xref: /openbmc/linux/arch/s390/kvm/interrupt.c (revision 2a9eb57e)
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 KMSG_COMPONENT "kvm-s390"
11 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
12 
13 #include <linux/interrupt.h>
14 #include <linux/kvm_host.h>
15 #include <linux/hrtimer.h>
16 #include <linux/mmu_context.h>
17 #include <linux/nospec.h>
18 #include <linux/signal.h>
19 #include <linux/slab.h>
20 #include <linux/bitmap.h>
21 #include <linux/vmalloc.h>
22 #include <asm/asm-offsets.h>
23 #include <asm/dis.h>
24 #include <linux/uaccess.h>
25 #include <asm/sclp.h>
26 #include <asm/isc.h>
27 #include <asm/gmap.h>
28 #include <asm/switch_to.h>
29 #include <asm/nmi.h>
30 #include <asm/airq.h>
31 #include <asm/tpi.h>
32 #include "kvm-s390.h"
33 #include "gaccess.h"
34 #include "trace-s390.h"
35 #include "pci.h"
36 
37 #define PFAULT_INIT 0x0600
38 #define PFAULT_DONE 0x0680
39 #define VIRTIO_PARAM 0x0d00
40 
41 static struct kvm_s390_gib *gib;
42 
43 /* handle external calls via sigp interpretation facility */
44 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
45 {
46 	int c, scn;
47 
48 	if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND))
49 		return 0;
50 
51 	BUG_ON(!kvm_s390_use_sca_entries());
52 	read_lock(&vcpu->kvm->arch.sca_lock);
53 	if (vcpu->kvm->arch.use_esca) {
54 		struct esca_block *sca = vcpu->kvm->arch.sca;
55 		union esca_sigp_ctrl sigp_ctrl =
56 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
57 
58 		c = sigp_ctrl.c;
59 		scn = sigp_ctrl.scn;
60 	} else {
61 		struct bsca_block *sca = vcpu->kvm->arch.sca;
62 		union bsca_sigp_ctrl sigp_ctrl =
63 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
64 
65 		c = sigp_ctrl.c;
66 		scn = sigp_ctrl.scn;
67 	}
68 	read_unlock(&vcpu->kvm->arch.sca_lock);
69 
70 	if (src_id)
71 		*src_id = scn;
72 
73 	return c;
74 }
75 
76 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
77 {
78 	int expect, rc;
79 
80 	BUG_ON(!kvm_s390_use_sca_entries());
81 	read_lock(&vcpu->kvm->arch.sca_lock);
82 	if (vcpu->kvm->arch.use_esca) {
83 		struct esca_block *sca = vcpu->kvm->arch.sca;
84 		union esca_sigp_ctrl *sigp_ctrl =
85 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
86 		union esca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
87 
88 		new_val.scn = src_id;
89 		new_val.c = 1;
90 		old_val.c = 0;
91 
92 		expect = old_val.value;
93 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
94 	} else {
95 		struct bsca_block *sca = vcpu->kvm->arch.sca;
96 		union bsca_sigp_ctrl *sigp_ctrl =
97 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
98 		union bsca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
99 
100 		new_val.scn = src_id;
101 		new_val.c = 1;
102 		old_val.c = 0;
103 
104 		expect = old_val.value;
105 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
106 	}
107 	read_unlock(&vcpu->kvm->arch.sca_lock);
108 
109 	if (rc != expect) {
110 		/* another external call is pending */
111 		return -EBUSY;
112 	}
113 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
114 	return 0;
115 }
116 
117 static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
118 {
119 	int rc, expect;
120 
121 	if (!kvm_s390_use_sca_entries())
122 		return;
123 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
124 	read_lock(&vcpu->kvm->arch.sca_lock);
125 	if (vcpu->kvm->arch.use_esca) {
126 		struct esca_block *sca = vcpu->kvm->arch.sca;
127 		union esca_sigp_ctrl *sigp_ctrl =
128 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
129 		union esca_sigp_ctrl old = *sigp_ctrl;
130 
131 		expect = old.value;
132 		rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
133 	} else {
134 		struct bsca_block *sca = vcpu->kvm->arch.sca;
135 		union bsca_sigp_ctrl *sigp_ctrl =
136 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
137 		union bsca_sigp_ctrl old = *sigp_ctrl;
138 
139 		expect = old.value;
140 		rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
141 	}
142 	read_unlock(&vcpu->kvm->arch.sca_lock);
143 	WARN_ON(rc != expect); /* cannot clear? */
144 }
145 
146 int psw_extint_disabled(struct kvm_vcpu *vcpu)
147 {
148 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
149 }
150 
151 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
152 {
153 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
154 }
155 
156 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
157 {
158 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
159 }
160 
161 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
162 {
163 	return psw_extint_disabled(vcpu) &&
164 	       psw_ioint_disabled(vcpu) &&
165 	       psw_mchk_disabled(vcpu);
166 }
167 
168 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
169 {
170 	if (psw_extint_disabled(vcpu) ||
171 	    !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
172 		return 0;
173 	if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
174 		/* No timer interrupts when single stepping */
175 		return 0;
176 	return 1;
177 }
178 
179 static int ckc_irq_pending(struct kvm_vcpu *vcpu)
180 {
181 	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
182 	const u64 ckc = vcpu->arch.sie_block->ckc;
183 
184 	if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
185 		if ((s64)ckc >= (s64)now)
186 			return 0;
187 	} else if (ckc >= now) {
188 		return 0;
189 	}
190 	return ckc_interrupts_enabled(vcpu);
191 }
192 
193 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
194 {
195 	return !psw_extint_disabled(vcpu) &&
196 	       (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK);
197 }
198 
199 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
200 {
201 	if (!cpu_timer_interrupts_enabled(vcpu))
202 		return 0;
203 	return kvm_s390_get_cpu_timer(vcpu) >> 63;
204 }
205 
206 static uint64_t isc_to_isc_bits(int isc)
207 {
208 	return (0x80 >> isc) << 24;
209 }
210 
211 static inline u32 isc_to_int_word(u8 isc)
212 {
213 	return ((u32)isc << 27) | 0x80000000;
214 }
215 
216 static inline u8 int_word_to_isc(u32 int_word)
217 {
218 	return (int_word & 0x38000000) >> 27;
219 }
220 
221 /*
222  * To use atomic bitmap functions, we have to provide a bitmap address
223  * that is u64 aligned. However, the ipm might be u32 aligned.
224  * Therefore, we logically start the bitmap at the very beginning of the
225  * struct and fixup the bit number.
226  */
227 #define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE)
228 
229 /**
230  * gisa_set_iam - change the GISA interruption alert mask
231  *
232  * @gisa: gisa to operate on
233  * @iam: new IAM value to use
234  *
235  * Change the IAM atomically with the next alert address and the IPM
236  * of the GISA if the GISA is not part of the GIB alert list. All three
237  * fields are located in the first long word of the GISA.
238  *
239  * Returns: 0 on success
240  *          -EBUSY in case the gisa is part of the alert list
241  */
242 static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam)
243 {
244 	u64 word, _word;
245 
246 	do {
247 		word = READ_ONCE(gisa->u64.word[0]);
248 		if ((u64)gisa != word >> 32)
249 			return -EBUSY;
250 		_word = (word & ~0xffUL) | iam;
251 	} while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
252 
253 	return 0;
254 }
255 
256 /**
257  * gisa_clear_ipm - clear the GISA interruption pending mask
258  *
259  * @gisa: gisa to operate on
260  *
261  * Clear the IPM atomically with the next alert address and the IAM
262  * of the GISA unconditionally. All three fields are located in the
263  * first long word of the GISA.
264  */
265 static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa)
266 {
267 	u64 word, _word;
268 
269 	do {
270 		word = READ_ONCE(gisa->u64.word[0]);
271 		_word = word & ~(0xffUL << 24);
272 	} while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
273 }
274 
275 /**
276  * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM
277  *
278  * @gi: gisa interrupt struct to work on
279  *
280  * Atomically restores the interruption alert mask if none of the
281  * relevant ISCs are pending and return the IPM.
282  *
283  * Returns: the relevant pending ISCs
284  */
285 static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
286 {
287 	u8 pending_mask, alert_mask;
288 	u64 word, _word;
289 
290 	do {
291 		word = READ_ONCE(gi->origin->u64.word[0]);
292 		alert_mask = READ_ONCE(gi->alert.mask);
293 		pending_mask = (u8)(word >> 24) & alert_mask;
294 		if (pending_mask)
295 			return pending_mask;
296 		_word = (word & ~0xffUL) | alert_mask;
297 	} while (cmpxchg(&gi->origin->u64.word[0], word, _word) != word);
298 
299 	return 0;
300 }
301 
302 static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
303 {
304 	return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa;
305 }
306 
307 static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
308 {
309 	set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
310 }
311 
312 static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
313 {
314 	return READ_ONCE(gisa->ipm);
315 }
316 
317 static inline void gisa_clear_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
318 {
319 	clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
320 }
321 
322 static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
323 {
324 	return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
325 }
326 
327 static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
328 {
329 	unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs |
330 				vcpu->arch.local_int.pending_irqs;
331 
332 	pending &= ~vcpu->kvm->arch.float_int.masked_irqs;
333 	return pending;
334 }
335 
336 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
337 {
338 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
339 	unsigned long pending_mask;
340 
341 	pending_mask = pending_irqs_no_gisa(vcpu);
342 	if (gi->origin)
343 		pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
344 	return pending_mask;
345 }
346 
347 static inline int isc_to_irq_type(unsigned long isc)
348 {
349 	return IRQ_PEND_IO_ISC_0 - isc;
350 }
351 
352 static inline int irq_type_to_isc(unsigned long irq_type)
353 {
354 	return IRQ_PEND_IO_ISC_0 - irq_type;
355 }
356 
357 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
358 				   unsigned long active_mask)
359 {
360 	int i;
361 
362 	for (i = 0; i <= MAX_ISC; i++)
363 		if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
364 			active_mask &= ~(1UL << (isc_to_irq_type(i)));
365 
366 	return active_mask;
367 }
368 
369 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
370 {
371 	unsigned long active_mask;
372 
373 	active_mask = pending_irqs(vcpu);
374 	if (!active_mask)
375 		return 0;
376 
377 	if (psw_extint_disabled(vcpu))
378 		active_mask &= ~IRQ_PEND_EXT_MASK;
379 	if (psw_ioint_disabled(vcpu))
380 		active_mask &= ~IRQ_PEND_IO_MASK;
381 	else
382 		active_mask = disable_iscs(vcpu, active_mask);
383 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
384 		__clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
385 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
386 		__clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
387 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
388 		__clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
389 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
390 		__clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
391 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) {
392 		__clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
393 		__clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask);
394 	}
395 	if (psw_mchk_disabled(vcpu))
396 		active_mask &= ~IRQ_PEND_MCHK_MASK;
397 	/* PV guest cpus can have a single interruption injected at a time. */
398 	if (kvm_s390_pv_cpu_get_handle(vcpu) &&
399 	    vcpu->arch.sie_block->iictl != IICTL_CODE_NONE)
400 		active_mask &= ~(IRQ_PEND_EXT_II_MASK |
401 				 IRQ_PEND_IO_MASK |
402 				 IRQ_PEND_MCHK_MASK);
403 	/*
404 	 * Check both floating and local interrupt's cr14 because
405 	 * bit IRQ_PEND_MCHK_REP could be set in both cases.
406 	 */
407 	if (!(vcpu->arch.sie_block->gcr[14] &
408 	   (vcpu->kvm->arch.float_int.mchk.cr14 |
409 	   vcpu->arch.local_int.irq.mchk.cr14)))
410 		__clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
411 
412 	/*
413 	 * STOP irqs will never be actively delivered. They are triggered via
414 	 * intercept requests and cleared when the stop intercept is performed.
