xref: /openbmc/linux/arch/s390/kvm/interrupt.c (revision b2765275)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * handling kvm guest interrupts
4  *
5  * Copyright IBM Corp. 2008, 2015
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 "kvm-s390.h"
32 #include "gaccess.h"
33 #include "trace-s390.h"
34 
35 #define PFAULT_INIT 0x0600
36 #define PFAULT_DONE 0x0680
37 #define VIRTIO_PARAM 0x0d00
38 
39 static struct kvm_s390_gib *gib;
40 
41 /* handle external calls via sigp interpretation facility */
42 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
43 {
44 	int c, scn;
45 
46 	if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND))
47 		return 0;
48 
49 	BUG_ON(!kvm_s390_use_sca_entries());
50 	read_lock(&vcpu->kvm->arch.sca_lock);
51 	if (vcpu->kvm->arch.use_esca) {
52 		struct esca_block *sca = vcpu->kvm->arch.sca;
53 		union esca_sigp_ctrl sigp_ctrl =
54 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
55 
56 		c = sigp_ctrl.c;
57 		scn = sigp_ctrl.scn;
58 	} else {
59 		struct bsca_block *sca = vcpu->kvm->arch.sca;
60 		union bsca_sigp_ctrl sigp_ctrl =
61 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
62 
63 		c = sigp_ctrl.c;
64 		scn = sigp_ctrl.scn;
65 	}
66 	read_unlock(&vcpu->kvm->arch.sca_lock);
67 
68 	if (src_id)
69 		*src_id = scn;
70 
71 	return c;
72 }
73 
74 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
75 {
76 	int expect, rc;
77 
78 	BUG_ON(!kvm_s390_use_sca_entries());
79 	read_lock(&vcpu->kvm->arch.sca_lock);
80 	if (vcpu->kvm->arch.use_esca) {
81 		struct esca_block *sca = vcpu->kvm->arch.sca;
82 		union esca_sigp_ctrl *sigp_ctrl =
83 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
84 		union esca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
85 
86 		new_val.scn = src_id;
87 		new_val.c = 1;
88 		old_val.c = 0;
89 
90 		expect = old_val.value;
91 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
92 	} else {
93 		struct bsca_block *sca = vcpu->kvm->arch.sca;
94 		union bsca_sigp_ctrl *sigp_ctrl =
95 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
96 		union bsca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
97 
98 		new_val.scn = src_id;
99 		new_val.c = 1;
100 		old_val.c = 0;
101 
102 		expect = old_val.value;
103 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
104 	}
105 	read_unlock(&vcpu->kvm->arch.sca_lock);
106 
107 	if (rc != expect) {
108 		/* another external call is pending */
109 		return -EBUSY;
110 	}
111 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
112 	return 0;
113 }
114 
115 static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
116 {
117 	int rc, expect;
118 
119 	if (!kvm_s390_use_sca_entries())
120 		return;
121 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
122 	read_lock(&vcpu->kvm->arch.sca_lock);
123 	if (vcpu->kvm->arch.use_esca) {
124 		struct esca_block *sca = vcpu->kvm->arch.sca;
125 		union esca_sigp_ctrl *sigp_ctrl =
126 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
127 		union esca_sigp_ctrl old = *sigp_ctrl;
128 
129 		expect = old.value;
130 		rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
131 	} else {
132 		struct bsca_block *sca = vcpu->kvm->arch.sca;
133 		union bsca_sigp_ctrl *sigp_ctrl =
134 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
135 		union bsca_sigp_ctrl old = *sigp_ctrl;
136 
137 		expect = old.value;
138 		rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
139 	}
140 	read_unlock(&vcpu->kvm->arch.sca_lock);
141 	WARN_ON(rc != expect); /* cannot clear? */
142 }
143 
144 int psw_extint_disabled(struct kvm_vcpu *vcpu)
145 {
146 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
147 }
148 
149 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
150 {
151 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
152 }
153 
154 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
155 {
156 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
157 }
158 
159 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
160 {
161 	return psw_extint_disabled(vcpu) &&
162 	       psw_ioint_disabled(vcpu) &&
163 	       psw_mchk_disabled(vcpu);
164 }
165 
166 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
167 {
168 	if (psw_extint_disabled(vcpu) ||
169 	    !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
170 		return 0;
171 	if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
172 		/* No timer interrupts when single stepping */
173 		return 0;
174 	return 1;
175 }
176 
177 static int ckc_irq_pending(struct kvm_vcpu *vcpu)
178 {
179 	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
180 	const u64 ckc = vcpu->arch.sie_block->ckc;
181 
182 	if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
183 		if ((s64)ckc >= (s64)now)
184 			return 0;
185 	} else if (ckc >= now) {
186 		return 0;
187 	}
188 	return ckc_interrupts_enabled(vcpu);
189 }
190 
191 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
192 {
193 	return !psw_extint_disabled(vcpu) &&
194 	       (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK);
195 }
196 
197 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
198 {
199 	if (!cpu_timer_interrupts_enabled(vcpu))
200 		return 0;
201 	return kvm_s390_get_cpu_timer(vcpu) >> 63;
202 }
203 
204 static uint64_t isc_to_isc_bits(int isc)
205 {
206 	return (0x80 >> isc) << 24;
207 }
208 
209 static inline u32 isc_to_int_word(u8 isc)
210 {
211 	return ((u32)isc << 27) | 0x80000000;
212 }
213 
214 static inline u8 int_word_to_isc(u32 int_word)
215 {
216 	return (int_word & 0x38000000) >> 27;
217 }
218 
219 /*
220  * To use atomic bitmap functions, we have to provide a bitmap address
221  * that is u64 aligned. However, the ipm might be u32 aligned.
222  * Therefore, we logically start the bitmap at the very beginning of the
223  * struct and fixup the bit number.
224  */
225 #define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE)
226 
227 /**
228  * gisa_set_iam - change the GISA interruption alert mask
229  *
230  * @gisa: gisa to operate on
231  * @iam: new IAM value to use
232  *
233  * Change the IAM atomically with the next alert address and the IPM
234  * of the GISA if the GISA is not part of the GIB alert list. All three
235  * fields are located in the first long word of the GISA.
236  *
237  * Returns: 0 on success
238  *          -EBUSY in case the gisa is part of the alert list
239  */
240 static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam)
241 {
242 	u64 word, _word;
243 
244 	do {
245 		word = READ_ONCE(gisa->u64.word[0]);
246 		if ((u64)gisa != word >> 32)
247 			return -EBUSY;
248 		_word = (word & ~0xffUL) | iam;
249 	} while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
250 
251 	return 0;
252 }
253 
254 /**
255  * gisa_clear_ipm - clear the GISA interruption pending mask
256  *
257  * @gisa: gisa to operate on
258  *
259  * Clear the IPM atomically with the next alert address and the IAM
260  * of the GISA unconditionally. All three fields are located in the
261  * first long word of the GISA.
262  */
263 static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa)
264 {
265 	u64 word, _word;
266 
267 	do {
268 		word = READ_ONCE(gisa->u64.word[0]);
269 		_word = word & ~(0xffUL << 24);
270 	} while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
271 }
272 
273 /**
274  * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM
275  *
276  * @gi: gisa interrupt struct to work on
277  *
278  * Atomically restores the interruption alert mask if none of the
279  * relevant ISCs are pending and return the IPM.
280  *
281  * Returns: the relevant pending ISCs
282  */
283 static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
284 {
285 	u8 pending_mask, alert_mask;
286 	u64 word, _word;
287 
288 	do {
289 		word = READ_ONCE(gi->origin->u64.word[0]);
290 		alert_mask = READ_ONCE(gi->alert.mask);
291 		pending_mask = (u8)(word >> 24) & alert_mask;
292 		if (pending_mask)
293 			return pending_mask;
294 		_word = (word & ~0xffUL) | alert_mask;
295 	} while (cmpxchg(&gi->origin->u64.word[0], word, _word) != word);
296 
297 	return 0;
298 }
299 
300 static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
301 {
302 	return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa;
303 }
304 
305 static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
306 {
307 	set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
308 }
309 
310 static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
311 {
312 	return READ_ONCE(gisa->ipm);
313 }
314 
315 static inline void gisa_clear_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
316 {
317 	clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
318 }
319 
320 static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
321 {
322 	return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
323 }
324 
325 static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
326 {
327 	return vcpu->kvm->arch.float_int.pending_irqs |
328 		vcpu->arch.local_int.pending_irqs;
329 }
330 
331 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
332 {
333 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
334 	unsigned long pending_mask;
335 
336 	pending_mask = pending_irqs_no_gisa(vcpu);
337 	if (gi->origin)
338 		pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
339 	return pending_mask;
340 }
341 
342 static inline int isc_to_irq_type(unsigned long isc)
343 {
344 	return IRQ_PEND_IO_ISC_0 - isc;
345 }
346 
347 static inline int irq_type_to_isc(unsigned long irq_type)
348 {
349 	return IRQ_PEND_IO_ISC_0 - irq_type;
350 }
351 
352 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
353 				   unsigned long active_mask)
354 {
355 	int i;
356 
357 	for (i = 0; i <= MAX_ISC; i++)
358 		if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
359 			active_mask &= ~(1UL << (isc_to_irq_type(i)));
360 
361 	return active_mask;
362 }
363 
364 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
365 {
366 	unsigned long active_mask;
367 
368 	active_mask = pending_irqs(vcpu);
369 	if (!active_mask)
370 		return 0;
371 
372 	if (psw_extint_disabled(vcpu))
373 		active_mask &= ~IRQ_PEND_EXT_MASK;
374 	if (psw_ioint_disabled(vcpu))
375 		active_mask &= ~IRQ_PEND_IO_MASK;
376 	else
377 		active_mask = disable_iscs(vcpu, active_mask);
378 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
379 		__clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
380 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
381 		__clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
382 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
383 		__clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
384 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
385 		__clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
386 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
387 		__clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
388 	if (psw_mchk_disabled(vcpu))
389 		active_mask &= ~IRQ_PEND_MCHK_MASK;
390 	/*
391 	 * Check both floating and local interrupt's cr14 because
392 	 * bit IRQ_PEND_MCHK_REP could be set in both cases.
393 	 */
394 	if (!(vcpu->arch.sie_block->gcr[14] &
395 	   (vcpu->kvm->arch.float_int.mchk.cr14 |
396 	   vcpu->arch.local_int.irq.mchk.cr14)))
397 		__clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
398 
399 	/*
400 	 * STOP irqs will never be actively delivered. They are triggered via
401 	 * intercept requests and cleared when the stop intercept is performed.
