1689 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1692 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1693 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1697 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1700 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1701 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1702 struct kvm_s390_interrupt_info *iter;
1703 u16 id = (schid & 0xffff0000U) >> 16;
1704 u16 nr = schid & 0x0000ffffU;
1706 spin_lock(&fi->lock);
1707 list_for_each_entry(iter, isc_list, list) {
1708 if (schid && (id != iter->io.subchannel_id ||
1709 nr != iter->io.subchannel_nr))
1711 /* found an appropriate entry */
1712 list_del_init(&iter->list);
1713 fi->counters[FIRQ_CNTR_IO] -= 1;
1714 if (list_empty(isc_list))
1715 clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1716 spin_unlock(&fi->lock);
1719 spin_unlock(&fi->lock);
1723 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1724 u64 isc_mask, u32 schid)
1726 struct kvm_s390_interrupt_info *inti = NULL;
1729 for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1730 if (isc_mask & isc_to_isc_bits(isc))
1731 inti = get_io_int(kvm, isc, schid);
1736 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1738 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1739 unsigned long active_mask;
1747 active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1748 while (active_mask) {
1749 isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1750 if (gisa_tac_ipm_gisc(gi->origin, isc))
1752 clear_bit_inv(isc, &active_mask);
1759 * Dequeue and return an I/O interrupt matching any of the interruption
1760 * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1761 * Take into account the interrupts pending in the interrupt list and in GISA.
1763 * Note that for a guest that does not enable I/O interrupts
1764 * but relies on TPI, a flood of classic interrupts may starve
1765 * out adapter interrupts on the same isc. Linux does not do
1766 * that, and it is possible to work around the issue by configuring
1767 * different iscs for classic and adapter interrupts in the guest,
1768 * but we may want to revisit this in the future.
1770 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1771 u64 isc_mask, u32 schid)
1773 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1774 struct kvm_s390_interrupt_info *inti, *tmp_inti;
1777 inti = get_top_io_int(kvm, isc_mask, schid);
1779 isc = get_top_gisa_isc(kvm, isc_mask, schid);
1785 /* AI in GISA but no classical IO int */
1788 /* both types of interrupts present */
1789 if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1790 /* classical IO int with higher priority */
1791 gisa_set_ipm_gisc(gi->origin, isc);
1795 tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
1797 tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1798 tmp_inti->io.io_int_word = isc_to_int_word(isc);
1800 kvm_s390_reinject_io_int(kvm, inti);
1803 gisa_set_ipm_gisc(gi->origin, isc);
1808 static int __inject_service(struct kvm *kvm,
1809 struct kvm_s390_interrupt_info *inti)
1811 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1813 kvm->stat.inject_service_signal++;
1814 spin_lock(&fi->lock);
1815 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1817 /* We always allow events, track them separately from the sccb ints */
1818 if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING)
1819 set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1822 * Early versions of the QEMU s390 bios will inject several
1823 * service interrupts after another without handling a
1824 * condition code indicating busy.
1825 * We will silently ignore those superfluous sccb values.
1826 * A future version of QEMU will take care of serialization
1829 if (fi->srv_signal.ext_params & SCCB_MASK)
1831 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1832 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1834 spin_unlock(&fi->lock);
1839 static int __inject_virtio(struct kvm *kvm,
1840 struct kvm_s390_interrupt_info *inti)
1842 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1844 kvm->stat.inject_virtio++;
1845 spin_lock(&fi->lock);
1846 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1847 spin_unlock(&fi->lock);
1850 fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1851 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1852 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1853 spin_unlock(&fi->lock);
1857 static int __inject_pfault_done(struct kvm *kvm,
1858 struct kvm_s390_interrupt_info *inti)
1860 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1862 kvm->stat.inject_pfault_done++;
1863 spin_lock(&fi->lock);
1864 if (fi->counters[FIRQ_CNTR_PFAULT] >=
1865 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1866 spin_unlock(&fi->lock);
1869 fi->counters[FIRQ_CNTR_PFAULT] += 1;
1870 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1871 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1872 spin_unlock(&fi->lock);
1876 #define CR_PENDING_SUBCLASS 28
1877 static int __inject_float_mchk(struct kvm *kvm,
1878 struct kvm_s390_interrupt_info *inti)
1880 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1882 kvm->stat.inject_float_mchk++;
1883 spin_lock(&fi->lock);
1884 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1885 fi->mchk.mcic |= inti->mchk.mcic;
1886 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1887 spin_unlock(&fi->lock);
1892 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1894 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1895 struct kvm_s390_float_interrupt *fi;
1896 struct list_head *list;
1899 kvm->stat.inject_io++;
1900 isc = int_word_to_isc(inti->io.io_int_word);
1903 * Do not make use of gisa in protected mode. We do not use the lock
1904 * checking variant as this is just a performance optimization and we
1905 * do not hold the lock here. This is ok as the code will pick
1906 * interrupts from both "lists" for delivery.
