1 // SPDX-License-Identifier: GPL-2.0
3 * handling kvm guest interrupts
5 * Copyright IBM Corp. 2008, 2015
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 return vcpu->kvm->arch.float_int.pending_irqs |
328 vcpu->arch.local_int.pending_irqs;
331 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
333 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
334 unsigned long pending_mask;
336 pending_mask = pending_irqs_no_gisa(vcpu);
338 pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
342 static inline int isc_to_irq_type(unsigned long isc)
344 return IRQ_PEND_IO_ISC_0 - isc;
347 static inline int irq_type_to_isc(unsigned long irq_type)
349 return IRQ_PEND_IO_ISC_0 - irq_type;
352 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
353 unsigned long active_mask)
357 for (i = 0; i <= MAX_ISC; i++)
358 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
359 active_mask &= ~(1UL << (isc_to_irq_type(i)));
364 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
366 unsigned long active_mask;
368 active_mask = pending_irqs(vcpu);
372 if (psw_extint_disabled(vcpu))
373 active_mask &= ~IRQ_PEND_EXT_MASK;
374 if (psw_ioint_disabled(vcpu))
375 active_mask &= ~IRQ_PEND_IO_MASK;
377 active_mask = disable_iscs(vcpu, active_mask);
378 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
379 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
380 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
381 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
382 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
383 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
384 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
385 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
386 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
387 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
388 if (psw_mchk_disabled(vcpu))
389 active_mask &= ~IRQ_PEND_MCHK_MASK;
391 * Check both floating and local interrupt's cr14 because
392 * bit IRQ_PEND_MCHK_REP could be set in both cases.
394 if (!(vcpu->arch.sie_block->gcr[14] &
395 (vcpu->kvm->arch.float_int.mchk.cr14 |
396 vcpu->arch.local_int.irq.mchk.cr14)))
397 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
400 * STOP irqs will never be actively delivered. They are triggered via
401 * intercept requests and cleared when the stop intercept is performed.
403 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
408 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
410 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
411 set_bit(vcpu->vcpu_id, vcpu->kvm->arch.idle_mask);
414 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
416 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
417 clear_bit(vcpu->vcpu_id, vcpu->kvm->arch.idle_mask);
420 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
422 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
424 vcpu->arch.sie_block->lctl = 0x0000;
425 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
427 if (guestdbg_enabled(vcpu)) {
428 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
429 LCTL_CR10 | LCTL_CR11);
430 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
434 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
436 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
438 if (psw_ioint_disabled(vcpu))
439 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
441 vcpu->arch.sie_block->lctl |= LCTL_CR6;
444 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
446 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
448 if (psw_extint_disabled(vcpu))
449 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
451 vcpu->arch.sie_block->lctl |= LCTL_CR0;
454 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
456 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
458 if (psw_mchk_disabled(vcpu))
459 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
461 vcpu->arch.sie_block->lctl |= LCTL_CR14;
464 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
466 if (kvm_s390_is_stop_irq_pending(vcpu))
467 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
470 /* Set interception request for non-deliverable interrupts */
471 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
473 set_intercept_indicators_io(vcpu);
474 set_intercept_indicators_ext(vcpu);
475 set_intercept_indicators_mchk(vcpu);
476 set_intercept_indicators_stop(vcpu);
479 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
481 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
484 vcpu->stat.deliver_cputm++;
485 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
488 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
489 (u16 *)__LC_EXT_INT_CODE);
490 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
491 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
492 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
493 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
494 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
495 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
496 return rc ? -EFAULT : 0;
499 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
501 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
504 vcpu->stat.deliver_ckc++;
505 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
508 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
509 (u16 __user *)__LC_EXT_INT_CODE);
510 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
511 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
512 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
513 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
514 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
515 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
516 return rc ? -EFAULT : 0;
519 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
521 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
522 struct kvm_s390_ext_info ext;
525 spin_lock(&li->lock);
527 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
528 li->irq.ext.ext_params2 = 0;
529 spin_unlock(&li->lock);
531 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
533 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
534 KVM_S390_INT_PFAULT_INIT,
537 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
538 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
539 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
540 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
541 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
542 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
543 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
544 return rc ? -EFAULT : 0;
547 static int __write_machine_check(struct kvm_vcpu *vcpu,
548 struct kvm_s390_mchk_info *mchk)
550 unsigned long ext_sa_addr;
552 freg_t fprs[NUM_FPRS];
556 mci.val = mchk->mcic;
557 /* take care of lazy register loading */
559 save_access_regs(vcpu->run->s.regs.acrs);
560 if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
561 save_gs_cb(current->thread.gs_cb);
563 /* Extended save area */
564 rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
565 sizeof(unsigned long));
566 /* Only bits 0 through 63-LC are used for address formation */
567 lc = ext_sa_addr & MCESA_LC_MASK;
568 if (test_kvm_facility(vcpu->kvm, 133)) {
572 ext_sa_addr &= ~0x3ffUL;
575 ext_sa_addr &= ~0x7ffUL;
578 ext_sa_addr &= ~0xfffUL;
585 ext_sa_addr &= ~0x3ffUL;
588 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
589 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
595 if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
596 && (lc == 11 || lc == 12)) {
597 if (write_guest_abs(vcpu, ext_sa_addr + 1024,
598 &vcpu->run->s.regs.gscb, 32))
604 /* General interruption information */
605 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
606 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
607 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
608 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
609 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
610 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
612 /* Register-save areas */
613 if (MACHINE_HAS_VX) {
614 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
615 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
617 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
618 vcpu->run->s.regs.fprs, 128);
620 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
621 vcpu->run->s.regs.gprs, 128);
622 rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
623 (u32 __user *) __LC_FP_CREG_SAVE_AREA);
624 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
625 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
626 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
627 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
628 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
629 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
630 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
631 &vcpu->run->s.regs.acrs, 64);
632 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
633 &vcpu->arch.sie_block->gcr, 128);
635 /* Extended interruption information */
636 rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
637 (u32 __user *) __LC_EXT_DAMAGE_CODE);
638 rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
639 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
640 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
641 sizeof(mchk->fixed_logout));
642 return rc ? -EFAULT : 0;
645 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
647 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
648 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
649 struct kvm_s390_mchk_info mchk = {};
653 spin_lock(&fi->lock);
654 spin_lock(&li->lock);
655 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
656 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
658 * If there was an exigent machine check pending, then any
659 * repressible machine checks that might have been pending
660 * are indicated along with it, so always clear bits for
661 * repressible and exigent interrupts
664 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
665 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
666 memset(&li->irq.mchk, 0, sizeof(mchk));
670 * We indicate floating repressible conditions along with
671 * other pending conditions. Channel Report Pending and Channel
672 * Subsystem damage are the only two and and are indicated by
673 * bits in mcic and masked in cr14.