415 	 */
416 	__clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
417 
418 	return active_mask;
419 }
420 
421 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
422 {
423 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
424 	set_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
425 }
426 
427 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
428 {
429 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
430 	clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
431 }
432 
433 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
434 {
435 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
436 				      CPUSTAT_STOP_INT);
437 	vcpu->arch.sie_block->lctl = 0x0000;
438 	vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
439 
440 	if (guestdbg_enabled(vcpu)) {
441 		vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
442 					       LCTL_CR10 | LCTL_CR11);
443 		vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
444 	}
445 }
446 
447 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
448 {
449 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
450 		return;
451 	if (psw_ioint_disabled(vcpu))
452 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
453 	else
454 		vcpu->arch.sie_block->lctl |= LCTL_CR6;
455 }
456 
457 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
458 {
459 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
460 		return;
461 	if (psw_extint_disabled(vcpu))
462 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
463 	else
464 		vcpu->arch.sie_block->lctl |= LCTL_CR0;
465 }
466 
467 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
468 {
469 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
470 		return;
471 	if (psw_mchk_disabled(vcpu))
472 		vcpu->arch.sie_block->ictl |= ICTL_LPSW;
473 	else
474 		vcpu->arch.sie_block->lctl |= LCTL_CR14;
475 }
476 
477 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
478 {
479 	if (kvm_s390_is_stop_irq_pending(vcpu))
480 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
481 }
482 
483 /* Set interception request for non-deliverable interrupts */
484 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
485 {
486 	set_intercept_indicators_io(vcpu);
487 	set_intercept_indicators_ext(vcpu);
488 	set_intercept_indicators_mchk(vcpu);
489 	set_intercept_indicators_stop(vcpu);
490 }
491 
492 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
493 {
494 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
495 	int rc = 0;
496 
497 	vcpu->stat.deliver_cputm++;
498 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
499 					 0, 0);
500 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
501 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
502 		vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER;
503 	} else {
504 		rc  = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
505 				   (u16 *)__LC_EXT_INT_CODE);
506 		rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
507 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
508 				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
509 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
510 				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
511 	}
512 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
513 	return rc ? -EFAULT : 0;
514 }
515 
516 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
517 {
518 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
519 	int rc = 0;
520 
521 	vcpu->stat.deliver_ckc++;
522 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
523 					 0, 0);
524 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
525 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
526 		vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP;
527 	} else {
528 		rc  = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
529 				   (u16 __user *)__LC_EXT_INT_CODE);
530 		rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
531 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
532 				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
533 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
534 				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
535 	}
536 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
537 	return rc ? -EFAULT : 0;
538 }
539 
540 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
541 {
542 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
543 	struct kvm_s390_ext_info ext;
544 	int rc;
545 
546 	spin_lock(&li->lock);
547 	ext = li->irq.ext;
548 	clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
549 	li->irq.ext.ext_params2 = 0;
550 	spin_unlock(&li->lock);
551 
552 	VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
553 		   ext.ext_params2);
554 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
555 					 KVM_S390_INT_PFAULT_INIT,
556 					 0, ext.ext_params2);
557 
558 	rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
559 	rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
560 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
561 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
562 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
563 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
564 	rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
565 	return rc ? -EFAULT : 0;
566 }
567 
568 static int __write_machine_check(struct kvm_vcpu *vcpu,
569 				 struct kvm_s390_mchk_info *mchk)
570 {
571 	unsigned long ext_sa_addr;
572 	unsigned long lc;
573 	freg_t fprs[NUM_FPRS];
574 	union mci mci;
575 	int rc;
576 
577 	/*
578 	 * All other possible payload for a machine check (e.g. the register
579 	 * contents in the save area) will be handled by the ultravisor, as
580 	 * the hypervisor does not not have the needed information for
581 	 * protected guests.
582 	 */
583 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
584 		vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK;
585 		vcpu->arch.sie_block->mcic = mchk->mcic;
586 		vcpu->arch.sie_block->faddr = mchk->failing_storage_address;
587 		vcpu->arch.sie_block->edc = mchk->ext_damage_code;
588 		return 0;
589 	}
590 
591 	mci.val = mchk->mcic;
592 	/* take care of lazy register loading */
593 	save_fpu_regs();
594 	save_access_regs(vcpu->run->s.regs.acrs);
595 	if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
596 		save_gs_cb(current->thread.gs_cb);
597 
598 	/* Extended save area */
599 	rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
600 			   sizeof(unsigned long));
601 	/* Only bits 0 through 63-LC are used for address formation */
602 	lc = ext_sa_addr & MCESA_LC_MASK;
603 	if (test_kvm_facility(vcpu->kvm, 133)) {
604 		switch (lc) {
605 		case 0:
606 		case 10:
607 			ext_sa_addr &= ~0x3ffUL;
608 			break;
609 		case 11:
610 			ext_sa_addr &= ~0x7ffUL;
611 			break;
612 		case 12:
613 			ext_sa_addr &= ~0xfffUL;
614 			break;
615 		default:
616 			ext_sa_addr = 0;
617 			break;
618 		}
619 	} else {
620 		ext_sa_addr &= ~0x3ffUL;
621 	}
622 
623 	if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
624 		if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
625 				    512))
626 			mci.vr = 0;
627 	} else {
628 		mci.vr = 0;
629 	}
630 	if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
631 	    && (lc == 11 || lc == 12)) {
632 		if (write_guest_abs(vcpu, ext_sa_addr + 1024,
633 				    &vcpu->run->s.regs.gscb, 32))
634 			mci.gs = 0;
635 	} else {
636 		mci.gs = 0;
637 	}
638 
639 	/* General interruption information */
640 	rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
641 	rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
642 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
643 	rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
644 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
645 	rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
646 
647 	/* Register-save areas */
648 	if (MACHINE_HAS_VX) {
649 		convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
650 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
651 	} else {
652 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
653 				     vcpu->run->s.regs.fprs, 128);
654 	}
655 	rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
656 			     vcpu->run->s.regs.gprs, 128);
657 	rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
658 			   (u32 __user *) __LC_FP_CREG_SAVE_AREA);
659 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
660 			   (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
661 	rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
662 			   (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
663 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
664 			   (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
665 	rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
666 			     &vcpu->run->s.regs.acrs, 64);
667 	rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
668 			     &vcpu->arch.sie_block->gcr, 128);
669 
670 	/* Extended interruption information */
671 	rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
672 			   (u32 __user *) __LC_EXT_DAMAGE_CODE);
673 	rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
674 			   (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
675 	rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
676 			     sizeof(mchk->fixed_logout));
677 	return rc ? -EFAULT : 0;
678 }
679 
680 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
681 {
682 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
683 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
684 	struct kvm_s390_mchk_info mchk = {};
685 	int deliver = 0;
686 	int rc = 0;
687 
688 	spin_lock(&fi->lock);
689 	spin_lock(&li->lock);
690 	if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
691 	    test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
692 		/*
693 		 * If there was an exigent machine check pending, then any
694 		 * repressible machine checks that might have been pending
695 		 * are indicated along with it, so always clear bits for
696 		 * repressible and exigent interrupts
697 		 */
698 		mchk = li->irq.mchk;
699 		clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
700 		clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
701 		memset(&li->irq.mchk, 0, sizeof(mchk));
702 		deliver = 1;
703 	}
704 	/*
705 	 * We indicate floating repressible conditions along with
706 	 * other pending conditions. Channel Report Pending and Channel
707 	 * Subsystem damage are the only two and are indicated by
708 	 * bits in mcic and masked in cr14.
709 	 */
710 	if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
711 		mchk.mcic |= fi->mchk.mcic;
712 		mchk.cr14 |= fi->mchk.cr14;
713 		memset(&fi->mchk, 0, sizeof(mchk));
714 		deliver = 1;
715 	}
716 	spin_unlock(&li->lock);
717 	spin_unlock(&fi->lock);
718 
719 	if (deliver) {
720 		VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
721 			   mchk.mcic);
722 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
723 						 KVM_S390_MCHK,
724 						 mchk.cr14, mchk.mcic);
725 		vcpu->stat.deliver_machine_check++;
726 		rc = __write_machine_check(vcpu, &mchk);
727 	}
728 	return rc;
729 }
730 
731 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
732 {
733 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
734 	int rc = 0;
735 
736 	VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
737 	vcpu->stat.deliver_restart_signal++;
738 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
739 
740 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
741 		vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART;
742 	} else {
743 		rc  = write_guest_lc(vcpu,
744 				     offsetof(struct lowcore, restart_old_psw),
745 				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
746 		rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
747 				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
748 	}
749 	clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
750 	return rc ? -EFAULT : 0;
751 }
752 
753 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
754 {
755 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
756 	struct kvm_s390_prefix_info prefix;
757 
758 	spin_lock(&li->lock);
759 	prefix = li->irq.prefix;
760 	li->irq.prefix.address = 0;
761 	clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
762 	spin_unlock(&li->lock);
763 
764 	vcpu->stat.deliver_prefix_signal++;
765 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
766 					 KVM_S390_SIGP_SET_PREFIX,
767 					 prefix.address, 0);
768 
769 	kvm_s390_set_prefix(vcpu, prefix.address);
770 	return 0;
771 }
772 
773 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
774 {
775 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
776 	int rc;
777 	int cpu_addr;
778 
779 	spin_lock(&li->lock);
780 	cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
781 	clear_bit(cpu_addr, li->sigp_emerg_pending);
782 	if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
783 		clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
784 	spin_unlock(&li->lock);
785 
786 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
787 	vcpu->stat.deliver_emergency_signal++;