1908 if (!kvm_s390_pv_get_handle(kvm) &&
1909 gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1910 VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1911 gisa_set_ipm_gisc(gi->origin, isc);
1916 fi = &kvm->arch.float_int;
1917 spin_lock(&fi->lock);
1918 if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1919 spin_unlock(&fi->lock);
1922 fi->counters[FIRQ_CNTR_IO] += 1;
1924 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1925 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1927 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1928 inti->io.subchannel_id >> 8,
1929 inti->io.subchannel_id >> 1 & 0x3,
1930 inti->io.subchannel_nr);
1931 list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1932 list_add_tail(&inti->list, list);
1933 set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1934 spin_unlock(&fi->lock);
1939 * Find a destination VCPU for a floating irq and kick it.
1941 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1943 struct kvm_vcpu *dst_vcpu;
1944 int sigcpu, online_vcpus, nr_tries = 0;
1946 online_vcpus = atomic_read(&kvm->online_vcpus);
1950 /* find idle VCPUs first, then round robin */
1951 sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1952 if (sigcpu == online_vcpus) {
1954 sigcpu = kvm->arch.float_int.next_rr_cpu++;
1955 kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1956 /* avoid endless loops if all vcpus are stopped */
1957 if (nr_tries++ >= online_vcpus)
1959 } while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1961 dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1963 /* make the VCPU drop out of the SIE, or wake it up if sleeping */
1966 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1968 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1969 if (!(type & KVM_S390_INT_IO_AI_MASK &&
1970 kvm->arch.gisa_int.origin) ||
1971 kvm_s390_pv_cpu_get_handle(dst_vcpu))
1972 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1975 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1978 kvm_s390_vcpu_wakeup(dst_vcpu);
1981 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1983 u64 type = READ_ONCE(inti->type);
1988 rc = __inject_float_mchk(kvm, inti);
1990 case KVM_S390_INT_VIRTIO:
1991 rc = __inject_virtio(kvm, inti);
1993 case KVM_S390_INT_SERVICE:
1994 rc = __inject_service(kvm, inti);
1996 case KVM_S390_INT_PFAULT_DONE:
1997 rc = __inject_pfault_done(kvm, inti);
1999 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2000 rc = __inject_io(kvm, inti);
2008 __floating_irq_kick(kvm, type);
2012 int kvm_s390_inject_vm(struct kvm *kvm,
2013 struct kvm_s390_interrupt *s390int)
2015 struct kvm_s390_interrupt_info *inti;
2018 inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2022 inti->type = s390int->type;
2023 switch (inti->type) {
2024 case KVM_S390_INT_VIRTIO:
2025 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
2026 s390int->parm, s390int->parm64);
2027 inti->ext.ext_params = s390int->parm;
2028 inti->ext.ext_params2 = s390int->parm64;
2030 case KVM_S390_INT_SERVICE:
2031 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
2032 inti->ext.ext_params = s390int->parm;
2034 case KVM_S390_INT_PFAULT_DONE:
2035 inti->ext.ext_params2 = s390int->parm64;
2038 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
2040 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
2041 inti->mchk.mcic = s390int->parm64;
2043 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2044 inti->io.subchannel_id = s390int->parm >> 16;
2045 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
2046 inti->io.io_int_parm = s390int->parm64 >> 32;
2047 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
2053 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
2056 rc = __inject_vm(kvm, inti);
2062 int kvm_s390_reinject_io_int(struct kvm *kvm,
2063 struct kvm_s390_interrupt_info *inti)
2065 return __inject_vm(kvm, inti);
2068 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
2069 struct kvm_s390_irq *irq)
2071 irq->type = s390int->type;
2072 switch (irq->type) {
2073 case KVM_S390_PROGRAM_INT:
2074 if (s390int->parm & 0xffff0000)
2076 irq->u.pgm.code = s390int->parm;
2078 case KVM_S390_SIGP_SET_PREFIX:
2079 irq->u.prefix.address = s390int->parm;
2081 case KVM_S390_SIGP_STOP:
2082 irq->u.stop.flags = s390int->parm;
2084 case KVM_S390_INT_EXTERNAL_CALL:
2085 if (s390int->parm & 0xffff0000)
2087 irq->u.extcall.code = s390int->parm;
2089 case KVM_S390_INT_EMERGENCY:
2090 if (s390int->parm & 0xffff0000)
2092 irq->u.emerg.code = s390int->parm;
2095 irq->u.mchk.mcic = s390int->parm64;
2097 case KVM_S390_INT_PFAULT_INIT:
2098 irq->u.ext.ext_params = s390int->parm;
2099 irq->u.ext.ext_params2 = s390int->parm64;
2101 case KVM_S390_RESTART:
2102 case KVM_S390_INT_CLOCK_COMP:
2103 case KVM_S390_INT_CPU_TIMER:
2111 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
2113 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2115 return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2118 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
2120 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2122 spin_lock(&li->lock);
2123 li->irq.stop.flags = 0;
2124 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2125 spin_unlock(&li->lock);
2128 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2132 switch (irq->type) {
2133 case KVM_S390_PROGRAM_INT:
2134 rc = __inject_prog(vcpu, irq);
2136 case KVM_S390_SIGP_SET_PREFIX:
2137 rc = __inject_set_prefix(vcpu, irq);
2139 case KVM_S390_SIGP_STOP:
2140 rc = __inject_sigp_stop(vcpu, irq);
2142 case KVM_S390_RESTART:
2143 rc = __inject_sigp_restart(vcpu);
2145 case KVM_S390_INT_CLOCK_COMP:
2146 rc = __inject_ckc(vcpu);
2148 case KVM_S390_INT_CPU_TIMER:
2149 rc = __inject_cpu_timer(vcpu);
2151 case KVM_S390_INT_EXTERNAL_CALL:
2152 rc = __inject_extcall(vcpu, irq);
2154 case KVM_S390_INT_EMERGENCY:
2155 rc = __inject_sigp_emergency(vcpu, irq);
2158 rc = __inject_mchk(vcpu, irq);
2160 case KVM_S390_INT_PFAULT_INIT:
2161 rc = __inject_pfault_init(vcpu, irq);
2163 case KVM_S390_INT_VIRTIO:
2164 case KVM_S390_INT_SERVICE:
2165 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2173 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2175 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2178 spin_lock(&li->lock);
2179 rc = do_inject_vcpu(vcpu, irq);
2180 spin_unlock(&li->lock);
2182 kvm_s390_vcpu_wakeup(vcpu);
2186 static inline void clear_irq_list(struct list_head *_list)
2188 struct kvm_s390_interrupt_info *inti, *n;
2190 list_for_each_entry_safe(inti, n, _list, list) {
2191 list_del(&inti->list);
2196 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2197 struct kvm_s390_irq *irq)
2199 irq->type = inti->type;
2200 switch (inti->type) {
2201 case KVM_S390_INT_PFAULT_INIT:
2202 case KVM_S390_INT_PFAULT_DONE:
2203 case KVM_S390_INT_VIRTIO:
2204 irq->u.ext = inti->ext;
2206 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2207 irq->u.io = inti->io;
2212 void kvm_s390_clear_float_irqs(struct kvm *kvm)
2214 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2217 mutex_lock(&kvm->lock);
2218 if (!kvm_s390_pv_is_protected(kvm))
2219 fi->masked_irqs = 0;
2220 mutex_unlock(&kvm->lock);
2221 spin_lock(&fi->lock);
2222 fi->pending_irqs = 0;
2223 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2224 memset(&fi->mchk, 0, sizeof(fi->mchk));
2225 for (i = 0; i < FIRQ_LIST_COUNT; i++)
2226 clear_irq_list(&fi->lists[i]);
2227 for (i = 0; i < FIRQ_MAX_COUNT; i++)
2228 fi->counters[i] = 0;
2229 spin_unlock(&fi->lock);
2230 kvm_s390_gisa_clear(kvm);
2233 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2235 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2236 struct kvm_s390_interrupt_info *inti;
2237 struct kvm_s390_float_interrupt *fi;
2238 struct kvm_s390_irq *buf;
2239 struct kvm_s390_irq *irq;
2245 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2249 * We are already using -ENOMEM to signal
2250 * userspace it may retry with a bigger buffer,
2251 * so we need to use something else for this case
2257 max_irqs = len / sizeof(struct kvm_s390_irq);
2259 if (gi->origin && gisa_get_ipm(gi->origin)) {
2260 for (i = 0; i <= MAX_ISC; i++) {
2261 if (n == max_irqs) {
2262 /* signal userspace to try again */
2266 if (gisa_tac_ipm_gisc(gi->origin, i)) {
2267 irq = (struct kvm_s390_irq *) &buf[n];
2268 irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2269 irq->u.io.io_int_word = isc_to_int_word(i);
2274 fi = &kvm->arch.float_int;
2275 spin_lock(&fi->lock);
2276 for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2277 list_for_each_entry(inti, &fi->lists[i], list) {
2278 if (n == max_irqs) {
2279 /* signal userspace to try again */
2283 inti_to_irq(inti, &buf[n]);
2287 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) ||
2288 test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) {
2289 if (n == max_irqs) {
2290 /* signal userspace to try again */