675 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
676 mchk.mcic |= fi->mchk.mcic;
677 mchk.cr14 |= fi->mchk.cr14;
678 memset(&fi->mchk, 0, sizeof(mchk));
681 spin_unlock(&li->lock);
682 spin_unlock(&fi->lock);
685 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
687 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
689 mchk.cr14, mchk.mcic);
690 vcpu->stat.deliver_machine_check++;
691 rc = __write_machine_check(vcpu, &mchk);
696 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
698 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
701 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
702 vcpu->stat.deliver_restart_signal++;
703 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
705 rc = write_guest_lc(vcpu,
706 offsetof(struct lowcore, restart_old_psw),
707 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
708 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
709 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
710 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
711 return rc ? -EFAULT : 0;
714 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
716 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
717 struct kvm_s390_prefix_info prefix;
719 spin_lock(&li->lock);
720 prefix = li->irq.prefix;
721 li->irq.prefix.address = 0;
722 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
723 spin_unlock(&li->lock);
725 vcpu->stat.deliver_prefix_signal++;
726 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
727 KVM_S390_SIGP_SET_PREFIX,
730 kvm_s390_set_prefix(vcpu, prefix.address);
734 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
736 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
740 spin_lock(&li->lock);
741 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
742 clear_bit(cpu_addr, li->sigp_emerg_pending);
743 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
744 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
745 spin_unlock(&li->lock);
747 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
748 vcpu->stat.deliver_emergency_signal++;
749 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
752 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
753 (u16 *)__LC_EXT_INT_CODE);
754 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
755 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
756 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
757 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
758 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
759 return rc ? -EFAULT : 0;
762 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
764 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
765 struct kvm_s390_extcall_info extcall;
768 spin_lock(&li->lock);
769 extcall = li->irq.extcall;
770 li->irq.extcall.code = 0;
771 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
772 spin_unlock(&li->lock);
774 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
775 vcpu->stat.deliver_external_call++;
776 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
777 KVM_S390_INT_EXTERNAL_CALL,
780 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
781 (u16 *)__LC_EXT_INT_CODE);
782 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
783 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
784 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
785 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
787 return rc ? -EFAULT : 0;
790 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
792 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
793 struct kvm_s390_pgm_info pgm_info;
794 int rc = 0, nullifying = false;
797 spin_lock(&li->lock);
798 pgm_info = li->irq.pgm;
799 clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
800 memset(&li->irq.pgm, 0, sizeof(pgm_info));
801 spin_unlock(&li->lock);
803 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
804 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
805 pgm_info.code, ilen);
806 vcpu->stat.deliver_program++;
807 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
810 switch (pgm_info.code & ~PGM_PER) {
811 case PGM_AFX_TRANSLATION:
812 case PGM_ASX_TRANSLATION:
813 case PGM_EX_TRANSLATION:
814 case PGM_LFX_TRANSLATION:
815 case PGM_LSTE_SEQUENCE:
816 case PGM_LSX_TRANSLATION:
817 case PGM_LX_TRANSLATION:
818 case PGM_PRIMARY_AUTHORITY:
819 case PGM_SECONDARY_AUTHORITY:
822 case PGM_SPACE_SWITCH:
823 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
824 (u64 *)__LC_TRANS_EXC_CODE);
826 case PGM_ALEN_TRANSLATION:
827 case PGM_ALE_SEQUENCE:
828 case PGM_ASTE_INSTANCE:
829 case PGM_ASTE_SEQUENCE:
830 case PGM_ASTE_VALIDITY:
831 case PGM_EXTENDED_AUTHORITY:
832 rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
833 (u8 *)__LC_EXC_ACCESS_ID);
837 case PGM_PAGE_TRANSLATION:
838 case PGM_REGION_FIRST_TRANS:
839 case PGM_REGION_SECOND_TRANS:
840 case PGM_REGION_THIRD_TRANS:
841 case PGM_SEGMENT_TRANSLATION:
842 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
843 (u64 *)__LC_TRANS_EXC_CODE);
844 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
845 (u8 *)__LC_EXC_ACCESS_ID);
846 rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
847 (u8 *)__LC_OP_ACCESS_ID);
851 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
852 (u16 *)__LC_MON_CLASS_NR);
853 rc |= put_guest_lc(vcpu, pgm_info.mon_code,
854 (u64 *)__LC_MON_CODE);
856 case PGM_VECTOR_PROCESSING:
858 rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
859 (u32 *)__LC_DATA_EXC_CODE);
862 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
863 (u64 *)__LC_TRANS_EXC_CODE);
864 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
865 (u8 *)__LC_EXC_ACCESS_ID);
868 case PGM_STACK_EMPTY:
869 case PGM_STACK_SPECIFICATION:
871 case PGM_STACK_OPERATION:
872 case PGM_TRACE_TABEL:
873 case PGM_CRYPTO_OPERATION:
878 if (pgm_info.code & PGM_PER) {
879 rc |= put_guest_lc(vcpu, pgm_info.per_code,
880 (u8 *) __LC_PER_CODE);
881 rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
882 (u8 *)__LC_PER_ATMID);
883 rc |= put_guest_lc(vcpu, pgm_info.per_address,
884 (u64 *) __LC_PER_ADDRESS);
885 rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
886 (u8 *) __LC_PER_ACCESS_ID);
889 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
890 kvm_s390_rewind_psw(vcpu, ilen);
892 /* bit 1+2 of the target are the ilc, so we can directly use ilen */
893 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
894 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
895 (u64 *) __LC_LAST_BREAK);
896 rc |= put_guest_lc(vcpu, pgm_info.code,
897 (u16 *)__LC_PGM_INT_CODE);
898 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
899 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
900 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
901 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
902 return rc ? -EFAULT : 0;
905 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
907 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
908 struct kvm_s390_ext_info ext;
911 spin_lock(&fi->lock);
912 if (!(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
913 spin_unlock(&fi->lock);
916 ext = fi->srv_signal;
917 memset(&fi->srv_signal, 0, sizeof(ext));
918 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
919 spin_unlock(&fi->lock);
921 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
923 vcpu->stat.deliver_service_signal++;
924 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
927 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
928 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
929 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
930 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
931 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
932 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
933 rc |= put_guest_lc(vcpu, ext.ext_params,
934 (u32 *)__LC_EXT_PARAMS);
936 return rc ? -EFAULT : 0;
939 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
941 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
942 struct kvm_s390_interrupt_info *inti;
945 spin_lock(&fi->lock);
946 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
947 struct kvm_s390_interrupt_info,
950 list_del(&inti->list);
951 fi->counters[FIRQ_CNTR_PFAULT] -= 1;
953 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
954 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
955 spin_unlock(&fi->lock);
958 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
959 KVM_S390_INT_PFAULT_DONE, 0,
960 inti->ext.ext_params2);
961 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
962 inti->ext.ext_params2);
964 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
965 (u16 *)__LC_EXT_INT_CODE);
966 rc |= put_guest_lc(vcpu, PFAULT_DONE,
967 (u16 *)__LC_EXT_CPU_ADDR);
968 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
969 &vcpu->arch.sie_block->gpsw,
971 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
972 &vcpu->arch.sie_block->gpsw,
974 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
975 (u64 *)__LC_EXT_PARAMS2);
978 return rc ? -EFAULT : 0;
981 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
983 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
984 struct kvm_s390_interrupt_info *inti;
987 spin_lock(&fi->lock);
988 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
989 struct kvm_s390_interrupt_info,
993 "deliver: virtio parm: 0x%x,parm64: 0x%llx",
994 inti->ext.ext_params, inti->ext.ext_params2);
995 vcpu->stat.deliver_virtio++;
996 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
998 inti->ext.ext_params,
999 inti->ext.ext_params2);
1000 list_del(&inti->list);
1001 fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1003 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1004 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1005 spin_unlock(&fi->lock);
1008 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1009 (u16 *)__LC_EXT_INT_CODE);
1010 rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1011 (u16 *)__LC_EXT_CPU_ADDR);
1012 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1013 &vcpu->arch.sie_block->gpsw,
1015 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1016 &vcpu->arch.sie_block->gpsw,
1018 rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1019 (u32 *)__LC_EXT_PARAMS);
1020 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1021 (u64 *)__LC_EXT_PARAMS2);
1024 return rc ? -EFAULT : 0;
1027 static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1031 rc = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1032 rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1033 rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1034 rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1035 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1036 &vcpu->arch.sie_block->gpsw,
1038 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1039 &vcpu->arch.sie_block->gpsw,
1041 return rc ? -EFAULT : 0;
1044 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1045 unsigned long irq_type)
1047 struct list_head *isc_list;
1048 struct kvm_s390_float_interrupt *fi;
1049 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1050 struct kvm_s390_interrupt_info *inti = NULL;
1051 struct kvm_s390_io_info io;
1055 fi = &vcpu->kvm->arch.float_int;
1057 spin_lock(&fi->lock);
1058 isc = irq_type_to_isc(irq_type);
1059 isc_list = &fi->lists[isc];
1060 inti = list_first_entry_or_null(isc_list,
1061 struct kvm_s390_interrupt_info,
1064 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1065 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1067 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1068 inti->io.subchannel_id >> 8,
1069 inti->io.subchannel_id >> 1 & 0x3,
1070 inti->io.subchannel_nr);
1072 vcpu->stat.deliver_io++;
1073 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1075 ((__u32)inti->io.subchannel_id << 16) |
1076 inti->io.subchannel_nr,
1077 ((__u64)inti->io.io_int_parm << 32) |
1078 inti->io.io_int_word);
1079 list_del(&inti->list);
1080 fi->counters[FIRQ_CNTR_IO] -= 1;