402 	 */
403 	__clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
404 
405 	return active_mask;
406 }
407 
408 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
409 {
410 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
411 	set_bit(vcpu->vcpu_id, vcpu->kvm->arch.idle_mask);
412 }
413 
414 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
415 {
416 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
417 	clear_bit(vcpu->vcpu_id, vcpu->kvm->arch.idle_mask);
418 }
419 
420 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
421 {
422 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
423 				      CPUSTAT_STOP_INT);
424 	vcpu->arch.sie_block->lctl = 0x0000;
425 	vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
426 
427 	if (guestdbg_enabled(vcpu)) {
428 		vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
429 					       LCTL_CR10 | LCTL_CR11);
430 		vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
431 	}
432 }
433 
434 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
435 {
436 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
437 		return;
438 	if (psw_ioint_disabled(vcpu))
439 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
440 	else
441 		vcpu->arch.sie_block->lctl |= LCTL_CR6;
442 }
443 
444 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
445 {
446 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
447 		return;
448 	if (psw_extint_disabled(vcpu))
449 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
450 	else
451 		vcpu->arch.sie_block->lctl |= LCTL_CR0;
452 }
453 
454 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
455 {
456 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
457 		return;
458 	if (psw_mchk_disabled(vcpu))
459 		vcpu->arch.sie_block->ictl |= ICTL_LPSW;
460 	else
461 		vcpu->arch.sie_block->lctl |= LCTL_CR14;
462 }
463 
464 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
465 {
466 	if (kvm_s390_is_stop_irq_pending(vcpu))
467 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
468 }
469 
470 /* Set interception request for non-deliverable interrupts */
471 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
472 {
473 	set_intercept_indicators_io(vcpu);
474 	set_intercept_indicators_ext(vcpu);
475 	set_intercept_indicators_mchk(vcpu);
476 	set_intercept_indicators_stop(vcpu);
477 }
478 
479 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
480 {
481 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
482 	int rc;
483 
484 	vcpu->stat.deliver_cputm++;
485 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
486 					 0, 0);
487 
488 	rc  = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
489 			   (u16 *)__LC_EXT_INT_CODE);
490 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
491 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
492 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
493 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
494 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
495 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
496 	return rc ? -EFAULT : 0;
497 }
498 
499 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
500 {
501 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
502 	int rc;
503 
504 	vcpu->stat.deliver_ckc++;
505 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
506 					 0, 0);
507 
508 	rc  = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
509 			   (u16 __user *)__LC_EXT_INT_CODE);
510 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
511 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
512 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
513 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
514 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
515 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
516 	return rc ? -EFAULT : 0;
517 }
518 
519 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
520 {
521 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
522 	struct kvm_s390_ext_info ext;
523 	int rc;
524 
525 	spin_lock(&li->lock);
526 	ext = li->irq.ext;
527 	clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
528 	li->irq.ext.ext_params2 = 0;
529 	spin_unlock(&li->lock);
530 
531 	VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
532 		   ext.ext_params2);
533 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
534 					 KVM_S390_INT_PFAULT_INIT,
535 					 0, ext.ext_params2);
536 
537 	rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
538 	rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
539 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
540 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
541 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
542 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
543 	rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
544 	return rc ? -EFAULT : 0;
545 }
546 
547 static int __write_machine_check(struct kvm_vcpu *vcpu,
548 				 struct kvm_s390_mchk_info *mchk)
549 {
550 	unsigned long ext_sa_addr;
551 	unsigned long lc;
552 	freg_t fprs[NUM_FPRS];
553 	union mci mci;
554 	int rc;
555 
556 	mci.val = mchk->mcic;
557 	/* take care of lazy register loading */
558 	save_fpu_regs();
559 	save_access_regs(vcpu->run->s.regs.acrs);
560 	if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
561 		save_gs_cb(current->thread.gs_cb);
562 
563 	/* Extended save area */
564 	rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
565 			   sizeof(unsigned long));
566 	/* Only bits 0 through 63-LC are used for address formation */
567 	lc = ext_sa_addr & MCESA_LC_MASK;
568 	if (test_kvm_facility(vcpu->kvm, 133)) {
569 		switch (lc) {
570 		case 0:
571 		case 10:
572 			ext_sa_addr &= ~0x3ffUL;
573 			break;
574 		case 11:
575 			ext_sa_addr &= ~0x7ffUL;
576 			break;
577 		case 12:
578 			ext_sa_addr &= ~0xfffUL;
579 			break;
580 		default:
581 			ext_sa_addr = 0;
582 			break;
583 		}
584 	} else {
585 		ext_sa_addr &= ~0x3ffUL;
586 	}
587 
588 	if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
589 		if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
590 				    512))
591 			mci.vr = 0;
592 	} else {
593 		mci.vr = 0;
594 	}
595 	if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
596 	    && (lc == 11 || lc == 12)) {
597 		if (write_guest_abs(vcpu, ext_sa_addr + 1024,
598 				    &vcpu->run->s.regs.gscb, 32))
599 			mci.gs = 0;
600 	} else {
601 		mci.gs = 0;
602 	}
603 
604 	/* General interruption information */
605 	rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
606 	rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
607 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
608 	rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
609 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
610 	rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
611 
612 	/* Register-save areas */
613 	if (MACHINE_HAS_VX) {
614 		convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
615 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
616 	} else {
617 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
618 				     vcpu->run->s.regs.fprs, 128);
619 	}
620 	rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
621 			     vcpu->run->s.regs.gprs, 128);
622 	rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
623 			   (u32 __user *) __LC_FP_CREG_SAVE_AREA);
624 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
625 			   (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
626 	rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
627 			   (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
628 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
629 			   (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
630 	rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
631 			     &vcpu->run->s.regs.acrs, 64);
632 	rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
633 			     &vcpu->arch.sie_block->gcr, 128);
634 
635 	/* Extended interruption information */
636 	rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
637 			   (u32 __user *) __LC_EXT_DAMAGE_CODE);
638 	rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
639 			   (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
640 	rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
641 			     sizeof(mchk->fixed_logout));
642 	return rc ? -EFAULT : 0;
643 }
644 
645 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
646 {
647 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
648 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
649 	struct kvm_s390_mchk_info mchk = {};
650 	int deliver = 0;
651 	int rc = 0;
652 
653 	spin_lock(&fi->lock);
654 	spin_lock(&li->lock);
655 	if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
656 	    test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
657 		/*
658 		 * If there was an exigent machine check pending, then any
659 		 * repressible machine checks that might have been pending
660 		 * are indicated along with it, so always clear bits for
661 		 * repressible and exigent interrupts
662 		 */
663 		mchk = li->irq.mchk;
664 		clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
665 		clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
666 		memset(&li->irq.mchk, 0, sizeof(mchk));
667 		deliver = 1;
668 	}
669 	/*
670 	 * We indicate floating repressible conditions along with
671 	 * other pending conditions. Channel Report Pending and Channel
672 	 * Subsystem damage are the only two and and are indicated by
673 	 * bits in mcic and masked in cr14.
674 	 */
675 	if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
676 		mchk.mcic |= fi->mchk.mcic;
677 		mchk.cr14 |= fi->mchk.cr14;
678 		memset(&fi->mchk, 0, sizeof(mchk));
679 		deliver = 1;
680 	}
681 	spin_unlock(&li->lock);
682 	spin_unlock(&fi->lock);
683 
684 	if (deliver) {
685 		VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
686 			   mchk.mcic);
687 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
688 						 KVM_S390_MCHK,
689 						 mchk.cr14, mchk.mcic);
690 		vcpu->stat.deliver_machine_check++;
691 		rc = __write_machine_check(vcpu, &mchk);
692 	}
693 	return rc;
694 }
695 
696 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
697 {
698 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
699 	int rc;
700 
701 	VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
702 	vcpu->stat.deliver_restart_signal++;
703 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
704 
705 	rc  = write_guest_lc(vcpu,
706 			     offsetof(struct lowcore, restart_old_psw),
707 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
708 	rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
709 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
710 	clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
711 	return rc ? -EFAULT : 0;
712 }
713 
714 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
715 {
716 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
717 	struct kvm_s390_prefix_info prefix;
718 
719 	spin_lock(&li->lock);
720 	prefix = li->irq.prefix;
721 	li->irq.prefix.address = 0;
722 	clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
723 	spin_unlock(&li->lock);
724 
725 	vcpu->stat.deliver_prefix_signal++;
726 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
727 					 KVM_S390_SIGP_SET_PREFIX,
728 					 prefix.address, 0);
729 
730 	kvm_s390_set_prefix(vcpu, prefix.address);
731 	return 0;
732 }
733 
734 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
735 {
736 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
737 	int rc;
738 	int cpu_addr;
739 
740 	spin_lock(&li->lock);
741 	cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
742 	clear_bit(cpu_addr, li->sigp_emerg_pending);
743 	if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
744 		clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
745 	spin_unlock(&li->lock);
746 
747 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
748 	vcpu->stat.deliver_emergency_signal++;
749 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
750 					 cpu_addr, 0);
751 
752 	rc  = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
753 			   (u16 *)__LC_EXT_INT_CODE);
754 	rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
755 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
756 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
757 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
758 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
759 	return rc ? -EFAULT : 0;
760 }
761 
762 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
763 {
764 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
765 	struct kvm_s390_extcall_info extcall;
766 	int rc;
767 
768 	spin_lock(&li->lock);
769 	extcall = li->irq.extcall;
770 	li->irq.extcall.code = 0;
771 	clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
772 	spin_unlock(&li->lock);
773 
774 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
775 	vcpu->stat.deliver_external_call++;
776 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
777 					 KVM_S390_INT_EXTERNAL_CALL,
778 					 extcall.code, 0);
779 
780 	rc  = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
781 			   (u16 *)__LC_EXT_INT_CODE);
782 	rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
783 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
784 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
785 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
786 			    sizeof(psw_t));
787 	return rc ? -EFAULT : 0;
788 }
789 
790 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
791 {
792 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
793 	struct kvm_s390_pgm_info pgm_info;
794 	int rc = 0, nullifying = false;
795 	u16 ilen;
796 
797 	spin_lock(&li->lock);
798 	pgm_info = li->irq.pgm;
799 	clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
800 	memset(&li->irq.pgm, 0, sizeof(pgm_info));
801 	spin_unlock(&li->lock);
802 
803 	ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
804 	VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
805 		   pgm_info.code, ilen);
806 	vcpu->stat.deliver_program++;
807 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
808 					 pgm_info.code, 0);
809 
810 	switch (pgm_info.code & ~PGM_PER) {
811 	case PGM_AFX_TRANSLATION:
812 	case PGM_ASX_TRANSLATION:
813 	case PGM_EX_TRANSLATION:
814 	case PGM_LFX_TRANSLATION:
815 	case PGM_LSTE_SEQUENCE:
816 	case PGM_LSX_TRANSLATION:
817 	case PGM_LX_TRANSLATION:
818 	case PGM_PRIMARY_AUTHORITY:
819 	case PGM_SECONDARY_AUTHORITY:
820 		nullifying = true;
821 		/* fall through */
822 	case PGM_SPACE_SWITCH:
823 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
824 				  (u64 *)__LC_TRANS_EXC_CODE);
825 		break;
826 	case PGM_ALEN_TRANSLATION:
827 	case PGM_ALE_SEQUENCE:
828 	case PGM_ASTE_INSTANCE:
829 	case PGM_ASTE_SEQUENCE:
830 	case PGM_ASTE_VALIDITY:
831 	case PGM_EXTENDED_AUTHORITY:
832 		rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
833 				  (u8 *)__LC_EXC_ACCESS_ID);
834 		nullifying = true;
835 		break;
836 	case PGM_ASCE_TYPE:
837 	case PGM_PAGE_TRANSLATION:
838 	case PGM_REGION_FIRST_TRANS:
839 	case PGM_REGION_SECOND_TRANS:
840 	case PGM_REGION_THIRD_TRANS:
841 	case PGM_SEGMENT_TRANSLATION:
842 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
843 				  (u64 *)__LC_TRANS_EXC_CODE);
844 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
845 				   (u8 *)__LC_EXC_ACCESS_ID);
846 		rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
847 				   (u8 *)__LC_OP_ACCESS_ID);
848 		nullifying = true;
849 		break;
850 	case PGM_MONITOR:
851 		rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
852 				  (u16 *)__LC_MON_CLASS_NR);
853 		rc |= put_guest_lc(vcpu, pgm_info.mon_code,
854 				   (u64 *)__LC_MON_CODE);
855 		break;
856 	case PGM_VECTOR_PROCESSING:
857 	case PGM_DATA:
858 		rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
859 				  (u32 *)__LC_DATA_EXC_CODE);
860 		break;
861 	case PGM_PROTECTION:
862 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
863 				  (u64 *)__LC_TRANS_EXC_CODE);
864 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
865 				   (u8 *)__LC_EXC_ACCESS_ID);
866 		break;
867 	case PGM_STACK_FULL:
868 	case PGM_STACK_EMPTY:
869 	case PGM_STACK_SPECIFICATION:
870 	case PGM_STACK_TYPE:
871 	case PGM_STACK_OPERATION:
872 	case PGM_TRACE_TABEL:
873 	case PGM_CRYPTO_OPERATION:
874 		nullifying = true;
875 		break;
876 	}
877 
878 	if (pgm_info.code & PGM_PER) {
879 		rc |= put_guest_lc(vcpu, pgm_info.per_code,
880 				   (u8 *) __LC_PER_CODE);
881 		rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
882 				   (u8 *)__LC_PER_ATMID);
883 		rc |= put_guest_lc(vcpu, pgm_info.per_address,
884 				   (u64 *) __LC_PER_ADDRESS);
885 		rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
886 				   (u8 *) __LC_PER_ACCESS_ID);
887 	}
888 
889 	if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
890 		kvm_s390_rewind_psw(vcpu, ilen);
891 
892 	/* bit 1+2 of the target are the ilc, so we can directly use ilen */
893 	rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
894 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
895 				 (u64 *) __LC_LAST_BREAK);
896 	rc |= put_guest_lc(vcpu, pgm_info.code,
897 			   (u16 *)__LC_PGM_INT_CODE);
898 	rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
899 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
900 	rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
901 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
902 	return rc ? -EFAULT : 0;
903 }
904 
905 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
906 {
907 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
908 	struct kvm_s390_ext_info ext;
909 	int rc = 0;
910 
911 	spin_lock(&fi->lock);
912 	if (!(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
913 		spin_unlock(&fi->lock);
914 		return 0;
915 	}
916 	ext = fi->srv_signal;
917 	memset(&fi->srv_signal, 0, sizeof(ext));
918 	clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
919 	spin_unlock(&fi->lock);
920 
921 	VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
922 		   ext.ext_params);
923 	vcpu->stat.deliver_service_signal++;
924 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
925 					 ext.ext_params, 0);
926 
927 	rc  = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
928 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
929 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
930 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
931 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
932 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
933 	rc |= put_guest_lc(vcpu, ext.ext_params,
934 			   (u32 *)__LC_EXT_PARAMS);
935 
936 	return rc ? -EFAULT : 0;
937 }
938 
939 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
940 {
941 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
942 	struct kvm_s390_interrupt_info *inti;
943 	int rc = 0;
944 
945 	spin_lock(&fi->lock);
946 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
947 					struct kvm_s390_interrupt_info,
948 					list);
949 	if (inti) {
950 		list_del(&inti->list);
951 		fi->counters[FIRQ_CNTR_PFAULT] -= 1;
952 	}
953 	if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
954 		clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
955 	spin_unlock(&fi->lock);
956 
957 	if (inti) {
958 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
959 						 KVM_S390_INT_PFAULT_DONE, 0,
960 						 inti->ext.ext_params2);
961 		VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
962 			   inti->ext.ext_params2);
963 
964 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
965 				(u16 *)__LC_EXT_INT_CODE);
966 		rc |= put_guest_lc(vcpu, PFAULT_DONE,
967 				(u16 *)__LC_EXT_CPU_ADDR);
968 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
969 				&vcpu->arch.sie_block->gpsw,
970 				sizeof(psw_t));
971 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
972 				&vcpu->arch.sie_block->gpsw,
973 				sizeof(psw_t));
974 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
975 				(u64 *)__LC_EXT_PARAMS2);
976 		kfree(inti);
977 	}
978 	return rc ? -EFAULT : 0;
979 }
980 
981 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
982 {
983 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
984 	struct kvm_s390_interrupt_info *inti;
985 	int rc = 0;
986 
987 	spin_lock(&fi->lock);
988 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
989 					struct kvm_s390_interrupt_info,
990 					list);
991 	if (inti) {
992 		VCPU_EVENT(vcpu, 4,
993 			   "deliver: virtio parm: 0x%x,parm64: 0x%llx",
994 			   inti->ext.ext_params, inti->ext.ext_params2);
995 		vcpu->stat.deliver_virtio++;
996 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
997 				inti->type,
998 				inti->ext.ext_params,
999 				inti->ext.ext_params2);
1000 		list_del(&inti->list);
1001 		fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1002 	}
1003 	if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1004 		clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1005 	spin_unlock(&fi->lock);
1006 
1007 	if (inti) {
1008 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1009 				(u16 *)__LC_EXT_INT_CODE);
1010 		rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1011 				(u16 *)__LC_EXT_CPU_ADDR);
1012 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1013 				&vcpu->arch.sie_block->gpsw,
1014 				sizeof(psw_t));
1015 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1016 				&vcpu->arch.sie_block->gpsw,
1017 				sizeof(psw_t));
1018 		rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1019 				(u32 *)__LC_EXT_PARAMS);
1020 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1021 				(u64 *)__LC_EXT_PARAMS2);
1022 		kfree(inti);
1023 	}
1024 	return rc ? -EFAULT : 0;
1025 }
1026 
1027 static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1028 {
1029 	int rc;
1030 
1031 	rc  = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1032 	rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1033 	rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1034 	rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1035 	rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1036 			     &vcpu->arch.sie_block->gpsw,
1037 			     sizeof(psw_t));
1038 	rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1039 			    &vcpu->arch.sie_block->gpsw,
1040 			    sizeof(psw_t));
1041 	return rc ? -EFAULT : 0;
1042 }
1043 
1044 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1045 				     unsigned long irq_type)
1046 {
1047 	struct list_head *isc_list;
1048 	struct kvm_s390_float_interrupt *fi;
1049 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1050 	struct kvm_s390_interrupt_info *inti = NULL;
1051 	struct kvm_s390_io_info io;
1052 	u32 isc;
1053 	int rc = 0;
1054 
1055 	fi = &vcpu->kvm->arch.float_int;
1056 
1057 	spin_lock(&fi->lock);
1058 	isc = irq_type_to_isc(irq_type);
1059 	isc_list = &fi->lists[isc];
1060 	inti = list_first_entry_or_null(isc_list,
1061 					struct kvm_s390_interrupt_info,
1062 					list);
1063 	if (inti) {
1064 		if (inti->type & KVM_S390_INT_IO_AI_MASK)
1065 			VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1066 		else
1067 			VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1068 			inti->io.subchannel_id >> 8,
1069 			inti->io.subchannel_id >> 1 & 0x3,
1070 			inti->io.subchannel_nr);
1071 
1072 		vcpu->stat.deliver_io++;
1073 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1074 				inti->type,
1075 				((__u32)inti->io.subchannel_id << 16) |
1076 				inti->io.subchannel_nr,
1077 				((__u64)inti->io.io_int_parm << 32) |
1078 				inti->io.io_int_word);
1079 		list_del(&inti->list);
1080 		fi->counters[FIRQ_CNTR_IO] -= 1;
1081 	}
1082 	if (list_empty(isc_list))
1083 		clear_bit(irq_type, &fi->pending_irqs);
1084 	spin_unlock(&fi->lock);
1085 
1086 	if (inti) {
1087 		rc = __do_deliver_io(vcpu, &(inti->io));
1088 		kfree(inti);
1089 		goto out;
1090 	}
1091 
1092 	if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1093 		/*
1094 		 * in case an adapter interrupt was not delivered
1095 		 * in SIE context KVM will handle the delivery
1096 		 */
1097 		VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1098 		memset(&io, 0, sizeof(io));
1099 		io.io_int_word = isc_to_int_word(isc);
1100 		vcpu->stat.deliver_io++;
1101 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1102 			KVM_S390_INT_IO(1, 0, 0, 0),
1103 			((__u32)io.subchannel_id << 16) |
1104 			io.subchannel_nr,
1105 			((__u64)io.io_int_parm << 32) |
1106 			io.io_int_word);
1107 		rc = __do_deliver_io(vcpu, &io);
1108 	}
1109 out:
1110 	return rc;
1111 }
1112 
1113 /* Check whether an external call is pending (deliverable or not) */
1114 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1115 {
1116 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1117 
1118 	if (!sclp.has_sigpif)
1119 		return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1120 
1121 	return sca_ext_call_pending(vcpu, NULL);
1122 }
1123 
1124 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1125 {
1126 	if (deliverable_irqs(vcpu))
1127 		return 1;
1128 
1129 	if (kvm_cpu_has_pending_timer(vcpu))
1130 		return 1;
1131 
1132 	/* external call pending and deliverable */
1133 	if (kvm_s390_ext_call_pending(vcpu) &&
1134 	    !psw_extint_disabled(vcpu) &&
1135 	    (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1136 		return 1;
1137 
1138 	if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1139 		return 1;
1140 	return 0;
1141 }
1142 
1143 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1144 {
1145 	return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1146 }
1147 
1148 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1149 {
1150 	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1151 	const u64 ckc = vcpu->arch.sie_block->ckc;
1152 	u64 cputm, sltime = 0;
1153 
1154 	if (ckc_interrupts_enabled(vcpu)) {
1155 		if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1156 			if ((s64)now < (s64)ckc)
1157 				sltime = tod_to_ns((s64)ckc - (s64)now);
1158 		} else if (now < ckc) {
1159 			sltime = tod_to_ns(ckc - now);
1160 		}
1161 		/* already expired */
1162 		if (!sltime)
1163 			return 0;
1164 		if (cpu_timer_interrupts_enabled(vcpu)) {
1165 			cputm = kvm_s390_get_cpu_timer(vcpu);
1166 			/* already expired? */
1167 			if (cputm >> 63)
1168 				return 0;
1169 			return min(sltime, tod_to_ns(cputm));
1170 		}
1171 	} else if (cpu_timer_interrupts_enabled(vcpu)) {
1172 		sltime = kvm_s390_get_cpu_timer(vcpu);
1173 		/* already expired? */
1174 		if (sltime >> 63)
1175 			return 0;
1176 	}
1177 	return sltime;
1178 }
1179 
1180 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1181 {
1182 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1183 	u64 sltime;
1184 
1185 	vcpu->stat.exit_wait_state++;
1186 
1187 	/* fast path */
1188 	if (kvm_arch_vcpu_runnable(vcpu))
1189 		return 0;
1190 
1191 	if (psw_interrupts_disabled(vcpu)) {
1192 		VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1193 		return -EOPNOTSUPP; /* disabled wait */
1194 	}
1195 
1196 	if (gi->origin &&
1197 	    (gisa_get_ipm_or_restore_iam(gi) &
1198 	     vcpu->arch.sie_block->gcr[6] >> 24))
1199 		return 0;
1200 
1201 	if (!ckc_interrupts_enabled(vcpu) &&
1202 	    !cpu_timer_interrupts_enabled(vcpu)) {
1203 		VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1204 		__set_cpu_idle(vcpu);
1205 		goto no_timer;
1206 	}
1207 
1208 	sltime = __calculate_sltime(vcpu);
1209 	if (!sltime)
1210 		return 0;
1211 
1212 	__set_cpu_idle(vcpu);
1213 	hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1214 	VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1215 no_timer:
1216 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1217 	kvm_vcpu_block(vcpu);
1218 	__unset_cpu_idle(vcpu);
1219 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1220 
1221 	hrtimer_cancel(&vcpu->arch.ckc_timer);
1222 	return 0;
1223 }
1224 
1225 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1226 {
1227 	vcpu->valid_wakeup = true;
1228 	kvm_vcpu_wake_up(vcpu);
1229 
1230 	/*
1231 	 * The VCPU might not be sleeping but rather executing VSIE. Let's
1232 	 * kick it, so it leaves the SIE to process the request.
1233 	 */
1234 	kvm_s390_vsie_kick(vcpu);
1235 }
1236 
1237 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1238 {
1239 	struct kvm_vcpu *vcpu;
1240 	u64 sltime;
1241 
1242 	vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1243 	sltime = __calculate_sltime(vcpu);
1244 
1245 	/*
1246 	 * If the monotonic clock runs faster than the tod clock we might be
1247 	 * woken up too early and have to go back to sleep to avoid deadlocks.
