xref: /openbmc/linux/arch/s390/kvm/interrupt.c (revision 8f8d5745bb520c76b81abef4a2cb3023d0313bfd)
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/signal.h>
18 #include <linux/slab.h>
19 #include <linux/bitmap.h>
20 #include <linux/vmalloc.h>
21 #include <asm/asm-offsets.h>
22 #include <asm/dis.h>
23 #include <linux/uaccess.h>
24 #include <asm/sclp.h>
25 #include <asm/isc.h>
26 #include <asm/gmap.h>
27 #include <asm/switch_to.h>
28 #include <asm/nmi.h>
29 #include <asm/airq.h>
30 #include "kvm-s390.h"
31 #include "gaccess.h"
32 #include "trace-s390.h"
33 
34 #define PFAULT_INIT 0x0600
35 #define PFAULT_DONE 0x0680
36 #define VIRTIO_PARAM 0x0d00
37 
38 static struct kvm_s390_gib *gib;
39 
40 /* handle external calls via sigp interpretation facility */
41 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
42 {
43 	int c, scn;
44 
45 	if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND))
46 		return 0;
47 
48 	BUG_ON(!kvm_s390_use_sca_entries());
49 	read_lock(&vcpu->kvm->arch.sca_lock);
50 	if (vcpu->kvm->arch.use_esca) {
51 		struct esca_block *sca = vcpu->kvm->arch.sca;
52 		union esca_sigp_ctrl sigp_ctrl =
53 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
54 
55 		c = sigp_ctrl.c;
56 		scn = sigp_ctrl.scn;
57 	} else {
58 		struct bsca_block *sca = vcpu->kvm->arch.sca;
59 		union bsca_sigp_ctrl sigp_ctrl =
60 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
61 
62 		c = sigp_ctrl.c;
63 		scn = sigp_ctrl.scn;
64 	}
65 	read_unlock(&vcpu->kvm->arch.sca_lock);
66 
67 	if (src_id)
68 		*src_id = scn;
69 
70 	return c;
71 }
72 
73 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
74 {
75 	int expect, rc;
76 
77 	BUG_ON(!kvm_s390_use_sca_entries());
78 	read_lock(&vcpu->kvm->arch.sca_lock);
79 	if (vcpu->kvm->arch.use_esca) {
80 		struct esca_block *sca = vcpu->kvm->arch.sca;
81 		union esca_sigp_ctrl *sigp_ctrl =
82 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
83 		union esca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
84 
85 		new_val.scn = src_id;
86 		new_val.c = 1;
87 		old_val.c = 0;
88 
89 		expect = old_val.value;
90 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
91 	} else {
92 		struct bsca_block *sca = vcpu->kvm->arch.sca;
93 		union bsca_sigp_ctrl *sigp_ctrl =
94 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
95 		union bsca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
96 
97 		new_val.scn = src_id;
98 		new_val.c = 1;
99 		old_val.c = 0;
100 
101 		expect = old_val.value;
102 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
103 	}
104 	read_unlock(&vcpu->kvm->arch.sca_lock);
105 
106 	if (rc != expect) {
107 		/* another external call is pending */
108 		return -EBUSY;
109 	}
110 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
111 	return 0;
112 }
113 
114 static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
115 {
116 	int rc, expect;
117 
118 	if (!kvm_s390_use_sca_entries())
119 		return;
120 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
121 	read_lock(&vcpu->kvm->arch.sca_lock);
122 	if (vcpu->kvm->arch.use_esca) {
123 		struct esca_block *sca = vcpu->kvm->arch.sca;
124 		union esca_sigp_ctrl *sigp_ctrl =
125 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
126 		union esca_sigp_ctrl old = *sigp_ctrl;
127 
128 		expect = old.value;
129 		rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
130 	} else {
131 		struct bsca_block *sca = vcpu->kvm->arch.sca;
132 		union bsca_sigp_ctrl *sigp_ctrl =
133 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
134 		union bsca_sigp_ctrl old = *sigp_ctrl;
135 
136 		expect = old.value;
137 		rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
138 	}
139 	read_unlock(&vcpu->kvm->arch.sca_lock);
140 	WARN_ON(rc != expect); /* cannot clear? */
141 }
142 
143 int psw_extint_disabled(struct kvm_vcpu *vcpu)
144 {
145 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