788 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
789 					 cpu_addr, 0);
790 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
791 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
792 		vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG;
793 		vcpu->arch.sie_block->extcpuaddr = cpu_addr;
794 		return 0;
795 	}
796 
797 	rc  = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
798 			   (u16 *)__LC_EXT_INT_CODE);
799 	rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
800 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
801 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
802 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
803 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
804 	return rc ? -EFAULT : 0;
805 }
806 
807 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
808 {
809 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
810 	struct kvm_s390_extcall_info extcall;
811 	int rc;
812 
813 	spin_lock(&li->lock);
814 	extcall = li->irq.extcall;
815 	li->irq.extcall.code = 0;
816 	clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
817 	spin_unlock(&li->lock);
818 
819 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
820 	vcpu->stat.deliver_external_call++;
821 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
822 					 KVM_S390_INT_EXTERNAL_CALL,
823 					 extcall.code, 0);
824 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
825 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
826 		vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL;
827 		vcpu->arch.sie_block->extcpuaddr = extcall.code;
828 		return 0;
829 	}
830 
831 	rc  = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
832 			   (u16 *)__LC_EXT_INT_CODE);
833 	rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
834 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
835 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
836 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
837 			    sizeof(psw_t));
838 	return rc ? -EFAULT : 0;
839 }
840 
841 static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code)
842 {
843 	switch (code) {
844 	case PGM_SPECIFICATION:
845 		vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION;
846 		break;
847 	case PGM_OPERAND:
848 		vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND;
849 		break;
850 	default:
851 		return -EINVAL;
852 	}
853 	return 0;
854 }
855 
856 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
857 {
858 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
859 	struct kvm_s390_pgm_info pgm_info;
860 	int rc = 0, nullifying = false;
861 	u16 ilen;
862 
863 	spin_lock(&li->lock);
864 	pgm_info = li->irq.pgm;
865 	clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
866 	memset(&li->irq.pgm, 0, sizeof(pgm_info));
867 	spin_unlock(&li->lock);
868 
869 	ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
870 	VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
871 		   pgm_info.code, ilen);
872 	vcpu->stat.deliver_program++;
873 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
874 					 pgm_info.code, 0);
875 
876 	/* PER is handled by the ultravisor */
877 	if (kvm_s390_pv_cpu_is_protected(vcpu))
878 		return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER);
879 
880 	switch (pgm_info.code & ~PGM_PER) {
881 	case PGM_AFX_TRANSLATION:
882 	case PGM_ASX_TRANSLATION:
883 	case PGM_EX_TRANSLATION:
884 	case PGM_LFX_TRANSLATION:
885 	case PGM_LSTE_SEQUENCE:
886 	case PGM_LSX_TRANSLATION:
887 	case PGM_LX_TRANSLATION:
888 	case PGM_PRIMARY_AUTHORITY:
889 	case PGM_SECONDARY_AUTHORITY:
890 		nullifying = true;
891 		fallthrough;
892 	case PGM_SPACE_SWITCH:
893 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
894 				  (u64 *)__LC_TRANS_EXC_CODE);
895 		break;
896 	case PGM_ALEN_TRANSLATION:
897 	case PGM_ALE_SEQUENCE:
898 	case PGM_ASTE_INSTANCE:
899 	case PGM_ASTE_SEQUENCE:
900 	case PGM_ASTE_VALIDITY:
901 	case PGM_EXTENDED_AUTHORITY:
902 		rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
903 				  (u8 *)__LC_EXC_ACCESS_ID);
904 		nullifying = true;
905 		break;
906 	case PGM_ASCE_TYPE:
907 	case PGM_PAGE_TRANSLATION:
908 	case PGM_REGION_FIRST_TRANS:
909 	case PGM_REGION_SECOND_TRANS:
910 	case PGM_REGION_THIRD_TRANS:
911 	case PGM_SEGMENT_TRANSLATION:
912 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
913 				  (u64 *)__LC_TRANS_EXC_CODE);
914 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
915 				   (u8 *)__LC_EXC_ACCESS_ID);
916 		rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
917 				   (u8 *)__LC_OP_ACCESS_ID);
918 		nullifying = true;
919 		break;
920 	case PGM_MONITOR:
921 		rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
922 				  (u16 *)__LC_MON_CLASS_NR);
923 		rc |= put_guest_lc(vcpu, pgm_info.mon_code,
924 				   (u64 *)__LC_MON_CODE);
925 		break;
926 	case PGM_VECTOR_PROCESSING:
927 	case PGM_DATA:
928 		rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
929 				  (u32 *)__LC_DATA_EXC_CODE);
930 		break;
931 	case PGM_PROTECTION:
932 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
933 				  (u64 *)__LC_TRANS_EXC_CODE);
934 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
935 				   (u8 *)__LC_EXC_ACCESS_ID);
936 		break;
937 	case PGM_STACK_FULL:
938 	case PGM_STACK_EMPTY:
939 	case PGM_STACK_SPECIFICATION:
940 	case PGM_STACK_TYPE:
941 	case PGM_STACK_OPERATION:
942 	case PGM_TRACE_TABEL:
943 	case PGM_CRYPTO_OPERATION:
944 		nullifying = true;
945 		break;
946 	}
947 
948 	if (pgm_info.code & PGM_PER) {
949 		rc |= put_guest_lc(vcpu, pgm_info.per_code,
950 				   (u8 *) __LC_PER_CODE);
951 		rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
952 				   (u8 *)__LC_PER_ATMID);
953 		rc |= put_guest_lc(vcpu, pgm_info.per_address,
954 				   (u64 *) __LC_PER_ADDRESS);
955 		rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
956 				   (u8 *) __LC_PER_ACCESS_ID);
957 	}
958 
959 	if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
960 		kvm_s390_rewind_psw(vcpu, ilen);
961 
962 	/* bit 1+2 of the target are the ilc, so we can directly use ilen */
963 	rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
964 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
965 				 (u64 *) __LC_PGM_LAST_BREAK);
966 	rc |= put_guest_lc(vcpu, pgm_info.code,
967 			   (u16 *)__LC_PGM_INT_CODE);
968 	rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
969 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
970 	rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
971 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
972 	return rc ? -EFAULT : 0;
973 }
974 
975 #define SCCB_MASK 0xFFFFFFF8
976 #define SCCB_EVENT_PENDING 0x3
977 
978 static int write_sclp(struct kvm_vcpu *vcpu, u32 parm)
979 {
980 	int rc;
981 
982 	if (kvm_s390_pv_cpu_get_handle(vcpu)) {
983 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
984 		vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG;
985 		vcpu->arch.sie_block->eiparams = parm;
986 		return 0;
987 	}
988 
989 	rc  = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
990 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
991 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
992 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
993 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
994 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
995 	rc |= put_guest_lc(vcpu, parm,
996 			   (u32 *)__LC_EXT_PARAMS);
997 
998 	return rc ? -EFAULT : 0;
999 }
1000 
1001 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
1002 {
1003 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1004 	struct kvm_s390_ext_info ext;
1005 
1006 	spin_lock(&fi->lock);
1007 	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) ||
1008 	    !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
1009 		spin_unlock(&fi->lock);
1010 		return 0;
1011 	}
1012 	ext = fi->srv_signal;
1013 	memset(&fi->srv_signal, 0, sizeof(ext));
1014 	clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1015 	clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1016 	if (kvm_s390_pv_cpu_is_protected(vcpu))
1017 		set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs);
1018 	spin_unlock(&fi->lock);
1019 
1020 	VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
1021 		   ext.ext_params);
1022 	vcpu->stat.deliver_service_signal++;
1023 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1024 					 ext.ext_params, 0);
1025 
1026 	return write_sclp(vcpu, ext.ext_params);
1027 }
1028 
1029 static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)
1030 {
1031 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1032 	struct kvm_s390_ext_info ext;
1033 
1034 	spin_lock(&fi->lock);
1035 	if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) {
1036 		spin_unlock(&fi->lock);
1037 		return 0;
1038 	}
1039 	ext = fi->srv_signal;
1040 	/* only clear the event bit */
1041 	fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING;
1042 	clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1043 	spin_unlock(&fi->lock);
1044 
1045 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event");
1046 	vcpu->stat.deliver_service_signal++;
1047 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1048 					 ext.ext_params, 0);
1049 
1050 	return write_sclp(vcpu, SCCB_EVENT_PENDING);
1051 }
1052 
1053 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
1054 {
1055 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1056 	struct kvm_s390_interrupt_info *inti;
1057 	int rc = 0;
1058 
1059 	spin_lock(&fi->lock);
1060 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
1061 					struct kvm_s390_interrupt_info,
1062 					list);
1063 	if (inti) {
1064 		list_del(&inti->list);
1065 		fi->counters[FIRQ_CNTR_PFAULT] -= 1;
1066 	}
1067 	if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
1068 		clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1069 	spin_unlock(&fi->lock);
1070 
1071 	if (inti) {
1072 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1073 						 KVM_S390_INT_PFAULT_DONE, 0,
1074 						 inti->ext.ext_params2);
1075 		VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
1076 			   inti->ext.ext_params2);
1077 
1078 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1079 				(u16 *)__LC_EXT_INT_CODE);
1080 		rc |= put_guest_lc(vcpu, PFAULT_DONE,
1081 				(u16 *)__LC_EXT_CPU_ADDR);
1082 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1083 				&vcpu->arch.sie_block->gpsw,
1084 				sizeof(psw_t));
1085 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1086 				&vcpu->arch.sie_block->gpsw,
1087 				sizeof(psw_t));
1088 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1089 				(u64 *)__LC_EXT_PARAMS2);
1090 		kfree(inti);
1091 	}
1092 	return rc ? -EFAULT : 0;
1093 }
1094 
1095 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
1096 {
1097 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1098 	struct kvm_s390_interrupt_info *inti;
1099 	int rc = 0;
1100 
1101 	spin_lock(&fi->lock);
1102 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
1103 					struct kvm_s390_interrupt_info,
1104 					list);
1105 	if (inti) {
1106 		VCPU_EVENT(vcpu, 4,
1107 			   "deliver: virtio parm: 0x%x,parm64: 0x%llx",
1108 			   inti->ext.ext_params, inti->ext.ext_params2);
1109 		vcpu->stat.deliver_virtio++;
1110 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1111 				inti->type,
1112 				inti->ext.ext_params,
1113 				inti->ext.ext_params2);
1114 		list_del(&inti->list);
1115 		fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1116 	}
1117 	if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1118 		clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1119 	spin_unlock(&fi->lock);
1120 
1121 	if (inti) {
1122 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1123 				(u16 *)__LC_EXT_INT_CODE);
1124 		rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1125 				(u16 *)__LC_EXT_CPU_ADDR);
1126 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1127 				&vcpu->arch.sie_block->gpsw,
1128 				sizeof(psw_t));
1129 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1130 				&vcpu->arch.sie_block->gpsw,
1131 				sizeof(psw_t));
1132 		rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1133 				(u32 *)__LC_EXT_PARAMS);
1134 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1135 				(u64 *)__LC_EXT_PARAMS2);
1136 		kfree(inti);
1137 	}
1138 	return rc ? -EFAULT : 0;
1139 }
1140 
1141 static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1142 {
1143 	int rc;
1144 
1145 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
1146 		vcpu->arch.sie_block->iictl = IICTL_CODE_IO;
1147 		vcpu->arch.sie_block->subchannel_id = io->subchannel_id;
1148 		vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr;
1149 		vcpu->arch.sie_block->io_int_parm = io->io_int_parm;
1150 		vcpu->arch.sie_block->io_int_word = io->io_int_word;
1151 		return 0;
1152 	}
1153 
1154 	rc  = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1155 	rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1156 	rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1157 	rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1158 	rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1159 			     &vcpu->arch.sie_block->gpsw,