1082 if (list_empty(isc_list))
1083 clear_bit(irq_type, &fi->pending_irqs);
1084 spin_unlock(&fi->lock);
1087 rc = __do_deliver_io(vcpu, &(inti->io));
1092 if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1094 * in case an adapter interrupt was not delivered
1095 * in SIE context KVM will handle the delivery
1097 VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1098 memset(&io, 0, sizeof(io));
1099 io.io_int_word = isc_to_int_word(isc);
1100 vcpu->stat.deliver_io++;
1101 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1102 KVM_S390_INT_IO(1, 0, 0, 0),
1103 ((__u32)io.subchannel_id << 16) |
1105 ((__u64)io.io_int_parm << 32) |
1107 rc = __do_deliver_io(vcpu, &io);
1113 /* Check whether an external call is pending (deliverable or not) */
1114 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1116 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1118 if (!sclp.has_sigpif)
1119 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1121 return sca_ext_call_pending(vcpu, NULL);
1124 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1126 if (deliverable_irqs(vcpu))
1129 if (kvm_cpu_has_pending_timer(vcpu))
1132 /* external call pending and deliverable */
1133 if (kvm_s390_ext_call_pending(vcpu) &&
1134 !psw_extint_disabled(vcpu) &&
1135 (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1138 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1143 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1145 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1148 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1150 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1151 const u64 ckc = vcpu->arch.sie_block->ckc;
1152 u64 cputm, sltime = 0;
1154 if (ckc_interrupts_enabled(vcpu)) {
1155 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1156 if ((s64)now < (s64)ckc)
1157 sltime = tod_to_ns((s64)ckc - (s64)now);
1158 } else if (now < ckc) {
1159 sltime = tod_to_ns(ckc - now);
1161 /* already expired */
1164 if (cpu_timer_interrupts_enabled(vcpu)) {
1165 cputm = kvm_s390_get_cpu_timer(vcpu);
1166 /* already expired? */
1169 return min(sltime, tod_to_ns(cputm));
1171 } else if (cpu_timer_interrupts_enabled(vcpu)) {
1172 sltime = kvm_s390_get_cpu_timer(vcpu);
1173 /* already expired? */
1180 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1182 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1185 vcpu->stat.exit_wait_state++;
1188 if (kvm_arch_vcpu_runnable(vcpu))
1191 if (psw_interrupts_disabled(vcpu)) {
1192 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1193 return -EOPNOTSUPP; /* disabled wait */
1197 (gisa_get_ipm_or_restore_iam(gi) &
1198 vcpu->arch.sie_block->gcr[6] >> 24))
1201 if (!ckc_interrupts_enabled(vcpu) &&
1202 !cpu_timer_interrupts_enabled(vcpu)) {
1203 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1204 __set_cpu_idle(vcpu);
1208 sltime = __calculate_sltime(vcpu);
1212 __set_cpu_idle(vcpu);
1213 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1214 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1216 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1217 kvm_vcpu_block(vcpu);
1218 __unset_cpu_idle(vcpu);
1219 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1221 hrtimer_cancel(&vcpu->arch.ckc_timer);
1225 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1227 vcpu->valid_wakeup = true;
1228 kvm_vcpu_wake_up(vcpu);
1231 * The VCPU might not be sleeping but rather executing VSIE. Let's
1232 * kick it, so it leaves the SIE to process the request.
1234 kvm_s390_vsie_kick(vcpu);
1237 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1239 struct kvm_vcpu *vcpu;
1242 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1243 sltime = __calculate_sltime(vcpu);
1246 * If the monotonic clock runs faster than the tod clock we might be
1247 * woken up too early and have to go back to sleep to avoid deadlocks.
1249 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1250 return HRTIMER_RESTART;
1251 kvm_s390_vcpu_wakeup(vcpu);
1252 return HRTIMER_NORESTART;
1255 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1257 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1259 spin_lock(&li->lock);
1260 li->pending_irqs = 0;
1261 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1262 memset(&li->irq, 0, sizeof(li->irq));
1263 spin_unlock(&li->lock);
1265 sca_clear_ext_call(vcpu);
1268 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1270 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1272 unsigned long irq_type;
1275 __reset_intercept_indicators(vcpu);
1277 /* pending ckc conditions might have been invalidated */
1278 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1279 if (ckc_irq_pending(vcpu))
1280 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1282 /* pending cpu timer conditions might have been invalidated */
1283 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1284 if (cpu_timer_irq_pending(vcpu))
1285 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1287 while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1288 /* bits are in the reverse order of interrupt priority */
1289 irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1291 case IRQ_PEND_IO_ISC_0:
1292 case IRQ_PEND_IO_ISC_1:
1293 case IRQ_PEND_IO_ISC_2:
1294 case IRQ_PEND_IO_ISC_3:
1295 case IRQ_PEND_IO_ISC_4:
1296 case IRQ_PEND_IO_ISC_5:
1297 case IRQ_PEND_IO_ISC_6:
1298 case IRQ_PEND_IO_ISC_7:
1299 rc = __deliver_io(vcpu, irq_type);
1301 case IRQ_PEND_MCHK_EX:
1302 case IRQ_PEND_MCHK_REP:
1303 rc = __deliver_machine_check(vcpu);
1306 rc = __deliver_prog(vcpu);
1308 case IRQ_PEND_EXT_EMERGENCY:
1309 rc = __deliver_emergency_signal(vcpu);
1311 case IRQ_PEND_EXT_EXTERNAL:
1312 rc = __deliver_external_call(vcpu);
1314 case IRQ_PEND_EXT_CLOCK_COMP:
1315 rc = __deliver_ckc(vcpu);
1317 case IRQ_PEND_EXT_CPU_TIMER:
1318 rc = __deliver_cpu_timer(vcpu);
1320 case IRQ_PEND_RESTART:
1321 rc = __deliver_restart(vcpu);
1323 case IRQ_PEND_SET_PREFIX:
1324 rc = __deliver_set_prefix(vcpu);
1326 case IRQ_PEND_PFAULT_INIT:
1327 rc = __deliver_pfault_init(vcpu);
1329 case IRQ_PEND_EXT_SERVICE:
1330 rc = __deliver_service(vcpu);
1332 case IRQ_PEND_PFAULT_DONE:
1333 rc = __deliver_pfault_done(vcpu);
1335 case IRQ_PEND_VIRTIO:
1336 rc = __deliver_virtio(vcpu);
1339 WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1340 clear_bit(irq_type, &li->pending_irqs);
1344 set_intercept_indicators(vcpu);
1349 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1351 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1353 vcpu->stat.inject_program++;
1354 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1355 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1356 irq->u.pgm.code, 0);
1358 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1359 /* auto detection if no valid ILC was given */
1360 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1361 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1362 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1365 if (irq->u.pgm.code == PGM_PER) {
1366 li->irq.pgm.code |= PGM_PER;
1367 li->irq.pgm.flags = irq->u.pgm.flags;
1368 /* only modify PER related information */
1369 li->irq.pgm.per_address = irq->u.pgm.per_address;
1370 li->irq.pgm.per_code = irq->u.pgm.per_code;
1371 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1372 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1373 } else if (!(irq->u.pgm.code & PGM_PER)) {
1374 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1376 li->irq.pgm.flags = irq->u.pgm.flags;
1377 /* only modify non-PER information */
1378 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1379 li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1380 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1381 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1382 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1383 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1385 li->irq.pgm = irq->u.pgm;
1387 set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1391 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1393 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1395 vcpu->stat.inject_pfault_init++;
1396 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1397 irq->u.ext.ext_params2);
1398 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1399 irq->u.ext.ext_params,
1400 irq->u.ext.ext_params2);
1402 li->irq.ext = irq->u.ext;
1403 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1404 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1408 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1410 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1411 struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1412 uint16_t src_id = irq->u.extcall.code;
1414 vcpu->stat.inject_external_call++;
1415 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1417 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1420 /* sending vcpu invalid */
1421 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1424 if (sclp.has_sigpif)
1425 return sca_inject_ext_call(vcpu, src_id);
1427 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1429 *extcall = irq->u.extcall;
1430 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1434 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1436 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1437 struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1439 vcpu->stat.inject_set_prefix++;
1440 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1441 irq->u.prefix.address);
1442 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1443 irq->u.prefix.address, 0);
1445 if (!is_vcpu_stopped(vcpu))
1448 *prefix = irq->u.prefix;
1449 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1453 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1454 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1456 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1457 struct kvm_s390_stop_info *stop = &li->irq.stop;
1460 vcpu->stat.inject_stop_signal++;
1461 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1463 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1466 if (is_vcpu_stopped(vcpu)) {
1467 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1468 rc = kvm_s390_store_status_unloaded(vcpu,
1469 KVM_S390_STORE_STATUS_NOADDR);
1473 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1475 stop->flags = irq->u.stop.flags;
1476 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1480 static int __inject_sigp_restart(struct kvm_vcpu *vcpu,
1481 struct kvm_s390_irq *irq)
1483 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1485 vcpu->stat.inject_restart++;
1486 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1487 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1489 set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1493 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1494 struct kvm_s390_irq *irq)
1496 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1498 vcpu->stat.inject_emergency_signal++;
1499 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1501 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1502 irq->u.emerg.code, 0);
1504 /* sending vcpu invalid */
1505 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1508 set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1509 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1510 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1514 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1516 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1517 struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1519 vcpu->stat.inject_mchk++;
1520 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1522 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1526 * Because repressible machine checks can be indicated along with
1527 * exigent machine checks (PoP, Chapter 11, Interruption action)
1528 * we need to combine cr14, mcic and external damage code.