1248 	 */
1249 	if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1250 		return HRTIMER_RESTART;
1251 	kvm_s390_vcpu_wakeup(vcpu);
1252 	return HRTIMER_NORESTART;
1253 }
1254 
1255 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1256 {
1257 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1258 
1259 	spin_lock(&li->lock);
1260 	li->pending_irqs = 0;
1261 	bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1262 	memset(&li->irq, 0, sizeof(li->irq));
1263 	spin_unlock(&li->lock);
1264 
1265 	sca_clear_ext_call(vcpu);
1266 }
1267 
1268 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1269 {
1270 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1271 	int rc = 0;
1272 	unsigned long irq_type;
1273 	unsigned long irqs;
1274 
1275 	__reset_intercept_indicators(vcpu);
1276 
1277 	/* pending ckc conditions might have been invalidated */
1278 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1279 	if (ckc_irq_pending(vcpu))
1280 		set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1281 
1282 	/* pending cpu timer conditions might have been invalidated */
1283 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1284 	if (cpu_timer_irq_pending(vcpu))
1285 		set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1286 
1287 	while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1288 		/* bits are in the reverse order of interrupt priority */
1289 		irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1290 		switch (irq_type) {
1291 		case IRQ_PEND_IO_ISC_0:
1292 		case IRQ_PEND_IO_ISC_1:
1293 		case IRQ_PEND_IO_ISC_2:
1294 		case IRQ_PEND_IO_ISC_3:
1295 		case IRQ_PEND_IO_ISC_4:
1296 		case IRQ_PEND_IO_ISC_5:
1297 		case IRQ_PEND_IO_ISC_6:
1298 		case IRQ_PEND_IO_ISC_7:
1299 			rc = __deliver_io(vcpu, irq_type);
1300 			break;
1301 		case IRQ_PEND_MCHK_EX:
1302 		case IRQ_PEND_MCHK_REP:
1303 			rc = __deliver_machine_check(vcpu);
1304 			break;
1305 		case IRQ_PEND_PROG:
1306 			rc = __deliver_prog(vcpu);
1307 			break;
1308 		case IRQ_PEND_EXT_EMERGENCY:
1309 			rc = __deliver_emergency_signal(vcpu);
1310 			break;
1311 		case IRQ_PEND_EXT_EXTERNAL:
1312 			rc = __deliver_external_call(vcpu);
1313 			break;
1314 		case IRQ_PEND_EXT_CLOCK_COMP:
1315 			rc = __deliver_ckc(vcpu);
1316 			break;
1317 		case IRQ_PEND_EXT_CPU_TIMER:
1318 			rc = __deliver_cpu_timer(vcpu);
1319 			break;
1320 		case IRQ_PEND_RESTART:
1321 			rc = __deliver_restart(vcpu);
1322 			break;
1323 		case IRQ_PEND_SET_PREFIX:
1324 			rc = __deliver_set_prefix(vcpu);
1325 			break;
1326 		case IRQ_PEND_PFAULT_INIT:
1327 			rc = __deliver_pfault_init(vcpu);
1328 			break;
1329 		case IRQ_PEND_EXT_SERVICE:
1330 			rc = __deliver_service(vcpu);
1331 			break;
1332 		case IRQ_PEND_PFAULT_DONE:
1333 			rc = __deliver_pfault_done(vcpu);
1334 			break;
1335 		case IRQ_PEND_VIRTIO:
1336 			rc = __deliver_virtio(vcpu);
1337 			break;
1338 		default:
1339 			WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1340 			clear_bit(irq_type, &li->pending_irqs);
1341 		}
1342 	}
1343 
1344 	set_intercept_indicators(vcpu);
1345 
1346 	return rc;
1347 }
1348 
1349 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1350 {
1351 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1352 
1353 	vcpu->stat.inject_program++;
1354 	VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1355 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1356 				   irq->u.pgm.code, 0);
1357 
1358 	if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1359 		/* auto detection if no valid ILC was given */
1360 		irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1361 		irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1362 		irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1363 	}
1364 
1365 	if (irq->u.pgm.code == PGM_PER) {
1366 		li->irq.pgm.code |= PGM_PER;
1367 		li->irq.pgm.flags = irq->u.pgm.flags;
1368 		/* only modify PER related information */
1369 		li->irq.pgm.per_address = irq->u.pgm.per_address;
1370 		li->irq.pgm.per_code = irq->u.pgm.per_code;
1371 		li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1372 		li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1373 	} else if (!(irq->u.pgm.code & PGM_PER)) {
1374 		li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1375 				   irq->u.pgm.code;
1376 		li->irq.pgm.flags = irq->u.pgm.flags;
1377 		/* only modify non-PER information */
1378 		li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1379 		li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1380 		li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1381 		li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1382 		li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1383 		li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1384 	} else {
1385 		li->irq.pgm = irq->u.pgm;
1386 	}
1387 	set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1388 	return 0;
1389 }
1390 
1391 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1392 {
1393 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1394 
1395 	vcpu->stat.inject_pfault_init++;
1396 	VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1397 		   irq->u.ext.ext_params2);
1398 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1399 				   irq->u.ext.ext_params,
1400 				   irq->u.ext.ext_params2);
1401 
1402 	li->irq.ext = irq->u.ext;
1403 	set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1404 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1405 	return 0;
1406 }
1407 
1408 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1409 {
1410 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1411 	struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1412 	uint16_t src_id = irq->u.extcall.code;
1413 
1414 	vcpu->stat.inject_external_call++;
1415 	VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1416 		   src_id);
1417 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1418 				   src_id, 0);
1419 
1420 	/* sending vcpu invalid */
1421 	if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1422 		return -EINVAL;
1423 
1424 	if (sclp.has_sigpif)
1425 		return sca_inject_ext_call(vcpu, src_id);
1426 
1427 	if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1428 		return -EBUSY;
1429 	*extcall = irq->u.extcall;
1430 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1431 	return 0;
1432 }
1433 
1434 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1435 {
1436 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1437 	struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1438 
1439 	vcpu->stat.inject_set_prefix++;
1440 	VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1441 		   irq->u.prefix.address);
1442 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1443 				   irq->u.prefix.address, 0);
1444 
1445 	if (!is_vcpu_stopped(vcpu))
1446 		return -EBUSY;
1447 
1448 	*prefix = irq->u.prefix;
1449 	set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1450 	return 0;
1451 }
1452 
1453 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1454 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1455 {
1456 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1457 	struct kvm_s390_stop_info *stop = &li->irq.stop;
1458 	int rc = 0;
1459 
1460 	vcpu->stat.inject_stop_signal++;
1461 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1462 
1463 	if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1464 		return -EINVAL;
1465 
1466 	if (is_vcpu_stopped(vcpu)) {
1467 		if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1468 			rc = kvm_s390_store_status_unloaded(vcpu,
1469 						KVM_S390_STORE_STATUS_NOADDR);
1470 		return rc;
1471 	}
1472 
1473 	if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1474 		return -EBUSY;
1475 	stop->flags = irq->u.stop.flags;
1476 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1477 	return 0;
1478 }
1479 
1480 static int __inject_sigp_restart(struct kvm_vcpu *vcpu)
1481 {
1482 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1483 
1484 	vcpu->stat.inject_restart++;
1485 	VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1486 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1487 
1488 	set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1489 	return 0;
1490 }
1491 
1492 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1493 				   struct kvm_s390_irq *irq)
1494 {
1495 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1496 
1497 	vcpu->stat.inject_emergency_signal++;
1498 	VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1499 		   irq->u.emerg.code);
1500 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1501 				   irq->u.emerg.code, 0);
1502 
1503 	/* sending vcpu invalid */
1504 	if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1505 		return -EINVAL;
1506 
1507 	set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1508 	set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1509 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1510 	return 0;
1511 }
1512 
1513 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1514 {
1515 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1516 	struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1517 
1518 	vcpu->stat.inject_mchk++;
1519 	VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1520 		   irq->u.mchk.mcic);
1521 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1522 				   irq->u.mchk.mcic);
1523 
1524 	/*
1525 	 * Because repressible machine checks can be indicated along with
1526 	 * exigent machine checks (PoP, Chapter 11, Interruption action)
1527 	 * we need to combine cr14, mcic and external damage code.
1528 	 * Failing storage address and the logout area should not be or'ed
1529 	 * together, we just indicate the last occurrence of the corresponding
1530 	 * machine check
1531 	 */
1532 	mchk->cr14 |= irq->u.mchk.cr14;
1533 	mchk->mcic |= irq->u.mchk.mcic;
1534 	mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1535 	mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1536 	memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1537 	       sizeof(mchk->fixed_logout));
1538 	if (mchk->mcic & MCHK_EX_MASK)
1539 		set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1540 	else if (mchk->mcic & MCHK_REP_MASK)
1541 		set_bit(IRQ_PEND_MCHK_REP,  &li->pending_irqs);
1542 	return 0;
1543 }
1544 
1545 static int __inject_ckc(struct kvm_vcpu *vcpu)
1546 {
1547 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1548 
1549 	vcpu->stat.inject_ckc++;
1550 	VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1551 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1552 				   0, 0);
1553 
1554 	set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1555 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1556 	return 0;
1557 }
1558 
1559 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1560 {
1561 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1562 
1563 	vcpu->stat.inject_cputm++;
1564 	VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1565 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1566 				   0, 0);
1567 
1568 	set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1569 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1570 	return 0;
1571 }
1572 
1573 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1574 						  int isc, u32 schid)
1575 {
1576 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1577 	struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1578 	struct kvm_s390_interrupt_info *iter;
1579 	u16 id = (schid & 0xffff0000U) >> 16;
1580 	u16 nr = schid & 0x0000ffffU;
1581 
1582 	spin_lock(&fi->lock);
1583 	list_for_each_entry(iter, isc_list, list) {
1584 		if (schid && (id != iter->io.subchannel_id ||
1585 			      nr != iter->io.subchannel_nr))
1586 			continue;
1587 		/* found an appropriate entry */
1588 		list_del_init(&iter->list);
1589 		fi->counters[FIRQ_CNTR_IO] -= 1;
1590 		if (list_empty(isc_list))
1591 			clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1592 		spin_unlock(&fi->lock);
1593 		return iter;
1594 	}
1595 	spin_unlock(&fi->lock);
1596 	return NULL;
1597 }
1598 
1599 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1600 						      u64 isc_mask, u32 schid)
1601 {
1602 	struct kvm_s390_interrupt_info *inti = NULL;
1603 	int isc;
1604 
1605 	for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1606 		if (isc_mask & isc_to_isc_bits(isc))
1607 			inti = get_io_int(kvm, isc, schid);
1608 	}
1609 	return inti;
1610 }
1611 
1612 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1613 {
1614 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1615 	unsigned long active_mask;
1616 	int isc;
1617 
1618 	if (schid)
1619 		goto out;
1620 	if (!gi->origin)
1621 		goto out;
1622 
1623 	active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1624 	while (active_mask) {
1625 		isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1626 		if (gisa_tac_ipm_gisc(gi->origin, isc))
1627 			return isc;
1628 		clear_bit_inv(isc, &active_mask);
1629 	}
1630 out:
1631 	return -EINVAL;
1632 }
1633 
1634 /*
1635  * Dequeue and return an I/O interrupt matching any of the interruption
1636  * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1637  * Take into account the interrupts pending in the interrupt list and in GISA.
1638  *
1639  * Note that for a guest that does not enable I/O interrupts
1640  * but relies on TPI, a flood of classic interrupts may starve
1641  * out adapter interrupts on the same isc. Linux does not do
1642  * that, and it is possible to work around the issue by configuring
1643  * different iscs for classic and adapter interrupts in the guest,
1644  * but we may want to revisit this in the future.
1645  */
1646 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1647 						    u64 isc_mask, u32 schid)
1648 {
1649 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1650 	struct kvm_s390_interrupt_info *inti, *tmp_inti;
1651 	int isc;
1652 
1653 	inti = get_top_io_int(kvm, isc_mask, schid);
1654 
1655 	isc = get_top_gisa_isc(kvm, isc_mask, schid);
1656 	if (isc < 0)
1657 		/* no AI in GISA */
1658 		goto out;
1659 
1660 	if (!inti)
1661 		/* AI in GISA but no classical IO int */
1662 		goto gisa_out;
1663 
1664 	/* both types of interrupts present */
1665 	if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1666 		/* classical IO int with higher priority */
1667 		gisa_set_ipm_gisc(gi->origin, isc);
1668 		goto out;
1669 	}
1670 gisa_out:
1671 	tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1672 	if (tmp_inti) {
1673 		tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1674 		tmp_inti->io.io_int_word = isc_to_int_word(isc);
1675 		if (inti)
1676 			kvm_s390_reinject_io_int(kvm, inti);
1677 		inti = tmp_inti;
1678 	} else
1679 		gisa_set_ipm_gisc(gi->origin, isc);
1680 out:
1681 	return inti;
1682 }
1683 
1684 #define SCCB_MASK 0xFFFFFFF8
1685 #define SCCB_EVENT_PENDING 0x3
1686 
1687 static int __inject_service(struct kvm *kvm,
1688 			     struct kvm_s390_interrupt_info *inti)
1689 {
1690 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1691 
1692 	kvm->stat.inject_service_signal++;
1693 	spin_lock(&fi->lock);
1694 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1695 	/*
1696 	 * Early versions of the QEMU s390 bios will inject several
1697 	 * service interrupts after another without handling a
1698 	 * condition code indicating busy.
1699 	 * We will silently ignore those superfluous sccb values.