1160 			     sizeof(psw_t));
1161 	rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1162 			    &vcpu->arch.sie_block->gpsw,
1163 			    sizeof(psw_t));
1164 	return rc ? -EFAULT : 0;
1165 }
1166 
1167 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1168 				     unsigned long irq_type)
1169 {
1170 	struct list_head *isc_list;
1171 	struct kvm_s390_float_interrupt *fi;
1172 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1173 	struct kvm_s390_interrupt_info *inti = NULL;
1174 	struct kvm_s390_io_info io;
1175 	u32 isc;
1176 	int rc = 0;
1177 
1178 	fi = &vcpu->kvm->arch.float_int;
1179 
1180 	spin_lock(&fi->lock);
1181 	isc = irq_type_to_isc(irq_type);
1182 	isc_list = &fi->lists[isc];
1183 	inti = list_first_entry_or_null(isc_list,
1184 					struct kvm_s390_interrupt_info,
1185 					list);
1186 	if (inti) {
1187 		if (inti->type & KVM_S390_INT_IO_AI_MASK)
1188 			VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1189 		else
1190 			VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1191 			inti->io.subchannel_id >> 8,
1192 			inti->io.subchannel_id >> 1 & 0x3,
1193 			inti->io.subchannel_nr);
1194 
1195 		vcpu->stat.deliver_io++;
1196 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1197 				inti->type,
1198 				((__u32)inti->io.subchannel_id << 16) |
1199 				inti->io.subchannel_nr,
1200 				((__u64)inti->io.io_int_parm << 32) |
1201 				inti->io.io_int_word);
1202 		list_del(&inti->list);
1203 		fi->counters[FIRQ_CNTR_IO] -= 1;
1204 	}
1205 	if (list_empty(isc_list))
1206 		clear_bit(irq_type, &fi->pending_irqs);
1207 	spin_unlock(&fi->lock);
1208 
1209 	if (inti) {
1210 		rc = __do_deliver_io(vcpu, &(inti->io));
1211 		kfree(inti);
1212 		goto out;
1213 	}
1214 
1215 	if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1216 		/*
1217 		 * in case an adapter interrupt was not delivered
1218 		 * in SIE context KVM will handle the delivery
1219 		 */
1220 		VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1221 		memset(&io, 0, sizeof(io));
1222 		io.io_int_word = isc_to_int_word(isc);
1223 		vcpu->stat.deliver_io++;
1224 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1225 			KVM_S390_INT_IO(1, 0, 0, 0),
1226 			((__u32)io.subchannel_id << 16) |
1227 			io.subchannel_nr,
1228 			((__u64)io.io_int_parm << 32) |
1229 			io.io_int_word);
1230 		rc = __do_deliver_io(vcpu, &io);
1231 	}
1232 out:
1233 	return rc;
1234 }
1235 
1236 /* Check whether an external call is pending (deliverable or not) */
1237 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1238 {
1239 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1240 
1241 	if (!sclp.has_sigpif)
1242 		return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1243 
1244 	return sca_ext_call_pending(vcpu, NULL);
1245 }
1246 
1247 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1248 {
1249 	if (deliverable_irqs(vcpu))
1250 		return 1;
1251 
1252 	if (kvm_cpu_has_pending_timer(vcpu))
1253 		return 1;
1254 
1255 	/* external call pending and deliverable */
1256 	if (kvm_s390_ext_call_pending(vcpu) &&
1257 	    !psw_extint_disabled(vcpu) &&
1258 	    (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1259 		return 1;
1260 
1261 	if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1262 		return 1;
1263 	return 0;
1264 }
1265 
1266 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1267 {
1268 	return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1269 }
1270 
1271 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1272 {
1273 	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1274 	const u64 ckc = vcpu->arch.sie_block->ckc;
1275 	u64 cputm, sltime = 0;
1276 
1277 	if (ckc_interrupts_enabled(vcpu)) {
1278 		if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1279 			if ((s64)now < (s64)ckc)
1280 				sltime = tod_to_ns((s64)ckc - (s64)now);
1281 		} else if (now < ckc) {
1282 			sltime = tod_to_ns(ckc - now);
1283 		}
1284 		/* already expired */
1285 		if (!sltime)
1286 			return 0;
1287 		if (cpu_timer_interrupts_enabled(vcpu)) {
1288 			cputm = kvm_s390_get_cpu_timer(vcpu);
1289 			/* already expired? */
1290 			if (cputm >> 63)
1291 				return 0;
1292 			return min_t(u64, sltime, tod_to_ns(cputm));
1293 		}
1294 	} else if (cpu_timer_interrupts_enabled(vcpu)) {
1295 		sltime = kvm_s390_get_cpu_timer(vcpu);
1296 		/* already expired? */
1297 		if (sltime >> 63)
1298 			return 0;
1299 	}
1300 	return sltime;
1301 }
1302 
1303 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1304 {
1305 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1306 	u64 sltime;
1307 
1308 	vcpu->stat.exit_wait_state++;
1309 
1310 	/* fast path */
1311 	if (kvm_arch_vcpu_runnable(vcpu))
1312 		return 0;
1313 
1314 	if (psw_interrupts_disabled(vcpu)) {
1315 		VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1316 		return -EOPNOTSUPP; /* disabled wait */
1317 	}
1318 
1319 	if (gi->origin &&
1320 	    (gisa_get_ipm_or_restore_iam(gi) &
1321 	     vcpu->arch.sie_block->gcr[6] >> 24))
1322 		return 0;
1323 
1324 	if (!ckc_interrupts_enabled(vcpu) &&
1325 	    !cpu_timer_interrupts_enabled(vcpu)) {
1326 		VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1327 		__set_cpu_idle(vcpu);
1328 		goto no_timer;
1329 	}
1330 
1331 	sltime = __calculate_sltime(vcpu);
1332 	if (!sltime)
1333 		return 0;
1334 
1335 	__set_cpu_idle(vcpu);
1336 	hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1337 	VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1338 no_timer:
1339 	kvm_vcpu_srcu_read_unlock(vcpu);
1340 	kvm_vcpu_halt(vcpu);
1341 	vcpu->valid_wakeup = false;
1342 	__unset_cpu_idle(vcpu);
1343 	kvm_vcpu_srcu_read_lock(vcpu);
1344 
1345 	hrtimer_cancel(&vcpu->arch.ckc_timer);
1346 	return 0;
1347 }
1348 
1349 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1350 {
1351 	vcpu->valid_wakeup = true;
1352 	kvm_vcpu_wake_up(vcpu);
1353 
1354 	/*
1355 	 * The VCPU might not be sleeping but rather executing VSIE. Let's
1356 	 * kick it, so it leaves the SIE to process the request.
1357 	 */
1358 	kvm_s390_vsie_kick(vcpu);
1359 }
1360 
1361 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1362 {
1363 	struct kvm_vcpu *vcpu;
1364 	u64 sltime;
1365 
1366 	vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1367 	sltime = __calculate_sltime(vcpu);
1368 
1369 	/*
1370 	 * If the monotonic clock runs faster than the tod clock we might be
1371 	 * woken up too early and have to go back to sleep to avoid deadlocks.
1372 	 */
1373 	if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1374 		return HRTIMER_RESTART;
1375 	kvm_s390_vcpu_wakeup(vcpu);
1376 	return HRTIMER_NORESTART;
1377 }
1378 
1379 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1380 {
1381 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1382 
1383 	spin_lock(&li->lock);
1384 	li->pending_irqs = 0;
1385 	bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1386 	memset(&li->irq, 0, sizeof(li->irq));
1387 	spin_unlock(&li->lock);
1388 
1389 	sca_clear_ext_call(vcpu);
1390 }
1391 
1392 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1393 {
1394 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1395 	int rc = 0;
1396 	unsigned long irq_type;
1397 	unsigned long irqs;
1398 
1399 	__reset_intercept_indicators(vcpu);
1400 
1401 	/* pending ckc conditions might have been invalidated */
1402 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1403 	if (ckc_irq_pending(vcpu))
1404 		set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1405 
1406 	/* pending cpu timer conditions might have been invalidated */
1407 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1408 	if (cpu_timer_irq_pending(vcpu))
1409 		set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1410 
1411 	while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1412 		/* bits are in the reverse order of interrupt priority */
1413 		irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1414 		switch (irq_type) {
1415 		case IRQ_PEND_IO_ISC_0:
1416 		case IRQ_PEND_IO_ISC_1:
1417 		case IRQ_PEND_IO_ISC_2:
1418 		case IRQ_PEND_IO_ISC_3:
1419 		case IRQ_PEND_IO_ISC_4:
1420 		case IRQ_PEND_IO_ISC_5:
1421 		case IRQ_PEND_IO_ISC_6:
1422 		case IRQ_PEND_IO_ISC_7:
1423 			rc = __deliver_io(vcpu, irq_type);
1424 			break;
1425 		case IRQ_PEND_MCHK_EX:
1426 		case IRQ_PEND_MCHK_REP:
1427 			rc = __deliver_machine_check(vcpu);
1428 			break;
1429 		case IRQ_PEND_PROG:
1430 			rc = __deliver_prog(vcpu);
1431 			break;
1432 		case IRQ_PEND_EXT_EMERGENCY:
1433 			rc = __deliver_emergency_signal(vcpu);
1434 			break;
1435 		case IRQ_PEND_EXT_EXTERNAL:
1436 			rc = __deliver_external_call(vcpu);
1437 			break;
1438 		case IRQ_PEND_EXT_CLOCK_COMP:
1439 			rc = __deliver_ckc(vcpu);
1440 			break;
1441 		case IRQ_PEND_EXT_CPU_TIMER:
1442 			rc = __deliver_cpu_timer(vcpu);
1443 			break;
1444 		case IRQ_PEND_RESTART:
1445 			rc = __deliver_restart(vcpu);
1446 			break;
1447 		case IRQ_PEND_SET_PREFIX:
1448 			rc = __deliver_set_prefix(vcpu);
1449 			break;
1450 		case IRQ_PEND_PFAULT_INIT:
1451 			rc = __deliver_pfault_init(vcpu);
1452 			break;
1453 		case IRQ_PEND_EXT_SERVICE:
1454 			rc = __deliver_service(vcpu);
1455 			break;
1456 		case IRQ_PEND_EXT_SERVICE_EV:
1457 			rc = __deliver_service_ev(vcpu);
1458 			break;
1459 		case IRQ_PEND_PFAULT_DONE:
1460 			rc = __deliver_pfault_done(vcpu);
1461 			break;
1462 		case IRQ_PEND_VIRTIO:
1463 			rc = __deliver_virtio(vcpu);
1464 			break;
1465 		default:
1466 			WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1467 			clear_bit(irq_type, &li->pending_irqs);
1468 		}
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 	/* find idle VCPUs first, then round robin */
1953 	sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1954 	if (sigcpu == online_vcpus) {
1955 		do {
1956 			sigcpu = kvm->arch.float_int.next_rr_cpu++;
1957 			kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1958 			/* avoid endless loops if all vcpus are stopped */
1959 			if (nr_tries++ >= online_vcpus)
1960 				return;
1961 		} while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1962 	}
1963 	dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1964 
1965 	/* make the VCPU drop out of the SIE, or wake it up if sleeping */
1966 	switch (type) {
1967 	case KVM_S390_MCHK:
1968 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1969 		break;
1970 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1971 		if (!(type & KVM_S390_INT_IO_AI_MASK &&
1972 		      kvm->arch.gisa_int.origin) ||
1973 		      kvm_s390_pv_cpu_get_handle(dst_vcpu))
1974 			kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1975 		break;
1976 	default:
1977 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1978 		break;
1979 	}
1980 	kvm_s390_vcpu_wakeup(dst_vcpu);
1981 }
1982 
1983 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1984 {
1985 	u64 type = READ_ONCE(inti->type);
1986 	int rc;
1987 
1988 	switch (type) {
1989 	case KVM_S390_MCHK:
1990 		rc = __inject_float_mchk(kvm, inti);
1991 		break;
1992 	case KVM_S390_INT_VIRTIO:
1993 		rc = __inject_virtio(kvm, inti);
1994 		break;
1995 	case KVM_S390_INT_SERVICE:
1996 		rc = __inject_service(kvm, inti);
1997 		break;
1998 	case KVM_S390_INT_PFAULT_DONE:
1999 		rc = __inject_pfault_done(kvm, inti);
2000 		break;
2001 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2002 		rc = __inject_io(kvm, inti);
2003 		break;
2004 	default:
2005 		rc = -EINVAL;
2006 	}
2007 	if (rc)
2008 		return rc;
2009 
2010 	__floating_irq_kick(kvm, type);
2011 	return 0;
2012 }
2013 
2014 int kvm_s390_inject_vm(struct kvm *kvm,
2015 		       struct kvm_s390_interrupt *s390int)
2016 {
2017 	struct kvm_s390_interrupt_info *inti;
2018 	int rc;
2019 
2020 	inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2021 	if (!inti)
2022 		return -ENOMEM;
2023 
2024 	inti->type = s390int->type;
2025 	switch (inti->type) {
2026 	case KVM_S390_INT_VIRTIO:
2027 		VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
2028 			 s390int->parm, s390int->parm64);
2029 		inti->ext.ext_params = s390int->parm;
2030 		inti->ext.ext_params2 = s390int->parm64;
2031 		break;
2032 	case KVM_S390_INT_SERVICE:
2033 		VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
2034 		inti->ext.ext_params = s390int->parm;
2035 		break;
2036 	case KVM_S390_INT_PFAULT_DONE:
2037 		inti->ext.ext_params2 = s390int->parm64;
2038 		break;
2039 	case KVM_S390_MCHK:
2040 		VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
2041 			 s390int->parm64);
2042 		inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
2043 		inti->mchk.mcic = s390int->parm64;
2044 		break;
2045 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2046 		inti->io.subchannel_id = s390int->parm >> 16;
2047 		inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
2048 		inti->io.io_int_parm = s390int->parm64 >> 32;
2049 		inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
2050 		break;
2051 	default:
2052 		kfree(inti);
2053 		return -EINVAL;
2054 	}
2055 	trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