146 }
147 
148 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
149 {
150 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
151 }
152 
153 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
154 {
155 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
156 }
157 
158 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
159 {
160 	return psw_extint_disabled(vcpu) &&
161 	       psw_ioint_disabled(vcpu) &&
162 	       psw_mchk_disabled(vcpu);
163 }
164 
165 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
166 {
167 	if (psw_extint_disabled(vcpu) ||
168 	    !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
169 		return 0;
170 	if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
171 		/* No timer interrupts when single stepping */
172 		return 0;
173 	return 1;
174 }
175 
176 static int ckc_irq_pending(struct kvm_vcpu *vcpu)
177 {
178 	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
179 	const u64 ckc = vcpu->arch.sie_block->ckc;
180 
181 	if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
182 		if ((s64)ckc >= (s64)now)
183 			return 0;
184 	} else if (ckc >= now) {
185 		return 0;
186 	}
187 	return ckc_interrupts_enabled(vcpu);
188 }
189 
190 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
191 {
192 	return !psw_extint_disabled(vcpu) &&
193 	       (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK);
194 }
195 
196 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
197 {
198 	if (!cpu_timer_interrupts_enabled(vcpu))
199 		return 0;
200 	return kvm_s390_get_cpu_timer(vcpu) >> 63;
201 }
202 
203 static uint64_t isc_to_isc_bits(int isc)
204 {
205 	return (0x80 >> isc) << 24;
206 }
207 
208 static inline u32 isc_to_int_word(u8 isc)
209 {
210 	return ((u32)isc << 27) | 0x80000000;
211 }
212 
213 static inline u8 int_word_to_isc(u32 int_word)
214 {
215 	return (int_word & 0x38000000) >> 27;
216 }
217 
218 /*
219  * To use atomic bitmap functions, we have to provide a bitmap address
220  * that is u64 aligned. However, the ipm might be u32 aligned.
221  * Therefore, we logically start the bitmap at the very beginning of the
222  * struct and fixup the bit number.
223  */
224 #define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE)
225 
226 /**
227  * gisa_set_iam - change the GISA interruption alert mask
228  *
229  * @gisa: gisa to operate on
230  * @iam: new IAM value to use
231  *
232  * Change the IAM atomically with the next alert address and the IPM
233  * of the GISA if the GISA is not part of the GIB alert list. All three
234  * fields are located in the first long word of the GISA.
235  *
236  * Returns: 0 on success
237  *          -EBUSY in case the gisa is part of the alert list
238  */
239 static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam)
240 {
241 	u64 word, _word;
242 
243 	do {
244 		word = READ_ONCE(gisa->u64.word[0]);
245 		if ((u64)gisa != word >> 32)
246 			return -EBUSY;
247 		_word = (word & ~0xffUL) | iam;
248 	} while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
249 
250 	return 0;
251 }
252 
253 /**
254  * gisa_clear_ipm - clear the GISA interruption pending mask
255  *
256  * @gisa: gisa to operate on
257  *
258  * Clear the IPM atomically with the next alert address and the IAM
259  * of the GISA unconditionally. All three fields are located in the
260  * first long word of the GISA.
261  */
262 static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa)
263 {
264 	u64 word, _word;
265 
266 	do {
267 		word = READ_ONCE(gisa->u64.word[0]);
268 		_word = word & ~(0xffUL << 24);
269 	} while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
270 }
271 
272 /**
273  * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM
274  *
275  * @gi: gisa interrupt struct to work on
276  *
277  * Atomically restores the interruption alert mask if none of the
278  * relevant ISCs are pending and return the IPM.