1529 * Failing storage address and the logout area should not be or'ed
1530 * together, we just indicate the last occurrence of the corresponding
1533 mchk->cr14 |= irq->u.mchk.cr14;
1534 mchk->mcic |= irq->u.mchk.mcic;
1535 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1536 mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1537 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1538 sizeof(mchk->fixed_logout));
1539 if (mchk->mcic & MCHK_EX_MASK)
1540 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1541 else if (mchk->mcic & MCHK_REP_MASK)
1542 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
1546 static int __inject_ckc(struct kvm_vcpu *vcpu)
1548 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1550 vcpu->stat.inject_ckc++;
1551 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1552 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1555 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1556 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1560 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1562 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1564 vcpu->stat.inject_cputm++;
1565 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1566 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1569 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1570 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1574 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1577 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1578 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1579 struct kvm_s390_interrupt_info *iter;
1580 u16 id = (schid & 0xffff0000U) >> 16;
1581 u16 nr = schid & 0x0000ffffU;
1583 spin_lock(&fi->lock);
1584 list_for_each_entry(iter, isc_list, list) {
1585 if (schid && (id != iter->io.subchannel_id ||
1586 nr != iter->io.subchannel_nr))
1588 /* found an appropriate entry */
1589 list_del_init(&iter->list);
1590 fi->counters[FIRQ_CNTR_IO] -= 1;
1591 if (list_empty(isc_list))
1592 clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1593 spin_unlock(&fi->lock);
1596 spin_unlock(&fi->lock);
1600 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1601 u64 isc_mask, u32 schid)
1603 struct kvm_s390_interrupt_info *inti = NULL;
1606 for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1607 if (isc_mask & isc_to_isc_bits(isc))
1608 inti = get_io_int(kvm, isc, schid);
1613 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1615 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1616 unsigned long active_mask;
1624 active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1625 while (active_mask) {
1626 isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1627 if (gisa_tac_ipm_gisc(gi->origin, isc))
1629 clear_bit_inv(isc, &active_mask);
1636 * Dequeue and return an I/O interrupt matching any of the interruption
1637 * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1638 * Take into account the interrupts pending in the interrupt list and in GISA.
1640 * Note that for a guest that does not enable I/O interrupts
1641 * but relies on TPI, a flood of classic interrupts may starve
1642 * out adapter interrupts on the same isc. Linux does not do
1643 * that, and it is possible to work around the issue by configuring
1644 * different iscs for classic and adapter interrupts in the guest,
1645 * but we may want to revisit this in the future.
1647 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1648 u64 isc_mask, u32 schid)
1650 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1651 struct kvm_s390_interrupt_info *inti, *tmp_inti;
1654 inti = get_top_io_int(kvm, isc_mask, schid);
1656 isc = get_top_gisa_isc(kvm, isc_mask, schid);
1662 /* AI in GISA but no classical IO int */
1665 /* both types of interrupts present */
1666 if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1667 /* classical IO int with higher priority */
1668 gisa_set_ipm_gisc(gi->origin, isc);
1672 tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1674 tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1675 tmp_inti->io.io_int_word = isc_to_int_word(isc);
1677 kvm_s390_reinject_io_int(kvm, inti);
1680 gisa_set_ipm_gisc(gi->origin, isc);
1685 #define SCCB_MASK 0xFFFFFFF8
1686 #define SCCB_EVENT_PENDING 0x3
1688 static int __inject_service(struct kvm *kvm,
1689 struct kvm_s390_interrupt_info *inti)
1691 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1693 kvm->stat.inject_service_signal++;
1694 spin_lock(&fi->lock);
1695 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1697 * Early versions of the QEMU s390 bios will inject several
1698 * service interrupts after another without handling a
1699 * condition code indicating busy.
1700 * We will silently ignore those superfluous sccb values.
1701 * A future version of QEMU will take care of serialization
1704 if (fi->srv_signal.ext_params & SCCB_MASK)
1706 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1707 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1709 spin_unlock(&fi->lock);
1714 static int __inject_virtio(struct kvm *kvm,
1715 struct kvm_s390_interrupt_info *inti)
1717 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1719 kvm->stat.inject_virtio++;
1720 spin_lock(&fi->lock);
1721 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1722 spin_unlock(&fi->lock);
1725 fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1726 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1727 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1728 spin_unlock(&fi->lock);
1732 static int __inject_pfault_done(struct kvm *kvm,
1733 struct kvm_s390_interrupt_info *inti)
1735 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1737 kvm->stat.inject_pfault_done++;
1738 spin_lock(&fi->lock);
1739 if (fi->counters[FIRQ_CNTR_PFAULT] >=
1740 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1741 spin_unlock(&fi->lock);
1744 fi->counters[FIRQ_CNTR_PFAULT] += 1;
1745 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1746 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1747 spin_unlock(&fi->lock);
1751 #define CR_PENDING_SUBCLASS 28
1752 static int __inject_float_mchk(struct kvm *kvm,
1753 struct kvm_s390_interrupt_info *inti)
1755 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1757 kvm->stat.inject_float_mchk++;
1758 spin_lock(&fi->lock);
1759 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1760 fi->mchk.mcic |= inti->mchk.mcic;
1761 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1762 spin_unlock(&fi->lock);
1767 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1769 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1770 struct kvm_s390_float_interrupt *fi;
1771 struct list_head *list;
1774 kvm->stat.inject_io++;
1775 isc = int_word_to_isc(inti->io.io_int_word);
1777 if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1778 VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1779 gisa_set_ipm_gisc(gi->origin, isc);
1784 fi = &kvm->arch.float_int;
1785 spin_lock(&fi->lock);
1786 if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1787 spin_unlock(&fi->lock);
1790 fi->counters[FIRQ_CNTR_IO] += 1;
1792 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1793 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1795 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1796 inti->io.subchannel_id >> 8,
1797 inti->io.subchannel_id >> 1 & 0x3,
1798 inti->io.subchannel_nr);
1799 list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1800 list_add_tail(&inti->list, list);
1801 set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1802 spin_unlock(&fi->lock);
1807 * Find a destination VCPU for a floating irq and kick it.