2056 				 2);
2057 
2058 	rc = __inject_vm(kvm, inti);
2059 	if (rc)
2060 		kfree(inti);
2061 	return rc;
2062 }
2063 
2064 int kvm_s390_reinject_io_int(struct kvm *kvm,
2065 			      struct kvm_s390_interrupt_info *inti)
2066 {
2067 	return __inject_vm(kvm, inti);
2068 }
2069 
2070 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
2071 		       struct kvm_s390_irq *irq)
2072 {
2073 	irq->type = s390int->type;
2074 	switch (irq->type) {
2075 	case KVM_S390_PROGRAM_INT:
2076 		if (s390int->parm & 0xffff0000)
2077 			return -EINVAL;
2078 		irq->u.pgm.code = s390int->parm;
2079 		break;
2080 	case KVM_S390_SIGP_SET_PREFIX:
2081 		irq->u.prefix.address = s390int->parm;
2082 		break;
2083 	case KVM_S390_SIGP_STOP:
2084 		irq->u.stop.flags = s390int->parm;
2085 		break;
2086 	case KVM_S390_INT_EXTERNAL_CALL:
2087 		if (s390int->parm & 0xffff0000)
2088 			return -EINVAL;
2089 		irq->u.extcall.code = s390int->parm;
2090 		break;
2091 	case KVM_S390_INT_EMERGENCY:
2092 		if (s390int->parm & 0xffff0000)
2093 			return -EINVAL;
2094 		irq->u.emerg.code = s390int->parm;
2095 		break;
2096 	case KVM_S390_MCHK:
2097 		irq->u.mchk.mcic = s390int->parm64;
2098 		break;
2099 	case KVM_S390_INT_PFAULT_INIT:
2100 		irq->u.ext.ext_params = s390int->parm;
2101 		irq->u.ext.ext_params2 = s390int->parm64;
2102 		break;
2103 	case KVM_S390_RESTART:
2104 	case KVM_S390_INT_CLOCK_COMP:
2105 	case KVM_S390_INT_CPU_TIMER:
2106 		break;
2107 	default:
2108 		return -EINVAL;
2109 	}
2110 	return 0;
2111 }
2112 
2113 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
2114 {
2115 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2116 
2117 	return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2118 }
2119 
2120 int kvm_s390_is_restart_irq_pending(struct kvm_vcpu *vcpu)
2121 {
2122 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2123 
2124 	return test_bit(IRQ_PEND_RESTART, &li->pending_irqs);
2125 }
2126 
2127 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
2128 {
2129 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2130 
2131 	spin_lock(&li->lock);
2132 	li->irq.stop.flags = 0;
2133 	clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2134 	spin_unlock(&li->lock);
2135 }
2136 
2137 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2138 {
2139 	int rc;
2140 
2141 	switch (irq->type) {
2142 	case KVM_S390_PROGRAM_INT:
2143 		rc = __inject_prog(vcpu, irq);
2144 		break;
2145 	case KVM_S390_SIGP_SET_PREFIX:
2146 		rc = __inject_set_prefix(vcpu, irq);
2147 		break;
2148 	case KVM_S390_SIGP_STOP:
2149 		rc = __inject_sigp_stop(vcpu, irq);
2150 		break;
2151 	case KVM_S390_RESTART:
2152 		rc = __inject_sigp_restart(vcpu);
2153 		break;
2154 	case KVM_S390_INT_CLOCK_COMP:
2155 		rc = __inject_ckc(vcpu);
2156 		break;
2157 	case KVM_S390_INT_CPU_TIMER:
2158 		rc = __inject_cpu_timer(vcpu);
2159 		break;
2160 	case KVM_S390_INT_EXTERNAL_CALL:
2161 		rc = __inject_extcall(vcpu, irq);
2162 		break;
2163 	case KVM_S390_INT_EMERGENCY:
2164 		rc = __inject_sigp_emergency(vcpu, irq);
2165 		break;
2166 	case KVM_S390_MCHK:
2167 		rc = __inject_mchk(vcpu, irq);
2168 		break;
2169 	case KVM_S390_INT_PFAULT_INIT:
2170 		rc = __inject_pfault_init(vcpu, irq);
2171 		break;
2172 	case KVM_S390_INT_VIRTIO:
2173 	case KVM_S390_INT_SERVICE:
2174 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2175 	default:
2176 		rc = -EINVAL;
2177 	}
2178 
2179 	return rc;
2180 }
2181 
2182 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2183 {
2184 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2185 	int rc;
2186 
2187 	spin_lock(&li->lock);
2188 	rc = do_inject_vcpu(vcpu, irq);
2189 	spin_unlock(&li->lock);
2190 	if (!rc)
2191 		kvm_s390_vcpu_wakeup(vcpu);
2192 	return rc;
2193 }
2194 
2195 static inline void clear_irq_list(struct list_head *_list)
2196 {
2197 	struct kvm_s390_interrupt_info *inti, *n;
2198 
2199 	list_for_each_entry_safe(inti, n, _list, list) {
2200 		list_del(&inti->list);
2201 		kfree(inti);
2202 	}
2203 }
2204 
2205 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2206 		       struct kvm_s390_irq *irq)
2207 {
2208 	irq->type = inti->type;
2209 	switch (inti->type) {
2210 	case KVM_S390_INT_PFAULT_INIT:
2211 	case KVM_S390_INT_PFAULT_DONE:
2212 	case KVM_S390_INT_VIRTIO:
2213 		irq->u.ext = inti->ext;
2214 		break;
2215 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2216 		irq->u.io = inti->io;
2217 		break;
2218 	}
2219 }
2220 
2221 void kvm_s390_clear_float_irqs(struct kvm *kvm)
2222 {
2223 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2224 	int i;
2225 
2226 	mutex_lock(&kvm->lock);
2227 	if (!kvm_s390_pv_is_protected(kvm))
2228 		fi->masked_irqs = 0;
2229 	mutex_unlock(&kvm->lock);
2230 	spin_lock(&fi->lock);
2231 	fi->pending_irqs = 0;
2232 	memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2233 	memset(&fi->mchk, 0, sizeof(fi->mchk));
2234 	for (i = 0; i < FIRQ_LIST_COUNT; i++)
2235 		clear_irq_list(&fi->lists[i]);
2236 	for (i = 0; i < FIRQ_MAX_COUNT; i++)
2237 		fi->counters[i] = 0;
2238 	spin_unlock(&fi->lock);
2239 	kvm_s390_gisa_clear(kvm);
2240 };
2241 
2242 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2243 {
2244 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2245 	struct kvm_s390_interrupt_info *inti;
2246 	struct kvm_s390_float_interrupt *fi;
2247 	struct kvm_s390_irq *buf;
2248 	struct kvm_s390_irq *irq;
2249 	int max_irqs;
2250 	int ret = 0;
2251 	int n = 0;
2252 	int i;
2253 
2254 	if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2255 		return -EINVAL;
2256 
2257 	/*
2258 	 * We are already using -ENOMEM to signal
2259 	 * userspace it may retry with a bigger buffer,
2260 	 * so we need to use something else for this case
2261 	 */
2262 	buf = vzalloc(len);
2263 	if (!buf)
2264 		return -ENOBUFS;
2265 
2266 	max_irqs = len / sizeof(struct kvm_s390_irq);
2267 
2268 	if (gi->origin && gisa_get_ipm(gi->origin)) {
2269 		for (i = 0; i <= MAX_ISC; i++) {
2270 			if (n == max_irqs) {
2271 				/* signal userspace to try again */
2272 				ret = -ENOMEM;
2273 				goto out_nolock;
2274 			}
2275 			if (gisa_tac_ipm_gisc(gi->origin, i)) {
2276 				irq = (struct kvm_s390_irq *) &buf[n];
2277 				irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2278 				irq->u.io.io_int_word = isc_to_int_word(i);
2279 				n++;
2280 			}
2281 		}
2282 	}
2283 	fi = &kvm->arch.float_int;
2284 	spin_lock(&fi->lock);
2285 	for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2286 		list_for_each_entry(inti, &fi->lists[i], list) {
2287 			if (n == max_irqs) {
2288 				/* signal userspace to try again */
2289 				ret = -ENOMEM;
2290 				goto out;
2291 			}
2292 			inti_to_irq(inti, &buf[n]);
2293 			n++;
2294 		}
2295 	}
2296 	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) ||
2297 	    test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) {
2298 		if (n == max_irqs) {
2299 			/* signal userspace to try again */
2300 			ret = -ENOMEM;
2301 			goto out;
2302 		}
2303 		irq = (struct kvm_s390_irq *) &buf[n];
2304 		irq->type = KVM_S390_INT_SERVICE;
2305 		irq->u.ext = fi->srv_signal;
2306 		n++;
2307 	}
2308 	if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2309 		if (n == max_irqs) {
2310 				/* signal userspace to try again */
2311 				ret = -ENOMEM;
2312 				goto out;
2313 		}
2314 		irq = (struct kvm_s390_irq *) &buf[n];
2315 		irq->type = KVM_S390_MCHK;
2316 		irq->u.mchk = fi->mchk;
2317 		n++;
2318 }
2319 
2320 out:
2321 	spin_unlock(&fi->lock);
2322 out_nolock:
2323 	if (!ret && n > 0) {
2324 		if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2325 			ret = -EFAULT;
2326 	}
2327 	vfree(buf);
2328 
2329 	return ret < 0 ? ret : n;
2330 }
2331 
2332 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2333 {
2334 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2335 	struct kvm_s390_ais_all ais;
2336 
2337 	if (attr->attr < sizeof(ais))
2338 		return -EINVAL;
2339 
2340 	if (!test_kvm_facility(kvm, 72))
2341 		return -EOPNOTSUPP;
2342 
2343 	mutex_lock(&fi->ais_lock);
2344 	ais.simm = fi->simm;
2345 	ais.nimm = fi->nimm;
2346 	mutex_unlock(&fi->ais_lock);
2347 
2348 	if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2349 		return -EFAULT;
2350 
2351 	return 0;
2352 }
2353 
2354 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2355 {
2356 	int r;
2357 
2358 	switch (attr->group) {
2359 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2360 		r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2361 					  attr->attr);
2362 		break;
2363 	case KVM_DEV_FLIC_AISM_ALL:
2364 		r = flic_ais_mode_get_all(dev->kvm, attr);
2365 		break;
2366 	default:
2367 		r = -EINVAL;
2368 	}
2369 
2370 	return r;
2371 }
2372 
2373 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2374 				     u64 addr)
2375 {
2376 	struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2377 	void *target = NULL;
2378 	void __user *source;
2379 	u64 size;
2380 
2381 	if (get_user(inti->type, (u64 __user *)addr))
2382 		return -EFAULT;
2383 
2384 	switch (inti->type) {
2385 	case KVM_S390_INT_PFAULT_INIT:
2386 	case KVM_S390_INT_PFAULT_DONE:
2387 	case KVM_S390_INT_VIRTIO:
2388 	case KVM_S390_INT_SERVICE:
2389 		target = (void *) &inti->ext;
2390 		source = &uptr->u.ext;
2391 		size = sizeof(inti->ext);
2392 		break;
2393 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2394 		target = (void *) &inti->io;
2395 		source = &uptr->u.io;
2396 		size = sizeof(inti->io);
2397 		break;
2398 	case KVM_S390_MCHK:
2399 		target = (void *) &inti->mchk;
2400 		source = &uptr->u.mchk;
2401 		size = sizeof(inti->mchk);
2402 		break;
2403 	default:
2404 		return -EINVAL;
2405 	}
2406 
2407 	if (copy_from_user(target, source, size))
2408 		return -EFAULT;
2409 
2410 	return 0;
2411 }
2412 
2413 static int enqueue_floating_irq(struct kvm_device *dev,
2414 				struct kvm_device_attr *attr)
2415 {
2416 	struct kvm_s390_interrupt_info *inti = NULL;
2417 	int r = 0;
2418 	int len = attr->attr;
2419 
2420 	if (len % sizeof(struct kvm_s390_irq) != 0)
2421 		return -EINVAL;
2422 	else if (len > KVM_S390_FLIC_MAX_BUFFER)
2423 		return -EINVAL;
2424 
2425 	while (len >= sizeof(struct kvm_s390_irq)) {
2426 		inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2427 		if (!inti)
2428 			return -ENOMEM;
2429 
2430 		r = copy_irq_from_user(inti, attr->addr);
2431 		if (r) {
2432 			kfree(inti);
2433 			return r;
2434 		}
2435 		r = __inject_vm(dev->kvm, inti);
2436 		if (r) {
2437 			kfree(inti);
2438 			return r;
2439 		}
2440 		len -= sizeof(struct kvm_s390_irq);
2441 		attr->addr += sizeof(struct kvm_s390_irq);
2442 	}
2443 
2444 	return r;
2445 }
2446 
2447 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2448 {
2449 	if (id >= MAX_S390_IO_ADAPTERS)
2450 		return NULL;
2451 	id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2452 	return kvm->arch.adapters[id];
2453 }
2454 
2455 static int register_io_adapter(struct kvm_device *dev,
2456 			       struct kvm_device_attr *attr)
2457 {
2458 	struct s390_io_adapter *adapter;
2459 	struct kvm_s390_io_adapter adapter_info;
2460 
2461 	if (copy_from_user(&adapter_info,
2462 			   (void __user *)attr->addr, sizeof(adapter_info)))
2463 		return -EFAULT;
2464 
2465 	if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2466 		return -EINVAL;
2467 
2468 	adapter_info.id = array_index_nospec(adapter_info.id,
2469 					     MAX_S390_IO_ADAPTERS);
2470 
2471 	if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2472 		return -EINVAL;
2473 
2474 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL_ACCOUNT);
2475 	if (!adapter)
2476 		return -ENOMEM;
2477 
2478 	adapter->id = adapter_info.id;
2479 	adapter->isc = adapter_info.isc;
2480 	adapter->maskable = adapter_info.maskable;
2481 	adapter->masked = false;
2482 	adapter->swap = adapter_info.swap;
2483 	adapter->suppressible = (adapter_info.flags) &
2484 				KVM_S390_ADAPTER_SUPPRESSIBLE;
2485 	dev->kvm->arch.adapters[adapter->id] = adapter;
2486 
2487 	return 0;
2488 }
2489 
2490 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2491 {
2492 	int ret;
2493 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2494 
2495 	if (!adapter || !adapter->maskable)
2496 		return -EINVAL;
2497 	ret = adapter->masked;
2498 	adapter->masked = masked;
2499 	return ret;
2500 }
2501 
2502 void kvm_s390_destroy_adapters(struct kvm *kvm)
2503 {
2504 	int i;
2505 
2506 	for (i = 0; i < MAX_S390_IO_ADAPTERS; i++)
2507 		kfree(kvm->arch.adapters[i]);
2508 }
2509 
2510 static int modify_io_adapter(struct kvm_device *dev,
2511 			     struct kvm_device_attr *attr)
2512 {
2513 	struct kvm_s390_io_adapter_req req;
2514 	struct s390_io_adapter *adapter;
2515 	int ret;
2516 
2517 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2518 		return -EFAULT;
2519 
2520 	adapter = get_io_adapter(dev->kvm, req.id);
2521 	if (!adapter)
2522 		return -EINVAL;
2523 	switch (req.type) {
2524 	case KVM_S390_IO_ADAPTER_MASK:
2525 		ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2526 		if (ret > 0)
2527 			ret = 0;
2528 		break;
2529 	/*
2530 	 * The following operations are no longer needed and therefore no-ops.