1700 	 * A future version of QEMU will take care of serialization
1701 	 * of servc requests
1702 	 */
1703 	if (fi->srv_signal.ext_params & SCCB_MASK)
1704 		goto out;
1705 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1706 	set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1707 out:
1708 	spin_unlock(&fi->lock);
1709 	kfree(inti);
1710 	return 0;
1711 }
1712 
1713 static int __inject_virtio(struct kvm *kvm,
1714 			    struct kvm_s390_interrupt_info *inti)
1715 {
1716 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1717 
1718 	kvm->stat.inject_virtio++;
1719 	spin_lock(&fi->lock);
1720 	if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1721 		spin_unlock(&fi->lock);
1722 		return -EBUSY;
1723 	}
1724 	fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1725 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1726 	set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1727 	spin_unlock(&fi->lock);
1728 	return 0;
1729 }
1730 
1731 static int __inject_pfault_done(struct kvm *kvm,
1732 				 struct kvm_s390_interrupt_info *inti)
1733 {
1734 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1735 
1736 	kvm->stat.inject_pfault_done++;
1737 	spin_lock(&fi->lock);
1738 	if (fi->counters[FIRQ_CNTR_PFAULT] >=
1739 		(ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1740 		spin_unlock(&fi->lock);
1741 		return -EBUSY;
1742 	}
1743 	fi->counters[FIRQ_CNTR_PFAULT] += 1;
1744 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1745 	set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1746 	spin_unlock(&fi->lock);
1747 	return 0;
1748 }
1749 
1750 #define CR_PENDING_SUBCLASS 28
1751 static int __inject_float_mchk(struct kvm *kvm,
1752 				struct kvm_s390_interrupt_info *inti)
1753 {
1754 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1755 
1756 	kvm->stat.inject_float_mchk++;
1757 	spin_lock(&fi->lock);
1758 	fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1759 	fi->mchk.mcic |= inti->mchk.mcic;
1760 	set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1761 	spin_unlock(&fi->lock);
1762 	kfree(inti);
1763 	return 0;
1764 }
1765 
1766 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1767 {
1768 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1769 	struct kvm_s390_float_interrupt *fi;
1770 	struct list_head *list;
1771 	int isc;
1772 
1773 	kvm->stat.inject_io++;
1774 	isc = int_word_to_isc(inti->io.io_int_word);
1775 
1776 	if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1777 		VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1778 		gisa_set_ipm_gisc(gi->origin, isc);
1779 		kfree(inti);
1780 		return 0;
1781 	}
1782 
1783 	fi = &kvm->arch.float_int;
1784 	spin_lock(&fi->lock);
1785 	if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1786 		spin_unlock(&fi->lock);
1787 		return -EBUSY;
1788 	}
1789 	fi->counters[FIRQ_CNTR_IO] += 1;
1790 
1791 	if (inti->type & KVM_S390_INT_IO_AI_MASK)
1792 		VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1793 	else
1794 		VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1795 			inti->io.subchannel_id >> 8,
1796 			inti->io.subchannel_id >> 1 & 0x3,
1797 			inti->io.subchannel_nr);
1798 	list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1799 	list_add_tail(&inti->list, list);
1800 	set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1801 	spin_unlock(&fi->lock);
1802 	return 0;
1803 }
1804 
1805 /*
1806  * Find a destination VCPU for a floating irq and kick it.
1807  */
1808 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1809 {
1810 	struct kvm_vcpu *dst_vcpu;
1811 	int sigcpu, online_vcpus, nr_tries = 0;
1812 
1813 	online_vcpus = atomic_read(&kvm->online_vcpus);
1814 	if (!online_vcpus)
1815 		return;
1816 
1817 	/* find idle VCPUs first, then round robin */
1818 	sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1819 	if (sigcpu == online_vcpus) {
1820 		do {
1821 			sigcpu = kvm->arch.float_int.next_rr_cpu++;
1822 			kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1823 			/* avoid endless loops if all vcpus are stopped */
1824 			if (nr_tries++ >= online_vcpus)
1825 				return;
1826 		} while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1827 	}
1828 	dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1829 
1830 	/* make the VCPU drop out of the SIE, or wake it up if sleeping */
1831 	switch (type) {
1832 	case KVM_S390_MCHK:
1833 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1834 		break;
1835 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1836 		if (!(type & KVM_S390_INT_IO_AI_MASK &&
1837 		      kvm->arch.gisa_int.origin))
1838 			kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1839 		break;
1840 	default:
1841 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1842 		break;
1843 	}
1844 	kvm_s390_vcpu_wakeup(dst_vcpu);
1845 }
1846 
1847 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1848 {
1849 	u64 type = READ_ONCE(inti->type);
1850 	int rc;
1851 
1852 	switch (type) {
1853 	case KVM_S390_MCHK:
1854 		rc = __inject_float_mchk(kvm, inti);
1855 		break;
1856 	case KVM_S390_INT_VIRTIO:
1857 		rc = __inject_virtio(kvm, inti);
1858 		break;
1859 	case KVM_S390_INT_SERVICE:
1860 		rc = __inject_service(kvm, inti);
1861 		break;
1862 	case KVM_S390_INT_PFAULT_DONE:
1863 		rc = __inject_pfault_done(kvm, inti);
1864 		break;
1865 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1866 		rc = __inject_io(kvm, inti);
1867 		break;
1868 	default:
1869 		rc = -EINVAL;
1870 	}
1871 	if (rc)
1872 		return rc;
1873 
1874 	__floating_irq_kick(kvm, type);
1875 	return 0;
1876 }
1877 
1878 int kvm_s390_inject_vm(struct kvm *kvm,
1879 		       struct kvm_s390_interrupt *s390int)
1880 {
1881 	struct kvm_s390_interrupt_info *inti;
1882 	int rc;
1883 
1884 	inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1885 	if (!inti)
1886 		return -ENOMEM;
1887 
1888 	inti->type = s390int->type;
1889 	switch (inti->type) {
1890 	case KVM_S390_INT_VIRTIO:
1891 		VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
1892 			 s390int->parm, s390int->parm64);
1893 		inti->ext.ext_params = s390int->parm;
1894 		inti->ext.ext_params2 = s390int->parm64;
1895 		break;
1896 	case KVM_S390_INT_SERVICE:
1897 		VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
1898 		inti->ext.ext_params = s390int->parm;
1899 		break;
1900 	case KVM_S390_INT_PFAULT_DONE:
1901 		inti->ext.ext_params2 = s390int->parm64;
1902 		break;
1903 	case KVM_S390_MCHK:
1904 		VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
1905 			 s390int->parm64);
1906 		inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
1907 		inti->mchk.mcic = s390int->parm64;
1908 		break;
1909 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1910 		inti->io.subchannel_id = s390int->parm >> 16;
1911 		inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
1912 		inti->io.io_int_parm = s390int->parm64 >> 32;
1913 		inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
1914 		break;
1915 	default:
1916 		kfree(inti);
1917 		return -EINVAL;
1918 	}
1919 	trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
1920 				 2);
1921 
1922 	rc = __inject_vm(kvm, inti);
1923 	if (rc)
1924 		kfree(inti);
1925 	return rc;
1926 }
1927 
1928 int kvm_s390_reinject_io_int(struct kvm *kvm,
1929 			      struct kvm_s390_interrupt_info *inti)
1930 {
1931 	return __inject_vm(kvm, inti);
1932 }
1933 
1934 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
1935 		       struct kvm_s390_irq *irq)
1936 {
1937 	irq->type = s390int->type;
1938 	switch (irq->type) {
1939 	case KVM_S390_PROGRAM_INT:
1940 		if (s390int->parm & 0xffff0000)
1941 			return -EINVAL;
1942 		irq->u.pgm.code = s390int->parm;
1943 		break;
1944 	case KVM_S390_SIGP_SET_PREFIX:
1945 		irq->u.prefix.address = s390int->parm;
1946 		break;
1947 	case KVM_S390_SIGP_STOP:
1948 		irq->u.stop.flags = s390int->parm;
1949 		break;
1950 	case KVM_S390_INT_EXTERNAL_CALL:
1951 		if (s390int->parm & 0xffff0000)
1952 			return -EINVAL;
1953 		irq->u.extcall.code = s390int->parm;
1954 		break;
1955 	case KVM_S390_INT_EMERGENCY:
1956 		if (s390int->parm & 0xffff0000)
1957 			return -EINVAL;
1958 		irq->u.emerg.code = s390int->parm;
1959 		break;
1960 	case KVM_S390_MCHK:
1961 		irq->u.mchk.mcic = s390int->parm64;
1962 		break;
1963 	case KVM_S390_INT_PFAULT_INIT:
1964 		irq->u.ext.ext_params = s390int->parm;
1965 		irq->u.ext.ext_params2 = s390int->parm64;
1966 		break;
1967 	case KVM_S390_RESTART:
1968 	case KVM_S390_INT_CLOCK_COMP:
1969 	case KVM_S390_INT_CPU_TIMER:
1970 		break;
1971 	default:
1972 		return -EINVAL;
1973 	}
1974 	return 0;
1975 }
1976 
1977 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
1978 {
1979 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1980 
1981 	return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1982 }
1983 
1984 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
1985 {
1986 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1987 
1988 	spin_lock(&li->lock);
1989 	li->irq.stop.flags = 0;
1990 	clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1991 	spin_unlock(&li->lock);
1992 }
1993 
1994 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1995 {
1996 	int rc;
1997 
1998 	switch (irq->type) {
1999 	case KVM_S390_PROGRAM_INT:
2000 		rc = __inject_prog(vcpu, irq);
2001 		break;
2002 	case KVM_S390_SIGP_SET_PREFIX:
2003 		rc = __inject_set_prefix(vcpu, irq);
2004 		break;
2005 	case KVM_S390_SIGP_STOP:
2006 		rc = __inject_sigp_stop(vcpu, irq);
2007 		break;
2008 	case KVM_S390_RESTART:
2009 		rc = __inject_sigp_restart(vcpu);
2010 		break;
2011 	case KVM_S390_INT_CLOCK_COMP:
2012 		rc = __inject_ckc(vcpu);
2013 		break;
2014 	case KVM_S390_INT_CPU_TIMER:
2015 		rc = __inject_cpu_timer(vcpu);
2016 		break;
2017 	case KVM_S390_INT_EXTERNAL_CALL:
2018 		rc = __inject_extcall(vcpu, irq);
2019 		break;
2020 	case KVM_S390_INT_EMERGENCY:
2021 		rc = __inject_sigp_emergency(vcpu, irq);
2022 		break;
2023 	case KVM_S390_MCHK:
2024 		rc = __inject_mchk(vcpu, irq);
2025 		break;
2026 	case KVM_S390_INT_PFAULT_INIT:
2027 		rc = __inject_pfault_init(vcpu, irq);
2028 		break;
2029 	case KVM_S390_INT_VIRTIO:
2030 	case KVM_S390_INT_SERVICE:
2031 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2032 	default:
2033 		rc = -EINVAL;
2034 	}
2035 
2036 	return rc;
2037 }
2038 
2039 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2040 {
2041 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2042 	int rc;
2043 
2044 	spin_lock(&li->lock);
2045 	rc = do_inject_vcpu(vcpu, irq);
2046 	spin_unlock(&li->lock);
2047 	if (!rc)
2048 		kvm_s390_vcpu_wakeup(vcpu);
2049 	return rc;
2050 }
2051 
2052 static inline void clear_irq_list(struct list_head *_list)
2053 {
2054 	struct kvm_s390_interrupt_info *inti, *n;
2055 
2056 	list_for_each_entry_safe(inti, n, _list, list) {
2057 		list_del(&inti->list);
2058 		kfree(inti);
2059 	}
2060 }
2061 
2062 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2063 		       struct kvm_s390_irq *irq)
2064 {
2065 	irq->type = inti->type;
2066 	switch (inti->type) {
2067 	case KVM_S390_INT_PFAULT_INIT:
2068 	case KVM_S390_INT_PFAULT_DONE:
2069 	case KVM_S390_INT_VIRTIO:
2070 		irq->u.ext = inti->ext;
2071 		break;
2072 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2073 		irq->u.io = inti->io;
2074 		break;
2075 	}
2076 }
2077 
2078 void kvm_s390_clear_float_irqs(struct kvm *kvm)
2079 {
2080 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2081 	int i;
2082 
2083 	spin_lock(&fi->lock);
2084 	fi->pending_irqs = 0;
2085 	memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2086 	memset(&fi->mchk, 0, sizeof(fi->mchk));
2087 	for (i = 0; i < FIRQ_LIST_COUNT; i++)