1809 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1811 struct kvm_vcpu *dst_vcpu;
1812 int sigcpu, online_vcpus, nr_tries = 0;
1814 online_vcpus = atomic_read(&kvm->online_vcpus);
1818 /* find idle VCPUs first, then round robin */
1819 sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1820 if (sigcpu == online_vcpus) {
1822 sigcpu = kvm->arch.float_int.next_rr_cpu++;
1823 kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1824 /* avoid endless loops if all vcpus are stopped */
1825 if (nr_tries++ >= online_vcpus)
1827 } while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1829 dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1831 /* make the VCPU drop out of the SIE, or wake it up if sleeping */
1834 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1836 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1837 if (!(type & KVM_S390_INT_IO_AI_MASK &&
1838 kvm->arch.gisa_int.origin))
1839 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1842 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1845 kvm_s390_vcpu_wakeup(dst_vcpu);
1848 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1850 u64 type = READ_ONCE(inti->type);
1855 rc = __inject_float_mchk(kvm, inti);
1857 case KVM_S390_INT_VIRTIO:
1858 rc = __inject_virtio(kvm, inti);
1860 case KVM_S390_INT_SERVICE:
1861 rc = __inject_service(kvm, inti);
1863 case KVM_S390_INT_PFAULT_DONE:
1864 rc = __inject_pfault_done(kvm, inti);
1866 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1867 rc = __inject_io(kvm, inti);
1875 __floating_irq_kick(kvm, type);
1879 int kvm_s390_inject_vm(struct kvm *kvm,
1880 struct kvm_s390_interrupt *s390int)
1882 struct kvm_s390_interrupt_info *inti;
1885 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1889 inti->type = s390int->type;
1890 switch (inti->type) {
1891 case KVM_S390_INT_VIRTIO:
1892 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
1893 s390int->parm, s390int->parm64);
1894 inti->ext.ext_params = s390int->parm;
1895 inti->ext.ext_params2 = s390int->parm64;
1897 case KVM_S390_INT_SERVICE:
1898 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
1899 inti->ext.ext_params = s390int->parm;
1901 case KVM_S390_INT_PFAULT_DONE:
1902 inti->ext.ext_params2 = s390int->parm64;
1905 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
1907 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
1908 inti->mchk.mcic = s390int->parm64;
1910 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1911 inti->io.subchannel_id = s390int->parm >> 16;
1912 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
1913 inti->io.io_int_parm = s390int->parm64 >> 32;
1914 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
1920 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
1923 rc = __inject_vm(kvm, inti);
1929 int kvm_s390_reinject_io_int(struct kvm *kvm,
1930 struct kvm_s390_interrupt_info *inti)
1932 return __inject_vm(kvm, inti);
1935 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
1936 struct kvm_s390_irq *irq)
1938 irq->type = s390int->type;
1939 switch (irq->type) {
1940 case KVM_S390_PROGRAM_INT:
1941 if (s390int->parm & 0xffff0000)
1943 irq->u.pgm.code = s390int->parm;
1945 case KVM_S390_SIGP_SET_PREFIX:
1946 irq->u.prefix.address = s390int->parm;
1948 case KVM_S390_SIGP_STOP:
1949 irq->u.stop.flags = s390int->parm;
1951 case KVM_S390_INT_EXTERNAL_CALL:
1952 if (s390int->parm & 0xffff0000)
1954 irq->u.extcall.code = s390int->parm;
1956 case KVM_S390_INT_EMERGENCY:
1957 if (s390int->parm & 0xffff0000)
1959 irq->u.emerg.code = s390int->parm;
1962 irq->u.mchk.mcic = s390int->parm64;
1964 case KVM_S390_INT_PFAULT_INIT:
1965 irq->u.ext.ext_params = s390int->parm;
1966 irq->u.ext.ext_params2 = s390int->parm64;
1968 case KVM_S390_RESTART:
1969 case KVM_S390_INT_CLOCK_COMP:
1970 case KVM_S390_INT_CPU_TIMER:
1978 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
1980 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1982 return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1985 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
1987 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1989 spin_lock(&li->lock);
1990 li->irq.stop.flags = 0;
1991 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1992 spin_unlock(&li->lock);
1995 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1999 switch (irq->type) {
2000 case KVM_S390_PROGRAM_INT:
2001 rc = __inject_prog(vcpu, irq);
2003 case KVM_S390_SIGP_SET_PREFIX:
2004 rc = __inject_set_prefix(vcpu, irq);
2006 case KVM_S390_SIGP_STOP:
2007 rc = __inject_sigp_stop(vcpu, irq);
2009 case KVM_S390_RESTART:
2010 rc = __inject_sigp_restart(vcpu, irq);
2012 case KVM_S390_INT_CLOCK_COMP:
2013 rc = __inject_ckc(vcpu);
2015 case KVM_S390_INT_CPU_TIMER:
2016 rc = __inject_cpu_timer(vcpu);
2018 case KVM_S390_INT_EXTERNAL_CALL:
2019 rc = __inject_extcall(vcpu, irq);
2021 case KVM_S390_INT_EMERGENCY:
2022 rc = __inject_sigp_emergency(vcpu, irq);
2025 rc = __inject_mchk(vcpu, irq);
2027 case KVM_S390_INT_PFAULT_INIT:
2028 rc = __inject_pfault_init(vcpu, irq);
2030 case KVM_S390_INT_VIRTIO:
2031 case KVM_S390_INT_SERVICE:
2032 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2040 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2042 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2045 spin_lock(&li->lock);
2046 rc = do_inject_vcpu(vcpu, irq);
2047 spin_unlock(&li->lock);
2049 kvm_s390_vcpu_wakeup(vcpu);
2053 static inline void clear_irq_list(struct list_head *_list)
2055 struct kvm_s390_interrupt_info *inti, *n;
2057 list_for_each_entry_safe(inti, n, _list, list) {
2058 list_del(&inti->list);
2063 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2064 struct kvm_s390_irq *irq)
2066 irq->type = inti->type;
2067 switch (inti->type) {
2068 case KVM_S390_INT_PFAULT_INIT:
2069 case KVM_S390_INT_PFAULT_DONE:
2070 case KVM_S390_INT_VIRTIO:
2071 irq->u.ext = inti->ext;
2073 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2074 irq->u.io = inti->io;
2079 void kvm_s390_clear_float_irqs(struct kvm *kvm)
2081 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2084 spin_lock(&fi->lock);
2085 fi->pending_irqs = 0;
2086 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2087 memset(&fi->mchk, 0, sizeof(fi->mchk));
2088 for (i = 0; i < FIRQ_LIST_COUNT; i++)
2089 clear_irq_list(&fi->lists[i]);
2090 for (i = 0; i < FIRQ_MAX_COUNT; i++)
2091 fi->counters[i] = 0;
2092 spin_unlock(&fi->lock);
2093 kvm_s390_gisa_clear(kvm);
2096 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2098 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2099 struct kvm_s390_interrupt_info *inti;
2100 struct kvm_s390_float_interrupt *fi;
2101 struct kvm_s390_irq *buf;
2102 struct kvm_s390_irq *irq;
2108 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2112 * We are already using -ENOMEM to signal
2113 * userspace it may retry with a bigger buffer,
2114 * so we need to use something else for this case
2120 max_irqs = len / sizeof(struct kvm_s390_irq);
2122 if (gi->origin && gisa_get_ipm(gi->origin)) {
2123 for (i = 0; i <= MAX_ISC; i++) {
2124 if (n == max_irqs) {
2125 /* signal userspace to try again */
2129 if (gisa_tac_ipm_gisc(gi->origin, i)) {
2130 irq = (struct kvm_s390_irq *) &buf[n];
2131 irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2132 irq->u.io.io_int_word = isc_to_int_word(i);
2137 fi = &kvm->arch.float_int;
2138 spin_lock(&fi->lock);
2139 for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2140 list_for_each_entry(inti, &fi->lists[i], list) {
2141 if (n == max_irqs) {
2142 /* signal userspace to try again */
2146 inti_to_irq(inti, &buf[n]);
2150 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs)) {
2151 if (n == max_irqs) {
2152 /* signal userspace to try again */
2156 irq = (struct kvm_s390_irq *) &buf[n];
2157 irq->type = KVM_S390_INT_SERVICE;
2158 irq->u.ext = fi->srv_signal;
2161 if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2162 if (n == max_irqs) {
2163 /* signal userspace to try again */
2167 irq = (struct kvm_s390_irq *) &buf[n];
2168 irq->type = KVM_S390_MCHK;
2169 irq->u.mchk = fi->mchk;
2174 spin_unlock(&fi->lock);
2176 if (!ret && n > 0) {
2177 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2182 return ret < 0 ? ret : n;
2185 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2187 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2188 struct kvm_s390_ais_all ais;
2190 if (attr->attr < sizeof(ais))
2193 if (!test_kvm_facility(kvm, 72))
2196 mutex_lock(&fi->ais_lock);