2531 	 * The gpa to hva translation is done when an IRQ route is set up. The
2532 	 * set_irq code uses get_user_pages_remote() to do the actual write.
2533 	 */
2534 	case KVM_S390_IO_ADAPTER_MAP:
2535 	case KVM_S390_IO_ADAPTER_UNMAP:
2536 		ret = 0;
2537 		break;
2538 	default:
2539 		ret = -EINVAL;
2540 	}
2541 
2542 	return ret;
2543 }
2544 
2545 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2546 
2547 {
2548 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2549 	u32 schid;
2550 
2551 	if (attr->flags)
2552 		return -EINVAL;
2553 	if (attr->attr != sizeof(schid))
2554 		return -EINVAL;
2555 	if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2556 		return -EFAULT;
2557 	if (!schid)
2558 		return -EINVAL;
2559 	kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2560 	/*
2561 	 * If userspace is conforming to the architecture, we can have at most
2562 	 * one pending I/O interrupt per subchannel, so this is effectively a
2563 	 * clear all.
2564 	 */
2565 	return 0;
2566 }
2567 
2568 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2569 {
2570 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2571 	struct kvm_s390_ais_req req;
2572 	int ret = 0;
2573 
2574 	if (!test_kvm_facility(kvm, 72))
2575 		return -EOPNOTSUPP;
2576 
2577 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2578 		return -EFAULT;
2579 
2580 	if (req.isc > MAX_ISC)
2581 		return -EINVAL;
2582 
2583 	trace_kvm_s390_modify_ais_mode(req.isc,
2584 				       (fi->simm & AIS_MODE_MASK(req.isc)) ?
2585 				       (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2586 				       2 : KVM_S390_AIS_MODE_SINGLE :
2587 				       KVM_S390_AIS_MODE_ALL, req.mode);
2588 
2589 	mutex_lock(&fi->ais_lock);
2590 	switch (req.mode) {
2591 	case KVM_S390_AIS_MODE_ALL:
2592 		fi->simm &= ~AIS_MODE_MASK(req.isc);
2593 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2594 		break;
2595 	case KVM_S390_AIS_MODE_SINGLE:
2596 		fi->simm |= AIS_MODE_MASK(req.isc);
2597 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2598 		break;
2599 	default:
2600 		ret = -EINVAL;
2601 	}
2602 	mutex_unlock(&fi->ais_lock);
2603 
2604 	return ret;
2605 }
2606 
2607 static int kvm_s390_inject_airq(struct kvm *kvm,
2608 				struct s390_io_adapter *adapter)
2609 {
2610 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2611 	struct kvm_s390_interrupt s390int = {
2612 		.type = KVM_S390_INT_IO(1, 0, 0, 0),
2613 		.parm = 0,
2614 		.parm64 = isc_to_int_word(adapter->isc),
2615 	};
2616 	int ret = 0;
2617 
2618 	if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2619 		return kvm_s390_inject_vm(kvm, &s390int);
2620 
2621 	mutex_lock(&fi->ais_lock);
2622 	if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2623 		trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2624 		goto out;
2625 	}
2626 
2627 	ret = kvm_s390_inject_vm(kvm, &s390int);
2628 	if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2629 		fi->nimm |= AIS_MODE_MASK(adapter->isc);
2630 		trace_kvm_s390_modify_ais_mode(adapter->isc,
2631 					       KVM_S390_AIS_MODE_SINGLE, 2);
2632 	}
2633 out:
2634 	mutex_unlock(&fi->ais_lock);
2635 	return ret;
2636 }
2637 
2638 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2639 {
2640 	unsigned int id = attr->attr;
2641 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2642 
2643 	if (!adapter)
2644 		return -EINVAL;
2645 
2646 	return kvm_s390_inject_airq(kvm, adapter);
2647 }
2648 
2649 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2650 {
2651 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2652 	struct kvm_s390_ais_all ais;
2653 
2654 	if (!test_kvm_facility(kvm, 72))
2655 		return -EOPNOTSUPP;
2656 
2657 	if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2658 		return -EFAULT;
2659 
2660 	mutex_lock(&fi->ais_lock);
2661 	fi->simm = ais.simm;
2662 	fi->nimm = ais.nimm;
2663 	mutex_unlock(&fi->ais_lock);
2664 
2665 	return 0;
2666 }
2667 
2668 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2669 {
2670 	int r = 0;
2671 	unsigned long i;
2672 	struct kvm_vcpu *vcpu;
2673 
2674 	switch (attr->group) {
2675 	case KVM_DEV_FLIC_ENQUEUE:
2676 		r = enqueue_floating_irq(dev, attr);
2677 		break;
2678 	case KVM_DEV_FLIC_CLEAR_IRQS:
2679 		kvm_s390_clear_float_irqs(dev->kvm);
2680 		break;
2681 	case KVM_DEV_FLIC_APF_ENABLE:
2682 		dev->kvm->arch.gmap->pfault_enabled = 1;
2683 		break;
2684 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2685 		dev->kvm->arch.gmap->pfault_enabled = 0;
2686 		/*
2687 		 * Make sure no async faults are in transition when
2688 		 * clearing the queues. So we don't need to worry
2689 		 * about late coming workers.
2690 		 */
2691 		synchronize_srcu(&dev->kvm->srcu);
2692 		kvm_for_each_vcpu(i, vcpu, dev->kvm)
2693 			kvm_clear_async_pf_completion_queue(vcpu);
2694 		break;
2695 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2696 		r = register_io_adapter(dev, attr);
2697 		break;
2698 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2699 		r = modify_io_adapter(dev, attr);
2700 		break;
2701 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2702 		r = clear_io_irq(dev->kvm, attr);
2703 		break;
2704 	case KVM_DEV_FLIC_AISM:
2705 		r = modify_ais_mode(dev->kvm, attr);
2706 		break;
2707 	case KVM_DEV_FLIC_AIRQ_INJECT:
2708 		r = flic_inject_airq(dev->kvm, attr);
2709 		break;
2710 	case KVM_DEV_FLIC_AISM_ALL:
2711 		r = flic_ais_mode_set_all(dev->kvm, attr);
2712 		break;
2713 	default:
2714 		r = -EINVAL;
2715 	}
2716 
2717 	return r;
2718 }
2719 
2720 static int flic_has_attr(struct kvm_device *dev,
2721 			     struct kvm_device_attr *attr)
2722 {
2723 	switch (attr->group) {
2724 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2725 	case KVM_DEV_FLIC_ENQUEUE:
2726 	case KVM_DEV_FLIC_CLEAR_IRQS:
2727 	case KVM_DEV_FLIC_APF_ENABLE:
2728 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2729 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2730 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2731 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2732 	case KVM_DEV_FLIC_AISM:
2733 	case KVM_DEV_FLIC_AIRQ_INJECT:
2734 	case KVM_DEV_FLIC_AISM_ALL:
2735 		return 0;
2736 	}
2737 	return -ENXIO;
2738 }
2739 
2740 static int flic_create(struct kvm_device *dev, u32 type)
2741 {
2742 	if (!dev)
2743 		return -EINVAL;
2744 	if (dev->kvm->arch.flic)
2745 		return -EINVAL;
2746 	dev->kvm->arch.flic = dev;
2747 	return 0;
2748 }
2749 
2750 static void flic_destroy(struct kvm_device *dev)
2751 {
2752 	dev->kvm->arch.flic = NULL;
2753 	kfree(dev);
2754 }
2755 
2756 /* s390 floating irq controller (flic) */
2757 struct kvm_device_ops kvm_flic_ops = {
2758 	.name = "kvm-flic",
2759 	.get_attr = flic_get_attr,
2760 	.set_attr = flic_set_attr,
2761 	.has_attr = flic_has_attr,
2762 	.create = flic_create,
2763 	.destroy = flic_destroy,
2764 };
2765 
2766 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2767 {
2768 	unsigned long bit;
2769 
2770 	bit = bit_nr + (addr % PAGE_SIZE) * 8;
2771 
2772 	return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2773 }
2774 
2775 static struct page *get_map_page(struct kvm *kvm, u64 uaddr)
2776 {
2777 	struct page *page = NULL;
2778 
2779 	mmap_read_lock(kvm->mm);
2780 	get_user_pages_remote(kvm->mm, uaddr, 1, FOLL_WRITE,
2781 			      &page, NULL, NULL);
2782 	mmap_read_unlock(kvm->mm);
2783 	return page;
2784 }
2785 
2786 static int adapter_indicators_set(struct kvm *kvm,
2787 				  struct s390_io_adapter *adapter,
2788 				  struct kvm_s390_adapter_int *adapter_int)
2789 {
2790 	unsigned long bit;
2791 	int summary_set, idx;
2792 	struct page *ind_page, *summary_page;
2793 	void *map;
2794 
2795 	ind_page = get_map_page(kvm, adapter_int->ind_addr);
2796 	if (!ind_page)
2797 		return -1;
2798 	summary_page = get_map_page(kvm, adapter_int->summary_addr);
2799 	if (!summary_page) {
2800 		put_page(ind_page);
2801 		return -1;
2802 	}
2803 
2804 	idx = srcu_read_lock(&kvm->srcu);
2805 	map = page_address(ind_page);
2806 	bit = get_ind_bit(adapter_int->ind_addr,
2807 			  adapter_int->ind_offset, adapter->swap);
2808 	set_bit(bit, map);
2809 	mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT);
2810 	set_page_dirty_lock(ind_page);
2811 	map = page_address(summary_page);
2812 	bit = get_ind_bit(adapter_int->summary_addr,
2813 			  adapter_int->summary_offset, adapter->swap);
2814 	summary_set = test_and_set_bit(bit, map);
2815 	mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT);
2816 	set_page_dirty_lock(summary_page);
2817 	srcu_read_unlock(&kvm->srcu, idx);
2818 
2819 	put_page(ind_page);
2820 	put_page(summary_page);
2821 	return summary_set ? 0 : 1;
2822 }
2823 
2824 /*
2825  * < 0 - not injected due to error
2826  * = 0 - coalesced, summary indicator already active
2827  * > 0 - injected interrupt
2828  */
2829 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2830 			   struct kvm *kvm, int irq_source_id, int level,
2831 			   bool line_status)
2832 {
2833 	int ret;
2834 	struct s390_io_adapter *adapter;
2835 
2836 	/* We're only interested in the 0->1 transition. */
2837 	if (!level)
2838 		return 0;
2839 	adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2840 	if (!adapter)
2841 		return -1;
2842 	ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2843 	if ((ret > 0) && !adapter->masked) {
2844 		ret = kvm_s390_inject_airq(kvm, adapter);
2845 		if (ret == 0)
2846 			ret = 1;
2847 	}
2848 	return ret;
2849 }
2850 
2851 /*
2852  * Inject the machine check to the guest.