2088 		clear_irq_list(&fi->lists[i]);
2089 	for (i = 0; i < FIRQ_MAX_COUNT; i++)
2090 		fi->counters[i] = 0;
2091 	spin_unlock(&fi->lock);
2092 	kvm_s390_gisa_clear(kvm);
2093 };
2094 
2095 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2096 {
2097 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2098 	struct kvm_s390_interrupt_info *inti;
2099 	struct kvm_s390_float_interrupt *fi;
2100 	struct kvm_s390_irq *buf;
2101 	struct kvm_s390_irq *irq;
2102 	int max_irqs;
2103 	int ret = 0;
2104 	int n = 0;
2105 	int i;
2106 
2107 	if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2108 		return -EINVAL;
2109 
2110 	/*
2111 	 * We are already using -ENOMEM to signal
2112 	 * userspace it may retry with a bigger buffer,
2113 	 * so we need to use something else for this case
2114 	 */
2115 	buf = vzalloc(len);
2116 	if (!buf)
2117 		return -ENOBUFS;
2118 
2119 	max_irqs = len / sizeof(struct kvm_s390_irq);
2120 
2121 	if (gi->origin && gisa_get_ipm(gi->origin)) {
2122 		for (i = 0; i <= MAX_ISC; i++) {
2123 			if (n == max_irqs) {
2124 				/* signal userspace to try again */
2125 				ret = -ENOMEM;
2126 				goto out_nolock;
2127 			}
2128 			if (gisa_tac_ipm_gisc(gi->origin, i)) {
2129 				irq = (struct kvm_s390_irq *) &buf[n];
2130 				irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2131 				irq->u.io.io_int_word = isc_to_int_word(i);
2132 				n++;
2133 			}
2134 		}
2135 	}
2136 	fi = &kvm->arch.float_int;
2137 	spin_lock(&fi->lock);
2138 	for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2139 		list_for_each_entry(inti, &fi->lists[i], list) {
2140 			if (n == max_irqs) {
2141 				/* signal userspace to try again */
2142 				ret = -ENOMEM;
2143 				goto out;
2144 			}
2145 			inti_to_irq(inti, &buf[n]);
2146 			n++;
2147 		}
2148 	}
2149 	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs)) {
2150 		if (n == max_irqs) {
2151 			/* signal userspace to try again */
2152 			ret = -ENOMEM;
2153 			goto out;
2154 		}
2155 		irq = (struct kvm_s390_irq *) &buf[n];
2156 		irq->type = KVM_S390_INT_SERVICE;
2157 		irq->u.ext = fi->srv_signal;
2158 		n++;
2159 	}
2160 	if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2161 		if (n == max_irqs) {
2162 				/* signal userspace to try again */
2163 				ret = -ENOMEM;
2164 				goto out;
2165 		}
2166 		irq = (struct kvm_s390_irq *) &buf[n];
2167 		irq->type = KVM_S390_MCHK;
2168 		irq->u.mchk = fi->mchk;
2169 		n++;
2170 }
2171 
2172 out:
2173 	spin_unlock(&fi->lock);
2174 out_nolock:
2175 	if (!ret && n > 0) {
2176 		if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2177 			ret = -EFAULT;
2178 	}
2179 	vfree(buf);
2180 
2181 	return ret < 0 ? ret : n;
2182 }
2183 
2184 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2185 {
2186 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2187 	struct kvm_s390_ais_all ais;
2188 
2189 	if (attr->attr < sizeof(ais))
2190 		return -EINVAL;
2191 
2192 	if (!test_kvm_facility(kvm, 72))
2193 		return -EOPNOTSUPP;
2194 
2195 	mutex_lock(&fi->ais_lock);
2196 	ais.simm = fi->simm;
2197 	ais.nimm = fi->nimm;
2198 	mutex_unlock(&fi->ais_lock);
2199 
2200 	if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2201 		return -EFAULT;
2202 
2203 	return 0;
2204 }
2205 
2206 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2207 {
2208 	int r;
2209 
2210 	switch (attr->group) {
2211 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2212 		r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2213 					  attr->attr);
2214 		break;
2215 	case KVM_DEV_FLIC_AISM_ALL:
2216 		r = flic_ais_mode_get_all(dev->kvm, attr);
2217 		break;
2218 	default:
2219 		r = -EINVAL;
2220 	}
2221 
2222 	return r;
2223 }
2224 
2225 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2226 				     u64 addr)
2227 {
2228 	struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2229 	void *target = NULL;
2230 	void __user *source;
2231 	u64 size;
2232 
2233 	if (get_user(inti->type, (u64 __user *)addr))
2234 		return -EFAULT;
2235 
2236 	switch (inti->type) {
2237 	case KVM_S390_INT_PFAULT_INIT:
2238 	case KVM_S390_INT_PFAULT_DONE:
2239 	case KVM_S390_INT_VIRTIO:
2240 	case KVM_S390_INT_SERVICE:
2241 		target = (void *) &inti->ext;
2242 		source = &uptr->u.ext;
2243 		size = sizeof(inti->ext);
2244 		break;
2245 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2246 		target = (void *) &inti->io;
2247 		source = &uptr->u.io;
2248 		size = sizeof(inti->io);
2249 		break;
2250 	case KVM_S390_MCHK:
2251 		target = (void *) &inti->mchk;
2252 		source = &uptr->u.mchk;
2253 		size = sizeof(inti->mchk);
2254 		break;
2255 	default:
2256 		return -EINVAL;
2257 	}
2258 
2259 	if (copy_from_user(target, source, size))
2260 		return -EFAULT;
2261 
2262 	return 0;
2263 }
2264 
2265 static int enqueue_floating_irq(struct kvm_device *dev,
2266 				struct kvm_device_attr *attr)
2267 {
2268 	struct kvm_s390_interrupt_info *inti = NULL;
2269 	int r = 0;
2270 	int len = attr->attr;
2271 
2272 	if (len % sizeof(struct kvm_s390_irq) != 0)
2273 		return -EINVAL;
2274 	else if (len > KVM_S390_FLIC_MAX_BUFFER)
2275 		return -EINVAL;
2276 
2277 	while (len >= sizeof(struct kvm_s390_irq)) {
2278 		inti = kzalloc(sizeof(*inti), GFP_KERNEL);
2279 		if (!inti)
2280 			return -ENOMEM;
2281 
2282 		r = copy_irq_from_user(inti, attr->addr);
2283 		if (r) {
2284 			kfree(inti);
2285 			return r;
2286 		}
2287 		r = __inject_vm(dev->kvm, inti);
2288 		if (r) {
2289 			kfree(inti);
2290 			return r;
2291 		}
2292 		len -= sizeof(struct kvm_s390_irq);
2293 		attr->addr += sizeof(struct kvm_s390_irq);
2294 	}
2295 
2296 	return r;
2297 }
2298 
2299 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2300 {
2301 	if (id >= MAX_S390_IO_ADAPTERS)
2302 		return NULL;
2303 	id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2304 	return kvm->arch.adapters[id];
2305 }
2306 
2307 static int register_io_adapter(struct kvm_device *dev,
2308 			       struct kvm_device_attr *attr)
2309 {
2310 	struct s390_io_adapter *adapter;
2311 	struct kvm_s390_io_adapter adapter_info;
2312 
2313 	if (copy_from_user(&adapter_info,
2314 			   (void __user *)attr->addr, sizeof(adapter_info)))
2315 		return -EFAULT;
2316 
2317 	if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2318 		return -EINVAL;
2319 
2320 	adapter_info.id = array_index_nospec(adapter_info.id,
2321 					     MAX_S390_IO_ADAPTERS);
2322 
2323 	if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2324 		return -EINVAL;
2325 
2326 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2327 	if (!adapter)
2328 		return -ENOMEM;
2329 
2330 	INIT_LIST_HEAD(&adapter->maps);
2331 	init_rwsem(&adapter->maps_lock);
2332 	atomic_set(&adapter->nr_maps, 0);
2333 	adapter->id = adapter_info.id;
2334 	adapter->isc = adapter_info.isc;
2335 	adapter->maskable = adapter_info.maskable;
2336 	adapter->masked = false;
2337 	adapter->swap = adapter_info.swap;
2338 	adapter->suppressible = (adapter_info.flags) &
2339 				KVM_S390_ADAPTER_SUPPRESSIBLE;
2340 	dev->kvm->arch.adapters[adapter->id] = adapter;
2341 
2342 	return 0;
2343 }
2344 
2345 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2346 {
2347 	int ret;
2348 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2349 
2350 	if (!adapter || !adapter->maskable)
2351 		return -EINVAL;
2352 	ret = adapter->masked;
2353 	adapter->masked = masked;
2354 	return ret;
2355 }
2356 
2357 static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr)
2358 {
2359 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2360 	struct s390_map_info *map;
2361 	int ret;
2362 
2363 	if (!adapter || !addr)
2364 		return -EINVAL;
2365 
2366 	map = kzalloc(sizeof(*map), GFP_KERNEL);
2367 	if (!map) {
2368 		ret = -ENOMEM;
2369 		goto out;
2370 	}
2371 	INIT_LIST_HEAD(&map->list);
2372 	map->guest_addr = addr;
2373 	map->addr = gmap_translate(kvm->arch.gmap, addr);
2374 	if (map->addr == -EFAULT) {
2375 		ret = -EFAULT;
2376 		goto out;
2377 	}
2378 	ret = get_user_pages_fast(map->addr, 1, FOLL_WRITE, &map->page);
2379 	if (ret < 0)
2380 		goto out;
2381 	BUG_ON(ret != 1);
2382 	down_write(&adapter->maps_lock);
2383 	if (atomic_inc_return(&adapter->nr_maps) < MAX_S390_ADAPTER_MAPS) {
2384 		list_add_tail(&map->list, &adapter->maps);
2385 		ret = 0;
2386 	} else {
2387 		put_page(map->page);
2388 		ret = -EINVAL;
2389 	}
2390 	up_write(&adapter->maps_lock);
2391 out:
2392 	if (ret)
2393 		kfree(map);
2394 	return ret;
2395 }
2396 
2397 static int kvm_s390_adapter_unmap(struct kvm *kvm, unsigned int id, __u64 addr)
2398 {
2399 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2400 	struct s390_map_info *map, *tmp;
2401 	int found = 0;
2402 
2403 	if (!adapter || !addr)
2404 		return -EINVAL;
2405 
2406 	down_write(&adapter->maps_lock);
2407 	list_for_each_entry_safe(map, tmp, &adapter->maps, list) {
2408 		if (map->guest_addr == addr) {
2409 			found = 1;
2410 			atomic_dec(&adapter->nr_maps);
2411 			list_del(&map->list);
2412 			put_page(map->page);
2413 			kfree(map);
2414 			break;
2415 		}
2416 	}
2417 	up_write(&adapter->maps_lock);
2418 
2419 	return found ? 0 : -EINVAL;
2420 }
2421 
2422 void kvm_s390_destroy_adapters(struct kvm *kvm)
2423 {
2424 	int i;
2425 	struct s390_map_info *map, *tmp;
2426 
2427 	for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) {
2428 		if (!kvm->arch.adapters[i])
2429 			continue;
2430 		list_for_each_entry_safe(map, tmp,
2431 					 &kvm->arch.adapters[i]->maps, list) {
2432 			list_del(&map->list);
2433 			put_page(map->page);
2434 			kfree(map);
2435 		}
2436 		kfree(kvm->arch.adapters[i]);
2437 	}
2438 }
2439 
2440 static int modify_io_adapter(struct kvm_device *dev,
2441 			     struct kvm_device_attr *attr)
2442 {
2443 	struct kvm_s390_io_adapter_req req;
2444 	struct s390_io_adapter *adapter;
2445 	int ret;
2446 
2447 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2448 		return -EFAULT;
2449 
2450 	adapter = get_io_adapter(dev->kvm, req.id);
2451 	if (!adapter)
2452 		return -EINVAL;
2453 	switch (req.type) {
2454 	case KVM_S390_IO_ADAPTER_MASK:
2455 		ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2456 		if (ret > 0)
2457 			ret = 0;
2458 		break;
2459 	case KVM_S390_IO_ADAPTER_MAP:
2460 		ret = kvm_s390_adapter_map(dev->kvm, req.id, req.addr);
2461 		break;
2462 	case KVM_S390_IO_ADAPTER_UNMAP:
2463 		ret = kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr);
2464 		break;
2465 	default:
2466 		ret = -EINVAL;
2467 	}
2468 
2469 	return ret;
2470 }
2471 
2472 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2473 
2474 {
2475 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2476 	u32 schid;
2477 
2478 	if (attr->flags)
2479 		return -EINVAL;
2480 	if (attr->attr != sizeof(schid))
2481 		return -EINVAL;
2482 	if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2483 		return -EFAULT;
2484 	if (!schid)
2485 		return -EINVAL;
2486 	kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2487 	/*
2488 	 * If userspace is conforming to the architecture, we can have at most
2489 	 * one pending I/O interrupt per subchannel, so this is effectively a
2490 	 * clear all.
2491 	 */
2492 	return 0;
2493 }
2494 
2495 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2496 {
2497 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2498 	struct kvm_s390_ais_req req;
2499 	int ret = 0;
2500 
2501 	if (!test_kvm_facility(kvm, 72))
2502 		return -EOPNOTSUPP;
2503 
2504 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2505 		return -EFAULT;
2506 
2507 	if (req.isc > MAX_ISC)
2508 		return -EINVAL;
2509 
2510 	trace_kvm_s390_modify_ais_mode(req.isc,
2511 				       (fi->simm & AIS_MODE_MASK(req.isc)) ?
