xref: /linux/arch/s390/kvm/interrupt.c (revision 7fc2cd2e4b398c57c9cf961cfea05eadbf34c05c)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * handling kvm guest interrupts
4  *
5  * Copyright IBM Corp. 2008, 2020
6  *
7  *    Author(s): Carsten Otte <cotte@de.ibm.com>
8  */
9 
10 #define pr_fmt(fmt) "kvm-s390: " fmt
11 
12 #include <linux/cpufeature.h>
13 #include <linux/interrupt.h>
14 #include <linux/kvm_host.h>
15 #include <linux/hrtimer.h>
16 #include <linux/export.h>
17 #include <linux/mmu_context.h>
18 #include <linux/nospec.h>
19 #include <linux/signal.h>
20 #include <linux/slab.h>
21 #include <linux/bitmap.h>
22 #include <linux/vmalloc.h>
23 #include <asm/access-regs.h>
24 #include <asm/asm-offsets.h>
25 #include <asm/dis.h>
26 #include <linux/uaccess.h>
27 #include <asm/sclp.h>
28 #include <asm/isc.h>
29 #include <asm/gmap.h>
30 #include <asm/nmi.h>
31 #include <asm/airq.h>
32 #include <asm/tpi.h>
33 #include "kvm-s390.h"
34 #include "gaccess.h"
35 #include "trace-s390.h"
36 #include "pci.h"
37 
38 #define PFAULT_INIT 0x0600
39 #define PFAULT_DONE 0x0680
40 #define VIRTIO_PARAM 0x0d00
41 
42 static struct kvm_s390_gib *gib;
43 
44 /* handle external calls via sigp interpretation facility */
45 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
46 {
47 	int c, scn;
48 
49 	if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND))
50 		return 0;
51 
52 	BUG_ON(!kvm_s390_use_sca_entries());
53 	read_lock(&vcpu->kvm->arch.sca_lock);
54 	if (vcpu->kvm->arch.use_esca) {
55 		struct esca_block *sca = vcpu->kvm->arch.sca;
56 		union esca_sigp_ctrl sigp_ctrl =
57 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
58 
59 		c = sigp_ctrl.c;
60 		scn = sigp_ctrl.scn;
61 	} else {
62 		struct bsca_block *sca = vcpu->kvm->arch.sca;
63 		union bsca_sigp_ctrl sigp_ctrl =
64 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
65 
66 		c = sigp_ctrl.c;
67 		scn = sigp_ctrl.scn;
68 	}
69 	read_unlock(&vcpu->kvm->arch.sca_lock);
70 
71 	if (src_id)
72 		*src_id = scn;
73 
74 	return c;
75 }
76 
77 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
78 {
79 	int expect, rc;
80 
81 	BUG_ON(!kvm_s390_use_sca_entries());
82 	read_lock(&vcpu->kvm->arch.sca_lock);
83 	if (vcpu->kvm->arch.use_esca) {
84 		struct esca_block *sca = vcpu->kvm->arch.sca;
85 		union esca_sigp_ctrl *sigp_ctrl =
86 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
87 		union esca_sigp_ctrl new_val = {0}, old_val;
88 
89 		old_val = READ_ONCE(*sigp_ctrl);
90 		new_val.scn = src_id;
91 		new_val.c = 1;
92 		old_val.c = 0;
93 
94 		expect = old_val.value;
95 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
96 	} else {
97 		struct bsca_block *sca = vcpu->kvm->arch.sca;
98 		union bsca_sigp_ctrl *sigp_ctrl =
99 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
100 		union bsca_sigp_ctrl new_val = {0}, old_val;
101 
102 		old_val = READ_ONCE(*sigp_ctrl);
103 		new_val.scn = src_id;
104 		new_val.c = 1;
105 		old_val.c = 0;
106 
107 		expect = old_val.value;
108 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
109 	}
110 	read_unlock(&vcpu->kvm->arch.sca_lock);
111 
112 	if (rc != expect) {
113 		/* another external call is pending */
114 		return -EBUSY;
115 	}
116 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
117 	return 0;
118 }
119 
120 static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
121 {
122 	if (!kvm_s390_use_sca_entries())
123 		return;
124 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
125 	read_lock(&vcpu->kvm->arch.sca_lock);
126 	if (vcpu->kvm->arch.use_esca) {
127 		struct esca_block *sca = vcpu->kvm->arch.sca;
128 		union esca_sigp_ctrl *sigp_ctrl =
129 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
130 
131 		WRITE_ONCE(sigp_ctrl->value, 0);
132 	} else {
133 		struct bsca_block *sca = vcpu->kvm->arch.sca;
134 		union bsca_sigp_ctrl *sigp_ctrl =
135 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
136 
137 		WRITE_ONCE(sigp_ctrl->value, 0);
138 	}
139 	read_unlock(&vcpu->kvm->arch.sca_lock);
140 }
141 
142 int psw_extint_disabled(struct kvm_vcpu *vcpu)
143 {
144 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
145 }
146 
147 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
148 {
149 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
150 }
151 
152 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
153 {
154 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
155 }
156 
157 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
158 {
159 	return psw_extint_disabled(vcpu) &&
160 	       psw_ioint_disabled(vcpu) &&
161 	       psw_mchk_disabled(vcpu);
162 }
163 
164 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
165 {
166 	if (psw_extint_disabled(vcpu) ||
167 	    !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
168 		return 0;
169 	if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
170 		/* No timer interrupts when single stepping */
171 		return 0;
172 	return 1;
173 }
174 
175 static int ckc_irq_pending(struct kvm_vcpu *vcpu)
176 {
177 	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
178 	const u64 ckc = vcpu->arch.sie_block->ckc;
179 
180 	if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
181 		if ((s64)ckc >= (s64)now)
182 			return 0;
183 	} else if (ckc >= now) {
184 		return 0;
185 	}
186 	return ckc_interrupts_enabled(vcpu);
187 }
188 
189 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
190 {
191 	return !psw_extint_disabled(vcpu) &&
192 	       (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK);
193 }
194 
195 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
196 {
197 	if (!cpu_timer_interrupts_enabled(vcpu))
198 		return 0;
199 	return kvm_s390_get_cpu_timer(vcpu) >> 63;
200 }
201 
202 static uint64_t isc_to_isc_bits(int isc)
203 {
204 	return (0x80 >> isc) << 24;
205 }
206 
207 static inline u32 isc_to_int_word(u8 isc)
208 {
209 	return ((u32)isc << 27) | 0x80000000;
210 }
211 
212 static inline u8 int_word_to_isc(u32 int_word)
213 {
214 	return (int_word & 0x38000000) >> 27;
215 }
216 
217 /*
218  * To use atomic bitmap functions, we have to provide a bitmap address
219  * that is u64 aligned. However, the ipm might be u32 aligned.
220  * Therefore, we logically start the bitmap at the very beginning of the
221  * struct and fixup the bit number.
222  */
223 #define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE)
224 
225 /**
226  * gisa_set_iam - change the GISA interruption alert mask
227  *
228  * @gisa: gisa to operate on
229  * @iam: new IAM value to use
230  *
231  * Change the IAM atomically with the next alert address and the IPM
232  * of the GISA if the GISA is not part of the GIB alert list. All three
233  * fields are located in the first long word of the GISA.
234  *
235  * Returns: 0 on success
236  *          -EBUSY in case the gisa is part of the alert list
237  */
238 static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam)
239 {
240 	u64 word, _word;
241 
242 	word = READ_ONCE(gisa->u64.word[0]);
243 	do {
244 		if ((u64)gisa != word >> 32)
245 			return -EBUSY;
246 		_word = (word & ~0xffUL) | iam;
247 	} while (!try_cmpxchg(&gisa->u64.word[0], &word, _word));
248 
249 	return 0;
250 }
251 
252 /**
253  * gisa_clear_ipm - clear the GISA interruption pending mask
254  *
255  * @gisa: gisa to operate on
256  *
257  * Clear the IPM atomically with the next alert address and the IAM
258  * of the GISA unconditionally. All three fields are located in the
259  * first long word of the GISA.
260  */
261 static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa)
262 {
263 	u64 word, _word;
264 
265 	word = READ_ONCE(gisa->u64.word[0]);
266 	do {
267 		_word = word & ~(0xffUL << 24);
268 	} while (!try_cmpxchg(&gisa->u64.word[0], &word, _word));
269 }
270 
271 /**
272  * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM
273  *
274  * @gi: gisa interrupt struct to work on
275  *
276  * Atomically restores the interruption alert mask if none of the
277  * relevant ISCs are pending and return the IPM.
278  *
279  * Returns: the relevant pending ISCs
280  */
281 static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
282 {
283 	u8 pending_mask, alert_mask;
284 	u64 word, _word;
285 
286 	word = READ_ONCE(gi->origin->u64.word[0]);
287 	do {
288 		alert_mask = READ_ONCE(gi->alert.mask);
289 		pending_mask = (u8)(word >> 24) & alert_mask;
290 		if (pending_mask)
291 			return pending_mask;
292 		_word = (word & ~0xffUL) | alert_mask;
293 	} while (!try_cmpxchg(&gi->origin->u64.word[0], &word, _word));
294 
295 	return 0;
296 }
297 
298 static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
299 {
300 	set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
301 }
302 
303 static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
304 {
305 	return READ_ONCE(gisa->ipm);
306 }
307 
308 static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
309 {
310 	return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
311 }
312 
313 static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
314 {
315 	unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs |
316 				vcpu->arch.local_int.pending_irqs;
317 
318 	pending &= ~vcpu->kvm->arch.float_int.masked_irqs;
319 	return pending;
320 }
321 
322 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
323 {
324 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
325 	unsigned long pending_mask;
326 
327 	pending_mask = pending_irqs_no_gisa(vcpu);
328 	if (gi->origin)
329 		pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
330 	return pending_mask;
331 }
332 
333 static inline int isc_to_irq_type(unsigned long isc)
334 {
335 	return IRQ_PEND_IO_ISC_0 - isc;
336 }
337 
338 static inline int irq_type_to_isc(unsigned long irq_type)
339 {
340 	return IRQ_PEND_IO_ISC_0 - irq_type;
341 }
342 
343 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
344 				   unsigned long active_mask)
345 {
346 	int i;
347 
348 	for (i = 0; i <= MAX_ISC; i++)
349 		if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
350 			active_mask &= ~(1UL << (isc_to_irq_type(i)));
351 
352 	return active_mask;
353 }
354 
355 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
356 {
357 	unsigned long active_mask;
358 
359 	active_mask = pending_irqs(vcpu);
360 	if (!active_mask)
361 		return 0;
362 
363 	if (psw_extint_disabled(vcpu))
364 		active_mask &= ~IRQ_PEND_EXT_MASK;
365 	if (psw_ioint_disabled(vcpu))
366 		active_mask &= ~IRQ_PEND_IO_MASK;
367 	else
368 		active_mask = disable_iscs(vcpu, active_mask);
369 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
370 		__clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
371 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
372 		__clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
373 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
374 		__clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
375 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
376 		__clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
377 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) {
378 		__clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
379 		__clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask);
380 	}
381 	if (psw_mchk_disabled(vcpu))
382 		active_mask &= ~IRQ_PEND_MCHK_MASK;
383 	/* PV guest cpus can have a single interruption injected at a time. */
384 	if (kvm_s390_pv_cpu_get_handle(vcpu) &&
385 	    vcpu->arch.sie_block->iictl != IICTL_CODE_NONE)
386 		active_mask &= ~(IRQ_PEND_EXT_II_MASK |
387 				 IRQ_PEND_IO_MASK |
388 				 IRQ_PEND_MCHK_MASK);
389 	/*
390 	 * Check both floating and local interrupt's cr14 because
391 	 * bit IRQ_PEND_MCHK_REP could be set in both cases.
392 	 */
393 	if (!(vcpu->arch.sie_block->gcr[14] &
394 	   (vcpu->kvm->arch.float_int.mchk.cr14 |
395 	   vcpu->arch.local_int.irq.mchk.cr14)))
396 		__clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
397 
398 	/*
399 	 * STOP irqs will never be actively delivered. They are triggered via
400 	 * intercept requests and cleared when the stop intercept is performed.
401 	 */
402 	__clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
403 
404 	return active_mask;
405 }
406 
407 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
408 {
409 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
410 	set_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
411 }
412 
413 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
414 {
415 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
416 	clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
417 }
418 
419 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
420 {
421 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
422 				      CPUSTAT_STOP_INT);
423 	vcpu->arch.sie_block->lctl = 0x0000;
424 	vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
425 
426 	if (guestdbg_enabled(vcpu)) {
427 		vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
428 					       LCTL_CR10 | LCTL_CR11);
429 		vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
430 	}
431 }
432 
433 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
434 {
435 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
436 		return;
437 	if (psw_ioint_disabled(vcpu))
438 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
439 	else
440 		vcpu->arch.sie_block->lctl |= LCTL_CR6;
441 }
442 
443 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
444 {
445 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
446 		return;
447 	if (psw_extint_disabled(vcpu))
448 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
449 	else
450 		vcpu->arch.sie_block->lctl |= LCTL_CR0;
451 }
452 
453 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
454 {
455 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
456 		return;
457 	if (psw_mchk_disabled(vcpu))
458 		vcpu->arch.sie_block->ictl |= ICTL_LPSW;
459 	else
460 		vcpu->arch.sie_block->lctl |= LCTL_CR14;
461 }
462 
463 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
464 {
465 	if (kvm_s390_is_stop_irq_pending(vcpu))
466 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
467 }
468 
469 /* Set interception request for non-deliverable interrupts */
470 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
471 {
472 	set_intercept_indicators_io(vcpu);
473 	set_intercept_indicators_ext(vcpu);
474 	set_intercept_indicators_mchk(vcpu);
475 	set_intercept_indicators_stop(vcpu);
476 }
477 
478 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
479 {
480 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
481 	int rc = 0;
482 
483 	vcpu->stat.deliver_cputm++;
484 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
485 					 0, 0);
486 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
487 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
488 		vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER;
489 	} else {
490 		rc  = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
491 				   (u16 *)__LC_EXT_INT_CODE);
492 		rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
493 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
494 				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
495 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
496 				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
497 	}
498 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
499 	return rc ? -EFAULT : 0;
500 }
501 
502 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
503 {
504 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
505 	int rc = 0;
506 
507 	vcpu->stat.deliver_ckc++;
508 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
509 					 0, 0);
510 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
511 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
512 		vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP;
513 	} else {
514 		rc  = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
515 				   (u16 __user *)__LC_EXT_INT_CODE);
516 		rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
517 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
518 				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
519 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
520 				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
521 	}
522 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
523 	return rc ? -EFAULT : 0;
524 }
525 
526 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
527 {
528 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
529 	struct kvm_s390_ext_info ext;
530 	int rc;
531 
532 	spin_lock(&li->lock);
533 	ext = li->irq.ext;
534 	clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
535 	li->irq.ext.ext_params2 = 0;
536 	spin_unlock(&li->lock);
537 
538 	VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
539 		   ext.ext_params2);
540 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
541 					 KVM_S390_INT_PFAULT_INIT,
542 					 0, ext.ext_params2);
543 
544 	rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
545 	rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
546 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
547 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
548 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
549 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
550 	rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
551 	return rc ? -EFAULT : 0;
552 }
553 
554 static int __write_machine_check(struct kvm_vcpu *vcpu,
555 				 struct kvm_s390_mchk_info *mchk)
556 {
557 	unsigned long ext_sa_addr;
558 	unsigned long lc;
559 	freg_t fprs[NUM_FPRS];
560 	union mci mci;
561 	int rc;
562 
563 	/*
564 	 * All other possible payload for a machine check (e.g. the register
565 	 * contents in the save area) will be handled by the ultravisor, as
566 	 * the hypervisor does not not have the needed information for
567 	 * protected guests.