279  *
280  * Returns: the relevant pending ISCs
281  */
282 static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
283 {
284 	u8 pending_mask, alert_mask;
285 	u64 word, _word;
286 
287 	do {
288 		word = READ_ONCE(gi->origin->u64.word[0]);
289 		alert_mask = READ_ONCE(gi->alert.mask);
290 		pending_mask = (u8)(word >> 24) & alert_mask;
291 		if (pending_mask)
292 			return pending_mask;
293 		_word = (word & ~0xffUL) | alert_mask;
294 	} while (cmpxchg(&gi->origin->u64.word[0], word, _word) != word);
295 
296 	return 0;
297 }
298 
299 static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
300 {
301 	return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa;
302 }
303 
304 static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
305 {
306 	set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
307 }
308 
309 static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
310 {
311 	return READ_ONCE(gisa->ipm);
312 }
313 
314 static inline void gisa_clear_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
315 {
316 	clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
317 }
318 
319 static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
320 {
321 	return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
322 }
323 
324 static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
325 {
326 	return vcpu->kvm->arch.float_int.pending_irqs |
327 		vcpu->arch.local_int.pending_irqs;
328 }
329 
330 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
331 {
332 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
333 	unsigned long pending_mask;
334 
335 	pending_mask = pending_irqs_no_gisa(vcpu);
336 	if (gi->origin)
337 		pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
338 	return pending_mask;
339 }
340 
341 static inline int isc_to_irq_type(unsigned long isc)
342 {
343 	return IRQ_PEND_IO_ISC_0 - isc;
344 }
345 
346 static inline int irq_type_to_isc(unsigned long irq_type)
347 {
348 	return IRQ_PEND_IO_ISC_0 - irq_type;
349 }
350 
351 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
352 				   unsigned long active_mask)
353 {
354 	int i;
355 
356 	for (i = 0; i <= MAX_ISC; i++)
357 		if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
358 			active_mask &= ~(1UL << (isc_to_irq_type(i)));
359 
360 	return active_mask;
361 }
362 
363 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
364 {
365 	unsigned long active_mask;
366 
367 	active_mask = pending_irqs(vcpu);
368 	if (!active_mask)
369 		return 0;
370 
371 	if (psw_extint_disabled(vcpu))
372 		active_mask &= ~IRQ_PEND_EXT_MASK;
373 	if (psw_ioint_disabled(vcpu))
374 		active_mask &= ~IRQ_PEND_IO_MASK;
375 	else
376 		active_mask = disable_iscs(vcpu, active_mask);
377 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
378 		__clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
379 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
380 		__clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
381 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
382 		__clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
383 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
384 		__clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
385 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
386 		__clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
387 	if (psw_mchk_disabled(vcpu))
388 		active_mask &= ~IRQ_PEND_MCHK_MASK;
389 	/*
390 	 * Check both floating and local interrupt's cr14 because
391 	 * bit IRQ_PEND_MCHK_REP could be set in both cases.
392 	 */
393 	if (!(vcpu->arch.sie_block->gcr[14] &
394 	   (vcpu->kvm->arch.float_int.mchk.cr14 |
395 	   vcpu->arch.local_int.irq.mchk.cr14)))
396 		__clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
397 
398 	/*
399 	 * STOP irqs will never be actively delivered. They are triggered via
400 	 * intercept requests and cleared when the stop intercept is performed.