2197 ais.simm = fi->simm;
2198 ais.nimm = fi->nimm;
2199 mutex_unlock(&fi->ais_lock);
2201 if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2207 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2211 switch (attr->group) {
2212 case KVM_DEV_FLIC_GET_ALL_IRQS:
2213 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2216 case KVM_DEV_FLIC_AISM_ALL:
2217 r = flic_ais_mode_get_all(dev->kvm, attr);
2226 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2229 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2230 void *target = NULL;
2231 void __user *source;
2234 if (get_user(inti->type, (u64 __user *)addr))
2237 switch (inti->type) {
2238 case KVM_S390_INT_PFAULT_INIT:
2239 case KVM_S390_INT_PFAULT_DONE:
2240 case KVM_S390_INT_VIRTIO:
2241 case KVM_S390_INT_SERVICE:
2242 target = (void *) &inti->ext;
2243 source = &uptr->u.ext;
2244 size = sizeof(inti->ext);
2246 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2247 target = (void *) &inti->io;
2248 source = &uptr->u.io;
2249 size = sizeof(inti->io);
2252 target = (void *) &inti->mchk;
2253 source = &uptr->u.mchk;
2254 size = sizeof(inti->mchk);
2260 if (copy_from_user(target, source, size))
2266 static int enqueue_floating_irq(struct kvm_device *dev,
2267 struct kvm_device_attr *attr)
2269 struct kvm_s390_interrupt_info *inti = NULL;
2271 int len = attr->attr;
2273 if (len % sizeof(struct kvm_s390_irq) != 0)
2275 else if (len > KVM_S390_FLIC_MAX_BUFFER)
2278 while (len >= sizeof(struct kvm_s390_irq)) {
2279 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
2283 r = copy_irq_from_user(inti, attr->addr);
2288 r = __inject_vm(dev->kvm, inti);
2293 len -= sizeof(struct kvm_s390_irq);
2294 attr->addr += sizeof(struct kvm_s390_irq);
2300 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2302 if (id >= MAX_S390_IO_ADAPTERS)
2304 id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2305 return kvm->arch.adapters[id];
2308 static int register_io_adapter(struct kvm_device *dev,
2309 struct kvm_device_attr *attr)
2311 struct s390_io_adapter *adapter;
2312 struct kvm_s390_io_adapter adapter_info;
2314 if (copy_from_user(&adapter_info,
2315 (void __user *)attr->addr, sizeof(adapter_info)))
2318 if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2321 adapter_info.id = array_index_nospec(adapter_info.id,
2322 MAX_S390_IO_ADAPTERS);
2324 if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2327 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2331 INIT_LIST_HEAD(&adapter->maps);
2332 init_rwsem(&adapter->maps_lock);
2333 atomic_set(&adapter->nr_maps, 0);
2334 adapter->id = adapter_info.id;
2335 adapter->isc = adapter_info.isc;
2336 adapter->maskable = adapter_info.maskable;
2337 adapter->masked = false;
2338 adapter->swap = adapter_info.swap;
2339 adapter->suppressible = (adapter_info.flags) &
2340 KVM_S390_ADAPTER_SUPPRESSIBLE;
2341 dev->kvm->arch.adapters[adapter->id] = adapter;
2346 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2349 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2351 if (!adapter || !adapter->maskable)
2353 ret = adapter->masked;
2354 adapter->masked = masked;
2358 static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr)
2360 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2361 struct s390_map_info *map;
2364 if (!adapter || !addr)
2367 map = kzalloc(sizeof(*map), GFP_KERNEL);
2372 INIT_LIST_HEAD(&map->list);
2373 map->guest_addr = addr;
2374 map->addr = gmap_translate(kvm->arch.gmap, addr);
2375 if (map->addr == -EFAULT) {
2379 ret = get_user_pages_fast(map->addr, 1, FOLL_WRITE, &map->page);
2383 down_write(&adapter->maps_lock);
2384 if (atomic_inc_return(&adapter->nr_maps) < MAX_S390_ADAPTER_MAPS) {
2385 list_add_tail(&map->list, &adapter->maps);
2388 put_page(map->page);
2391 up_write(&adapter->maps_lock);
2398 static int kvm_s390_adapter_unmap(struct kvm *kvm, unsigned int id, __u64 addr)
2400 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2401 struct s390_map_info *map, *tmp;
2404 if (!adapter || !addr)
2407 down_write(&adapter->maps_lock);
2408 list_for_each_entry_safe(map, tmp, &adapter->maps, list) {
2409 if (map->guest_addr == addr) {
2411 atomic_dec(&adapter->nr_maps);
2412 list_del(&map->list);
2413 put_page(map->page);
2418 up_write(&adapter->maps_lock);
2420 return found ? 0 : -EINVAL;
2423 void kvm_s390_destroy_adapters(struct kvm *kvm)
2426 struct s390_map_info *map, *tmp;
2428 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) {
2429 if (!kvm->arch.adapters[i])
2431 list_for_each_entry_safe(map, tmp,
2432 &kvm->arch.adapters[i]->maps, list) {
2433 list_del(&map->list);
2434 put_page(map->page);
2437 kfree(kvm->arch.adapters[i]);
2441 static int modify_io_adapter(struct kvm_device *dev,
2442 struct kvm_device_attr *attr)
2444 struct kvm_s390_io_adapter_req req;
2445 struct s390_io_adapter *adapter;
2448 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2451 adapter = get_io_adapter(dev->kvm, req.id);
2455 case KVM_S390_IO_ADAPTER_MASK:
2456 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2460 case KVM_S390_IO_ADAPTER_MAP:
2461 ret = kvm_s390_adapter_map(dev->kvm, req.id, req.addr);
2463 case KVM_S390_IO_ADAPTER_UNMAP:
2464 ret = kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr);
2473 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2476 const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2481 if (attr->attr != sizeof(schid))
2483 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2487 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2489 * If userspace is conforming to the architecture, we can have at most
2490 * one pending I/O interrupt per subchannel, so this is effectively a
2496 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2498 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2499 struct kvm_s390_ais_req req;
2502 if (!test_kvm_facility(kvm, 72))
2505 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2508 if (req.isc > MAX_ISC)
2511 trace_kvm_s390_modify_ais_mode(req.isc,
2512 (fi->simm & AIS_MODE_MASK(req.isc)) ?
2513 (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2514 2 : KVM_S390_AIS_MODE_SINGLE :
2515 KVM_S390_AIS_MODE_ALL, req.mode);
2517 mutex_lock(&fi->ais_lock);
2519 case KVM_S390_AIS_MODE_ALL:
2520 fi->simm &= ~AIS_MODE_MASK(req.isc);
2521 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2523 case KVM_S390_AIS_MODE_SINGLE:
2524 fi->simm |= AIS_MODE_MASK(req.isc);
2525 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2530 mutex_unlock(&fi->ais_lock);
2535 static int kvm_s390_inject_airq(struct kvm *kvm,
2536 struct s390_io_adapter *adapter)
2538 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2539 struct kvm_s390_interrupt s390int = {
2540 .type = KVM_S390_INT_IO(1, 0, 0, 0),
2542 .parm64 = isc_to_int_word(adapter->isc),
2546 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2547 return kvm_s390_inject_vm(kvm, &s390int);
2549 mutex_lock(&fi->ais_lock);
2550 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2551 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2555 ret = kvm_s390_inject_vm(kvm, &s390int);
2556 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2557 fi->nimm |= AIS_MODE_MASK(adapter->isc);
2558 trace_kvm_s390_modify_ais_mode(adapter->isc,
2559 KVM_S390_AIS_MODE_SINGLE, 2);
2562 mutex_unlock(&fi->ais_lock);
2566 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2568 unsigned int id = attr->attr;
2569 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2574 return kvm_s390_inject_airq(kvm, adapter);
2577 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2579 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2580 struct kvm_s390_ais_all ais;
2582 if (!test_kvm_facility(kvm, 72))
2585 if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2588 mutex_lock(&fi->ais_lock);
2589 fi->simm = ais.simm;
2590 fi->nimm = ais.nimm;
2591 mutex_unlock(&fi->ais_lock);
2596 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2600 struct kvm_vcpu *vcpu;
2602 switch (attr->group) {
2603 case KVM_DEV_FLIC_ENQUEUE:
2604 r = enqueue_floating_irq(dev, attr);
2606 case KVM_DEV_FLIC_CLEAR_IRQS:
2607 kvm_s390_clear_float_irqs(dev->kvm);
2609 case KVM_DEV_FLIC_APF_ENABLE:
2610 dev->kvm->arch.gmap->pfault_enabled = 1;
2612 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2613 dev->kvm->arch.gmap->pfault_enabled = 0;
2615 * Make sure no async faults are in transition when
2616 * clearing the queues. So we don't need to worry
2617 * about late coming workers.