568 	 */
569 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
570 		vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK;
571 		vcpu->arch.sie_block->mcic = mchk->mcic;
572 		vcpu->arch.sie_block->faddr = mchk->failing_storage_address;
573 		vcpu->arch.sie_block->edc = mchk->ext_damage_code;
574 		return 0;
575 	}
576 
577 	mci.val = mchk->mcic;
578 	/* take care of lazy register loading */
579 	kvm_s390_fpu_store(vcpu->run);
580 	save_access_regs(vcpu->run->s.regs.acrs);
581 	if (cpu_has_gs() && vcpu->arch.gs_enabled)
582 		save_gs_cb(current->thread.gs_cb);
583 
584 	/* Extended save area */
585 	rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
586 			   sizeof(unsigned long));
587 	/* Only bits 0 through 63-LC are used for address formation */
588 	lc = ext_sa_addr & MCESA_LC_MASK;
589 	if (test_kvm_facility(vcpu->kvm, 133)) {
590 		switch (lc) {
591 		case 0:
592 		case 10:
593 			ext_sa_addr &= ~0x3ffUL;
594 			break;
595 		case 11:
596 			ext_sa_addr &= ~0x7ffUL;
597 			break;
598 		case 12:
599 			ext_sa_addr &= ~0xfffUL;
600 			break;
601 		default:
602 			ext_sa_addr = 0;
603 			break;
604 		}
605 	} else {
606 		ext_sa_addr &= ~0x3ffUL;
607 	}
608 
609 	if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
610 		if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
611 				    512))
612 			mci.vr = 0;
613 	} else {
614 		mci.vr = 0;
615 	}
616 	if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
617 	    && (lc == 11 || lc == 12)) {
618 		if (write_guest_abs(vcpu, ext_sa_addr + 1024,
619 				    &vcpu->run->s.regs.gscb, 32))
620 			mci.gs = 0;
621 	} else {
622 		mci.gs = 0;
623 	}
624 
625 	/* General interruption information */
626 	rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
627 	rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
628 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
629 	rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
630 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
631 	rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
632 
633 	/* Register-save areas */
634 	if (cpu_has_vx()) {
635 		convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
636 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
637 	} else {
638 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
639 				     vcpu->run->s.regs.fprs, 128);
640 	}
641 	rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
642 			     vcpu->run->s.regs.gprs, 128);
643 	rc |= put_guest_lc(vcpu, vcpu->run->s.regs.fpc,
644 			   (u32 __user *) __LC_FP_CREG_SAVE_AREA);
645 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
646 			   (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
647 	rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
648 			   (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
649 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
650 			   (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
651 	rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
652 			     &vcpu->run->s.regs.acrs, 64);
653 	rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
654 			     &vcpu->arch.sie_block->gcr, 128);
655 
656 	/* Extended interruption information */
657 	rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
658 			   (u32 __user *) __LC_EXT_DAMAGE_CODE);
659 	rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
660 			   (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
661 	rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
662 			     sizeof(mchk->fixed_logout));
663 	return rc ? -EFAULT : 0;
664 }
665 
666 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
667 {
668 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
669 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
670 	struct kvm_s390_mchk_info mchk = {};
671 	int deliver = 0;
672 	int rc = 0;
673 
674 	spin_lock(&fi->lock);
675 	spin_lock(&li->lock);
676 	if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
677 	    test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
678 		/*
679 		 * If there was an exigent machine check pending, then any
680 		 * repressible machine checks that might have been pending
681 		 * are indicated along with it, so always clear bits for
682 		 * repressible and exigent interrupts
683 		 */
684 		mchk = li->irq.mchk;
685 		clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
686 		clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
687 		memset(&li->irq.mchk, 0, sizeof(mchk));
688 		deliver = 1;
689 	}
690 	/*
691 	 * We indicate floating repressible conditions along with
692 	 * other pending conditions. Channel Report Pending and Channel
693 	 * Subsystem damage are the only two and are indicated by
694 	 * bits in mcic and masked in cr14.
695 	 */
696 	if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
697 		mchk.mcic |= fi->mchk.mcic;
698 		mchk.cr14 |= fi->mchk.cr14;
699 		memset(&fi->mchk, 0, sizeof(mchk));
700 		deliver = 1;
701 	}
702 	spin_unlock(&li->lock);
703 	spin_unlock(&fi->lock);
704 
705 	if (deliver) {
706 		VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
707 			   mchk.mcic);
708 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
709 						 KVM_S390_MCHK,
710 						 mchk.cr14, mchk.mcic);
711 		vcpu->stat.deliver_machine_check++;
712 		rc = __write_machine_check(vcpu, &mchk);
713 	}
714 	return rc;
715 }
716 
717 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
718 {
719 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
720 	int rc = 0;
721 
722 	VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
723 	vcpu->stat.deliver_restart_signal++;
724 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
725 
726 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
727 		vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART;
728 	} else {
729 		rc  = write_guest_lc(vcpu,
730 				     offsetof(struct lowcore, restart_old_psw),
731 				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
732 		rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
733 				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
734 	}
735 	clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
736 	return rc ? -EFAULT : 0;
737 }
738 
739 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
740 {
741 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
742 	struct kvm_s390_prefix_info prefix;
743 
744 	spin_lock(&li->lock);
745 	prefix = li->irq.prefix;
746 	li->irq.prefix.address = 0;
747 	clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
748 	spin_unlock(&li->lock);
749 
750 	vcpu->stat.deliver_prefix_signal++;
751 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
752 					 KVM_S390_SIGP_SET_PREFIX,
753 					 prefix.address, 0);
754 
755 	kvm_s390_set_prefix(vcpu, prefix.address);
756 	return 0;
757 }
758 
759 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
760 {
761 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
762 	int rc;
763 	int cpu_addr;
764 
765 	spin_lock(&li->lock);
766 	cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
767 	clear_bit(cpu_addr, li->sigp_emerg_pending);
768 	if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
769 		clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
770 	spin_unlock(&li->lock);
771 
772 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
773 	vcpu->stat.deliver_emergency_signal++;
774 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
775 					 cpu_addr, 0);
776 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
777 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
778 		vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG;
779 		vcpu->arch.sie_block->extcpuaddr = cpu_addr;
780 		return 0;
781 	}
782 
783 	rc  = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
784 			   (u16 *)__LC_EXT_INT_CODE);
785 	rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
786 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
787 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
788 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
789 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
790 	return rc ? -EFAULT : 0;
791 }
792 
793 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
794 {
795 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
796 	struct kvm_s390_extcall_info extcall;
797 	int rc;
798 
799 	spin_lock(&li->lock);
800 	extcall = li->irq.extcall;
801 	li->irq.extcall.code = 0;
802 	clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
803 	spin_unlock(&li->lock);
804 
805 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
806 	vcpu->stat.deliver_external_call++;
807 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
808 					 KVM_S390_INT_EXTERNAL_CALL,
809 					 extcall.code, 0);
810 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
811 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
812 		vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL;
813 		vcpu->arch.sie_block->extcpuaddr = extcall.code;
814 		return 0;
815 	}
816 
817 	rc  = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
818 			   (u16 *)__LC_EXT_INT_CODE);
819 	rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
820 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
821 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
822 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
823 			    sizeof(psw_t));
824 	return rc ? -EFAULT : 0;
825 }
826 
827 static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code)
828 {
829 	switch (code) {
830 	case PGM_SPECIFICATION:
831 		vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION;
832 		break;
833 	case PGM_OPERAND:
834 		vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND;
835 		break;
836 	default:
837 		return -EINVAL;
838 	}
839 	return 0;
840 }
841 
842 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
843 {
844 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
845 	struct kvm_s390_pgm_info pgm_info;
846 	int rc = 0, nullifying = false;
847 	u16 ilen;
848 
849 	spin_lock(&li->lock);
850 	pgm_info = li->irq.pgm;
851 	clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
852 	memset(&li->irq.pgm, 0, sizeof(pgm_info));
853 	spin_unlock(&li->lock);
854 
855 	ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
856 	VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
857 		   pgm_info.code, ilen);
858 	vcpu->stat.deliver_program++;
859 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
860 					 pgm_info.code, 0);
861 
862 	/* PER is handled by the ultravisor */
863 	if (kvm_s390_pv_cpu_is_protected(vcpu))
864 		return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER);
865 
866 	switch (pgm_info.code & ~PGM_PER) {
867 	case PGM_AFX_TRANSLATION:
868 	case PGM_ASX_TRANSLATION:
869 	case PGM_EX_TRANSLATION:
870 	case PGM_LFX_TRANSLATION:
871 	case PGM_LSTE_SEQUENCE:
872 	case PGM_LSX_TRANSLATION:
873 	case PGM_LX_TRANSLATION:
874 	case PGM_PRIMARY_AUTHORITY:
875 	case PGM_SECONDARY_AUTHORITY:
876 		nullifying = true;
877 		fallthrough;
878 	case PGM_SPACE_SWITCH:
879 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
880 				  (u64 *)__LC_TRANS_EXC_CODE);
881 		break;
882 	case PGM_ALEN_TRANSLATION:
883 	case PGM_ALE_SEQUENCE:
884 	case PGM_ASTE_INSTANCE:
885 	case PGM_ASTE_SEQUENCE:
886 	case PGM_ASTE_VALIDITY:
887 	case PGM_EXTENDED_AUTHORITY:
888 		rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
889 				  (u8 *)__LC_EXC_ACCESS_ID);
890 		nullifying = true;
891 		break;
892 	case PGM_ASCE_TYPE:
893 	case PGM_PAGE_TRANSLATION:
894 	case PGM_REGION_FIRST_TRANS:
895 	case PGM_REGION_SECOND_TRANS:
896 	case PGM_REGION_THIRD_TRANS:
897 	case PGM_SEGMENT_TRANSLATION:
898 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
899 				  (u64 *)__LC_TRANS_EXC_CODE);
900 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
901 				   (u8 *)__LC_EXC_ACCESS_ID);
902 		rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
903 				   (u8 *)__LC_OP_ACCESS_ID);
904 		nullifying = true;
905 		break;
906 	case PGM_MONITOR:
907 		rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
908 				  (u16 *)__LC_MON_CLASS_NR);
909 		rc |= put_guest_lc(vcpu, pgm_info.mon_code,
910 				   (u64 *)__LC_MON_CODE);
911 		break;
912 	case PGM_VECTOR_PROCESSING:
913 	case PGM_DATA:
914 		rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
915 				  (u32 *)__LC_DATA_EXC_CODE);
916 		break;
917 	case PGM_PROTECTION:
918 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
919 				  (u64 *)__LC_TRANS_EXC_CODE);
920 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
921 				   (u8 *)__LC_EXC_ACCESS_ID);
922 		break;
923 	case PGM_STACK_FULL:
924 	case PGM_STACK_EMPTY:
925 	case PGM_STACK_SPECIFICATION:
926 	case PGM_STACK_TYPE:
927 	case PGM_STACK_OPERATION:
928 	case PGM_TRACE_TABEL:
929 	case PGM_CRYPTO_OPERATION:
930 		nullifying = true;
931 		break;
932 	}
933 
934 	if (pgm_info.code & PGM_PER) {
935 		rc |= put_guest_lc(vcpu, pgm_info.per_code,
936 				   (u8 *) __LC_PER_CODE);
937 		rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
938 				   (u8 *)__LC_PER_ATMID);
939 		rc |= put_guest_lc(vcpu, pgm_info.per_address,
940 				   (u64 *) __LC_PER_ADDRESS);
941 		rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
942 				   (u8 *) __LC_PER_ACCESS_ID);
943 	}
944 
945 	if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
946 		kvm_s390_rewind_psw(vcpu, ilen);
947 
948 	/* bit 1+2 of the target are the ilc, so we can directly use ilen */
949 	rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
950 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