2294 irq = (struct kvm_s390_irq *) &buf[n];
2295 irq->type = KVM_S390_INT_SERVICE;
2296 irq->u.ext = fi->srv_signal;
2299 if (test_bit(IRQ_PEND_MCHK_REP, &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_MCHK;
2307 irq->u.mchk = fi->mchk;
2312 spin_unlock(&fi->lock);
2314 if (!ret && n > 0) {
2315 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2320 return ret < 0 ? ret : n;
2323 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2325 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2326 struct kvm_s390_ais_all ais;
2328 if (attr->attr < sizeof(ais))
2331 if (!test_kvm_facility(kvm, 72))
2334 mutex_lock(&fi->ais_lock);
2335 ais.simm = fi->simm;
2336 ais.nimm = fi->nimm;
2337 mutex_unlock(&fi->ais_lock);
2339 if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2345 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2349 switch (attr->group) {
2350 case KVM_DEV_FLIC_GET_ALL_IRQS:
2351 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2354 case KVM_DEV_FLIC_AISM_ALL:
2355 r = flic_ais_mode_get_all(dev->kvm, attr);
2364 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2367 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2368 void *target = NULL;
2369 void __user *source;
2372 if (get_user(inti->type, (u64 __user *)addr))
2375 switch (inti->type) {
2376 case KVM_S390_INT_PFAULT_INIT:
2377 case KVM_S390_INT_PFAULT_DONE:
2378 case KVM_S390_INT_VIRTIO:
2379 case KVM_S390_INT_SERVICE:
2380 target = (void *) &inti->ext;
2381 source = &uptr->u.ext;
2382 size = sizeof(inti->ext);
2384 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2385 target = (void *) &inti->io;
2386 source = &uptr->u.io;
2387 size = sizeof(inti->io);
2390 target = (void *) &inti->mchk;
2391 source = &uptr->u.mchk;
2392 size = sizeof(inti->mchk);
2398 if (copy_from_user(target, source, size))
2404 static int enqueue_floating_irq(struct kvm_device *dev,
2405 struct kvm_device_attr *attr)
2407 struct kvm_s390_interrupt_info *inti = NULL;
2409 int len = attr->attr;
2411 if (len % sizeof(struct kvm_s390_irq) != 0)
2413 else if (len > KVM_S390_FLIC_MAX_BUFFER)
2416 while (len >= sizeof(struct kvm_s390_irq)) {
2417 inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2421 r = copy_irq_from_user(inti, attr->addr);
2426 r = __inject_vm(dev->kvm, inti);
2431 len -= sizeof(struct kvm_s390_irq);
2432 attr->addr += sizeof(struct kvm_s390_irq);
2438 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2440 if (id >= MAX_S390_IO_ADAPTERS)
2442 id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2443 return kvm->arch.adapters[id];
2446 static int register_io_adapter(struct kvm_device *dev,
2447 struct kvm_device_attr *attr)
2449 struct s390_io_adapter *adapter;
2450 struct kvm_s390_io_adapter adapter_info;
2452 if (copy_from_user(&adapter_info,
2453 (void __user *)attr->addr, sizeof(adapter_info)))
2456 if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2459 adapter_info.id = array_index_nospec(adapter_info.id,
2460 MAX_S390_IO_ADAPTERS);
2462 if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2465 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL_ACCOUNT);
2469 adapter->id = adapter_info.id;
2470 adapter->isc = adapter_info.isc;
2471 adapter->maskable = adapter_info.maskable;
2472 adapter->masked = false;
2473 adapter->swap = adapter_info.swap;
2474 adapter->suppressible = (adapter_info.flags) &
2475 KVM_S390_ADAPTER_SUPPRESSIBLE;
2476 dev->kvm->arch.adapters[adapter->id] = adapter;
2481 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2484 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2486 if (!adapter || !adapter->maskable)
2488 ret = adapter->masked;
2489 adapter->masked = masked;
2493 void kvm_s390_destroy_adapters(struct kvm *kvm)
2497 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++)
2498 kfree(kvm->arch.adapters[i]);
2501 static int modify_io_adapter(struct kvm_device *dev,
2502 struct kvm_device_attr *attr)
2504 struct kvm_s390_io_adapter_req req;
2505 struct s390_io_adapter *adapter;
2508 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2511 adapter = get_io_adapter(dev->kvm, req.id);
2515 case KVM_S390_IO_ADAPTER_MASK:
2516 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2521 * The following operations are no longer needed and therefore no-ops.
2522 * The gpa to hva translation is done when an IRQ route is set up. The
2523 * set_irq code uses get_user_pages_remote() to do the actual write.