2512 				       (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2513 				       2 : KVM_S390_AIS_MODE_SINGLE :
2514 				       KVM_S390_AIS_MODE_ALL, req.mode);
2515 
2516 	mutex_lock(&fi->ais_lock);
2517 	switch (req.mode) {
2518 	case KVM_S390_AIS_MODE_ALL:
2519 		fi->simm &= ~AIS_MODE_MASK(req.isc);
2520 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2521 		break;
2522 	case KVM_S390_AIS_MODE_SINGLE:
2523 		fi->simm |= AIS_MODE_MASK(req.isc);
2524 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2525 		break;
2526 	default:
2527 		ret = -EINVAL;
2528 	}
2529 	mutex_unlock(&fi->ais_lock);
2530 
2531 	return ret;
2532 }
2533 
2534 static int kvm_s390_inject_airq(struct kvm *kvm,
2535 				struct s390_io_adapter *adapter)
2536 {
2537 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2538 	struct kvm_s390_interrupt s390int = {
2539 		.type = KVM_S390_INT_IO(1, 0, 0, 0),
2540 		.parm = 0,
2541 		.parm64 = isc_to_int_word(adapter->isc),
2542 	};
2543 	int ret = 0;
2544 
2545 	if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2546 		return kvm_s390_inject_vm(kvm, &s390int);
2547 
2548 	mutex_lock(&fi->ais_lock);
2549 	if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2550 		trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2551 		goto out;
2552 	}
2553 
2554 	ret = kvm_s390_inject_vm(kvm, &s390int);
2555 	if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2556 		fi->nimm |= AIS_MODE_MASK(adapter->isc);
2557 		trace_kvm_s390_modify_ais_mode(adapter->isc,
2558 					       KVM_S390_AIS_MODE_SINGLE, 2);
2559 	}
2560 out:
2561 	mutex_unlock(&fi->ais_lock);
2562 	return ret;
2563 }
2564 
2565 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2566 {
2567 	unsigned int id = attr->attr;
2568 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2569 
2570 	if (!adapter)
2571 		return -EINVAL;
2572 
2573 	return kvm_s390_inject_airq(kvm, adapter);
2574 }
2575 
2576 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2577 {
2578 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2579 	struct kvm_s390_ais_all ais;
2580 
2581 	if (!test_kvm_facility(kvm, 72))
2582 		return -EOPNOTSUPP;
2583 
2584 	if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2585 		return -EFAULT;
2586 
2587 	mutex_lock(&fi->ais_lock);
2588 	fi->simm = ais.simm;
2589 	fi->nimm = ais.nimm;
2590 	mutex_unlock(&fi->ais_lock);
2591 
2592 	return 0;
2593 }
2594 
2595 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2596 {
2597 	int r = 0;
2598 	unsigned int i;
2599 	struct kvm_vcpu *vcpu;
2600 
2601 	switch (attr->group) {
2602 	case KVM_DEV_FLIC_ENQUEUE:
2603 		r = enqueue_floating_irq(dev, attr);
2604 		break;
2605 	case KVM_DEV_FLIC_CLEAR_IRQS:
2606 		kvm_s390_clear_float_irqs(dev->kvm);
2607 		break;
2608 	case KVM_DEV_FLIC_APF_ENABLE:
2609 		dev->kvm->arch.gmap->pfault_enabled = 1;
2610 		break;
2611 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2612 		dev->kvm->arch.gmap->pfault_enabled = 0;
2613 		/*
2614 		 * Make sure no async faults are in transition when
2615 		 * clearing the queues. So we don't need to worry
2616 		 * about late coming workers.
2617 		 */
2618 		synchronize_srcu(&dev->kvm->srcu);
2619 		kvm_for_each_vcpu(i, vcpu, dev->kvm)
2620 			kvm_clear_async_pf_completion_queue(vcpu);
2621 		break;
2622 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2623 		r = register_io_adapter(dev, attr);
2624 		break;
2625 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2626 		r = modify_io_adapter(dev, attr);
2627 		break;
2628 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2629 		r = clear_io_irq(dev->kvm, attr);
2630 		break;
2631 	case KVM_DEV_FLIC_AISM:
2632 		r = modify_ais_mode(dev->kvm, attr);
2633 		break;
2634 	case KVM_DEV_FLIC_AIRQ_INJECT:
2635 		r = flic_inject_airq(dev->kvm, attr);
2636 		break;
2637 	case KVM_DEV_FLIC_AISM_ALL:
2638 		r = flic_ais_mode_set_all(dev->kvm, attr);
2639 		break;
2640 	default:
2641 		r = -EINVAL;
2642 	}
2643 
2644 	return r;
2645 }
2646 
2647 static int flic_has_attr(struct kvm_device *dev,
2648 			     struct kvm_device_attr *attr)
2649 {
2650 	switch (attr->group) {
2651 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2652 	case KVM_DEV_FLIC_ENQUEUE:
2653 	case KVM_DEV_FLIC_CLEAR_IRQS:
2654 	case KVM_DEV_FLIC_APF_ENABLE:
2655 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2656 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2657 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2658 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2659 	case KVM_DEV_FLIC_AISM:
2660 	case KVM_DEV_FLIC_AIRQ_INJECT:
2661 	case KVM_DEV_FLIC_AISM_ALL:
2662 		return 0;
2663 	}
2664 	return -ENXIO;
2665 }
2666 
2667 static int flic_create(struct kvm_device *dev, u32 type)
2668 {
2669 	if (!dev)
2670 		return -EINVAL;
2671 	if (dev->kvm->arch.flic)
2672 		return -EINVAL;
2673 	dev->kvm->arch.flic = dev;
2674 	return 0;
2675 }
2676 
2677 static void flic_destroy(struct kvm_device *dev)
2678 {
2679 	dev->kvm->arch.flic = NULL;
2680 	kfree(dev);
2681 }
2682 
2683 /* s390 floating irq controller (flic) */
2684 struct kvm_device_ops kvm_flic_ops = {
2685 	.name = "kvm-flic",
2686 	.get_attr = flic_get_attr,
2687 	.set_attr = flic_set_attr,
2688 	.has_attr = flic_has_attr,
2689 	.create = flic_create,
2690 	.destroy = flic_destroy,
2691 };
2692 
2693 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2694 {
2695 	unsigned long bit;
2696 
2697 	bit = bit_nr + (addr % PAGE_SIZE) * 8;
2698 
2699 	return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2700 }
2701 
2702 static struct s390_map_info *get_map_info(struct s390_io_adapter *adapter,
2703 					  u64 addr)
2704 {
2705 	struct s390_map_info *map;
2706 
2707 	if (!adapter)
2708 		return NULL;
2709 
2710 	list_for_each_entry(map, &adapter->maps, list) {
2711 		if (map->guest_addr == addr)
2712 			return map;
2713 	}
2714 	return NULL;
2715 }
2716 
2717 static int adapter_indicators_set(struct kvm *kvm,
2718 				  struct s390_io_adapter *adapter,
2719 				  struct kvm_s390_adapter_int *adapter_int)
2720 {
2721 	unsigned long bit;
2722 	int summary_set, idx;
2723 	struct s390_map_info *info;
2724 	void *map;
2725 
2726 	info = get_map_info(adapter, adapter_int->ind_addr);
2727 	if (!info)
2728 		return -1;
2729 	map = page_address(info->page);
2730 	bit = get_ind_bit(info->addr, adapter_int->ind_offset, adapter->swap);
2731 	set_bit(bit, map);
2732 	idx = srcu_read_lock(&kvm->srcu);
2733 	mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2734 	set_page_dirty_lock(info->page);
2735 	info = get_map_info(adapter, adapter_int->summary_addr);
2736 	if (!info) {
2737 		srcu_read_unlock(&kvm->srcu, idx);
2738 		return -1;
2739 	}
2740 	map = page_address(info->page);
2741 	bit = get_ind_bit(info->addr, adapter_int->summary_offset,
2742 			  adapter->swap);
2743 	summary_set = test_and_set_bit(bit, map);
2744 	mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2745 	set_page_dirty_lock(info->page);
2746 	srcu_read_unlock(&kvm->srcu, idx);
2747 	return summary_set ? 0 : 1;
2748 }
2749 
2750 /*
2751  * < 0 - not injected due to error
2752  * = 0 - coalesced, summary indicator already active
2753  * > 0 - injected interrupt
2754  */
2755 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2756 			   struct kvm *kvm, int irq_source_id, int level,
2757 			   bool line_status)
2758 {
2759 	int ret;
2760 	struct s390_io_adapter *adapter;
2761 
2762 	/* We're only interested in the 0->1 transition. */
2763 	if (!level)
2764 		return 0;
2765 	adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2766 	if (!adapter)
2767 		return -1;
2768 	down_read(&adapter->maps_lock);
2769 	ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2770 	up_read(&adapter->maps_lock);
2771 	if ((ret > 0) && !adapter->masked) {
2772 		ret = kvm_s390_inject_airq(kvm, adapter);
2773 		if (ret == 0)
2774 			ret = 1;
2775 	}
2776 	return ret;
2777 }
2778 
2779 /*
2780  * Inject the machine check to the guest.
2781  */
2782 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2783 				     struct mcck_volatile_info *mcck_info)
2784 {
2785 	struct kvm_s390_interrupt_info inti;
2786 	struct kvm_s390_irq irq;
2787 	struct kvm_s390_mchk_info *mchk;
2788 	union mci mci;
2789 	__u64 cr14 = 0;         /* upper bits are not used */
2790 	int rc;
2791 
2792 	mci.val = mcck_info->mcic;
2793 	if (mci.sr)
2794 		cr14 |= CR14_RECOVERY_SUBMASK;
2795 	if (mci.dg)
2796 		cr14 |= CR14_DEGRADATION_SUBMASK;
2797 	if (mci.w)
2798 		cr14 |= CR14_WARNING_SUBMASK;
2799 
2800 	mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2801 	mchk->cr14 = cr14;
2802 	mchk->mcic = mcck_info->mcic;
2803 	mchk->ext_damage_code = mcck_info->ext_damage_code;
2804 	mchk->failing_storage_address = mcck_info->failing_storage_address;
2805 	if (mci.ck) {
2806 		/* Inject the floating machine check */
2807 		inti.type = KVM_S390_MCHK;
2808 		rc = __inject_vm(vcpu->kvm, &inti);
2809 	} else {
2810 		/* Inject the machine check to specified vcpu */
2811 		irq.type = KVM_S390_MCHK;
2812 		rc = kvm_s390_inject_vcpu(vcpu, &irq);
2813 	}
2814 	WARN_ON_ONCE(rc);
2815 }
2816 
2817 int kvm_set_routing_entry(struct kvm *kvm,
2818 			  struct kvm_kernel_irq_routing_entry *e,
2819 			  const struct kvm_irq_routing_entry *ue)
2820 {
2821 	int ret;
2822 
2823 	switch (ue->type) {
2824 	case KVM_IRQ_ROUTING_S390_ADAPTER:
2825 		e->set = set_adapter_int;
2826 		e->adapter.summary_addr = ue->u.adapter.summary_addr;
2827 		e->adapter.ind_addr = ue->u.adapter.ind_addr;
2828 		e->adapter.summary_offset = ue->u.adapter.summary_offset;
2829 		e->adapter.ind_offset = ue->u.adapter.ind_offset;
2830 		e->adapter.adapter_id = ue->u.adapter.adapter_id;
2831 		ret = 0;
2832 		break;
2833 	default:
2834 		ret = -EINVAL;
2835 	}
2836 
2837 	return ret;
2838 }
2839 
2840 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2841 		int irq_source_id, int level, bool line_status)
2842 {
2843 	return -EINVAL;
2844 }
2845 
2846 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2847 {
2848 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2849 	struct kvm_s390_irq *buf;
2850 	int r = 0;
2851 	int n;
2852 
2853 	buf = vmalloc(len);
2854 	if (!buf)
2855 		return -ENOMEM;
2856 
2857 	if (copy_from_user((void *) buf, irqstate, len)) {
2858 		r = -EFAULT;
2859 		goto out_free;
2860 	}
2861 
2862 	/*
2863 	 * Don't allow setting the interrupt state
2864 	 * when there are already interrupts pending
2865 	 */
2866 	spin_lock(&li->lock);
2867 	if (li->pending_irqs) {
2868 		r = -EBUSY;
2869 		goto out_unlock;
2870 	}
2871 
2872 	for (n = 0; n < len / sizeof(*buf); n++) {
2873 		r = do_inject_vcpu(vcpu, &buf[n]);
2874 		if (r)
2875 			break;
2876 	}
2877 
2878 out_unlock:
2879 	spin_unlock(&li->lock);
2880 out_free:
2881 	vfree(buf);
2882 
2883 	return r;
2884 }
2885 
2886 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2887 			    struct kvm_s390_irq *irq,
2888 			    unsigned long irq_type)
2889 {
2890 	switch (irq_type) {
2891 	case IRQ_PEND_MCHK_EX:
2892 	case IRQ_PEND_MCHK_REP:
2893 		irq->type = KVM_S390_MCHK;
2894 		irq->u.mchk = li->irq.mchk;
2895 		break;
2896 	case IRQ_PEND_PROG:
2897 		irq->type = KVM_S390_PROGRAM_INT;
2898 		irq->u.pgm = li->irq.pgm;
2899 		break;
2900 	case IRQ_PEND_PFAULT_INIT:
2901 		irq->type = KVM_S390_INT_PFAULT_INIT;