951 				 (u64 *) __LC_PGM_LAST_BREAK);
952 	rc |= put_guest_lc(vcpu, pgm_info.code, (u16 *)__LC_PGM_CODE);
953 	rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
954 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
955 	rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
956 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
957 	return rc ? -EFAULT : 0;
958 }
959 
960 #define SCCB_MASK 0xFFFFFFF8
961 #define SCCB_EVENT_PENDING 0x3
962 
963 static int write_sclp(struct kvm_vcpu *vcpu, u32 parm)
964 {
965 	int rc;
966 
967 	if (kvm_s390_pv_cpu_get_handle(vcpu)) {
968 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
969 		vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG;
970 		vcpu->arch.sie_block->eiparams = parm;
971 		return 0;
972 	}
973 
974 	rc  = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
975 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
976 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
977 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
978 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
979 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
980 	rc |= put_guest_lc(vcpu, parm,
981 			   (u32 *)__LC_EXT_PARAMS);
982 
983 	return rc ? -EFAULT : 0;
984 }
985 
986 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
987 {
988 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
989 	struct kvm_s390_ext_info ext;
990 
991 	spin_lock(&fi->lock);
992 	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) ||
993 	    !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
994 		spin_unlock(&fi->lock);
995 		return 0;
996 	}
997 	ext = fi->srv_signal;
998 	memset(&fi->srv_signal, 0, sizeof(ext));
999 	clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1000 	clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1001 	if (kvm_s390_pv_cpu_is_protected(vcpu))
1002 		set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs);
1003 	spin_unlock(&fi->lock);
1004 
1005 	VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
1006 		   ext.ext_params);
1007 	vcpu->stat.deliver_service_signal++;
1008 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1009 					 ext.ext_params, 0);
1010 
1011 	return write_sclp(vcpu, ext.ext_params);
1012 }
1013 
1014 static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)
1015 {
1016 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1017 	struct kvm_s390_ext_info ext;
1018 
1019 	spin_lock(&fi->lock);
1020 	if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) {
1021 		spin_unlock(&fi->lock);
1022 		return 0;
1023 	}
1024 	ext = fi->srv_signal;
1025 	/* only clear the event bits */
1026 	fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING;
1027 	clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1028 	spin_unlock(&fi->lock);
1029 
1030 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event");
1031 	vcpu->stat.deliver_service_signal++;
1032 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1033 					 ext.ext_params, 0);
1034 
1035 	return write_sclp(vcpu, ext.ext_params & SCCB_EVENT_PENDING);
1036 }
1037 
1038 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
1039 {
1040 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1041 	struct kvm_s390_interrupt_info *inti;
1042 	int rc = 0;
1043 
1044 	spin_lock(&fi->lock);
1045 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
1046 					struct kvm_s390_interrupt_info,
1047 					list);
1048 	if (inti) {
1049 		list_del(&inti->list);
1050 		fi->counters[FIRQ_CNTR_PFAULT] -= 1;
1051 	}
1052 	if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
1053 		clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1054 	spin_unlock(&fi->lock);
1055 
1056 	if (inti) {
1057 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1058 						 KVM_S390_INT_PFAULT_DONE, 0,
1059 						 inti->ext.ext_params2);
1060 		VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
1061 			   inti->ext.ext_params2);
1062 
1063 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1064 				(u16 *)__LC_EXT_INT_CODE);
1065 		rc |= put_guest_lc(vcpu, PFAULT_DONE,
1066 				(u16 *)__LC_EXT_CPU_ADDR);
1067 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1068 				&vcpu->arch.sie_block->gpsw,
1069 				sizeof(psw_t));
1070 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1071 				&vcpu->arch.sie_block->gpsw,
1072 				sizeof(psw_t));
1073 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1074 				(u64 *)__LC_EXT_PARAMS2);
1075 		kfree(inti);
1076 	}
1077 	return rc ? -EFAULT : 0;
1078 }
1079 
1080 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
1081 {
1082 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1083 	struct kvm_s390_interrupt_info *inti;
1084 	int rc = 0;
1085 
1086 	spin_lock(&fi->lock);
1087 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
1088 					struct kvm_s390_interrupt_info,
1089 					list);
1090 	if (inti) {
1091 		VCPU_EVENT(vcpu, 4,
1092 			   "deliver: virtio parm: 0x%x,parm64: 0x%llx",
1093 			   inti->ext.ext_params, inti->ext.ext_params2);
1094 		vcpu->stat.deliver_virtio++;
1095 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1096 				inti->type,
1097 				inti->ext.ext_params,
1098 				inti->ext.ext_params2);
1099 		list_del(&inti->list);
1100 		fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1101 	}
1102 	if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1103 		clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1104 	spin_unlock(&fi->lock);
1105 
1106 	if (inti) {
1107 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1108 				(u16 *)__LC_EXT_INT_CODE);
1109 		rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1110 				(u16 *)__LC_EXT_CPU_ADDR);
1111 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1112 				&vcpu->arch.sie_block->gpsw,
1113 				sizeof(psw_t));
1114 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1115 				&vcpu->arch.sie_block->gpsw,
1116 				sizeof(psw_t));
1117 		rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1118 				(u32 *)__LC_EXT_PARAMS);
1119 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1120 				(u64 *)__LC_EXT_PARAMS2);
1121 		kfree(inti);
1122 	}
1123 	return rc ? -EFAULT : 0;
1124 }
1125 
1126 static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1127 {
1128 	int rc;
1129 
1130 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
1131 		vcpu->arch.sie_block->iictl = IICTL_CODE_IO;
1132 		vcpu->arch.sie_block->subchannel_id = io->subchannel_id;
1133 		vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr;
1134 		vcpu->arch.sie_block->io_int_parm = io->io_int_parm;
1135 		vcpu->arch.sie_block->io_int_word = io->io_int_word;
1136 		return 0;
1137 	}
1138 
1139 	rc  = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1140 	rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1141 	rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1142 	rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1143 	rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1144 			     &vcpu->arch.sie_block->gpsw,
1145 			     sizeof(psw_t));
1146 	rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1147 			    &vcpu->arch.sie_block->gpsw,
1148 			    sizeof(psw_t));
1149 	return rc ? -EFAULT : 0;
1150 }
1151 
1152 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1153 				     unsigned long irq_type)
1154 {
1155 	struct list_head *isc_list;
1156 	struct kvm_s390_float_interrupt *fi;
1157 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1158 	struct kvm_s390_interrupt_info *inti = NULL;
1159 	struct kvm_s390_io_info io;
1160 	u32 isc;
1161 	int rc = 0;
1162 
1163 	fi = &vcpu->kvm->arch.float_int;
1164 
1165 	spin_lock(&fi->lock);
1166 	isc = irq_type_to_isc(irq_type);
1167 	isc_list = &fi->lists[isc];
1168 	inti = list_first_entry_or_null(isc_list,
1169 					struct kvm_s390_interrupt_info,
1170 					list);
1171 	if (inti) {
1172 		if (inti->type & KVM_S390_INT_IO_AI_MASK)
1173 			VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1174 		else
1175 			VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1176 			inti->io.subchannel_id >> 8,
1177 			inti->io.subchannel_id >> 1 & 0x3,
1178 			inti->io.subchannel_nr);
1179 
1180 		vcpu->stat.deliver_io++;
1181 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1182 				inti->type,
1183 				((__u32)inti->io.subchannel_id << 16) |
1184 				inti->io.subchannel_nr,
1185 				((__u64)inti->io.io_int_parm << 32) |
1186 				inti->io.io_int_word);
1187 		list_del(&inti->list);
1188 		fi->counters[FIRQ_CNTR_IO] -= 1;
1189 	}
1190 	if (list_empty(isc_list))
1191 		clear_bit(irq_type, &fi->pending_irqs);
1192 	spin_unlock(&fi->lock);
1193 
1194 	if (inti) {
1195 		rc = __do_deliver_io(vcpu, &(inti->io));
1196 		kfree(inti);
1197 		goto out;
1198 	}
1199 
1200 	if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1201 		/*
1202 		 * in case an adapter interrupt was not delivered
1203 		 * in SIE context KVM will handle the delivery
1204 		 */
1205 		VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1206 		memset(&io, 0, sizeof(io));
1207 		io.io_int_word = isc_to_int_word(isc);
1208 		vcpu->stat.deliver_io++;
1209 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1210 			KVM_S390_INT_IO(1, 0, 0, 0),
1211 			((__u32)io.subchannel_id << 16) |
1212 			io.subchannel_nr,
1213 			((__u64)io.io_int_parm << 32) |
1214 			io.io_int_word);
1215 		rc = __do_deliver_io(vcpu, &io);
1216 	}
1217 out:
1218 	return rc;
1219 }
1220 
1221 /* Check whether an external call is pending (deliverable or not) */
1222 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1223 {
1224 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1225 
1226 	if (!sclp.has_sigpif)
1227 		return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1228 
1229 	return sca_ext_call_pending(vcpu, NULL);
1230 }
1231 
1232 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1233 {
1234 	if (deliverable_irqs(vcpu))
1235 		return 1;
1236 
1237 	if (kvm_cpu_has_pending_timer(vcpu))
1238 		return 1;
1239 
1240 	/* external call pending and deliverable */
1241 	if (kvm_s390_ext_call_pending(vcpu) &&
1242 	    !psw_extint_disabled(vcpu) &&
1243 	    (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1244 		return 1;
1245 
1246 	if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1247 		return 1;
1248 	return 0;
1249 }
1250 
1251 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1252 {
1253 	return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1254 }
1255 
1256 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1257 {
1258 	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1259 	const u64 ckc = vcpu->arch.sie_block->ckc;
1260 	u64 cputm, sltime = 0;
1261 
1262 	if (ckc_interrupts_enabled(vcpu)) {
1263 		if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1264 			if ((s64)now < (s64)ckc)
1265 				sltime = tod_to_ns((s64)ckc - (s64)now);
1266 		} else if (now < ckc) {
1267 			sltime = tod_to_ns(ckc - now);
1268 		}
1269 		/* already expired */
1270 		if (!sltime)
1271 			return 0;
1272 		if (cpu_timer_interrupts_enabled(vcpu)) {
1273 			cputm = kvm_s390_get_cpu_timer(vcpu);
1274 			/* already expired? */
1275 			if (cputm >> 63)
1276 				return 0;
1277 			return min_t(u64, sltime, tod_to_ns(cputm));
1278 		}
1279 	} else if (cpu_timer_interrupts_enabled(vcpu)) {
1280 		sltime = kvm_s390_get_cpu_timer(vcpu);
1281 		/* already expired? */
1282 		if (sltime >> 63)
1283 			return 0;
1284 	}
1285 	return sltime;
1286 }
1287 
1288 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1289 {
1290 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1291 	u64 sltime;
1292 
1293 	vcpu->stat.exit_wait_state++;
1294 
1295 	/* fast path */
1296 	if (kvm_arch_vcpu_runnable(vcpu))
1297 		return 0;
1298 
1299 	if (psw_interrupts_disabled(vcpu)) {
1300 		VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1301 		return -EOPNOTSUPP; /* disabled wait */
1302 	}
1303 
1304 	if (gi->origin &&
1305 	    (gisa_get_ipm_or_restore_iam(gi) &
1306 	     vcpu->arch.sie_block->gcr[6] >> 24))
1307 		return 0;
1308 
1309 	if (!ckc_interrupts_enabled(vcpu) &&
1310 	    !cpu_timer_interrupts_enabled(vcpu)) {
1311 		VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1312 		__set_cpu_idle(vcpu);
1313 		goto no_timer;
1314 	}
1315 
1316 	sltime = __calculate_sltime(vcpu);
1317 	if (!sltime)
1318 		return 0;
1319 
1320 	__set_cpu_idle(vcpu);
1321 	hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1322 	VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1323 no_timer:
1324 	kvm_vcpu_srcu_read_unlock(vcpu);
1325 	vcpu->kvm->arch.float_int.last_sleep_cpu = vcpu->vcpu_idx;
1326 	kvm_vcpu_halt(vcpu);
1327 	vcpu->valid_wakeup = false;
1328 	__unset_cpu_idle(vcpu);
1329 	kvm_vcpu_srcu_read_lock(vcpu);
1330 
1331 	hrtimer_cancel(&vcpu->arch.ckc_timer);
1332 	return 0;
1333 }
1334 
1335 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1336 {
1337 	vcpu->valid_wakeup = true;
1338 	kvm_vcpu_wake_up(vcpu);
1339 
1340 	/*
1341 	 * The VCPU might not be sleeping but rather executing VSIE. Let's
1342 	 * kick it, so it leaves the SIE to process the request.
1343 	 */
1344 	kvm_s390_vsie_kick(vcpu);
1345 }
1346 
1347 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1348 {
1349 	struct kvm_vcpu *vcpu;
1350 	u64 sltime;
1351 
1352 	vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1353 	sltime = __calculate_sltime(vcpu);
1354 
1355 	/*
1356 	 * If the monotonic clock runs faster than the tod clock we might be
1357 	 * woken up too early and have to go back to sleep to avoid deadlocks.