2525 case KVM_S390_IO_ADAPTER_MAP:
2526 case KVM_S390_IO_ADAPTER_UNMAP:
2536 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2539 const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2544 if (attr->attr != sizeof(schid))
2546 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2550 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2552 * If userspace is conforming to the architecture, we can have at most
2553 * one pending I/O interrupt per subchannel, so this is effectively a
2559 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2561 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2562 struct kvm_s390_ais_req req;
2565 if (!test_kvm_facility(kvm, 72))
2568 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2571 if (req.isc > MAX_ISC)
2574 trace_kvm_s390_modify_ais_mode(req.isc,
2575 (fi->simm & AIS_MODE_MASK(req.isc)) ?
2576 (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2577 2 : KVM_S390_AIS_MODE_SINGLE :
2578 KVM_S390_AIS_MODE_ALL, req.mode);
2580 mutex_lock(&fi->ais_lock);
2582 case KVM_S390_AIS_MODE_ALL:
2583 fi->simm &= ~AIS_MODE_MASK(req.isc);
2584 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2586 case KVM_S390_AIS_MODE_SINGLE:
2587 fi->simm |= AIS_MODE_MASK(req.isc);
2588 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2593 mutex_unlock(&fi->ais_lock);
2598 static int kvm_s390_inject_airq(struct kvm *kvm,
2599 struct s390_io_adapter *adapter)
2601 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2602 struct kvm_s390_interrupt s390int = {
2603 .type = KVM_S390_INT_IO(1, 0, 0, 0),
2605 .parm64 = isc_to_int_word(adapter->isc),
2609 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2610 return kvm_s390_inject_vm(kvm, &s390int);
2612 mutex_lock(&fi->ais_lock);
2613 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2614 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2618 ret = kvm_s390_inject_vm(kvm, &s390int);
2619 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2620 fi->nimm |= AIS_MODE_MASK(adapter->isc);
2621 trace_kvm_s390_modify_ais_mode(adapter->isc,
2622 KVM_S390_AIS_MODE_SINGLE, 2);
2625 mutex_unlock(&fi->ais_lock);
2629 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2631 unsigned int id = attr->attr;
2632 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2637 return kvm_s390_inject_airq(kvm, adapter);
2640 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2642 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2643 struct kvm_s390_ais_all ais;
2645 if (!test_kvm_facility(kvm, 72))
2648 if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2651 mutex_lock(&fi->ais_lock);
2652 fi->simm = ais.simm;
2653 fi->nimm = ais.nimm;
2654 mutex_unlock(&fi->ais_lock);
2659 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2663 struct kvm_vcpu *vcpu;
2665 switch (attr->group) {
2666 case KVM_DEV_FLIC_ENQUEUE:
2667 r = enqueue_floating_irq(dev, attr);
2669 case KVM_DEV_FLIC_CLEAR_IRQS:
2670 kvm_s390_clear_float_irqs(dev->kvm);
2672 case KVM_DEV_FLIC_APF_ENABLE:
2673 dev->kvm->arch.gmap->pfault_enabled = 1;
2675 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2676 dev->kvm->arch.gmap->pfault_enabled = 0;
2678 * Make sure no async faults are in transition when
2679 * clearing the queues. So we don't need to worry
2680 * about late coming workers.
2682 synchronize_srcu(&dev->kvm->srcu);
2683 kvm_for_each_vcpu(i, vcpu, dev->kvm)
2684 kvm_clear_async_pf_completion_queue(vcpu);
2686 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2687 r = register_io_adapter(dev, attr);
2689 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2690 r = modify_io_adapter(dev, attr);
2692 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2693 r = clear_io_irq(dev->kvm, attr);
2695 case KVM_DEV_FLIC_AISM:
2696 r = modify_ais_mode(dev->kvm, attr);
2698 case KVM_DEV_FLIC_AIRQ_INJECT:
2699 r = flic_inject_airq(dev->kvm, attr);
2701 case KVM_DEV_FLIC_AISM_ALL:
2702 r = flic_ais_mode_set_all(dev->kvm, attr);
2711 static int flic_has_attr(struct kvm_device *dev,
2712 struct kvm_device_attr *attr)
2714 switch (attr->group) {
2715 case KVM_DEV_FLIC_GET_ALL_IRQS:
2716 case KVM_DEV_FLIC_ENQUEUE:
2717 case KVM_DEV_FLIC_CLEAR_IRQS:
2718 case KVM_DEV_FLIC_APF_ENABLE:
2719 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2720 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2721 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2722 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2723 case KVM_DEV_FLIC_AISM:
2724 case KVM_DEV_FLIC_AIRQ_INJECT:
2725 case KVM_DEV_FLIC_AISM_ALL:
2731 static int flic_create(struct kvm_device *dev, u32 type)
2735 if (dev->kvm->arch.flic)
2737 dev->kvm->arch.flic = dev;
2741 static void flic_destroy(struct kvm_device *dev)
2743 dev->kvm->arch.flic = NULL;
2747 /* s390 floating irq controller (flic) */
2748 struct kvm_device_ops kvm_flic_ops = {
2750 .get_attr = flic_get_attr,
2751 .set_attr = flic_set_attr,
2752 .has_attr = flic_has_attr,
2753 .create = flic_create,
2754 .destroy = flic_destroy,
2757 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2761 bit = bit_nr + (addr % PAGE_SIZE) * 8;
2763 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2766 static struct page *get_map_page(struct kvm *kvm, u64 uaddr)
2768 struct page *page = NULL;
2770 mmap_read_lock(kvm->mm);
2771 get_user_pages_remote(kvm->mm, uaddr, 1, FOLL_WRITE,
2773 mmap_read_unlock(kvm->mm);
2777 static int adapter_indicators_set(struct kvm *kvm,
2778 struct s390_io_adapter *adapter,
2779 struct kvm_s390_adapter_int *adapter_int)
2782 int summary_set, idx;
2783 struct page *ind_page, *summary_page;
2786 ind_page = get_map_page(kvm, adapter_int->ind_addr);
2789 summary_page = get_map_page(kvm, adapter_int->summary_addr);
2790 if (!summary_page) {
2795 idx = srcu_read_lock(&kvm->srcu);
2796 map = page_address(ind_page);
2797 bit = get_ind_bit(adapter_int->ind_addr,
2798 adapter_int->ind_offset, adapter->swap);
2800 mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT);
2801 set_page_dirty_lock(ind_page);
2802 map = page_address(summary_page);
2803 bit = get_ind_bit(adapter_int->summary_addr,
2804 adapter_int->summary_offset, adapter->swap);
2805 summary_set = test_and_set_bit(bit, map);
2806 mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT);
2807 set_page_dirty_lock(summary_page);
2808 srcu_read_unlock(&kvm->srcu, idx);
2811 put_page(summary_page);
2812 return summary_set ? 0 : 1;
2816 * < 0 - not injected due to error
2817 * = 0 - coalesced, summary indicator already active
2818 * > 0 - injected interrupt
2820 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2821 struct kvm *kvm, int irq_source_id, int level,
2825 struct s390_io_adapter *adapter;
2827 /* We're only interested in the 0->1 transition. */
2830 adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2833 ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2834 if ((ret > 0) && !adapter->masked) {
2835 ret = kvm_s390_inject_airq(kvm, adapter);
2843 * Inject the machine check to the guest.
2845 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2846 struct mcck_volatile_info *mcck_info)
2848 struct kvm_s390_interrupt_info inti;
2849 struct kvm_s390_irq irq;
2850 struct kvm_s390_mchk_info *mchk;
2852 __u64 cr14 = 0; /* upper bits are not used */
2855 mci.val = mcck_info->mcic;
2857 cr14 |= CR14_RECOVERY_SUBMASK;
2859 cr14 |= CR14_DEGRADATION_SUBMASK;
2861 cr14 |= CR14_WARNING_SUBMASK;
2863 mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2865 mchk->mcic = mcck_info->mcic;
2866 mchk->ext_damage_code = mcck_info->ext_damage_code;
2867 mchk->failing_storage_address = mcck_info->failing_storage_address;
2869 /* Inject the floating machine check */
2870 inti.type = KVM_S390_MCHK;
2871 rc = __inject_vm(vcpu->kvm, &inti);
2873 /* Inject the machine check to specified vcpu */
2874 irq.type = KVM_S390_MCHK;
2875 rc = kvm_s390_inject_vcpu(vcpu, &irq);
2880 int kvm_set_routing_entry(struct kvm *kvm,
2881 struct kvm_kernel_irq_routing_entry *e,
2882 const struct kvm_irq_routing_entry *ue)
2887 /* we store the userspace addresses instead of the guest addresses */
2888 case KVM_IRQ_ROUTING_S390_ADAPTER:
2889 e->set = set_adapter_int;
2890 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.summary_addr);
2891 if (uaddr == -EFAULT)
2893 e->adapter.summary_addr = uaddr;
2894 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.ind_addr);
2895 if (uaddr == -EFAULT)
2897 e->adapter.ind_addr = uaddr;
2898 e->adapter.summary_offset = ue->u.adapter.summary_offset;
2899 e->adapter.ind_offset = ue->u.adapter.ind_offset;
2900 e->adapter.adapter_id = ue->u.adapter.adapter_id;
2907 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2908 int irq_source_id, int level, bool line_status)
2913 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2915 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2916 struct kvm_s390_irq *buf;
2924 if (copy_from_user((void *) buf, irqstate, len)) {
2930 * Don't allow setting the interrupt state