2619 synchronize_srcu(&dev->kvm->srcu);
2620 kvm_for_each_vcpu(i, vcpu, dev->kvm)
2621 kvm_clear_async_pf_completion_queue(vcpu);
2623 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2624 r = register_io_adapter(dev, attr);
2626 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2627 r = modify_io_adapter(dev, attr);
2629 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2630 r = clear_io_irq(dev->kvm, attr);
2632 case KVM_DEV_FLIC_AISM:
2633 r = modify_ais_mode(dev->kvm, attr);
2635 case KVM_DEV_FLIC_AIRQ_INJECT:
2636 r = flic_inject_airq(dev->kvm, attr);
2638 case KVM_DEV_FLIC_AISM_ALL:
2639 r = flic_ais_mode_set_all(dev->kvm, attr);
2648 static int flic_has_attr(struct kvm_device *dev,
2649 struct kvm_device_attr *attr)
2651 switch (attr->group) {
2652 case KVM_DEV_FLIC_GET_ALL_IRQS:
2653 case KVM_DEV_FLIC_ENQUEUE:
2654 case KVM_DEV_FLIC_CLEAR_IRQS:
2655 case KVM_DEV_FLIC_APF_ENABLE:
2656 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2657 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2658 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2659 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2660 case KVM_DEV_FLIC_AISM:
2661 case KVM_DEV_FLIC_AIRQ_INJECT:
2662 case KVM_DEV_FLIC_AISM_ALL:
2668 static int flic_create(struct kvm_device *dev, u32 type)
2672 if (dev->kvm->arch.flic)
2674 dev->kvm->arch.flic = dev;
2678 static void flic_destroy(struct kvm_device *dev)
2680 dev->kvm->arch.flic = NULL;
2684 /* s390 floating irq controller (flic) */
2685 struct kvm_device_ops kvm_flic_ops = {
2687 .get_attr = flic_get_attr,
2688 .set_attr = flic_set_attr,
2689 .has_attr = flic_has_attr,
2690 .create = flic_create,
2691 .destroy = flic_destroy,
2694 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2698 bit = bit_nr + (addr % PAGE_SIZE) * 8;
2700 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2703 static struct s390_map_info *get_map_info(struct s390_io_adapter *adapter,
2706 struct s390_map_info *map;
2711 list_for_each_entry(map, &adapter->maps, list) {
2712 if (map->guest_addr == addr)
2718 static int adapter_indicators_set(struct kvm *kvm,
2719 struct s390_io_adapter *adapter,
2720 struct kvm_s390_adapter_int *adapter_int)
2723 int summary_set, idx;
2724 struct s390_map_info *info;
2727 info = get_map_info(adapter, adapter_int->ind_addr);
2730 map = page_address(info->page);
2731 bit = get_ind_bit(info->addr, adapter_int->ind_offset, adapter->swap);
2733 idx = srcu_read_lock(&kvm->srcu);
2734 mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2735 set_page_dirty_lock(info->page);
2736 info = get_map_info(adapter, adapter_int->summary_addr);
2738 srcu_read_unlock(&kvm->srcu, idx);
2741 map = page_address(info->page);
2742 bit = get_ind_bit(info->addr, adapter_int->summary_offset,
2744 summary_set = test_and_set_bit(bit, map);
2745 mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2746 set_page_dirty_lock(info->page);
2747 srcu_read_unlock(&kvm->srcu, idx);
2748 return summary_set ? 0 : 1;
2752 * < 0 - not injected due to error
2753 * = 0 - coalesced, summary indicator already active
2754 * > 0 - injected interrupt
2756 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2757 struct kvm *kvm, int irq_source_id, int level,
2761 struct s390_io_adapter *adapter;
2763 /* We're only interested in the 0->1 transition. */
2766 adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2769 down_read(&adapter->maps_lock);
2770 ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2771 up_read(&adapter->maps_lock);
2772 if ((ret > 0) && !adapter->masked) {
2773 ret = kvm_s390_inject_airq(kvm, adapter);
2781 * Inject the machine check to the guest.
2783 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2784 struct mcck_volatile_info *mcck_info)
2786 struct kvm_s390_interrupt_info inti;
2787 struct kvm_s390_irq irq;
2788 struct kvm_s390_mchk_info *mchk;
2790 __u64 cr14 = 0; /* upper bits are not used */
2793 mci.val = mcck_info->mcic;
2795 cr14 |= CR14_RECOVERY_SUBMASK;
2797 cr14 |= CR14_DEGRADATION_SUBMASK;
2799 cr14 |= CR14_WARNING_SUBMASK;
2801 mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2803 mchk->mcic = mcck_info->mcic;
2804 mchk->ext_damage_code = mcck_info->ext_damage_code;
2805 mchk->failing_storage_address = mcck_info->failing_storage_address;
2807 /* Inject the floating machine check */
2808 inti.type = KVM_S390_MCHK;
2809 rc = __inject_vm(vcpu->kvm, &inti);
2811 /* Inject the machine check to specified vcpu */
2812 irq.type = KVM_S390_MCHK;
2813 rc = kvm_s390_inject_vcpu(vcpu, &irq);
2818 int kvm_set_routing_entry(struct kvm *kvm,
2819 struct kvm_kernel_irq_routing_entry *e,
2820 const struct kvm_irq_routing_entry *ue)
2825 case KVM_IRQ_ROUTING_S390_ADAPTER:
2826 e->set = set_adapter_int;
2827 e->adapter.summary_addr = ue->u.adapter.summary_addr;
2828 e->adapter.ind_addr = ue->u.adapter.ind_addr;
2829 e->adapter.summary_offset = ue->u.adapter.summary_offset;
2830 e->adapter.ind_offset = ue->u.adapter.ind_offset;
2831 e->adapter.adapter_id = ue->u.adapter.adapter_id;
2841 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2842 int irq_source_id, int level, bool line_status)
2847 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2849 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2850 struct kvm_s390_irq *buf;
2858 if (copy_from_user((void *) buf, irqstate, len)) {
2864 * Don't allow setting the interrupt state
2865 * when there are already interrupts pending
2867 spin_lock(&li->lock);
2868 if (li->pending_irqs) {
2873 for (n = 0; n < len / sizeof(*buf); n++) {
2874 r = do_inject_vcpu(vcpu, &buf[n]);
2880 spin_unlock(&li->lock);
2887 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2888 struct kvm_s390_irq *irq,
2889 unsigned long irq_type)
2892 case IRQ_PEND_MCHK_EX:
2893 case IRQ_PEND_MCHK_REP:
2894 irq->type = KVM_S390_MCHK;
2895 irq->u.mchk = li->irq.mchk;
2898 irq->type = KVM_S390_PROGRAM_INT;
2899 irq->u.pgm = li->irq.pgm;
2901 case IRQ_PEND_PFAULT_INIT:
2902 irq->type = KVM_S390_INT_PFAULT_INIT;
2903 irq->u.ext = li->irq.ext;
2905 case IRQ_PEND_EXT_EXTERNAL:
2906 irq->type = KVM_S390_INT_EXTERNAL_CALL;
2907 irq->u.extcall = li->irq.extcall;
2909 case IRQ_PEND_EXT_CLOCK_COMP:
2910 irq->type = KVM_S390_INT_CLOCK_COMP;
2912 case IRQ_PEND_EXT_CPU_TIMER:
2913 irq->type = KVM_S390_INT_CPU_TIMER;
2915 case IRQ_PEND_SIGP_STOP:
2916 irq->type = KVM_S390_SIGP_STOP;
2917 irq->u.stop = li->irq.stop;
2919 case IRQ_PEND_RESTART:
2920 irq->type = KVM_S390_RESTART;
2922 case IRQ_PEND_SET_PREFIX:
2923 irq->type = KVM_S390_SIGP_SET_PREFIX;
2924 irq->u.prefix = li->irq.prefix;
2929 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
2932 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
2933 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2934 unsigned long pending_irqs;
2935 struct kvm_s390_irq irq;
2936 unsigned long irq_type;
2940 spin_lock(&li->lock);
2941 pending_irqs = li->pending_irqs;
2942 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
2943 sizeof(sigp_emerg_pending));
2944 spin_unlock(&li->lock);
2946 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
2947 memset(&irq, 0, sizeof(irq));
2948 if (irq_type == IRQ_PEND_EXT_EMERGENCY)
2950 if (n + sizeof(irq) > len)
2952 store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
2953 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2958 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