1358 	 */
1359 	if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1360 		return HRTIMER_RESTART;
1361 	kvm_s390_vcpu_wakeup(vcpu);
1362 	return HRTIMER_NORESTART;
1363 }
1364 
1365 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1366 {
1367 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1368 
1369 	spin_lock(&li->lock);
1370 	li->pending_irqs = 0;
1371 	bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1372 	memset(&li->irq, 0, sizeof(li->irq));
1373 	spin_unlock(&li->lock);
1374 
1375 	sca_clear_ext_call(vcpu);
1376 }
1377 
1378 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1379 {
1380 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1381 	int rc = 0;
1382 	bool delivered = false;
1383 	unsigned long irq_type;
1384 	unsigned long irqs;
1385 
1386 	__reset_intercept_indicators(vcpu);
1387 
1388 	/* pending ckc conditions might have been invalidated */
1389 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1390 	if (ckc_irq_pending(vcpu))
1391 		set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1392 
1393 	/* pending cpu timer conditions might have been invalidated */
1394 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1395 	if (cpu_timer_irq_pending(vcpu))
1396 		set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1397 
1398 	while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1399 		/* bits are in the reverse order of interrupt priority */
1400 		irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1401 		switch (irq_type) {
1402 		case IRQ_PEND_IO_ISC_0:
1403 		case IRQ_PEND_IO_ISC_1:
1404 		case IRQ_PEND_IO_ISC_2:
1405 		case IRQ_PEND_IO_ISC_3:
1406 		case IRQ_PEND_IO_ISC_4:
1407 		case IRQ_PEND_IO_ISC_5:
1408 		case IRQ_PEND_IO_ISC_6:
1409 		case IRQ_PEND_IO_ISC_7:
1410 			rc = __deliver_io(vcpu, irq_type);
1411 			break;
1412 		case IRQ_PEND_MCHK_EX:
1413 		case IRQ_PEND_MCHK_REP:
1414 			rc = __deliver_machine_check(vcpu);
1415 			break;
1416 		case IRQ_PEND_PROG:
1417 			rc = __deliver_prog(vcpu);
1418 			break;
1419 		case IRQ_PEND_EXT_EMERGENCY:
1420 			rc = __deliver_emergency_signal(vcpu);
1421 			break;
1422 		case IRQ_PEND_EXT_EXTERNAL:
1423 			rc = __deliver_external_call(vcpu);
1424 			break;
1425 		case IRQ_PEND_EXT_CLOCK_COMP:
1426 			rc = __deliver_ckc(vcpu);
1427 			break;
1428 		case IRQ_PEND_EXT_CPU_TIMER:
1429 			rc = __deliver_cpu_timer(vcpu);
1430 			break;
1431 		case IRQ_PEND_RESTART:
1432 			rc = __deliver_restart(vcpu);
1433 			break;
1434 		case IRQ_PEND_SET_PREFIX:
1435 			rc = __deliver_set_prefix(vcpu);
1436 			break;
1437 		case IRQ_PEND_PFAULT_INIT:
1438 			rc = __deliver_pfault_init(vcpu);
1439 			break;
1440 		case IRQ_PEND_EXT_SERVICE:
1441 			rc = __deliver_service(vcpu);
1442 			break;
1443 		case IRQ_PEND_EXT_SERVICE_EV:
1444 			rc = __deliver_service_ev(vcpu);
1445 			break;
1446 		case IRQ_PEND_PFAULT_DONE:
1447 			rc = __deliver_pfault_done(vcpu);
1448 			break;
1449 		case IRQ_PEND_VIRTIO:
1450 			rc = __deliver_virtio(vcpu);
1451 			break;
1452 		default:
1453 			WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1454 			clear_bit(irq_type, &li->pending_irqs);
1455 		}
1456 		delivered |= !rc;
1457 	}
1458 
1459 	/*
1460 	 * We delivered at least one interrupt and modified the PC. Force a
1461 	 * singlestep event now.
1462 	 */
1463 	if (delivered && guestdbg_sstep_enabled(vcpu)) {
1464 		struct kvm_debug_exit_arch *debug_exit = &vcpu->run->debug.arch;
1465 
1466 		debug_exit->addr = vcpu->arch.sie_block->gpsw.addr;
1467 		debug_exit->type = KVM_SINGLESTEP;
1468 		vcpu->guest_debug |= KVM_GUESTDBG_EXIT_PENDING;
1469 	}
1470 
1471 	set_intercept_indicators(vcpu);
1472 
1473 	return rc;
1474 }
1475 
1476 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1477 {
1478 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1479 
1480 	vcpu->stat.inject_program++;
1481 	VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1482 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1483 				   irq->u.pgm.code, 0);
1484 
1485 	if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1486 		/* auto detection if no valid ILC was given */
1487 		irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1488 		irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1489 		irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1490 	}
1491 
1492 	if (irq->u.pgm.code == PGM_PER) {
1493 		li->irq.pgm.code |= PGM_PER;
1494 		li->irq.pgm.flags = irq->u.pgm.flags;
1495 		/* only modify PER related information */
1496 		li->irq.pgm.per_address = irq->u.pgm.per_address;
1497 		li->irq.pgm.per_code = irq->u.pgm.per_code;
1498 		li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1499 		li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1500 	} else if (!(irq->u.pgm.code & PGM_PER)) {
1501 		li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1502 				   irq->u.pgm.code;
1503 		li->irq.pgm.flags = irq->u.pgm.flags;
1504 		/* only modify non-PER information */
1505 		li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1506 		li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1507 		li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1508 		li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1509 		li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1510 		li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1511 	} else {
1512 		li->irq.pgm = irq->u.pgm;
1513 	}
1514 	set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1515 	return 0;
1516 }
1517 
1518 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1519 {
1520 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1521 
1522 	vcpu->stat.inject_pfault_init++;
1523 	VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1524 		   irq->u.ext.ext_params2);
1525 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1526 				   irq->u.ext.ext_params,
1527 				   irq->u.ext.ext_params2);
1528 
1529 	li->irq.ext = irq->u.ext;
1530 	set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1531 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1532 	return 0;
1533 }
1534 
1535 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1536 {
1537 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1538 	struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1539 	uint16_t src_id = irq->u.extcall.code;
1540 
1541 	vcpu->stat.inject_external_call++;
1542 	VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1543 		   src_id);
1544 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1545 				   src_id, 0);
1546 
1547 	/* sending vcpu invalid */
1548 	if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1549 		return -EINVAL;
1550 
1551 	if (sclp.has_sigpif && !kvm_s390_pv_cpu_get_handle(vcpu))
1552 		return sca_inject_ext_call(vcpu, src_id);
1553 
1554 	if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1555 		return -EBUSY;
1556 	*extcall = irq->u.extcall;
1557 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1558 	return 0;
1559 }
1560 
1561 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1562 {
1563 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1564 	struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1565 
1566 	vcpu->stat.inject_set_prefix++;
1567 	VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1568 		   irq->u.prefix.address);
1569 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1570 				   irq->u.prefix.address, 0);
1571 
1572 	if (!is_vcpu_stopped(vcpu))
1573 		return -EBUSY;
1574 
1575 	*prefix = irq->u.prefix;
1576 	set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1577 	return 0;
1578 }
1579 
1580 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1581 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1582 {
1583 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1584 	struct kvm_s390_stop_info *stop = &li->irq.stop;
1585 	int rc = 0;
1586 
1587 	vcpu->stat.inject_stop_signal++;
1588 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1589 
1590 	if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1591 		return -EINVAL;
1592 
1593 	if (is_vcpu_stopped(vcpu)) {
1594 		if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1595 			rc = kvm_s390_store_status_unloaded(vcpu,
1596 						KVM_S390_STORE_STATUS_NOADDR);
1597 		return rc;
1598 	}
1599 
1600 	if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1601 		return -EBUSY;
1602 	stop->flags = irq->u.stop.flags;
1603 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1604 	return 0;
1605 }
1606 
1607 static int __inject_sigp_restart(struct kvm_vcpu *vcpu)
1608 {
1609 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1610 
1611 	vcpu->stat.inject_restart++;
1612 	VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1613 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1614 
1615 	set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1616 	return 0;
1617 }
1618 
1619 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1620 				   struct kvm_s390_irq *irq)
1621 {
1622 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1623 
1624 	vcpu->stat.inject_emergency_signal++;
1625 	VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1626 		   irq->u.emerg.code);
1627 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1628 				   irq->u.emerg.code, 0);
1629 
1630 	/* sending vcpu invalid */
1631 	if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1632 		return -EINVAL;
1633 
1634 	set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1635 	set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1636 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1637 	return 0;
1638 }
1639 
1640 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1641 {
1642 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1643 	struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1644 
1645 	vcpu->stat.inject_mchk++;
1646 	VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1647 		   irq->u.mchk.mcic);
1648 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1649 				   irq->u.mchk.mcic);
1650 
1651 	/*
1652 	 * Because repressible machine checks can be indicated along with
1653 	 * exigent machine checks (PoP, Chapter 11, Interruption action)
1654 	 * we need to combine cr14, mcic and external damage code.
1655 	 * Failing storage address and the logout area should not be or'ed
1656 	 * together, we just indicate the last occurrence of the corresponding
1657 	 * machine check
1658 	 */
1659 	mchk->cr14 |= irq->u.mchk.cr14;
1660 	mchk->mcic |= irq->u.mchk.mcic;
1661 	mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1662 	mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1663 	memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1664 	       sizeof(mchk->fixed_logout));
1665 	if (mchk->mcic & MCHK_EX_MASK)
1666 		set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1667 	else if (mchk->mcic & MCHK_REP_MASK)
1668 		set_bit(IRQ_PEND_MCHK_REP,  &li->pending_irqs);
1669 	return 0;
1670 }
1671 
1672 static int __inject_ckc(struct kvm_vcpu *vcpu)
1673 {
1674 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1675 
1676 	vcpu->stat.inject_ckc++;
1677 	VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1678 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1679 				   0, 0);
1680 
1681 	set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1682 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1683 	return 0;
1684 }
1685 
1686 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1687 {
1688 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1689 
1690 	vcpu->stat.inject_cputm++;
1691 	VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1692 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1693 				   0, 0);
1694 
1695 	set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1696 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1697 	return 0;
1698 }
1699 
1700 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1701 						  int isc, u32 schid)
1702 {
1703 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1704 	struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1705 	struct kvm_s390_interrupt_info *iter;
1706 	u16 id = (schid & 0xffff0000U) >> 16;
1707 	u16 nr = schid & 0x0000ffffU;
1708 
1709 	spin_lock(&fi->lock);
1710 	list_for_each_entry(iter, isc_list, list) {
1711 		if (schid && (id != iter->io.subchannel_id ||
1712 			      nr != iter->io.subchannel_nr))
1713 			continue;
1714 		/* found an appropriate entry */
1715 		list_del_init(&iter->list);
1716 		fi->counters[FIRQ_CNTR_IO] -= 1;
1717 		if (list_empty(isc_list))
1718 			clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1719 		spin_unlock(&fi->lock);
1720 		return iter;
1721 	}
1722 	spin_unlock(&fi->lock);
1723 	return NULL;
1724 }
1725 
1726 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1727 						      u64 isc_mask, u32 schid)
1728 {
1729 	struct kvm_s390_interrupt_info *inti = NULL;
1730 	int isc;
1731 
1732 	for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1733 		if (isc_mask & isc_to_isc_bits(isc))
1734 			inti = get_io_int(kvm, isc, schid);
1735 	}
1736 	return inti;
1737 }
1738 
1739 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1740 {
1741 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1742 	unsigned long active_mask;
1743 	int isc;
1744 
1745 	if (schid)
1746 		goto out;
1747 	if (!gi->origin)
1748 		goto out;
1749 
1750 	active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1751 	while (active_mask) {
1752 		isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1753 		if (gisa_tac_ipm_gisc(gi->origin, isc))
1754 			return isc;
1755 		clear_bit_inv(isc, &active_mask);
1756 	}
1757 out:
1758 	return -EINVAL;
1759 }
1760 
1761 /*
1762  * Dequeue and return an I/O interrupt matching any of the interruption
1763  * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1764  * Take into account the interrupts pending in the interrupt list and in GISA.
1765  *
1766  * Note that for a guest that does not enable I/O interrupts
1767  * but relies on TPI, a flood of classic interrupts may starve
1768  * out adapter interrupts on the same isc. Linux does not do
1769  * that, and it is possible to work around the issue by configuring
1770  * different iscs for classic and adapter interrupts in the guest,
1771  * but we may want to revisit this in the future.
1772  */
1773 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1774 						    u64 isc_mask, u32 schid)
1775 {
1776 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1777 	struct kvm_s390_interrupt_info *inti, *tmp_inti;
1778 	int isc;
1779 
1780 	inti = get_top_io_int(kvm, isc_mask, schid);
1781 
1782 	isc = get_top_gisa_isc(kvm, isc_mask, schid);
1783 	if (isc < 0)
1784 		/* no AI in GISA */
1785 		goto out;
1786 
1787 	if (!inti)
1788 		/* AI in GISA but no classical IO int */
1789 		goto gisa_out;
1790 
1791 	/* both types of interrupts present */
1792 	if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1793 		/* classical IO int with higher priority */
1794 		gisa_set_ipm_gisc(gi->origin, isc);
1795 		goto out;
1796 	}
1797 gisa_out:
1798 	tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
1799 	if (tmp_inti) {
1800 		tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1801 		tmp_inti->io.io_int_word = isc_to_int_word(isc);
1802 		if (inti)
1803 			kvm_s390_reinject_io_int(kvm, inti);
1804 		inti = tmp_inti;
1805 	} else
1806 		gisa_set_ipm_gisc(gi->origin, isc);
1807 out:
1808 	return inti;
1809 }
1810 
1811 static int __inject_service(struct kvm *kvm,
1812 			     struct kvm_s390_interrupt_info *inti)
1813 {
1814 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1815 
1816 	kvm->stat.inject_service_signal++;
1817 	spin_lock(&fi->lock);
1818 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1819 
1820 	/* We always allow events, track them separately from the sccb ints */
1821 	if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING)
1822 		set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1823 
1824 	/*
1825 	 * Early versions of the QEMU s390 bios will inject several
1826 	 * service interrupts after another without handling a
1827 	 * condition code indicating busy.
1828 	 * We will silently ignore those superfluous sccb values.