2931 * when there are already interrupts pending
2933 spin_lock(&li->lock);
2934 if (li->pending_irqs) {
2939 for (n = 0; n < len / sizeof(*buf); n++) {
2940 r = do_inject_vcpu(vcpu, &buf[n]);
2946 spin_unlock(&li->lock);
2953 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2954 struct kvm_s390_irq *irq,
2955 unsigned long irq_type)
2958 case IRQ_PEND_MCHK_EX:
2959 case IRQ_PEND_MCHK_REP:
2960 irq->type = KVM_S390_MCHK;
2961 irq->u.mchk = li->irq.mchk;
2964 irq->type = KVM_S390_PROGRAM_INT;
2965 irq->u.pgm = li->irq.pgm;
2967 case IRQ_PEND_PFAULT_INIT:
2968 irq->type = KVM_S390_INT_PFAULT_INIT;
2969 irq->u.ext = li->irq.ext;
2971 case IRQ_PEND_EXT_EXTERNAL:
2972 irq->type = KVM_S390_INT_EXTERNAL_CALL;
2973 irq->u.extcall = li->irq.extcall;
2975 case IRQ_PEND_EXT_CLOCK_COMP:
2976 irq->type = KVM_S390_INT_CLOCK_COMP;
2978 case IRQ_PEND_EXT_CPU_TIMER:
2979 irq->type = KVM_S390_INT_CPU_TIMER;
2981 case IRQ_PEND_SIGP_STOP:
2982 irq->type = KVM_S390_SIGP_STOP;
2983 irq->u.stop = li->irq.stop;
2985 case IRQ_PEND_RESTART:
2986 irq->type = KVM_S390_RESTART;
2988 case IRQ_PEND_SET_PREFIX:
2989 irq->type = KVM_S390_SIGP_SET_PREFIX;
2990 irq->u.prefix = li->irq.prefix;
2995 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
2998 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
2999 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
3000 unsigned long pending_irqs;
3001 struct kvm_s390_irq irq;
3002 unsigned long irq_type;
3006 spin_lock(&li->lock);
3007 pending_irqs = li->pending_irqs;
3008 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
3009 sizeof(sigp_emerg_pending));
3010 spin_unlock(&li->lock);
3012 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
3013 memset(&irq, 0, sizeof(irq));
3014 if (irq_type == IRQ_PEND_EXT_EMERGENCY)
3016 if (n + sizeof(irq) > len)
3018 store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
3019 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3024 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
3025 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
3026 memset(&irq, 0, sizeof(irq));
3027 if (n + sizeof(irq) > len)
3029 irq.type = KVM_S390_INT_EMERGENCY;
3030 irq.u.emerg.code = cpuaddr;
3031 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3037 if (sca_ext_call_pending(vcpu, &scn)) {
3038 if (n + sizeof(irq) > len)
3040 memset(&irq, 0, sizeof(irq));
3041 irq.type = KVM_S390_INT_EXTERNAL_CALL;
3042 irq.u.extcall.code = scn;
3043 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3051 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
3053 int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus);
3054 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3055 struct kvm_vcpu *vcpu;
3057 for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) {
3058 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
3059 if (psw_ioint_disabled(vcpu))
3061 deliverable_mask &= (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
3062 if (deliverable_mask) {
3063 /* lately kicked but not yet running */
3064 if (test_and_set_bit(vcpu_idx, gi->kicked_mask))
3066 kvm_s390_vcpu_wakeup(vcpu);
3072 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3074 struct kvm_s390_gisa_interrupt *gi =
3075 container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3077 container_of(gi->origin, struct sie_page2, gisa)->kvm;
3080 pending_mask = gisa_get_ipm_or_restore_iam(gi);
3082 __airqs_kick_single_vcpu(kvm, pending_mask);
3083 hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3084 return HRTIMER_RESTART;
3087 return HRTIMER_NORESTART;
3090 #define NULL_GISA_ADDR 0x00000000UL
3091 #define NONE_GISA_ADDR 0x00000001UL
3092 #define GISA_ADDR_MASK 0xfffff000UL
3094 static void process_gib_alert_list(void)
3096 struct kvm_s390_gisa_interrupt *gi;
3097 struct kvm_s390_gisa *gisa;
3099 u32 final, origin = 0UL;
3103 * If the NONE_GISA_ADDR is still stored in the alert list
3104 * origin, we will leave the outer loop. No further GISA has
3105 * been added to the alert list by millicode while processing
3106 * the current alert list.
3108 final = (origin & NONE_GISA_ADDR);
3110 * Cut off the alert list and store the NONE_GISA_ADDR in the
3111 * alert list origin to avoid further GAL interruptions.
3112 * A new alert list can be build up by millicode in parallel
3113 * for guests not in the yet cut-off alert list. When in the
3114 * final loop, store the NULL_GISA_ADDR instead. This will re-
3115 * enable GAL interruptions on the host again.