401 	 */
402 	__clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
403 
404 	return active_mask;
405 }
406 
407 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
408 {
409 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
410 	set_bit(vcpu->vcpu_id, vcpu->kvm->arch.idle_mask);
411 }
412 
413 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
414 {
415 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
416 	clear_bit(vcpu->vcpu_id, vcpu->kvm->arch.idle_mask);
417 }
418 
419 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
420 {
421 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
422 				      CPUSTAT_STOP_INT);
423 	vcpu->arch.sie_block->lctl = 0x0000;
424 	vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
425 
426 	if (guestdbg_enabled(vcpu)) {
427 		vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
428 					       LCTL_CR10 | LCTL_CR11);
429 		vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
430 	}
431 }
432 
433 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
434 {
435 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
436 		return;
437 	if (psw_ioint_disabled(vcpu))
438 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
439 	else
440 		vcpu->arch.sie_block->lctl |= LCTL_CR6;
441 }
442 
443 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
444 {
445 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
446 		return;
447 	if (psw_extint_disabled(vcpu))
448 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
449 	else
450 		vcpu->arch.sie_block->lctl |= LCTL_CR0;
451 }
452 
453 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
454 {
455 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
456 		return;
457 	if (psw_mchk_disabled(vcpu))
458 		vcpu->arch.sie_block->ictl |= ICTL_LPSW;
459 	else
460 		vcpu->arch.sie_block->lctl |= LCTL_CR14;
461 }
462 
463 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
464 {
465 	if (kvm_s390_is_stop_irq_pending(vcpu))
466 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
467 }
468 
469 /* Set interception request for non-deliverable interrupts */
470 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
471 {
472 	set_intercept_indicators_io(vcpu);
473 	set_intercept_indicators_ext(vcpu);
474 	set_intercept_indicators_mchk(vcpu);
475 	set_intercept_indicators_stop(vcpu);
476 }
477 
478 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
479 {
480 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
481 	int rc;
482 
483 	vcpu->stat.deliver_cputm++;
484 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
485 					 0, 0);
486 
487 	rc  = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
488 			   (u16 *)__LC_EXT_INT_CODE);
489 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
490 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
491 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
492 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
493 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
494 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
495 	return rc ? -EFAULT : 0;
496 }
497 
498 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
499 {
500 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
501 	int rc;
502 
503 	vcpu->stat.deliver_ckc++;
504 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
505 					 0, 0);
506 
507 	rc  = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
508 			   (u16 __user *)__LC_EXT_INT_CODE);
509 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
510 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
511 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
512 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
513 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
514 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
515 	return rc ? -EFAULT : 0;
516 }
517 
518 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
519 {
520 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
521 	struct kvm_s390_ext_info ext;
522 	int rc;
523 
524 	spin_lock(&li->lock);
525 	ext = li->irq.ext;
526 	clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
527 	li->irq.ext.ext_params2 = 0;
528 	spin_unlock(&li->lock);
529 
530 	VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
531 		   ext.ext_params2);
532 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
533 					 KVM_S390_INT_PFAULT_INIT,
534 					 0, ext.ext_params2);
535 
536 	rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
537 	rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
538 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
539 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
540 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
541 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
542 	rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
543 	return rc ? -EFAULT : 0;
544 }
545 
546 static int __write_machine_check(struct kvm_vcpu *vcpu,
547 				 struct kvm_s390_mchk_info *mchk)
548 {
549 	unsigned long ext_sa_addr;