1829 	 * A future version of QEMU will take care of serialization
1830 	 * of servc requests
1831 	 */
1832 	if (fi->srv_signal.ext_params & SCCB_MASK)
1833 		goto out;
1834 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1835 	set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1836 out:
1837 	spin_unlock(&fi->lock);
1838 	kfree(inti);
1839 	return 0;
1840 }
1841 
1842 static int __inject_virtio(struct kvm *kvm,
1843 			    struct kvm_s390_interrupt_info *inti)
1844 {
1845 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1846 
1847 	kvm->stat.inject_virtio++;
1848 	spin_lock(&fi->lock);
1849 	if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1850 		spin_unlock(&fi->lock);
1851 		return -EBUSY;
1852 	}
1853 	fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1854 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1855 	set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1856 	spin_unlock(&fi->lock);
1857 	return 0;
1858 }
1859 
1860 static int __inject_pfault_done(struct kvm *kvm,
1861 				 struct kvm_s390_interrupt_info *inti)
1862 {
1863 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1864 
1865 	kvm->stat.inject_pfault_done++;
1866 	spin_lock(&fi->lock);
1867 	if (fi->counters[FIRQ_CNTR_PFAULT] >=
1868 		(ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1869 		spin_unlock(&fi->lock);
1870 		return -EBUSY;
1871 	}
1872 	fi->counters[FIRQ_CNTR_PFAULT] += 1;
1873 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1874 	set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1875 	spin_unlock(&fi->lock);
1876 	return 0;
1877 }
1878 
1879 #define CR_PENDING_SUBCLASS 28
1880 static int __inject_float_mchk(struct kvm *kvm,
1881 				struct kvm_s390_interrupt_info *inti)
1882 {
1883 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1884 
1885 	kvm->stat.inject_float_mchk++;
1886 	spin_lock(&fi->lock);
1887 	fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1888 	fi->mchk.mcic |= inti->mchk.mcic;
1889 	set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1890 	spin_unlock(&fi->lock);
1891 	kfree(inti);
1892 	return 0;
1893 }
1894 
1895 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1896 {
1897 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1898 	struct kvm_s390_float_interrupt *fi;
1899 	struct list_head *list;
1900 	int isc;
1901 
1902 	kvm->stat.inject_io++;
1903 	isc = int_word_to_isc(inti->io.io_int_word);
1904 
1905 	/*
1906 	 * We do not use the lock checking variant as this is just a
1907 	 * performance optimization and we do not hold the lock here.
1908 	 * This is ok as the code will pick interrupts from both "lists"
1909 	 * for delivery.
1910 	 */
1911 	if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1912 		VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1913 		gisa_set_ipm_gisc(gi->origin, isc);
1914 		kfree(inti);
1915 		return 0;
1916 	}
1917 
1918 	fi = &kvm->arch.float_int;
1919 	spin_lock(&fi->lock);
1920 	if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1921 		spin_unlock(&fi->lock);
1922 		return -EBUSY;
1923 	}
1924 	fi->counters[FIRQ_CNTR_IO] += 1;
1925 
1926 	if (inti->type & KVM_S390_INT_IO_AI_MASK)
1927 		VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1928 	else
1929 		VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1930 			inti->io.subchannel_id >> 8,
1931 			inti->io.subchannel_id >> 1 & 0x3,
1932 			inti->io.subchannel_nr);
1933 	list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1934 	list_add_tail(&inti->list, list);
1935 	set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1936 	spin_unlock(&fi->lock);
1937 	return 0;
1938 }
1939 
1940 /*
1941  * Find a destination VCPU for a floating irq and kick it.
1942  */
1943 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1944 {
1945 	struct kvm_vcpu *dst_vcpu;
1946 	int sigcpu, online_vcpus, nr_tries = 0;
1947 
1948 	online_vcpus = atomic_read(&kvm->online_vcpus);
1949 	if (!online_vcpus)
1950 		return;
1951 
1952 	for (sigcpu = kvm->arch.float_int.last_sleep_cpu; ; sigcpu++) {
1953 		sigcpu %= online_vcpus;
1954 		dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1955 		if (!is_vcpu_stopped(dst_vcpu))
1956 			break;
1957 		/* avoid endless loops if all vcpus are stopped */
1958 		if (nr_tries++ >= online_vcpus)
1959 			return;
1960 	}
1961 
1962 	/* make the VCPU drop out of the SIE, or wake it up if sleeping */
1963 	switch (type) {
1964 	case KVM_S390_MCHK:
1965 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1966 		break;
1967 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1968 		if (!(type & KVM_S390_INT_IO_AI_MASK &&
1969 		      kvm->arch.gisa_int.origin) ||
1970 		      kvm_s390_pv_cpu_get_handle(dst_vcpu))
1971 			kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1972 		break;
1973 	default:
1974 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1975 		break;
1976 	}
1977 	kvm_s390_vcpu_wakeup(dst_vcpu);
1978 }
1979 
1980 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1981 {
1982 	u64 type = READ_ONCE(inti->type);
1983 	int rc;
1984 
1985 	switch (type) {
1986 	case KVM_S390_MCHK:
1987 		rc = __inject_float_mchk(kvm, inti);
1988 		break;
1989 	case KVM_S390_INT_VIRTIO:
1990 		rc = __inject_virtio(kvm, inti);
1991 		break;
1992 	case KVM_S390_INT_SERVICE:
1993 		rc = __inject_service(kvm, inti);
1994 		break;
1995 	case KVM_S390_INT_PFAULT_DONE:
1996 		rc = __inject_pfault_done(kvm, inti);
1997 		break;
1998 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1999 		rc = __inject_io(kvm, inti);
2000 		break;
2001 	default:
2002 		rc = -EINVAL;
2003 	}
2004 	if (rc)
2005 		return rc;
2006 
2007 	__floating_irq_kick(kvm, type);
2008 	return 0;
2009 }
2010 
2011 int kvm_s390_inject_vm(struct kvm *kvm,
2012 		       struct kvm_s390_interrupt *s390int)
2013 {
2014 	struct kvm_s390_interrupt_info *inti;
2015 	int rc;
2016 
2017 	inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2018 	if (!inti)
2019 		return -ENOMEM;
2020 
2021 	inti->type = s390int->type;
2022 	switch (inti->type) {
2023 	case KVM_S390_INT_VIRTIO:
2024 		VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
2025 			 s390int->parm, s390int->parm64);
2026 		inti->ext.ext_params = s390int->parm;
2027 		inti->ext.ext_params2 = s390int->parm64;
2028 		break;
2029 	case KVM_S390_INT_SERVICE:
2030 		VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
2031 		inti->ext.ext_params = s390int->parm;
2032 		break;
2033 	case KVM_S390_INT_PFAULT_DONE:
2034 		inti->ext.ext_params2 = s390int->parm64;
2035 		break;
2036 	case KVM_S390_MCHK:
2037 		VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
2038 			 s390int->parm64);
2039 		inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
2040 		inti->mchk.mcic = s390int->parm64;
2041 		break;
2042 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2043 		inti->io.subchannel_id = s390int->parm >> 16;
2044 		inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
2045 		inti->io.io_int_parm = s390int->parm64 >> 32;
2046 		inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
2047 		break;
2048 	default:
2049 		kfree(inti);
2050 		return -EINVAL;
2051 	}
2052 	trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
2053 				 2);
2054 
2055 	rc = __inject_vm(kvm, inti);
2056 	if (rc)
2057 		kfree(inti);
2058 	return rc;
2059 }
2060 
2061 int kvm_s390_reinject_io_int(struct kvm *kvm,
2062 			      struct kvm_s390_interrupt_info *inti)
2063 {
2064 	return __inject_vm(kvm, inti);
2065 }
2066 
2067 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
2068 		       struct kvm_s390_irq *irq)
2069 {
2070 	irq->type = s390int->type;
2071 	switch (irq->type) {
2072 	case KVM_S390_PROGRAM_INT:
2073 		if (s390int->parm & 0xffff0000)
2074 			return -EINVAL;
2075 		irq->u.pgm.code = s390int->parm;
2076 		break;
2077 	case KVM_S390_SIGP_SET_PREFIX:
2078 		irq->u.prefix.address = s390int->parm;
2079 		break;
2080 	case KVM_S390_SIGP_STOP:
2081 		irq->u.stop.flags = s390int->parm;
2082 		break;
2083 	case KVM_S390_INT_EXTERNAL_CALL:
2084 		if (s390int->parm & 0xffff0000)
2085 			return -EINVAL;
2086 		irq->u.extcall.code = s390int->parm;
2087 		break;
2088 	case KVM_S390_INT_EMERGENCY:
2089 		if (s390int->parm & 0xffff0000)
2090 			return -EINVAL;
2091 		irq->u.emerg.code = s390int->parm;
2092 		break;
2093 	case KVM_S390_MCHK:
2094 		irq->u.mchk.mcic = s390int->parm64;
2095 		break;
2096 	case KVM_S390_INT_PFAULT_INIT:
2097 		irq->u.ext.ext_params = s390int->parm;
2098 		irq->u.ext.ext_params2 = s390int->parm64;
2099 		break;
2100 	case KVM_S390_RESTART:
2101 	case KVM_S390_INT_CLOCK_COMP:
2102 	case KVM_S390_INT_CPU_TIMER:
2103 		break;
2104 	default:
2105 		return -EINVAL;
2106 	}
2107 	return 0;
2108 }
2109 
2110 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
2111 {
2112 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2113 
2114 	return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2115 }
2116 
2117 int kvm_s390_is_restart_irq_pending(struct kvm_vcpu *vcpu)
2118 {
2119 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2120 
2121 	return test_bit(IRQ_PEND_RESTART, &li->pending_irqs);
2122 }
2123 
2124 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
2125 {
2126 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2127 
2128 	spin_lock(&li->lock);
2129 	li->irq.stop.flags = 0;
2130 	clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2131 	spin_unlock(&li->lock);
2132 }
2133 
2134 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2135 {
2136 	int rc;
2137 
2138 	switch (irq->type) {
2139 	case KVM_S390_PROGRAM_INT:
2140 		rc = __inject_prog(vcpu, irq);
2141 		break;
2142 	case KVM_S390_SIGP_SET_PREFIX:
2143 		rc = __inject_set_prefix(vcpu, irq);
2144 		break;
2145 	case KVM_S390_SIGP_STOP:
2146 		rc = __inject_sigp_stop(vcpu, irq);
2147 		break;
2148 	case KVM_S390_RESTART:
2149 		rc = __inject_sigp_restart(vcpu);
2150 		break;
2151 	case KVM_S390_INT_CLOCK_COMP:
2152 		rc = __inject_ckc(vcpu);
2153 		break;
2154 	case KVM_S390_INT_CPU_TIMER:
2155 		rc = __inject_cpu_timer(vcpu);
2156 		break;
2157 	case KVM_S390_INT_EXTERNAL_CALL:
2158 		rc = __inject_extcall(vcpu, irq);
2159 		break;
2160 	case KVM_S390_INT_EMERGENCY:
2161 		rc = __inject_sigp_emergency(vcpu, irq);
2162 		break;
2163 	case KVM_S390_MCHK:
2164 		rc = __inject_mchk(vcpu, irq);
2165 		break;
2166 	case KVM_S390_INT_PFAULT_INIT:
2167 		rc = __inject_pfault_init(vcpu, irq);
2168 		break;
2169 	case KVM_S390_INT_VIRTIO:
2170 	case KVM_S390_INT_SERVICE:
2171 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2172 	default:
2173 		rc = -EINVAL;
2174 	}
2175 
2176 	return rc;
2177 }
2178 
2179 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2180 {
2181 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2182 	int rc;
2183 
2184 	spin_lock(&li->lock);
2185 	rc = do_inject_vcpu(vcpu, irq);
2186 	spin_unlock(&li->lock);
2187 	if (!rc)
2188 		kvm_s390_vcpu_wakeup(vcpu);
2189 	return rc;
2190 }
2191 
2192 static inline void clear_irq_list(struct list_head *_list)
2193 {
2194 	struct kvm_s390_interrupt_info *inti, *n;
2195 
2196 	list_for_each_entry_safe(inti, n, _list, list) {
2197 		list_del(&inti->list);
2198 		kfree(inti);
2199 	}
2200 }
2201 
2202 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2203 		       struct kvm_s390_irq *irq)
2204 {
2205 	irq->type = inti->type;
2206 	switch (inti->type) {
2207 	case KVM_S390_INT_PFAULT_INIT:
2208 	case KVM_S390_INT_PFAULT_DONE:
2209 	case KVM_S390_INT_VIRTIO:
2210 		irq->u.ext = inti->ext;
2211 		break;
2212 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2213 		irq->u.io = inti->io;
2214 		break;
2215 	}
2216 }
2217 
2218 void kvm_s390_clear_float_irqs(struct kvm *kvm)
2219 {
2220 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2221 	int i;
2222 
2223 	mutex_lock(&kvm->lock);
2224 	if (!kvm_s390_pv_is_protected(kvm))
2225 		fi->masked_irqs = 0;
2226 	mutex_unlock(&kvm->lock);
2227 	spin_lock(&fi->lock);
2228 	fi->pending_irqs = 0;
2229 	memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2230 	memset(&fi->mchk, 0, sizeof(fi->mchk));
2231 	for (i = 0; i < FIRQ_LIST_COUNT; i++)
2232 		clear_irq_list(&fi->lists[i]);
2233 	for (i = 0; i < FIRQ_MAX_COUNT; i++)
2234 		fi->counters[i] = 0;
2235 	spin_unlock(&fi->lock);
2236 	kvm_s390_gisa_clear(kvm);
2237 };
2238 
2239 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2240 {
2241 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2242 	struct kvm_s390_interrupt_info *inti;
2243 	struct kvm_s390_float_interrupt *fi;
2244 	struct kvm_s390_irq *buf;