2959 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
2960 memset(&irq, 0, sizeof(irq));
2961 if (n + sizeof(irq) > len)
2963 irq.type = KVM_S390_INT_EMERGENCY;
2964 irq.u.emerg.code = cpuaddr;
2965 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2971 if (sca_ext_call_pending(vcpu, &scn)) {
2972 if (n + sizeof(irq) > len)
2974 memset(&irq, 0, sizeof(irq));
2975 irq.type = KVM_S390_INT_EXTERNAL_CALL;
2976 irq.u.extcall.code = scn;
2977 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2985 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
2987 int vcpu_id, online_vcpus = atomic_read(&kvm->online_vcpus);
2988 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2989 struct kvm_vcpu *vcpu;
2991 for_each_set_bit(vcpu_id, kvm->arch.idle_mask, online_vcpus) {
2992 vcpu = kvm_get_vcpu(kvm, vcpu_id);
2993 if (psw_ioint_disabled(vcpu))
2995 deliverable_mask &= (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
2996 if (deliverable_mask) {
2997 /* lately kicked but not yet running */
2998 if (test_and_set_bit(vcpu_id, gi->kicked_mask))
3000 kvm_s390_vcpu_wakeup(vcpu);
3006 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3008 struct kvm_s390_gisa_interrupt *gi =
3009 container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3011 container_of(gi->origin, struct sie_page2, gisa)->kvm;
3014 pending_mask = gisa_get_ipm_or_restore_iam(gi);
3016 __airqs_kick_single_vcpu(kvm, pending_mask);
3017 hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3018 return HRTIMER_RESTART;
3021 return HRTIMER_NORESTART;
3024 #define NULL_GISA_ADDR 0x00000000UL
3025 #define NONE_GISA_ADDR 0x00000001UL
3026 #define GISA_ADDR_MASK 0xfffff000UL
3028 static void process_gib_alert_list(void)
3030 struct kvm_s390_gisa_interrupt *gi;
3031 struct kvm_s390_gisa *gisa;
3033 u32 final, origin = 0UL;
3037 * If the NONE_GISA_ADDR is still stored in the alert list
3038 * origin, we will leave the outer loop. No further GISA has
3039 * been added to the alert list by millicode while processing
3040 * the current alert list.
3042 final = (origin & NONE_GISA_ADDR);
3044 * Cut off the alert list and store the NONE_GISA_ADDR in the
3045 * alert list origin to avoid further GAL interruptions.
3046 * A new alert list can be build up by millicode in parallel
3047 * for guests not in the yet cut-off alert list. When in the
3048 * final loop, store the NULL_GISA_ADDR instead. This will re-
3049 * enable GAL interruptions on the host again.
3051 origin = xchg(&gib->alert_list_origin,
3052 (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3054 * Loop through the just cut-off alert list and start the
3055 * gisa timers to kick idle vcpus to consume the pending
3056 * interruptions asap.
3058 while (origin & GISA_ADDR_MASK) {
3059 gisa = (struct kvm_s390_gisa *)(u64)origin;
3060 origin = gisa->next_alert;
3061 gisa->next_alert = (u32)(u64)gisa;
3062 kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3063 gi = &kvm->arch.gisa_int;
3064 if (hrtimer_active(&gi->timer))
3065 hrtimer_cancel(&gi->timer);
3066 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3072 void kvm_s390_gisa_clear(struct kvm *kvm)
3074 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3078 gisa_clear_ipm(gi->origin);
3079 VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3082 void kvm_s390_gisa_init(struct kvm *kvm)
3084 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3086 if (!css_general_characteristics.aiv)
3088 gi->origin = &kvm->arch.sie_page2->gisa;
3090 spin_lock_init(&gi->alert.ref_lock);
3091 gi->expires = 50 * 1000; /* 50 usec */
3092 hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3093 gi->timer.function = gisa_vcpu_kicker;
3094 memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3095 gi->origin->next_alert = (u32)(u64)gi->origin;
3096 VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3099 void kvm_s390_gisa_destroy(struct kvm *kvm)
3101 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3106 KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
3107 kvm, gi->alert.mask);
3108 while (gisa_in_alert_list(gi->origin))
3110 hrtimer_cancel(&gi->timer);
3115 * kvm_s390_gisc_register - register a guest ISC
3117 * @kvm: the kernel vm to work with
3118 * @gisc: the guest interruption sub class to register
3120 * The function extends the vm specific alert mask to use.
3121 * The effective IAM mask in the GISA is updated as well
3122 * in case the GISA is not part of the GIB alert list.
3123 * It will be updated latest when the IAM gets restored
3124 * by gisa_get_ipm_or_restore_iam().
3126 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3127 * has registered with the channel subsystem.
3128 * -ENODEV in case the vm uses no GISA
3129 * -ERANGE in case the guest ISC is invalid
3131 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3133 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3140 spin_lock(&gi->alert.ref_lock);
3141 gi->alert.ref_count[gisc]++;
3142 if (gi->alert.ref_count[gisc] == 1) {
3143 gi->alert.mask |= 0x80 >> gisc;
3144 gisa_set_iam(gi->origin, gi->alert.mask);
3146 spin_unlock(&gi->alert.ref_lock);
3150 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3153 * kvm_s390_gisc_unregister - unregister a guest ISC
3155 * @kvm: the kernel vm to work with
3156 * @gisc: the guest interruption sub class to register
3158 * The function reduces the vm specific alert mask to use.
3159 * The effective IAM mask in the GISA is updated as well
3160 * in case the GISA is not part of the GIB alert list.
3161 * It will be updated latest when the IAM gets restored
3162 * by gisa_get_ipm_or_restore_iam().
3164 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3165 * has registered with the channel subsystem.
3166 * -ENODEV in case the vm uses no GISA
3167 * -ERANGE in case the guest ISC is invalid
3168 * -EINVAL in case the guest ISC is not registered
3170 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3172 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3180 spin_lock(&gi->alert.ref_lock);
3181 if (gi->alert.ref_count[gisc] == 0) {
3185 gi->alert.ref_count[gisc]--;
3186 if (gi->alert.ref_count[gisc] == 0) {
3187 gi->alert.mask &= ~(0x80 >> gisc);
3188 gisa_set_iam(gi->origin, gi->alert.mask);
3191 spin_unlock(&gi->alert.ref_lock);
3195 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3197 static void gib_alert_irq_handler(struct airq_struct *airq, bool floating)
3199 inc_irq_stat(IRQIO_GAL);
3200 process_gib_alert_list();
3203 static struct airq_struct gib_alert_irq = {
3204 .handler = gib_alert_irq_handler,
3205 .lsi_ptr = &gib_alert_irq.lsi_mask,
3208 void kvm_s390_gib_destroy(void)
3213 unregister_adapter_interrupt(&gib_alert_irq);
3214 free_page((unsigned long)gib);
3218 int kvm_s390_gib_init(u8 nisc)
3222 if (!css_general_characteristics.aiv) {
3223 KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3227 gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL | GFP_DMA);
3233 gib_alert_irq.isc = nisc;
3234 if (register_adapter_interrupt(&gib_alert_irq)) {
3235 pr_err("Registering the GIB alert interruption handler failed\n");
3241 if (chsc_sgib((u32)(u64)gib)) {
3242 pr_err("Associating the GIB with the AIV facility failed\n");
3243 free_page((unsigned long)gib);
3249 KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3253 unregister_adapter_interrupt(&gib_alert_irq);
3255 free_page((unsigned long)gib);