550 	unsigned long lc;
551 	freg_t fprs[NUM_FPRS];
552 	union mci mci;
553 	int rc;
554 
555 	mci.val = mchk->mcic;
556 	/* take care of lazy register loading */
557 	save_fpu_regs();
558 	save_access_regs(vcpu->run->s.regs.acrs);
559 	if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
560 		save_gs_cb(current->thread.gs_cb);
561 
562 	/* Extended save area */
563 	rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
564 			   sizeof(unsigned long));
565 	/* Only bits 0 through 63-LC are used for address formation */
566 	lc = ext_sa_addr & MCESA_LC_MASK;
567 	if (test_kvm_facility(vcpu->kvm, 133)) {
568 		switch (lc) {
569 		case 0:
570 		case 10:
571 			ext_sa_addr &= ~0x3ffUL;
572 			break;
573 		case 11:
574 			ext_sa_addr &= ~0x7ffUL;
575 			break;
576 		case 12:
577 			ext_sa_addr &= ~0xfffUL;
578 			break;
579 		default:
580 			ext_sa_addr = 0;
581 			break;
582 		}
583 	} else {
584 		ext_sa_addr &= ~0x3ffUL;
585 	}
586 
587 	if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
588 		if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
589 				    512))
590 			mci.vr = 0;
591 	} else {
592 		mci.vr = 0;
593 	}
594 	if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
595 	    && (lc == 11 || lc == 12)) {
596 		if (write_guest_abs(vcpu, ext_sa_addr + 1024,
597 				    &vcpu->run->s.regs.gscb, 32))
598 			mci.gs = 0;
599 	} else {
600 		mci.gs = 0;
601 	}
602 
603 	/* General interruption information */
604 	rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
605 	rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
606 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
607 	rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
608 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
609 	rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
610 
611 	/* Register-save areas */
612 	if (MACHINE_HAS_VX) {
613 		convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
614 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
615 	} else {
616 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
617 				     vcpu->run->s.regs.fprs, 128);
618 	}
619 	rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
620 			     vcpu->run->s.regs.gprs, 128);
621 	rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
622 			   (u32 __user *) __LC_FP_CREG_SAVE_AREA);
623 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
624 			   (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
625 	rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
626 			   (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
627 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
628 			   (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
629 	rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
630 			     &vcpu->run->s.regs.acrs, 64);
631 	rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
632 			     &vcpu->arch.sie_block->gcr, 128);
633 
634 	/* Extended interruption information */
635 	rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
636 			   (u32 __user *) __LC_EXT_DAMAGE_CODE);
637 	rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
638 			   (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
639 	rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
640 			     sizeof(mchk->fixed_logout));
641 	return rc ? -EFAULT : 0;
642 }
643 
644 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
645 {
646 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
647 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
648 	struct kvm_s390_mchk_info mchk = {};
649 	int deliver = 0;
650 	int rc = 0;
651 
652 	spin_lock(&fi->lock);
653 	spin_lock(&li->lock);
654 	if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
655 	    test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
656 		/*
657 		 * If there was an exigent machine check pending, then any
658 		 * repressible machine checks that might have been pending
659 		 * are indicated along with it, so always clear bits for
660 		 * repressible and exigent interrupts
661 		 */
662 		mchk = li->irq.mchk;
663 		clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
664 		clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
665 		memset(&li->irq.mchk, 0, sizeof(mchk));
666 		deliver = 1;
667 	}
668 	/*
669 	 * We indicate floating repressible conditions along with
670 	 * other pending conditions. Channel Report Pending and Channel
671 	 * Subsystem damage are the only two and and are indicated by
672 	 * bits in mcic and masked in cr14.
673 	 */
674 	if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
675 		mchk.mcic |= fi->mchk.mcic;
676 		mchk.cr14 |= fi->mchk.cr14;
677 		memset(&fi->mchk, 0, sizeof(mchk));
678 		deliver = 1;
679 	}
680 	spin_unlock(&li->lock);
681 	spin_unlock(&fi->lock);
682 
683 	if (deliver) {
684 		VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
685 			   mchk.mcic);
686 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
687 						 KVM_S390_MCHK,
688 						 mchk.cr14, mchk.mcic);