2245 	struct kvm_s390_irq *irq;
2246 	int max_irqs;
2247 	int ret = 0;
2248 	int n = 0;
2249 	int i;
2250 
2251 	if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2252 		return -EINVAL;
2253 
2254 	/*
2255 	 * We are already using -ENOMEM to signal
2256 	 * userspace it may retry with a bigger buffer,
2257 	 * so we need to use something else for this case
2258 	 */
2259 	buf = vzalloc(len);
2260 	if (!buf)
2261 		return -ENOBUFS;
2262 
2263 	max_irqs = len / sizeof(struct kvm_s390_irq);
2264 
2265 	if (gi->origin && gisa_get_ipm(gi->origin)) {
2266 		for (i = 0; i <= MAX_ISC; i++) {
2267 			if (n == max_irqs) {
2268 				/* signal userspace to try again */
2269 				ret = -ENOMEM;
2270 				goto out_nolock;
2271 			}
2272 			if (gisa_tac_ipm_gisc(gi->origin, i)) {
2273 				irq = (struct kvm_s390_irq *) &buf[n];
2274 				irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2275 				irq->u.io.io_int_word = isc_to_int_word(i);
2276 				n++;
2277 			}
2278 		}
2279 	}
2280 	fi = &kvm->arch.float_int;
2281 	spin_lock(&fi->lock);
2282 	for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2283 		list_for_each_entry(inti, &fi->lists[i], list) {
2284 			if (n == max_irqs) {
2285 				/* signal userspace to try again */
2286 				ret = -ENOMEM;
2287 				goto out;
2288 			}
2289 			inti_to_irq(inti, &buf[n]);
2290 			n++;
2291 		}
2292 	}
2293 	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) ||
2294 	    test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) {
2295 		if (n == max_irqs) {
2296 			/* signal userspace to try again */
2297 			ret = -ENOMEM;
2298 			goto out;
2299 		}
2300 		irq = (struct kvm_s390_irq *) &buf[n];
2301 		irq->type = KVM_S390_INT_SERVICE;
2302 		irq->u.ext = fi->srv_signal;
2303 		n++;
2304 	}
2305 	if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2306 		if (n == max_irqs) {
2307 				/* signal userspace to try again */
2308 				ret = -ENOMEM;
2309 				goto out;
2310 		}
2311 		irq = (struct kvm_s390_irq *) &buf[n];
2312 		irq->type = KVM_S390_MCHK;
2313 		irq->u.mchk = fi->mchk;
2314 		n++;
2315 }
2316 
2317 out:
2318 	spin_unlock(&fi->lock);
2319 out_nolock:
2320 	if (!ret && n > 0) {
2321 		if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2322 			ret = -EFAULT;
2323 	}
2324 	vfree(buf);
2325 
2326 	return ret < 0 ? ret : n;
2327 }
2328 
2329 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2330 {
2331 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2332 	struct kvm_s390_ais_all ais;
2333 
2334 	if (attr->attr < sizeof(ais))
2335 		return -EINVAL;
2336 
2337 	if (!test_kvm_facility(kvm, 72))
2338 		return -EOPNOTSUPP;
2339 
2340 	mutex_lock(&fi->ais_lock);
2341 	ais.simm = fi->simm;
2342 	ais.nimm = fi->nimm;
2343 	mutex_unlock(&fi->ais_lock);
2344 
2345 	if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2346 		return -EFAULT;
2347 
2348 	return 0;
2349 }
2350 
2351 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2352 {
2353 	int r;
2354 
2355 	switch (attr->group) {
2356 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2357 		r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2358 					  attr->attr);
2359 		break;
2360 	case KVM_DEV_FLIC_AISM_ALL:
2361 		r = flic_ais_mode_get_all(dev->kvm, attr);
2362 		break;
2363 	default:
2364 		r = -EINVAL;
2365 	}
2366 
2367 	return r;
2368 }
2369 
2370 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2371 				     u64 addr)
2372 {
2373 	struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2374 	void *target = NULL;
2375 	void __user *source;
2376 	u64 size;
2377 
2378 	if (get_user(inti->type, (u64 __user *)addr))
2379 		return -EFAULT;
2380 
2381 	switch (inti->type) {
2382 	case KVM_S390_INT_PFAULT_INIT:
2383 	case KVM_S390_INT_PFAULT_DONE:
2384 	case KVM_S390_INT_VIRTIO:
2385 	case KVM_S390_INT_SERVICE:
2386 		target = (void *) &inti->ext;
2387 		source = &uptr->u.ext;
2388 		size = sizeof(inti->ext);
2389 		break;
2390 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2391 		target = (void *) &inti->io;
2392 		source = &uptr->u.io;
2393 		size = sizeof(inti->io);
2394 		break;
2395 	case KVM_S390_MCHK:
2396 		target = (void *) &inti->mchk;
2397 		source = &uptr->u.mchk;
2398 		size = sizeof(inti->mchk);
2399 		break;
2400 	default:
2401 		return -EINVAL;
2402 	}
2403 
2404 	if (copy_from_user(target, source, size))
2405 		return -EFAULT;
2406 
2407 	return 0;
2408 }
2409 
2410 static int enqueue_floating_irq(struct kvm_device *dev,
2411 				struct kvm_device_attr *attr)
2412 {
2413 	struct kvm_s390_interrupt_info *inti = NULL;
2414 	int r = 0;
2415 	int len = attr->attr;
2416 
2417 	if (len % sizeof(struct kvm_s390_irq) != 0)
2418 		return -EINVAL;
2419 	else if (len > KVM_S390_FLIC_MAX_BUFFER)
2420 		return -EINVAL;
2421 
2422 	while (len >= sizeof(struct kvm_s390_irq)) {
2423 		inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2424 		if (!inti)
2425 			return -ENOMEM;
2426 
2427 		r = copy_irq_from_user(inti, attr->addr);
2428 		if (r) {
2429 			kfree(inti);
2430 			return r;
2431 		}
2432 		r = __inject_vm(dev->kvm, inti);
2433 		if (r) {
2434 			kfree(inti);
2435 			return r;
2436 		}
2437 		len -= sizeof(struct kvm_s390_irq);
2438 		attr->addr += sizeof(struct kvm_s390_irq);
2439 	}
2440 
2441 	return r;
2442 }
2443 
2444 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2445 {
2446 	if (id >= MAX_S390_IO_ADAPTERS)
2447 		return NULL;
2448 	id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2449 	return kvm->arch.adapters[id];
2450 }
2451 
2452 static int register_io_adapter(struct kvm_device *dev,
2453 			       struct kvm_device_attr *attr)
2454 {
2455 	struct s390_io_adapter *adapter;
2456 	struct kvm_s390_io_adapter adapter_info;
2457 
2458 	if (copy_from_user(&adapter_info,
2459 			   (void __user *)attr->addr, sizeof(adapter_info)))
2460 		return -EFAULT;
2461 
2462 	if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2463 		return -EINVAL;
2464 
2465 	adapter_info.id = array_index_nospec(adapter_info.id,
2466 					     MAX_S390_IO_ADAPTERS);
2467 
2468 	if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2469 		return -EINVAL;
2470 
2471 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL_ACCOUNT);
2472 	if (!adapter)
2473 		return -ENOMEM;
2474 
2475 	adapter->id = adapter_info.id;
2476 	adapter->isc = adapter_info.isc;
2477 	adapter->maskable = adapter_info.maskable;
2478 	adapter->masked = false;
2479 	adapter->swap = adapter_info.swap;
2480 	adapter->suppressible = (adapter_info.flags) &
2481 				KVM_S390_ADAPTER_SUPPRESSIBLE;
2482 	dev->kvm->arch.adapters[adapter->id] = adapter;
2483 
2484 	return 0;
2485 }
2486 
2487 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2488 {
2489 	int ret;
2490 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2491 
2492 	if (!adapter || !adapter->maskable)
2493 		return -EINVAL;
2494 	ret = adapter->masked;
2495 	adapter->masked = masked;
2496 	return ret;
2497 }
2498 
2499 void kvm_s390_destroy_adapters(struct kvm *kvm)
2500 {
2501 	int i;
2502 
2503 	for (i = 0; i < MAX_S390_IO_ADAPTERS; i++)
2504 		kfree(kvm->arch.adapters[i]);
2505 }
2506 
2507 static int modify_io_adapter(struct kvm_device *dev,
2508 			     struct kvm_device_attr *attr)
2509 {
2510 	struct kvm_s390_io_adapter_req req;
2511 	struct s390_io_adapter *adapter;
2512 	int ret;
2513 
2514 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2515 		return -EFAULT;
2516 
2517 	adapter = get_io_adapter(dev->kvm, req.id);
2518 	if (!adapter)
2519 		return -EINVAL;
2520 	switch (req.type) {
2521 	case KVM_S390_IO_ADAPTER_MASK:
2522 		ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2523 		if (ret > 0)
2524 			ret = 0;
2525 		break;
2526 	/*
2527 	 * The following operations are no longer needed and therefore no-ops.
2528 	 * The gpa to hva translation is done when an IRQ route is set up. The
2529 	 * set_irq code uses get_user_pages_remote() to do the actual write.
2530 	 */
2531 	case KVM_S390_IO_ADAPTER_MAP:
2532 	case KVM_S390_IO_ADAPTER_UNMAP:
2533 		ret = 0;
2534 		break;
2535 	default:
2536 		ret = -EINVAL;
2537 	}
2538 
2539 	return ret;
2540 }
2541 
2542 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2543 
2544 {
2545 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2546 	u32 schid;
2547 
2548 	if (attr->flags)
2549 		return -EINVAL;
2550 	if (attr->attr != sizeof(schid))
2551 		return -EINVAL;
2552 	if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2553 		return -EFAULT;
2554 	if (!schid)
2555 		return -EINVAL;
2556 	kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2557 	/*
2558 	 * If userspace is conforming to the architecture, we can have at most
2559 	 * one pending I/O interrupt per subchannel, so this is effectively a
2560 	 * clear all.
2561 	 */
2562 	return 0;
2563 }
2564 
2565 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2566 {
2567 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2568 	struct kvm_s390_ais_req req;
2569 	int ret = 0;
2570 
2571 	if (!test_kvm_facility(kvm, 72))
2572 		return -EOPNOTSUPP;
2573 
2574 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2575 		return -EFAULT;
2576 
2577 	if (req.isc > MAX_ISC)
2578 		return -EINVAL;
2579 
2580 	trace_kvm_s390_modify_ais_mode(req.isc,
2581 				       (fi->simm & AIS_MODE_MASK(req.isc)) ?
2582 				       (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2583 				       2 : KVM_S390_AIS_MODE_SINGLE :
2584 				       KVM_S390_AIS_MODE_ALL, req.mode);
2585 
2586 	mutex_lock(&fi->ais_lock);
2587 	switch (req.mode) {
2588 	case KVM_S390_AIS_MODE_ALL:
2589 		fi->simm &= ~AIS_MODE_MASK(req.isc);
2590 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2591 		break;
2592 	case KVM_S390_AIS_MODE_SINGLE:
2593 		fi->simm |= AIS_MODE_MASK(req.isc);
2594 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2595 		break;
2596 	default:
2597 		ret = -EINVAL;
2598 	}
2599 	mutex_unlock(&fi->ais_lock);
2600 
2601 	return ret;
2602 }
2603 
2604 static int kvm_s390_inject_airq(struct kvm *kvm,
2605 				struct s390_io_adapter *adapter)
2606 {
2607 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2608 	struct kvm_s390_interrupt s390int = {
2609 		.type = KVM_S390_INT_IO(1, 0, 0, 0),
2610 		.parm = 0,
2611 		.parm64 = isc_to_int_word(adapter->isc),
2612 	};
2613 	int ret = 0;
2614 
2615 	if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2616 		return kvm_s390_inject_vm(kvm, &s390int);
2617 
2618 	mutex_lock(&fi->ais_lock);
2619 	if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2620 		trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2621 		goto out;
2622 	}
2623 
2624 	ret = kvm_s390_inject_vm(kvm, &s390int);
2625 	if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2626 		fi->nimm |= AIS_MODE_MASK(adapter->isc);
2627 		trace_kvm_s390_modify_ais_mode(adapter->isc,
2628 					       KVM_S390_AIS_MODE_SINGLE, 2);
2629 	}
2630 out:
2631 	mutex_unlock(&fi->ais_lock);
2632 	return ret;
2633 }
2634 
2635 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2636 {
2637 	unsigned int id = attr->attr;
2638 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2639 
2640 	if (!adapter)
2641 		return -EINVAL;
2642 
2643 	return kvm_s390_inject_airq(kvm, adapter);
2644 }
2645 
2646 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2647 {
2648 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2649 	struct kvm_s390_ais_all ais;
2650 
2651 	if (!test_kvm_facility(kvm, 72))
2652 		return -EOPNOTSUPP;
2653 
2654 	if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2655 		return -EFAULT;
2656 
2657 	mutex_lock(&fi->ais_lock);
2658 	fi->simm = ais.simm;
2659 	fi->nimm = ais.nimm;
2660 	mutex_unlock(&fi->ais_lock);
2661 
2662 	return 0;
2663 }
2664 
2665 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2666 {
2667 	int r = 0;
2668 	unsigned long i;
2669 	struct kvm_vcpu *vcpu;
2670 
2671 	switch (attr->group) {
2672 	case KVM_DEV_FLIC_ENQUEUE:
2673 		r = enqueue_floating_irq(dev, attr);
2674 		break;
2675 	case KVM_DEV_FLIC_CLEAR_IRQS:
2676 		kvm_s390_clear_float_irqs(dev->kvm);
2677 		break;
2678 	case KVM_DEV_FLIC_APF_ENABLE:
2679 		if (kvm_is_ucontrol(dev->kvm))
2680 			return -EINVAL;
2681 		dev->kvm->arch.gmap->pfault_enabled = 1;
2682 		break;
2683 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2684 		if (kvm_is_ucontrol(dev->kvm))
2685 			return -EINVAL;
2686 		dev->kvm->arch.gmap->pfault_enabled = 0;
2687 		/*
2688 		 * Make sure no async faults are in transition when
2689 		 * clearing the queues. So we don't need to worry
2690 		 * about late coming workers.