2853  */
2854 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2855 				     struct mcck_volatile_info *mcck_info)
2856 {
2857 	struct kvm_s390_interrupt_info inti;
2858 	struct kvm_s390_irq irq;
2859 	struct kvm_s390_mchk_info *mchk;
2860 	union mci mci;
2861 	__u64 cr14 = 0;         /* upper bits are not used */
2862 	int rc;
2863 
2864 	mci.val = mcck_info->mcic;
2865 	if (mci.sr)
2866 		cr14 |= CR14_RECOVERY_SUBMASK;
2867 	if (mci.dg)
2868 		cr14 |= CR14_DEGRADATION_SUBMASK;
2869 	if (mci.w)
2870 		cr14 |= CR14_WARNING_SUBMASK;
2871 
2872 	mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2873 	mchk->cr14 = cr14;
2874 	mchk->mcic = mcck_info->mcic;
2875 	mchk->ext_damage_code = mcck_info->ext_damage_code;
2876 	mchk->failing_storage_address = mcck_info->failing_storage_address;
2877 	if (mci.ck) {
2878 		/* Inject the floating machine check */
2879 		inti.type = KVM_S390_MCHK;
2880 		rc = __inject_vm(vcpu->kvm, &inti);
2881 	} else {
2882 		/* Inject the machine check to specified vcpu */
2883 		irq.type = KVM_S390_MCHK;
2884 		rc = kvm_s390_inject_vcpu(vcpu, &irq);
2885 	}
2886 	WARN_ON_ONCE(rc);
2887 }
2888 
2889 int kvm_set_routing_entry(struct kvm *kvm,
2890 			  struct kvm_kernel_irq_routing_entry *e,
2891 			  const struct kvm_irq_routing_entry *ue)
2892 {
2893 	u64 uaddr;
2894 
2895 	switch (ue->type) {
2896 	/* we store the userspace addresses instead of the guest addresses */
2897 	case KVM_IRQ_ROUTING_S390_ADAPTER:
2898 		e->set = set_adapter_int;
2899 		uaddr =  gmap_translate(kvm->arch.gmap, ue->u.adapter.summary_addr);
2900 		if (uaddr == -EFAULT)
2901 			return -EFAULT;
2902 		e->adapter.summary_addr = uaddr;
2903 		uaddr =  gmap_translate(kvm->arch.gmap, ue->u.adapter.ind_addr);
2904 		if (uaddr == -EFAULT)
2905 			return -EFAULT;
2906 		e->adapter.ind_addr = uaddr;
2907 		e->adapter.summary_offset = ue->u.adapter.summary_offset;
2908 		e->adapter.ind_offset = ue->u.adapter.ind_offset;
2909 		e->adapter.adapter_id = ue->u.adapter.adapter_id;
2910 		return 0;
2911 	default:
2912 		return -EINVAL;
2913 	}
2914 }
2915 
2916 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2917 		int irq_source_id, int level, bool line_status)
2918 {
2919 	return -EINVAL;
2920 }
2921 
2922 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2923 {
2924 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2925 	struct kvm_s390_irq *buf;
2926 	int r = 0;
2927 	int n;
2928 
2929 	buf = vmalloc(len);
2930 	if (!buf)
2931 		return -ENOMEM;
2932 
2933 	if (copy_from_user((void *) buf, irqstate, len)) {
2934 		r = -EFAULT;
2935 		goto out_free;
2936 	}
2937 
2938 	/*
2939 	 * Don't allow setting the interrupt state
2940 	 * when there are already interrupts pending
2941 	 */
2942 	spin_lock(&li->lock);
2943 	if (li->pending_irqs) {
2944 		r = -EBUSY;
2945 		goto out_unlock;
2946 	}
2947 
2948 	for (n = 0; n < len / sizeof(*buf); n++) {
2949 		r = do_inject_vcpu(vcpu, &buf[n]);
2950 		if (r)
2951 			break;
2952 	}
2953 
2954 out_unlock:
2955 	spin_unlock(&li->lock);
2956 out_free:
2957 	vfree(buf);
2958 
2959 	return r;
2960 }
2961 
2962 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2963 			    struct kvm_s390_irq *irq,
2964 			    unsigned long irq_type)
2965 {
2966 	switch (irq_type) {
2967 	case IRQ_PEND_MCHK_EX:
2968 	case IRQ_PEND_MCHK_REP:
2969 		irq->type = KVM_S390_MCHK;
2970 		irq->u.mchk = li->irq.mchk;
2971 		break;
2972 	case IRQ_PEND_PROG:
2973 		irq->type = KVM_S390_PROGRAM_INT;
2974 		irq->u.pgm = li->irq.pgm;
2975 		break;
2976 	case IRQ_PEND_PFAULT_INIT:
2977 		irq->type = KVM_S390_INT_PFAULT_INIT;
2978 		irq->u.ext = li->irq.ext;
2979 		break;
2980 	case IRQ_PEND_EXT_EXTERNAL:
2981 		irq->type = KVM_S390_INT_EXTERNAL_CALL;
2982 		irq->u.extcall = li->irq.extcall;
2983 		break;
2984 	case IRQ_PEND_EXT_CLOCK_COMP:
2985 		irq->type = KVM_S390_INT_CLOCK_COMP;
2986 		break;
2987 	case IRQ_PEND_EXT_CPU_TIMER:
2988 		irq->type = KVM_S390_INT_CPU_TIMER;
2989 		break;
2990 	case IRQ_PEND_SIGP_STOP:
2991 		irq->type = KVM_S390_SIGP_STOP;
2992 		irq->u.stop = li->irq.stop;
2993 		break;
2994 	case IRQ_PEND_RESTART:
2995 		irq->type = KVM_S390_RESTART;
2996 		break;
2997 	case IRQ_PEND_SET_PREFIX:
2998 		irq->type = KVM_S390_SIGP_SET_PREFIX;
2999 		irq->u.prefix = li->irq.prefix;
3000 		break;
3001 	}
3002 }
3003 
3004 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
3005 {
3006 	int scn;
3007 	DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
3008 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
3009 	unsigned long pending_irqs;
3010 	struct kvm_s390_irq irq;
3011 	unsigned long irq_type;
3012 	int cpuaddr;
3013 	int n = 0;
3014 
3015 	spin_lock(&li->lock);
3016 	pending_irqs = li->pending_irqs;
3017 	memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
3018 	       sizeof(sigp_emerg_pending));
3019 	spin_unlock(&li->lock);
3020 
3021 	for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
3022 		memset(&irq, 0, sizeof(irq));
3023 		if (irq_type == IRQ_PEND_EXT_EMERGENCY)
3024 			continue;
3025 		if (n + sizeof(irq) > len)
3026 			return -ENOBUFS;
3027 		store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
3028 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3029 			return -EFAULT;
3030 		n += sizeof(irq);
3031 	}
3032 
3033 	if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
3034 		for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
3035 			memset(&irq, 0, sizeof(irq));
3036 			if (n + sizeof(irq) > len)
3037 				return -ENOBUFS;
3038 			irq.type = KVM_S390_INT_EMERGENCY;
3039 			irq.u.emerg.code = cpuaddr;
3040 			if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3041 				return -EFAULT;
3042 			n += sizeof(irq);
3043 		}
3044 	}
3045 
3046 	if (sca_ext_call_pending(vcpu, &scn)) {
3047 		if (n + sizeof(irq) > len)
3048 			return -ENOBUFS;
3049 		memset(&irq, 0, sizeof(irq));
3050 		irq.type = KVM_S390_INT_EXTERNAL_CALL;
3051 		irq.u.extcall.code = scn;
3052 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3053 			return -EFAULT;
3054 		n += sizeof(irq);
3055 	}
3056 
3057 	return n;
3058 }
3059 
3060 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
3061 {
3062 	int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus);
3063 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3064 	struct kvm_vcpu *vcpu;
3065 	u8 vcpu_isc_mask;
3066 
3067 	for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) {
3068 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
3069 		if (psw_ioint_disabled(vcpu))
3070 			continue;
3071 		vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
3072 		if (deliverable_mask & vcpu_isc_mask) {
3073 			/* lately kicked but not yet running */
3074 			if (test_and_set_bit(vcpu_idx, gi->kicked_mask))
3075 				return;
3076 			kvm_s390_vcpu_wakeup(vcpu);
3077 			return;
3078 		}
3079 	}
3080 }
3081 
3082 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3083 {
3084 	struct kvm_s390_gisa_interrupt *gi =
3085 		container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3086 	struct kvm *kvm =
3087 		container_of(gi->origin, struct sie_page2, gisa)->kvm;
3088 	u8 pending_mask;
3089 
3090 	pending_mask = gisa_get_ipm_or_restore_iam(gi);
3091 	if (pending_mask) {
3092 		__airqs_kick_single_vcpu(kvm, pending_mask);
3093 		hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3094 		return HRTIMER_RESTART;
3095 	}
3096 
3097 	return HRTIMER_NORESTART;
3098 }
3099 
3100 #define NULL_GISA_ADDR 0x00000000UL
3101 #define NONE_GISA_ADDR 0x00000001UL
3102 #define GISA_ADDR_MASK 0xfffff000UL
3103 
3104 static void process_gib_alert_list(void)
3105 {
3106 	struct kvm_s390_gisa_interrupt *gi;
3107 	struct kvm_s390_gisa *gisa;
3108 	struct kvm *kvm;
3109 	u32 final, origin = 0UL;
3110 
3111 	do {
3112 		/*
3113 		 * If the NONE_GISA_ADDR is still stored in the alert list
3114 		 * origin, we will leave the outer loop. No further GISA has
3115 		 * been added to the alert list by millicode while processing
3116 		 * the current alert list.
3117 		 */
3118 		final = (origin & NONE_GISA_ADDR);
3119 		/*
3120 		 * Cut off the alert list and store the NONE_GISA_ADDR in the
3121 		 * alert list origin to avoid further GAL interruptions.
3122 		 * A new alert list can be build up by millicode in parallel
3123 		 * for guests not in the yet cut-off alert list. When in the
3124 		 * final loop, store the NULL_GISA_ADDR instead. This will re-
3125 		 * enable GAL interruptions on the host again.
3126 		 */
3127 		origin = xchg(&gib->alert_list_origin,
3128 			      (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3129 		/*
3130 		 * Loop through the just cut-off alert list and start the
3131 		 * gisa timers to kick idle vcpus to consume the pending
3132 		 * interruptions asap.