2902 		irq->u.ext = li->irq.ext;
2903 		break;
2904 	case IRQ_PEND_EXT_EXTERNAL:
2905 		irq->type = KVM_S390_INT_EXTERNAL_CALL;
2906 		irq->u.extcall = li->irq.extcall;
2907 		break;
2908 	case IRQ_PEND_EXT_CLOCK_COMP:
2909 		irq->type = KVM_S390_INT_CLOCK_COMP;
2910 		break;
2911 	case IRQ_PEND_EXT_CPU_TIMER:
2912 		irq->type = KVM_S390_INT_CPU_TIMER;
2913 		break;
2914 	case IRQ_PEND_SIGP_STOP:
2915 		irq->type = KVM_S390_SIGP_STOP;
2916 		irq->u.stop = li->irq.stop;
2917 		break;
2918 	case IRQ_PEND_RESTART:
2919 		irq->type = KVM_S390_RESTART;
2920 		break;
2921 	case IRQ_PEND_SET_PREFIX:
2922 		irq->type = KVM_S390_SIGP_SET_PREFIX;
2923 		irq->u.prefix = li->irq.prefix;
2924 		break;
2925 	}
2926 }
2927 
2928 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
2929 {
2930 	int scn;
2931 	DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
2932 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2933 	unsigned long pending_irqs;
2934 	struct kvm_s390_irq irq;
2935 	unsigned long irq_type;
2936 	int cpuaddr;
2937 	int n = 0;
2938 
2939 	spin_lock(&li->lock);
2940 	pending_irqs = li->pending_irqs;
2941 	memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
2942 	       sizeof(sigp_emerg_pending));
2943 	spin_unlock(&li->lock);
2944 
2945 	for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
2946 		memset(&irq, 0, sizeof(irq));
2947 		if (irq_type == IRQ_PEND_EXT_EMERGENCY)
2948 			continue;
2949 		if (n + sizeof(irq) > len)
2950 			return -ENOBUFS;
2951 		store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
2952 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2953 			return -EFAULT;
2954 		n += sizeof(irq);
2955 	}
2956 
2957 	if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
2958 		for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
2959 			memset(&irq, 0, sizeof(irq));
2960 			if (n + sizeof(irq) > len)
2961 				return -ENOBUFS;
2962 			irq.type = KVM_S390_INT_EMERGENCY;
2963 			irq.u.emerg.code = cpuaddr;
2964 			if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2965 				return -EFAULT;
2966 			n += sizeof(irq);
2967 		}
2968 	}
2969 
2970 	if (sca_ext_call_pending(vcpu, &scn)) {
2971 		if (n + sizeof(irq) > len)
2972 			return -ENOBUFS;
2973 		memset(&irq, 0, sizeof(irq));
2974 		irq.type = KVM_S390_INT_EXTERNAL_CALL;
2975 		irq.u.extcall.code = scn;
2976 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2977 			return -EFAULT;
2978 		n += sizeof(irq);
2979 	}
2980 
2981 	return n;
2982 }
2983 
2984 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
2985 {
2986 	int vcpu_id, online_vcpus = atomic_read(&kvm->online_vcpus);
2987 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2988 	struct kvm_vcpu *vcpu;
2989 
2990 	for_each_set_bit(vcpu_id, kvm->arch.idle_mask, online_vcpus) {
2991 		vcpu = kvm_get_vcpu(kvm, vcpu_id);
2992 		if (psw_ioint_disabled(vcpu))
2993 			continue;
2994 		deliverable_mask &= (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
2995 		if (deliverable_mask) {
2996 			/* lately kicked but not yet running */
2997 			if (test_and_set_bit(vcpu_id, gi->kicked_mask))
2998 				return;
2999 			kvm_s390_vcpu_wakeup(vcpu);
3000 			return;
3001 		}
3002 	}
3003 }
3004 
3005 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3006 {
3007 	struct kvm_s390_gisa_interrupt *gi =
3008 		container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3009 	struct kvm *kvm =
3010 		container_of(gi->origin, struct sie_page2, gisa)->kvm;
3011 	u8 pending_mask;
3012 
3013 	pending_mask = gisa_get_ipm_or_restore_iam(gi);
3014 	if (pending_mask) {
3015 		__airqs_kick_single_vcpu(kvm, pending_mask);
3016 		hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3017 		return HRTIMER_RESTART;
3018 	};
3019 
3020 	return HRTIMER_NORESTART;
3021 }
3022 
3023 #define NULL_GISA_ADDR 0x00000000UL
3024 #define NONE_GISA_ADDR 0x00000001UL
3025 #define GISA_ADDR_MASK 0xfffff000UL
3026 
3027 static void process_gib_alert_list(void)
3028 {
3029 	struct kvm_s390_gisa_interrupt *gi;
3030 	struct kvm_s390_gisa *gisa;
3031 	struct kvm *kvm;
3032 	u32 final, origin = 0UL;
3033 
3034 	do {
3035 		/*
3036 		 * If the NONE_GISA_ADDR is still stored in the alert list
3037 		 * origin, we will leave the outer loop. No further GISA has
3038 		 * been added to the alert list by millicode while processing
3039 		 * the current alert list.
3040 		 */
3041 		final = (origin & NONE_GISA_ADDR);
3042 		/*
3043 		 * Cut off the alert list and store the NONE_GISA_ADDR in the
3044 		 * alert list origin to avoid further GAL interruptions.
3045 		 * A new alert list can be build up by millicode in parallel
3046 		 * for guests not in the yet cut-off alert list. When in the
3047 		 * final loop, store the NULL_GISA_ADDR instead. This will re-
3048 		 * enable GAL interruptions on the host again.
3049 		 */
3050 		origin = xchg(&gib->alert_list_origin,
3051 			      (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3052 		/*
3053 		 * Loop through the just cut-off alert list and start the
3054 		 * gisa timers to kick idle vcpus to consume the pending
3055 		 * interruptions asap.
3056 		 */
3057 		while (origin & GISA_ADDR_MASK) {
3058 			gisa = (struct kvm_s390_gisa *)(u64)origin;
3059 			origin = gisa->next_alert;
3060 			gisa->next_alert = (u32)(u64)gisa;
3061 			kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3062 			gi = &kvm->arch.gisa_int;
3063 			if (hrtimer_active(&gi->timer))
3064 				hrtimer_cancel(&gi->timer);
3065 			hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3066 		}
3067 	} while (!final);
3068 
3069 }
3070 
3071 void kvm_s390_gisa_clear(struct kvm *kvm)
3072 {
3073 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3074 
3075 	if (!gi->origin)
3076 		return;
3077 	gisa_clear_ipm(gi->origin);
3078 	VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3079 }
3080 
3081 void kvm_s390_gisa_init(struct kvm *kvm)
3082 {
3083 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3084 
3085 	if (!css_general_characteristics.aiv)
3086 		return;
3087 	gi->origin = &kvm->arch.sie_page2->gisa;
3088 	gi->alert.mask = 0;
3089 	spin_lock_init(&gi->alert.ref_lock);
3090 	gi->expires = 50 * 1000; /* 50 usec */
3091 	hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3092 	gi->timer.function = gisa_vcpu_kicker;
3093 	memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3094 	gi->origin->next_alert = (u32)(u64)gi->origin;
3095 	VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3096 }
3097 
3098 void kvm_s390_gisa_destroy(struct kvm *kvm)
3099 {
3100 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3101 
3102 	if (!gi->origin)
3103 		return;
3104 	if (gi->alert.mask)
3105 		KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
3106 			  kvm, gi->alert.mask);
3107 	while (gisa_in_alert_list(gi->origin))
3108 		cpu_relax();
3109 	hrtimer_cancel(&gi->timer);
3110 	gi->origin = NULL;
3111 }
3112 
3113 /**
3114  * kvm_s390_gisc_register - register a guest ISC
3115  *
3116  * @kvm:  the kernel vm to work with
3117  * @gisc: the guest interruption sub class to register
3118  *
3119  * The function extends the vm specific alert mask to use.
3120  * The effective IAM mask in the GISA is updated as well
3121  * in case the GISA is not part of the GIB alert list.
3122  * It will be updated latest when the IAM gets restored
3123  * by gisa_get_ipm_or_restore_iam().
3124  *
3125  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3126  *          has registered with the channel subsystem.
3127  *          -ENODEV in case the vm uses no GISA
3128  *          -ERANGE in case the guest ISC is invalid
3129  */
3130 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3131 {
3132 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3133 
3134 	if (!gi->origin)
3135 		return -ENODEV;
3136 	if (gisc > MAX_ISC)
3137 		return -ERANGE;
3138 
3139 	spin_lock(&gi->alert.ref_lock);
3140 	gi->alert.ref_count[gisc]++;
3141 	if (gi->alert.ref_count[gisc] == 1) {
3142 		gi->alert.mask |= 0x80 >> gisc;
3143 		gisa_set_iam(gi->origin, gi->alert.mask);
3144 	}
3145 	spin_unlock(&gi->alert.ref_lock);
3146 
3147 	return gib->nisc;
3148 }
3149 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3150 
3151 /**
3152  * kvm_s390_gisc_unregister - unregister a guest ISC
3153  *
3154  * @kvm:  the kernel vm to work with
3155  * @gisc: the guest interruption sub class to register
3156  *
3157  * The function reduces the vm specific alert mask to use.
3158  * The effective IAM mask in the GISA is updated as well
3159  * in case the GISA is not part of the GIB alert list.
3160  * It will be updated latest when the IAM gets restored
3161  * by gisa_get_ipm_or_restore_iam().
3162  *
3163  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3164  *          has registered with the channel subsystem.
3165  *          -ENODEV in case the vm uses no GISA
3166  *          -ERANGE in case the guest ISC is invalid
3167  *          -EINVAL in case the guest ISC is not registered
3168  */
3169 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3170 {
3171 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3172 	int rc = 0;
3173 
3174 	if (!gi->origin)
3175 		return -ENODEV;
3176 	if (gisc > MAX_ISC)
3177 		return -ERANGE;
3178 
3179 	spin_lock(&gi->alert.ref_lock);
3180 	if (gi->alert.ref_count[gisc] == 0) {
3181 		rc = -EINVAL;
3182 		goto out;
3183 	}
3184 	gi->alert.ref_count[gisc]--;
3185 	if (gi->alert.ref_count[gisc] == 0) {
3186 		gi->alert.mask &= ~(0x80 >> gisc);
3187 		gisa_set_iam(gi->origin, gi->alert.mask);
3188 	}
3189 out:
3190 	spin_unlock(&gi->alert.ref_lock);
3191 
3192 	return rc;
3193 }
3194 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3195 
3196 static void gib_alert_irq_handler(struct airq_struct *airq, bool floating)
3197 {
3198 	inc_irq_stat(IRQIO_GAL);
3199 	process_gib_alert_list();
3200 }
3201 
3202 static struct airq_struct gib_alert_irq = {
3203 	.handler = gib_alert_irq_handler,
3204 	.lsi_ptr = &gib_alert_irq.lsi_mask,
3205 };
3206 
3207 void kvm_s390_gib_destroy(void)
3208 {
3209 	if (!gib)
3210 		return;
3211 	chsc_sgib(0);
3212 	unregister_adapter_interrupt(&gib_alert_irq);
3213 	free_page((unsigned long)gib);
3214 	gib = NULL;
3215 }
3216 
3217 int kvm_s390_gib_init(u8 nisc)
3218 {
3219 	int rc = 0;
3220 
3221 	if (!css_general_characteristics.aiv) {
3222 		KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3223 		goto out;
3224 	}
3225 
3226 	gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL | GFP_DMA);
3227 	if (!gib) {
3228 		rc = -ENOMEM;
3229 		goto out;
3230 	}
3231 
3232 	gib_alert_irq.isc = nisc;
3233 	if (register_adapter_interrupt(&gib_alert_irq)) {
3234 		pr_err("Registering the GIB alert interruption handler failed\n");
3235 		rc = -EIO;
3236 		goto out_free_gib;
3237 	}
3238 
3239 	gib->nisc = nisc;
3240 	if (chsc_sgib((u32)(u64)gib)) {
3241 		pr_err("Associating the GIB with the AIV facility failed\n");
3242 		free_page((unsigned long)gib);
3243 		gib = NULL;
3244 		rc = -EIO;
3245 		goto out_unreg_gal;
3246 	}
3247 
3248 	KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3249 	goto out;
3250 
3251 out_unreg_gal:
3252 	unregister_adapter_interrupt(&gib_alert_irq);
3253 out_free_gib:
3254 	free_page((unsigned long)gib);
3255 	gib = NULL;
3256 out:
3257 	return rc;
3258 }
3259