2691 		 */
2692 		synchronize_srcu(&dev->kvm->srcu);
2693 		kvm_for_each_vcpu(i, vcpu, dev->kvm)
2694 			kvm_clear_async_pf_completion_queue(vcpu);
2695 		break;
2696 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2697 		r = register_io_adapter(dev, attr);
2698 		break;
2699 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2700 		r = modify_io_adapter(dev, attr);
2701 		break;
2702 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2703 		r = clear_io_irq(dev->kvm, attr);
2704 		break;
2705 	case KVM_DEV_FLIC_AISM:
2706 		r = modify_ais_mode(dev->kvm, attr);
2707 		break;
2708 	case KVM_DEV_FLIC_AIRQ_INJECT:
2709 		r = flic_inject_airq(dev->kvm, attr);
2710 		break;
2711 	case KVM_DEV_FLIC_AISM_ALL:
2712 		r = flic_ais_mode_set_all(dev->kvm, attr);
2713 		break;
2714 	default:
2715 		r = -EINVAL;
2716 	}
2717 
2718 	return r;
2719 }
2720 
2721 static int flic_has_attr(struct kvm_device *dev,
2722 			     struct kvm_device_attr *attr)
2723 {
2724 	switch (attr->group) {
2725 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2726 	case KVM_DEV_FLIC_ENQUEUE:
2727 	case KVM_DEV_FLIC_CLEAR_IRQS:
2728 	case KVM_DEV_FLIC_APF_ENABLE:
2729 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2730 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2731 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2732 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2733 	case KVM_DEV_FLIC_AISM:
2734 	case KVM_DEV_FLIC_AIRQ_INJECT:
2735 	case KVM_DEV_FLIC_AISM_ALL:
2736 		return 0;
2737 	}
2738 	return -ENXIO;
2739 }
2740 
2741 static int flic_create(struct kvm_device *dev, u32 type)
2742 {
2743 	if (!dev)
2744 		return -EINVAL;
2745 	if (dev->kvm->arch.flic)
2746 		return -EINVAL;
2747 	dev->kvm->arch.flic = dev;
2748 	return 0;
2749 }
2750 
2751 static void flic_destroy(struct kvm_device *dev)
2752 {
2753 	dev->kvm->arch.flic = NULL;
2754 	kfree(dev);
2755 }
2756 
2757 /* s390 floating irq controller (flic) */
2758 struct kvm_device_ops kvm_flic_ops = {
2759 	.name = "kvm-flic",
2760 	.get_attr = flic_get_attr,
2761 	.set_attr = flic_set_attr,
2762 	.has_attr = flic_has_attr,
2763 	.create = flic_create,
2764 	.destroy = flic_destroy,
2765 };
2766 
2767 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2768 {
2769 	unsigned long bit;
2770 
2771 	bit = bit_nr + (addr % PAGE_SIZE) * 8;
2772 
2773 	return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2774 }
2775 
2776 static struct page *get_map_page(struct kvm *kvm, u64 uaddr)
2777 {
2778 	struct mm_struct *mm = kvm->mm;
2779 	struct page *page = NULL;
2780 	int locked = 1;
2781 
2782 	if (mmget_not_zero(mm)) {
2783 		mmap_read_lock(mm);
2784 		get_user_pages_remote(mm, uaddr, 1, FOLL_WRITE,
2785 				      &page, &locked);
2786 		if (locked)
2787 			mmap_read_unlock(mm);
2788 		mmput(mm);
2789 	}
2790 
2791 	return page;
2792 }
2793 
2794 static int adapter_indicators_set(struct kvm *kvm,
2795 				  struct s390_io_adapter *adapter,
2796 				  struct kvm_s390_adapter_int *adapter_int)
2797 {
2798 	unsigned long bit;
2799 	int summary_set, idx;
2800 	struct page *ind_page, *summary_page;
2801 	void *map;
2802 
2803 	ind_page = get_map_page(kvm, adapter_int->ind_addr);
2804 	if (!ind_page)
2805 		return -1;
2806 	summary_page = get_map_page(kvm, adapter_int->summary_addr);
2807 	if (!summary_page) {
2808 		put_page(ind_page);
2809 		return -1;
2810 	}
2811 
2812 	idx = srcu_read_lock(&kvm->srcu);
2813 	map = page_address(ind_page);
2814 	bit = get_ind_bit(adapter_int->ind_addr,
2815 			  adapter_int->ind_offset, adapter->swap);
2816 	set_bit(bit, map);
2817 	mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT);
2818 	set_page_dirty_lock(ind_page);
2819 	map = page_address(summary_page);
2820 	bit = get_ind_bit(adapter_int->summary_addr,
2821 			  adapter_int->summary_offset, adapter->swap);
2822 	summary_set = test_and_set_bit(bit, map);
2823 	mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT);
2824 	set_page_dirty_lock(summary_page);
2825 	srcu_read_unlock(&kvm->srcu, idx);
2826 
2827 	put_page(ind_page);
2828 	put_page(summary_page);
2829 	return summary_set ? 0 : 1;
2830 }
2831 
2832 /*
2833  * < 0 - not injected due to error
2834  * = 0 - coalesced, summary indicator already active
2835  * > 0 - injected interrupt
2836  */
2837 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2838 			   struct kvm *kvm, int irq_source_id, int level,
2839 			   bool line_status)
2840 {
2841 	int ret;
2842 	struct s390_io_adapter *adapter;
2843 
2844 	/* We're only interested in the 0->1 transition. */
2845 	if (!level)
2846 		return 0;
2847 	adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2848 	if (!adapter)
2849 		return -1;
2850 	ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2851 	if ((ret > 0) && !adapter->masked) {
2852 		ret = kvm_s390_inject_airq(kvm, adapter);
2853 		if (ret == 0)
2854 			ret = 1;
2855 	}
2856 	return ret;
2857 }
2858 
2859 /*
2860  * Inject the machine check to the guest.
2861  */
2862 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2863 				     struct mcck_volatile_info *mcck_info)
2864 {
2865 	struct kvm_s390_interrupt_info inti;
2866 	struct kvm_s390_irq irq;
2867 	struct kvm_s390_mchk_info *mchk;
2868 	union mci mci;
2869 	__u64 cr14 = 0;         /* upper bits are not used */
2870 	int rc;
2871 
2872 	mci.val = mcck_info->mcic;
2873 	if (mci.sr)
2874 		cr14 |= CR14_RECOVERY_SUBMASK;
2875 	if (mci.dg)
2876 		cr14 |= CR14_DEGRADATION_SUBMASK;
2877 	if (mci.w)
2878 		cr14 |= CR14_WARNING_SUBMASK;
2879 
2880 	mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2881 	mchk->cr14 = cr14;
2882 	mchk->mcic = mcck_info->mcic;
2883 	mchk->ext_damage_code = mcck_info->ext_damage_code;
2884 	mchk->failing_storage_address = mcck_info->failing_storage_address;
2885 	if (mci.ck) {
2886 		/* Inject the floating machine check */
2887 		inti.type = KVM_S390_MCHK;
2888 		rc = __inject_vm(vcpu->kvm, &inti);
2889 	} else {
2890 		/* Inject the machine check to specified vcpu */
2891 		irq.type = KVM_S390_MCHK;
2892 		rc = kvm_s390_inject_vcpu(vcpu, &irq);
2893 	}
2894 	WARN_ON_ONCE(rc);
2895 }
2896 
2897 int kvm_set_routing_entry(struct kvm *kvm,
2898 			  struct kvm_kernel_irq_routing_entry *e,
2899 			  const struct kvm_irq_routing_entry *ue)
2900 {
2901 	u64 uaddr_s, uaddr_i;
2902 	int idx;
2903 
2904 	switch (ue->type) {
2905 	/* we store the userspace addresses instead of the guest addresses */
2906 	case KVM_IRQ_ROUTING_S390_ADAPTER:
2907 		if (kvm_is_ucontrol(kvm))
2908 			return -EINVAL;
2909 		e->set = set_adapter_int;
2910 
2911 		idx = srcu_read_lock(&kvm->srcu);
2912 		uaddr_s = gpa_to_hva(kvm, ue->u.adapter.summary_addr);
2913 		uaddr_i = gpa_to_hva(kvm, ue->u.adapter.ind_addr);
2914 		srcu_read_unlock(&kvm->srcu, idx);
2915 
2916 		if (kvm_is_error_hva(uaddr_s) || kvm_is_error_hva(uaddr_i))
2917 			return -EFAULT;
2918 		e->adapter.summary_addr = uaddr_s;
2919 		e->adapter.ind_addr = uaddr_i;
2920 		e->adapter.summary_offset = ue->u.adapter.summary_offset;
2921 		e->adapter.ind_offset = ue->u.adapter.ind_offset;
2922 		e->adapter.adapter_id = ue->u.adapter.adapter_id;
2923 		return 0;
2924 	default:
2925 		return -EINVAL;
2926 	}
2927 }
2928 
2929 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2930 		int irq_source_id, int level, bool line_status)
2931 {
2932 	return -EINVAL;
2933 }
2934 
2935 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2936 {
2937 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2938 	struct kvm_s390_irq *buf;
2939 	int r = 0;
2940 	int n;
2941 
2942 	buf = vmalloc(len);
2943 	if (!buf)
2944 		return -ENOMEM;
2945 
2946 	if (copy_from_user((void *) buf, irqstate, len)) {
2947 		r = -EFAULT;
2948 		goto out_free;
2949 	}
2950 
2951 	/*
2952 	 * Don't allow setting the interrupt state
2953 	 * when there are already interrupts pending
2954 	 */
2955 	spin_lock(&li->lock);
2956 	if (li->pending_irqs) {
2957 		r = -EBUSY;
2958 		goto out_unlock;
2959 	}
2960 
2961 	for (n = 0; n < len / sizeof(*buf); n++) {
2962 		r = do_inject_vcpu(vcpu, &buf[n]);
2963 		if (r)
2964 			break;
2965 	}
2966 
2967 out_unlock:
2968 	spin_unlock(&li->lock);
2969 out_free:
2970 	vfree(buf);
2971 
2972 	return r;
2973 }
2974 
2975 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2976 			    struct kvm_s390_irq *irq,
2977 			    unsigned long irq_type)
2978 {
2979 	switch (irq_type) {
2980 	case IRQ_PEND_MCHK_EX:
2981 	case IRQ_PEND_MCHK_REP:
2982 		irq->type = KVM_S390_MCHK;
2983 		irq->u.mchk = li->irq.mchk;
2984 		break;
2985 	case IRQ_PEND_PROG:
2986 		irq->type = KVM_S390_PROGRAM_INT;
2987 		irq->u.pgm = li->irq.pgm;
2988 		break;
2989 	case IRQ_PEND_PFAULT_INIT:
2990 		irq->type = KVM_S390_INT_PFAULT_INIT;
2991 		irq->u.ext = li->irq.ext;
2992 		break;
2993 	case IRQ_PEND_EXT_EXTERNAL:
2994 		irq->type = KVM_S390_INT_EXTERNAL_CALL;
2995 		irq->u.extcall = li->irq.extcall;
2996 		break;
2997 	case IRQ_PEND_EXT_CLOCK_COMP:
2998 		irq->type = KVM_S390_INT_CLOCK_COMP;
2999 		break;
3000 	case IRQ_PEND_EXT_CPU_TIMER:
3001 		irq->type = KVM_S390_INT_CPU_TIMER;
3002 		break;
3003 	case IRQ_PEND_SIGP_STOP:
3004 		irq->type = KVM_S390_SIGP_STOP;
3005 		irq->u.stop = li->irq.stop;
3006 		break;
3007 	case IRQ_PEND_RESTART:
3008 		irq->type = KVM_S390_RESTART;
3009 		break;
3010 	case IRQ_PEND_SET_PREFIX:
3011 		irq->type = KVM_S390_SIGP_SET_PREFIX;
3012 		irq->u.prefix = li->irq.prefix;
3013 		break;
3014 	}
3015 }
3016 
3017 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
3018 {
3019 	int scn;
3020 	DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
3021 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
3022 	unsigned long pending_irqs;
3023 	struct kvm_s390_irq irq;
3024 	unsigned long irq_type;
3025 	int cpuaddr;
3026 	int n = 0;
3027 
3028 	spin_lock(&li->lock);
3029 	pending_irqs = li->pending_irqs;
3030 	memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
3031 	       sizeof(sigp_emerg_pending));
3032 	spin_unlock(&li->lock);
3033 
3034 	for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
3035 		memset(&irq, 0, sizeof(irq));
3036 		if (irq_type == IRQ_PEND_EXT_EMERGENCY)
3037 			continue;
3038 		if (n + sizeof(irq) > len)
3039 			return -ENOBUFS;
3040 		store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
3041 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3042 			return -EFAULT;
3043 		n += sizeof(irq);
3044 	}
3045 
3046 	if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
3047 		for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
3048 			memset(&irq, 0, sizeof(irq));
3049 			if (n + sizeof(irq) > len)
3050 				return -ENOBUFS;
3051 			irq.type = KVM_S390_INT_EMERGENCY;
3052 			irq.u.emerg.code = cpuaddr;
3053 			if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3054 				return -EFAULT;
3055 			n += sizeof(irq);
3056 		}
3057 	}
3058 
3059 	if (sca_ext_call_pending(vcpu, &scn)) {
3060 		if (n + sizeof(irq) > len)
3061 			return -ENOBUFS;
3062 		memset(&irq, 0, sizeof(irq));
3063 		irq.type = KVM_S390_INT_EXTERNAL_CALL;
3064 		irq.u.extcall.code = scn;
3065 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3066 			return -EFAULT;
3067 		n += sizeof(irq);
3068 	}
3069 
3070 	return n;
3071 }
3072 
3073 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
3074 {
3075 	int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus);
3076 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3077 	struct kvm_vcpu *vcpu;
3078 	u8 vcpu_isc_mask;
3079 
3080 	for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) {
3081 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
3082 		if (psw_ioint_disabled(vcpu))
3083 			continue;
3084 		vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
3085 		if (deliverable_mask & vcpu_isc_mask) {
3086 			/* lately kicked but not yet running */
3087 			if (test_and_set_bit(vcpu_idx, gi->kicked_mask))
3088 				return;
3089 			kvm_s390_vcpu_wakeup(vcpu);
3090 			return;
3091 		}
3092 	}
3093 }
3094 
3095 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3096 {
3097 	struct kvm_s390_gisa_interrupt *gi =
3098 		container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3099 	struct kvm *kvm =
3100 		container_of(gi->origin, struct sie_page2, gisa)->kvm;
3101 	u8 pending_mask;
3102 
3103 	pending_mask = gisa_get_ipm_or_restore_iam(gi);
3104 	if (pending_mask) {
3105 		__airqs_kick_single_vcpu(kvm, pending_mask);
3106 		hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3107 		return HRTIMER_RESTART;
3108 	}
3109 
3110 	return HRTIMER_NORESTART;
3111 }
3112 
3113 #define NULL_GISA_ADDR 0x00000000UL
3114 #define NONE_GISA_ADDR 0x00000001UL
3115 #define GISA_ADDR_MASK 0xfffff000UL
3116 
3117 static void process_gib_alert_list(void)
3118 {
3119 	struct kvm_s390_gisa_interrupt *gi;
3120 	u32 final, gisa_phys, origin = 0UL;
3121 	struct kvm_s390_gisa *gisa;
3122 	struct kvm *kvm;
3123 
3124 	do {
3125 		/*
3126 		 * If the NONE_GISA_ADDR is still stored in the alert list
3127 		 * origin, we will leave the outer loop. No further GISA has
3128 		 * been added to the alert list by millicode while processing
3129 		 * the current alert list.