3117 origin = xchg(&gib->alert_list_origin,
3118 (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3120 * Loop through the just cut-off alert list and start the
3121 * gisa timers to kick idle vcpus to consume the pending
3122 * interruptions asap.
3124 while (origin & GISA_ADDR_MASK) {
3125 gisa = (struct kvm_s390_gisa *)(u64)origin;
3126 origin = gisa->next_alert;
3127 gisa->next_alert = (u32)(u64)gisa;
3128 kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3129 gi = &kvm->arch.gisa_int;
3130 if (hrtimer_active(&gi->timer))
3131 hrtimer_cancel(&gi->timer);
3132 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3138 void kvm_s390_gisa_clear(struct kvm *kvm)
3140 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3144 gisa_clear_ipm(gi->origin);
3145 VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3148 void kvm_s390_gisa_init(struct kvm *kvm)
3150 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3152 if (!css_general_characteristics.aiv)
3154 gi->origin = &kvm->arch.sie_page2->gisa;
3156 spin_lock_init(&gi->alert.ref_lock);
3157 gi->expires = 50 * 1000; /* 50 usec */
3158 hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3159 gi->timer.function = gisa_vcpu_kicker;
3160 memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3161 gi->origin->next_alert = (u32)(u64)gi->origin;
3162 VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3165 void kvm_s390_gisa_destroy(struct kvm *kvm)
3167 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3172 KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
3173 kvm, gi->alert.mask);
3174 while (gisa_in_alert_list(gi->origin))
3176 hrtimer_cancel(&gi->timer);
3181 * kvm_s390_gisc_register - register a guest ISC
3183 * @kvm: the kernel vm to work with
3184 * @gisc: the guest interruption sub class to register
3186 * The function extends the vm specific alert mask to use.
3187 * The effective IAM mask in the GISA is updated as well
3188 * in case the GISA is not part of the GIB alert list.
3189 * It will be updated latest when the IAM gets restored
3190 * by gisa_get_ipm_or_restore_iam().
3192 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3193 * has registered with the channel subsystem.
3194 * -ENODEV in case the vm uses no GISA
3195 * -ERANGE in case the guest ISC is invalid
3197 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3199 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3206 spin_lock(&gi->alert.ref_lock);
3207 gi->alert.ref_count[gisc]++;
3208 if (gi->alert.ref_count[gisc] == 1) {
3209 gi->alert.mask |= 0x80 >> gisc;
3210 gisa_set_iam(gi->origin, gi->alert.mask);
3212 spin_unlock(&gi->alert.ref_lock);
3216 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3219 * kvm_s390_gisc_unregister - unregister a guest ISC
3221 * @kvm: the kernel vm to work with
3222 * @gisc: the guest interruption sub class to register
3224 * The function reduces the vm specific alert mask to use.
3225 * The effective IAM mask in the GISA is updated as well
3226 * in case the GISA is not part of the GIB alert list.
3227 * It will be updated latest when the IAM gets restored
3228 * by gisa_get_ipm_or_restore_iam().
3230 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3231 * has registered with the channel subsystem.
3232 * -ENODEV in case the vm uses no GISA
3233 * -ERANGE in case the guest ISC is invalid
3234 * -EINVAL in case the guest ISC is not registered
3236 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3238 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3246 spin_lock(&gi->alert.ref_lock);
3247 if (gi->alert.ref_count[gisc] == 0) {
3251 gi->alert.ref_count[gisc]--;
3252 if (gi->alert.ref_count[gisc] == 0) {
3253 gi->alert.mask &= ~(0x80 >> gisc);
3254 gisa_set_iam(gi->origin, gi->alert.mask);
3257 spin_unlock(&gi->alert.ref_lock);
3261 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3263 static void gib_alert_irq_handler(struct airq_struct *airq, bool floating)
3265 inc_irq_stat(IRQIO_GAL);
3266 process_gib_alert_list();
3269 static struct airq_struct gib_alert_irq = {
3270 .handler = gib_alert_irq_handler,
3271 .lsi_ptr = &gib_alert_irq.lsi_mask,
3274 void kvm_s390_gib_destroy(void)
3279 unregister_adapter_interrupt(&gib_alert_irq);
3280 free_page((unsigned long)gib);
3284 int kvm_s390_gib_init(u8 nisc)
3288 if (!css_general_characteristics.aiv) {
3289 KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3293 gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3299 gib_alert_irq.isc = nisc;
3300 if (register_adapter_interrupt(&gib_alert_irq)) {
3301 pr_err("Registering the GIB alert interruption handler failed\n");
3307 if (chsc_sgib((u32)(u64)gib)) {
3308 pr_err("Associating the GIB with the AIV facility failed\n");
3309 free_page((unsigned long)gib);
3315 KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3319 unregister_adapter_interrupt(&gib_alert_irq);
3321 free_page((unsigned long)gib);