3133 		 */
3134 		while (origin & GISA_ADDR_MASK) {
3135 			gisa = (struct kvm_s390_gisa *)(u64)origin;
3136 			origin = gisa->next_alert;
3137 			gisa->next_alert = (u32)(u64)gisa;
3138 			kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3139 			gi = &kvm->arch.gisa_int;
3140 			if (hrtimer_active(&gi->timer))
3141 				hrtimer_cancel(&gi->timer);
3142 			hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3143 		}
3144 	} while (!final);
3145 
3146 }
3147 
3148 void kvm_s390_gisa_clear(struct kvm *kvm)
3149 {
3150 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3151 
3152 	if (!gi->origin)
3153 		return;
3154 	gisa_clear_ipm(gi->origin);
3155 	VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3156 }
3157 
3158 void kvm_s390_gisa_init(struct kvm *kvm)
3159 {
3160 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3161 
3162 	if (!css_general_characteristics.aiv)
3163 		return;
3164 	gi->origin = &kvm->arch.sie_page2->gisa;
3165 	gi->alert.mask = 0;
3166 	spin_lock_init(&gi->alert.ref_lock);
3167 	gi->expires = 50 * 1000; /* 50 usec */
3168 	hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3169 	gi->timer.function = gisa_vcpu_kicker;
3170 	memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3171 	gi->origin->next_alert = (u32)(u64)gi->origin;
3172 	VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3173 }
3174 
3175 void kvm_s390_gisa_enable(struct kvm *kvm)
3176 {
3177 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3178 	struct kvm_vcpu *vcpu;
3179 	unsigned long i;
3180 	u32 gisa_desc;
3181 
3182 	if (gi->origin)
3183 		return;
3184 	kvm_s390_gisa_init(kvm);
3185 	gisa_desc = kvm_s390_get_gisa_desc(kvm);
3186 	if (!gisa_desc)
3187 		return;
3188 	kvm_for_each_vcpu(i, vcpu, kvm) {
3189 		mutex_lock(&vcpu->mutex);
3190 		vcpu->arch.sie_block->gd = gisa_desc;
3191 		vcpu->arch.sie_block->eca |= ECA_AIV;
3192 		VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3193 			   vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3194 		mutex_unlock(&vcpu->mutex);
3195 	}
3196 }
3197 
3198 void kvm_s390_gisa_destroy(struct kvm *kvm)
3199 {
3200 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3201 	struct kvm_s390_gisa *gisa = gi->origin;
3202 
3203 	if (!gi->origin)
3204 		return;
3205 	if (gi->alert.mask)
3206 		KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
3207 			  kvm, gi->alert.mask);
3208 	while (gisa_in_alert_list(gi->origin))
3209 		cpu_relax();
3210 	hrtimer_cancel(&gi->timer);
3211 	gi->origin = NULL;
3212 	VM_EVENT(kvm, 3, "gisa 0x%pK destroyed", gisa);
3213 }
3214 
3215 void kvm_s390_gisa_disable(struct kvm *kvm)
3216 {
3217 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3218 	struct kvm_vcpu *vcpu;
3219 	unsigned long i;
3220 
3221 	if (!gi->origin)
3222 		return;
3223 	kvm_for_each_vcpu(i, vcpu, kvm) {
3224 		mutex_lock(&vcpu->mutex);
3225 		vcpu->arch.sie_block->eca &= ~ECA_AIV;
3226 		vcpu->arch.sie_block->gd = 0U;
3227 		mutex_unlock(&vcpu->mutex);
3228 		VCPU_EVENT(vcpu, 3, "AIV disabled for cpu %03u", vcpu->vcpu_id);
3229 	}
3230 	kvm_s390_gisa_destroy(kvm);
3231 }
3232 
3233 /**
3234  * kvm_s390_gisc_register - register a guest ISC
3235  *
3236  * @kvm:  the kernel vm to work with
3237  * @gisc: the guest interruption sub class to register
3238  *
3239  * The function extends the vm specific alert mask to use.
3240  * The effective IAM mask in the GISA is updated as well
3241  * in case the GISA is not part of the GIB alert list.
3242  * It will be updated latest when the IAM gets restored
3243  * by gisa_get_ipm_or_restore_iam().
3244  *
3245  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3246  *          has registered with the channel subsystem.
3247  *          -ENODEV in case the vm uses no GISA
3248  *          -ERANGE in case the guest ISC is invalid
3249  */
3250 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3251 {
3252 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3253 
3254 	if (!gi->origin)
3255 		return -ENODEV;
3256 	if (gisc > MAX_ISC)
3257 		return -ERANGE;
3258 
3259 	spin_lock(&gi->alert.ref_lock);
3260 	gi->alert.ref_count[gisc]++;
3261 	if (gi->alert.ref_count[gisc] == 1) {
3262 		gi->alert.mask |= 0x80 >> gisc;
3263 		gisa_set_iam(gi->origin, gi->alert.mask);
3264 	}
3265 	spin_unlock(&gi->alert.ref_lock);
3266 
3267 	return gib->nisc;
3268 }
3269 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3270 
3271 /**
3272  * kvm_s390_gisc_unregister - unregister a guest ISC
3273  *
3274  * @kvm:  the kernel vm to work with
3275  * @gisc: the guest interruption sub class to register
3276  *
3277  * The function reduces the vm specific alert mask to use.
3278  * The effective IAM mask in the GISA is updated as well
3279  * in case the GISA is not part of the GIB alert list.
3280  * It will be updated latest when the IAM gets restored
3281  * by gisa_get_ipm_or_restore_iam().
3282  *
3283  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3284  *          has registered with the channel subsystem.
3285  *          -ENODEV in case the vm uses no GISA
3286  *          -ERANGE in case the guest ISC is invalid
3287  *          -EINVAL in case the guest ISC is not registered
3288  */
3289 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3290 {
3291 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3292 	int rc = 0;
3293 
3294 	if (!gi->origin)
3295 		return -ENODEV;
3296 	if (gisc > MAX_ISC)
3297 		return -ERANGE;
3298 
3299 	spin_lock(&gi->alert.ref_lock);
3300 	if (gi->alert.ref_count[gisc] == 0) {
3301 		rc = -EINVAL;
3302 		goto out;
3303 	}
3304 	gi->alert.ref_count[gisc]--;
3305 	if (gi->alert.ref_count[gisc] == 0) {
3306 		gi->alert.mask &= ~(0x80 >> gisc);
3307 		gisa_set_iam(gi->origin, gi->alert.mask);
3308 	}
3309 out:
3310 	spin_unlock(&gi->alert.ref_lock);
3311 
3312 	return rc;
3313 }
3314 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3315 
3316 static void aen_host_forward(unsigned long si)
3317 {
3318 	struct kvm_s390_gisa_interrupt *gi;
3319 	struct zpci_gaite *gaite;
3320 	struct kvm *kvm;
3321 
3322 	gaite = (struct zpci_gaite *)aift->gait +
3323 		(si * sizeof(struct zpci_gaite));
3324 	if (gaite->count == 0)
3325 		return;
3326 	if (gaite->aisb != 0)
3327 		set_bit_inv(gaite->aisbo, (unsigned long *)gaite->aisb);
3328 
3329 	kvm = kvm_s390_pci_si_to_kvm(aift, si);
3330 	if (!kvm)
3331 		return;
3332 	gi = &kvm->arch.gisa_int;
3333 
3334 	if (!(gi->origin->g1.simm & AIS_MODE_MASK(gaite->gisc)) ||
3335 	    !(gi->origin->g1.nimm & AIS_MODE_MASK(gaite->gisc))) {
3336 		gisa_set_ipm_gisc(gi->origin, gaite->gisc);
3337 		if (hrtimer_active(&gi->timer))
3338 			hrtimer_cancel(&gi->timer);
3339 		hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3340 		kvm->stat.aen_forward++;
3341 	}
3342 }
3343 
3344 static void aen_process_gait(u8 isc)
3345 {
3346 	bool found = false, first = true;
3347 	union zpci_sic_iib iib = {{0}};
3348 	unsigned long si, flags;
3349 
3350 	spin_lock_irqsave(&aift->gait_lock, flags);
3351 
3352 	if (!aift->gait) {
3353 		spin_unlock_irqrestore(&aift->gait_lock, flags);
3354 		return;
3355 	}
3356 
3357 	for (si = 0;;) {
3358 		/* Scan adapter summary indicator bit vector */
3359 		si = airq_iv_scan(aift->sbv, si, airq_iv_end(aift->sbv));
3360 		if (si == -1UL) {
3361 			if (first || found) {
3362 				/* Re-enable interrupts. */
3363 				zpci_set_irq_ctrl(SIC_IRQ_MODE_SINGLE, isc,
3364 						  &iib);
3365 				first = found = false;
3366 			} else {
3367 				/* Interrupts on and all bits processed */
3368 				break;
3369 			}
3370 			found = false;
3371 			si = 0;
3372 			/* Scan again after re-enabling interrupts */
3373 			continue;
3374 		}
3375 		found = true;
3376 		aen_host_forward(si);
3377 	}
3378 
3379 	spin_unlock_irqrestore(&aift->gait_lock, flags);
3380 }
3381 
3382 static void gib_alert_irq_handler(struct airq_struct *airq,
3383 				  struct tpi_info *tpi_info)
3384 {
3385 	struct tpi_adapter_info *info = (struct tpi_adapter_info *)tpi_info;
3386 
3387 	inc_irq_stat(IRQIO_GAL);
3388 
3389 	if ((info->forward || info->error) &&
3390 	    IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3391 		aen_process_gait(info->isc);
3392 		if (info->aism != 0)
3393 			process_gib_alert_list();
3394 	} else {
3395 		process_gib_alert_list();
3396 	}
3397 }
3398 
3399 static struct airq_struct gib_alert_irq = {
3400 	.handler = gib_alert_irq_handler,
3401 	.lsi_ptr = &gib_alert_irq.lsi_mask,
3402 };
3403 
3404 void kvm_s390_gib_destroy(void)
3405 {
3406 	if (!gib)
3407 		return;
3408 	if (kvm_s390_pci_interp_allowed() && aift) {
3409 		mutex_lock(&aift->aift_lock);
3410 		kvm_s390_pci_aen_exit();
3411 		mutex_unlock(&aift->aift_lock);
3412 	}
3413 	chsc_sgib(0);
3414 	unregister_adapter_interrupt(&gib_alert_irq);
3415 	free_page((unsigned long)gib);
3416 	gib = NULL;
3417 }
3418 
3419 int kvm_s390_gib_init(u8 nisc)
3420 {
3421 	int rc = 0;
3422 
3423 	if (!css_general_characteristics.aiv) {
3424 		KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3425 		goto out;
3426 	}
3427 
3428 	gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3429 	if (!gib) {
3430 		rc = -ENOMEM;
3431 		goto out;
3432 	}
3433 
3434 	gib_alert_irq.isc = nisc;
3435 	if (register_adapter_interrupt(&gib_alert_irq)) {
3436 		pr_err("Registering the GIB alert interruption handler failed\n");
3437 		rc = -EIO;
3438 		goto out_free_gib;
3439 	}
3440 
3441 	gib->nisc = nisc;
3442 	if (chsc_sgib((u32)(u64)gib)) {
3443 		pr_err("Associating the GIB with the AIV facility failed\n");
3444 		free_page((unsigned long)gib);
3445 		gib = NULL;
3446 		rc = -EIO;
3447 		goto out_unreg_gal;
3448 	}
3449 
3450 	if (kvm_s390_pci_interp_allowed()) {
3451 		if (kvm_s390_pci_aen_init(nisc)) {
3452 			pr_err("Initializing AEN for PCI failed\n");
3453 			rc = -EIO;
3454 			goto out_unreg_gal;
3455 		}
3456 	}
3457 
3458 	KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3459 	goto out;
3460 
3461 out_unreg_gal:
3462 	unregister_adapter_interrupt(&gib_alert_irq);
3463 out_free_gib:
3464 	free_page((unsigned long)gib);
3465 	gib = NULL;
3466 out:
3467 	return rc;
3468 }
3469