3130 		 */
3131 		final = (origin & NONE_GISA_ADDR);
3132 		/*
3133 		 * Cut off the alert list and store the NONE_GISA_ADDR in the
3134 		 * alert list origin to avoid further GAL interruptions.
3135 		 * A new alert list can be build up by millicode in parallel
3136 		 * for guests not in the yet cut-off alert list. When in the
3137 		 * final loop, store the NULL_GISA_ADDR instead. This will re-
3138 		 * enable GAL interruptions on the host again.
3139 		 */
3140 		origin = xchg(&gib->alert_list_origin,
3141 			      (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3142 		/*
3143 		 * Loop through the just cut-off alert list and start the
3144 		 * gisa timers to kick idle vcpus to consume the pending
3145 		 * interruptions asap.
3146 		 */
3147 		while (origin & GISA_ADDR_MASK) {
3148 			gisa_phys = origin;
3149 			gisa = phys_to_virt(gisa_phys);
3150 			origin = gisa->next_alert;
3151 			gisa->next_alert = gisa_phys;
3152 			kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3153 			gi = &kvm->arch.gisa_int;
3154 			if (hrtimer_active(&gi->timer))
3155 				hrtimer_cancel(&gi->timer);
3156 			hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3157 		}
3158 	} while (!final);
3159 
3160 }
3161 
3162 void kvm_s390_gisa_clear(struct kvm *kvm)
3163 {
3164 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3165 
3166 	if (!gi->origin)
3167 		return;
3168 	gisa_clear_ipm(gi->origin);
3169 	VM_EVENT(kvm, 3, "gisa 0x%p cleared", gi->origin);
3170 }
3171 
3172 void kvm_s390_gisa_init(struct kvm *kvm)
3173 {
3174 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3175 
3176 	if (!css_general_characteristics.aiv)
3177 		return;
3178 	gi->origin = &kvm->arch.sie_page2->gisa;
3179 	gi->alert.mask = 0;
3180 	spin_lock_init(&gi->alert.ref_lock);
3181 	gi->expires = 50 * 1000; /* 50 usec */
3182 	hrtimer_setup(&gi->timer, gisa_vcpu_kicker, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3183 	memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3184 	gi->origin->next_alert = (u32)virt_to_phys(gi->origin);
3185 	VM_EVENT(kvm, 3, "gisa 0x%p initialized", gi->origin);
3186 }
3187 
3188 void kvm_s390_gisa_enable(struct kvm *kvm)
3189 {
3190 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3191 	struct kvm_vcpu *vcpu;
3192 	unsigned long i;
3193 	u32 gisa_desc;
3194 
3195 	if (gi->origin)
3196 		return;
3197 	kvm_s390_gisa_init(kvm);
3198 	gisa_desc = kvm_s390_get_gisa_desc(kvm);
3199 	if (!gisa_desc)
3200 		return;
3201 	kvm_for_each_vcpu(i, vcpu, kvm) {
3202 		mutex_lock(&vcpu->mutex);
3203 		vcpu->arch.sie_block->gd = gisa_desc;
3204 		vcpu->arch.sie_block->eca |= ECA_AIV;
3205 		VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3206 			   vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3207 		mutex_unlock(&vcpu->mutex);
3208 	}
3209 }
3210 
3211 void kvm_s390_gisa_destroy(struct kvm *kvm)
3212 {
3213 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3214 	struct kvm_s390_gisa *gisa = gi->origin;
3215 
3216 	if (!gi->origin)
3217 		return;
3218 	WARN(gi->alert.mask != 0x00,
3219 	     "unexpected non zero alert.mask 0x%02x",
3220 	     gi->alert.mask);
3221 	gi->alert.mask = 0x00;
3222 	if (gisa_set_iam(gi->origin, gi->alert.mask))
3223 		process_gib_alert_list();
3224 	hrtimer_cancel(&gi->timer);
3225 	gi->origin = NULL;
3226 	VM_EVENT(kvm, 3, "gisa 0x%p destroyed", gisa);
3227 }
3228 
3229 void kvm_s390_gisa_disable(struct kvm *kvm)
3230 {
3231 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3232 	struct kvm_vcpu *vcpu;
3233 	unsigned long i;
3234 
3235 	if (!gi->origin)
3236 		return;
3237 	kvm_for_each_vcpu(i, vcpu, kvm) {
3238 		mutex_lock(&vcpu->mutex);
3239 		vcpu->arch.sie_block->eca &= ~ECA_AIV;
3240 		vcpu->arch.sie_block->gd = 0U;
3241 		mutex_unlock(&vcpu->mutex);
3242 		VCPU_EVENT(vcpu, 3, "AIV disabled for cpu %03u", vcpu->vcpu_id);
3243 	}
3244 	kvm_s390_gisa_destroy(kvm);
3245 }
3246 
3247 /**
3248  * kvm_s390_gisc_register - register a guest ISC
3249  *
3250  * @kvm:  the kernel vm to work with
3251  * @gisc: the guest interruption sub class to register
3252  *
3253  * The function extends the vm specific alert mask to use.
3254  * The effective IAM mask in the GISA is updated as well
3255  * in case the GISA is not part of the GIB alert list.
3256  * It will be updated latest when the IAM gets restored
3257  * by gisa_get_ipm_or_restore_iam().
3258  *
3259  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3260  *          has registered with the channel subsystem.
3261  *          -ENODEV in case the vm uses no GISA
3262  *          -ERANGE in case the guest ISC is invalid
3263  */
3264 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3265 {
3266 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3267 
3268 	if (!gi->origin)
3269 		return -ENODEV;
3270 	if (gisc > MAX_ISC)
3271 		return -ERANGE;
3272 
3273 	spin_lock(&gi->alert.ref_lock);
3274 	gi->alert.ref_count[gisc]++;
3275 	if (gi->alert.ref_count[gisc] == 1) {
3276 		gi->alert.mask |= 0x80 >> gisc;
3277 		gisa_set_iam(gi->origin, gi->alert.mask);
3278 	}
3279 	spin_unlock(&gi->alert.ref_lock);
3280 
3281 	return gib->nisc;
3282 }
3283 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3284 
3285 /**
3286  * kvm_s390_gisc_unregister - unregister a guest ISC
3287  *
3288  * @kvm:  the kernel vm to work with
3289  * @gisc: the guest interruption sub class to register
3290  *
3291  * The function reduces the vm specific alert mask to use.
3292  * The effective IAM mask in the GISA is updated as well
3293  * in case the GISA is not part of the GIB alert list.
3294  * It will be updated latest when the IAM gets restored
3295  * by gisa_get_ipm_or_restore_iam().
3296  *
3297  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3298  *          has registered with the channel subsystem.
3299  *          -ENODEV in case the vm uses no GISA
3300  *          -ERANGE in case the guest ISC is invalid
3301  *          -EINVAL in case the guest ISC is not registered
3302  */
3303 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3304 {
3305 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3306 	int rc = 0;
3307 
3308 	if (!gi->origin)
3309 		return -ENODEV;
3310 	if (gisc > MAX_ISC)
3311 		return -ERANGE;
3312 
3313 	spin_lock(&gi->alert.ref_lock);
3314 	if (gi->alert.ref_count[gisc] == 0) {
3315 		rc = -EINVAL;
3316 		goto out;
3317 	}
3318 	gi->alert.ref_count[gisc]--;
3319 	if (gi->alert.ref_count[gisc] == 0) {
3320 		gi->alert.mask &= ~(0x80 >> gisc);
3321 		gisa_set_iam(gi->origin, gi->alert.mask);
3322 	}
3323 out:
3324 	spin_unlock(&gi->alert.ref_lock);
3325 
3326 	return rc;
3327 }
3328 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3329 
3330 static void aen_host_forward(unsigned long si)
3331 {
3332 	struct kvm_s390_gisa_interrupt *gi;
3333 	struct zpci_gaite *gaite;
3334 	struct kvm *kvm;
3335 
3336 	gaite = (struct zpci_gaite *)aift->gait +
3337 		(si * sizeof(struct zpci_gaite));
3338 	if (gaite->count == 0)
3339 		return;
3340 	if (gaite->aisb != 0)
3341 		set_bit_inv(gaite->aisbo, phys_to_virt(gaite->aisb));
3342 
3343 	kvm = kvm_s390_pci_si_to_kvm(aift, si);
3344 	if (!kvm)
3345 		return;
3346 	gi = &kvm->arch.gisa_int;
3347 
3348 	if (!(gi->origin->g1.simm & AIS_MODE_MASK(gaite->gisc)) ||
3349 	    !(gi->origin->g1.nimm & AIS_MODE_MASK(gaite->gisc))) {
3350 		gisa_set_ipm_gisc(gi->origin, gaite->gisc);
3351 		if (hrtimer_active(&gi->timer))
3352 			hrtimer_cancel(&gi->timer);
3353 		hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3354 		kvm->stat.aen_forward++;
3355 	}
3356 }
3357 
3358 static void aen_process_gait(u8 isc)
3359 {
3360 	bool found = false, first = true;
3361 	union zpci_sic_iib iib = {{0}};
3362 	unsigned long si, flags;
3363 
3364 	spin_lock_irqsave(&aift->gait_lock, flags);
3365 
3366 	if (!aift->gait) {
3367 		spin_unlock_irqrestore(&aift->gait_lock, flags);
3368 		return;
3369 	}
3370 
3371 	for (si = 0;;) {
3372 		/* Scan adapter summary indicator bit vector */
3373 		si = airq_iv_scan(aift->sbv, si, airq_iv_end(aift->sbv));
3374 		if (si == -1UL) {
3375 			if (first || found) {
3376 				/* Re-enable interrupts. */
3377 				zpci_set_irq_ctrl(SIC_IRQ_MODE_SINGLE, isc,
3378 						  &iib);
3379 				first = found = false;
3380 			} else {
3381 				/* Interrupts on and all bits processed */
3382 				break;
3383 			}
3384 			found = false;
3385 			si = 0;
3386 			/* Scan again after re-enabling interrupts */
3387 			continue;
3388 		}
3389 		found = true;
3390 		aen_host_forward(si);
3391 	}
3392 
3393 	spin_unlock_irqrestore(&aift->gait_lock, flags);
3394 }
3395 
3396 static void gib_alert_irq_handler(struct airq_struct *airq,
3397 				  struct tpi_info *tpi_info)
3398 {
3399 	struct tpi_adapter_info *info = (struct tpi_adapter_info *)tpi_info;
3400 
3401 	inc_irq_stat(IRQIO_GAL);
3402 
3403 	if ((info->forward || info->error) &&
3404 	    IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3405 		aen_process_gait(info->isc);
3406 		if (info->aism != 0)
3407 			process_gib_alert_list();
3408 	} else {
3409 		process_gib_alert_list();
3410 	}
3411 }
3412 
3413 static struct airq_struct gib_alert_irq = {
3414 	.handler = gib_alert_irq_handler,
3415 };
3416 
3417 void kvm_s390_gib_destroy(void)
3418 {
3419 	if (!gib)
3420 		return;
3421 	if (kvm_s390_pci_interp_allowed() && aift) {
3422 		mutex_lock(&aift->aift_lock);
3423 		kvm_s390_pci_aen_exit();
3424 		mutex_unlock(&aift->aift_lock);
3425 	}
3426 	chsc_sgib(0);
3427 	unregister_adapter_interrupt(&gib_alert_irq);
3428 	free_page((unsigned long)gib);
3429 	gib = NULL;
3430 }
3431 
3432 int __init kvm_s390_gib_init(u8 nisc)
3433 {
3434 	u32 gib_origin;
3435 	int rc = 0;
3436 
3437 	if (!css_general_characteristics.aiv) {
3438 		KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3439 		goto out;
3440 	}
3441 
3442 	gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3443 	if (!gib) {
3444 		rc = -ENOMEM;
3445 		goto out;
3446 	}
3447 
3448 	gib_alert_irq.isc = nisc;
3449 	if (register_adapter_interrupt(&gib_alert_irq)) {
3450 		pr_err("Registering the GIB alert interruption handler failed\n");
3451 		rc = -EIO;
3452 		goto out_free_gib;
3453 	}
3454 	/* adapter interrupts used for AP (applicable here) don't use the LSI */
3455 	*gib_alert_irq.lsi_ptr = 0xff;
3456 
3457 	gib->nisc = nisc;
3458 	gib_origin = virt_to_phys(gib);
3459 	if (chsc_sgib(gib_origin)) {
3460 		pr_err("Associating the GIB with the AIV facility failed\n");
3461 		free_page((unsigned long)gib);
3462 		gib = NULL;
3463 		rc = -EIO;
3464 		goto out_unreg_gal;
3465 	}
3466 
3467 	if (kvm_s390_pci_interp_allowed()) {
3468 		if (kvm_s390_pci_aen_init(nisc)) {
3469 			pr_err("Initializing AEN for PCI failed\n");
3470 			rc = -EIO;
3471 			goto out_unreg_gal;
3472 		}
3473 	}
3474 
3475 	KVM_EVENT(3, "gib 0x%p (nisc=%d) initialized", gib, gib->nisc);
3476 	goto out;
3477 
3478 out_unreg_gal:
3479 	unregister_adapter_interrupt(&gib_alert_irq);
3480 out_free_gib:
3481 	free_page((unsigned long)gib);
3482 	gib = NULL;
3483 out:
3484 	return rc;
3485 }
3486