689 		vcpu->stat.deliver_machine_check++;
690 		rc = __write_machine_check(vcpu, &mchk);
691 	}
692 	return rc;
693 }
694 
695 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
696 {
697 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
698 	int rc;
699 
700 	VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
701 	vcpu->stat.deliver_restart_signal++;
702 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
703 
704 	rc  = write_guest_lc(vcpu,
705 			     offsetof(struct lowcore, restart_old_psw),
706 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
707 	rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
708 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
709 	clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
710 	return rc ? -EFAULT : 0;
711 }
712 
713 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
714 {
715 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
716 	struct kvm_s390_prefix_info prefix;
717 
718 	spin_lock(&li->lock);
719 	prefix = li->irq.prefix;
720 	li->irq.prefix.address = 0;
721 	clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
722 	spin_unlock(&li->lock);
723 
724 	vcpu->stat.deliver_prefix_signal++;
725 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
726 					 KVM_S390_SIGP_SET_PREFIX,
727 					 prefix.address, 0);
728 
729 	kvm_s390_set_prefix(vcpu, prefix.address);
730 	return 0;
731 }
732 
733 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
734 {
735 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
736 	int rc;
737 	int cpu_addr;
738 
739 	spin_lock(&li->lock);
740 	cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
741 	clear_bit(cpu_addr, li->sigp_emerg_pending);
742 	if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
743 		clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
744 	spin_unlock(&li->lock);
745 
746 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
747 	vcpu->stat.deliver_emergency_signal++;
748 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
749 					 cpu_addr, 0);
750 
751 	rc  = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
752 			   (u16 *)__LC_EXT_INT_CODE);
753 	rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
754 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
755 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
756 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
757 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
758 	return rc ? -EFAULT : 0;
759 }
760 
761 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
762 {
763 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
764 	struct kvm_s390_extcall_info extcall;
765 	int rc;
766 
767 	spin_lock(&li->lock);
768 	extcall = li->irq.extcall;
769 	li->irq.extcall.code = 0;
770 	clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
771 	spin_unlock(&li->lock);
772 
773 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
774 	vcpu->stat.deliver_external_call++;
775 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
776 					 KVM_S390_INT_EXTERNAL_CALL,
777 					 extcall.code, 0);
778 
779 	rc  = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
780 			   (u16 *)__LC_EXT_INT_CODE);
781 	rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
782 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
783 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
784 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
785 			    sizeof(psw_t));
786 	return rc ? -EFAULT : 0;
787 }
788 
789 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
790 {
791 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
792 	struct kvm_s390_pgm_info pgm_info;
793 	int rc = 0, nullifying = false;
794 	u16 ilen;
795 
796 	spin_lock(&li->lock);
797 	pgm_info = li->irq.pgm;
798 	clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
799 	memset(&li->irq.pgm, 0, sizeof(pgm_info));
800 	spin_unlock(&li->lock);
801 
802 	ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
803 	VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
804 		   pgm_info.code, ilen);
805 	vcpu->stat.deliver_program++;
806 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
807 					 pgm_info.code, 0);
808 
809 	switch (pgm_info.code & ~PGM_PER) {
810 	case PGM_AFX_TRANSLATION:
811 	case PGM_ASX_TRANSLATION:
812 	case PGM_EX_TRANSLATION:
813 	case PGM_LFX_TRANSLATION:
814 	case PGM_LSTE_SEQUENCE:
815 	case PGM_LSX_TRANSLATION:
816 	case PGM_LX_TRANSLATION:
817 	case PGM_PRIMARY_AUTHORITY:
818 	case PGM_SECONDARY_AUTHORITY:
819 		nullifying = true;
820 		/* fall through */
821 	case PGM_SPACE_SWITCH:
822 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
823 				  (u64 *)__LC_TRANS_EXC_CODE);
824 		break;
825 	case PGM_ALEN_TRANSLATION:
826 	case PGM_ALE_SEQUENCE:
827 	case PGM_ASTE_INSTANCE:
828 	case PGM_ASTE_SEQUENCE:
829 	case PGM_ASTE_VALIDITY:
830 	case PGM_EXTENDED_AUTHORITY:
831 		rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
832 				  (u8 *)__LC_EXC_ACCESS_ID);
833 		nullifying = true;
834 		break;
835 	case PGM_ASCE_TYPE:
836 	case PGM_PAGE_TRANSLATION:
837 	case PGM_REGION_FIRST_TRANS:
838 	case PGM_REGION_SECOND_TRANS:
839 	case PGM_REGION_THIRD_TRANS:
840 	case PGM_SEGMENT_TRANSLATION:
841 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
842 				  (u64 *)__LC_TRANS_EXC_CODE);
843 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
844 				   (u8 *)__LC_EXC_ACCESS_ID);
845 		rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
846 				   (u8 *)__LC_OP_ACCESS_ID);
847 		nullifying = true;
848 		break;
849 	case PGM_MONITOR:
850 		rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
851 				  (u16 *)__LC_MON_CLASS_NR);
852 		rc |= put_guest_lc(vcpu, pgm_info.mon_code,
853 				   (u64 *)__LC_MON_CODE);
854 		break;
855 	case PGM_VECTOR_PROCESSING:
856 	case PGM_DATA:
857 		rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
858 				  (u32 *)__LC_DATA_EXC_CODE);
859 		break;
860 	case PGM_PROTECTION:
861 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
862 				  (u64 *)__LC_TRANS_EXC_CODE);
863 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
864 				   (u8 *)__LC_EXC_ACCESS_ID);
865 		break;
866 	case PGM_STACK_FULL:
867 	case PGM_STACK_EMPTY:
868 	case PGM_STACK_SPECIFICATION:
869 	case PGM_STACK_TYPE:
870 	case PGM_STACK_OPERATION:
871 	case PGM_TRACE_TABEL:
872 	case PGM_CRYPTO_OPERATION:
873 		nullifying = true;
874 		break;
875 	}
876 
877 	if (pgm_info.code & PGM_PER) {
878 		rc |= put_guest_lc(vcpu, pgm_info.per_code,
879 				   (u8 *) __LC_PER_CODE);
880 		rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
881 				   (u8 *)__LC_PER_ATMID);
882 		rc |= put_guest_lc(vcpu, pgm_info.per_address,
883 				   (u64 *) __LC_PER_ADDRESS);
884 		rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
885 				   (u8 *) __LC_PER_ACCESS_ID);
886 	}
887 
888 	if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
889 		kvm_s390_rewind_psw(vcpu, ilen);
890 
891 	/* bit 1+2 of the target are the ilc, so we can directly use ilen */
892 	rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
893 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
894 				 (u64 *) __LC_LAST_BREAK);
895 	rc |= put_guest_lc(vcpu, pgm_info.code,
896 			   (u16 *)__LC_PGM_INT_CODE);
897 	rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
898 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
899 	rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
900 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
901 	return rc ? -EFAULT : 0;
902 }
903 
904 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
905 {
906 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
907 	struct kvm_s390_ext_info ext;
908 	int rc = 0;
909 
910 	spin_lock(&fi->lock);
911 	if (!(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
912 		spin_unlock(&fi->lock);
913 		return 0;
914 	}
915 	ext = fi->srv_signal;
916 	memset(&fi->srv_signal, 0, sizeof(ext));
917 	clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
918 	spin_unlock(&fi->lock);
919 
920 	VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
921 		   ext.ext_params);
922 	vcpu->stat.deliver_service_signal++;
923 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
924 					 ext.ext_params, 0);
925 
926 	rc  = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
927 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
928 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
929 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
930 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
931 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
932 	rc |= put_guest_lc(vcpu, ext.ext_params,
933 			   (u32 *)__LC_EXT_PARAMS);
934 
935 	return rc ? -EFAULT : 0;
936 }
937 
938 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
939 {
940 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
941 	struct kvm_s390_interrupt_info *inti;
942 	int rc = 0;
943 
944 	spin_lock(&fi->lock);
945 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
946 					struct kvm_s390_interrupt_info,
947 					list);
948 	if (inti) {
949 		list_del(&inti->list);
950 		fi->counters[FIRQ_CNTR_PFAULT] -= 1;
951 	}
952 	if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
953 		clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
954 	spin_unlock(&fi->lock);
955 
956 	if (inti) {
957 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
958 						 KVM_S390_INT_PFAULT_DONE, 0,
959 						 inti->ext.ext_params2);
960 		VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
961 			   inti->ext.ext_params2);
962 
963 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
964 				(u16 *)__LC_EXT_INT_CODE);
965 		rc |= put_guest_lc(vcpu, PFAULT_DONE,
966 				(u16 *)__LC_EXT_CPU_ADDR);
967 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
968 				&vcpu->arch.sie_block->gpsw,
969 				sizeof(psw_t));
970 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
971 				&vcpu->arch.sie_block->gpsw,
972 				sizeof(psw_t));
973 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
974 				(u64 *)__LC_EXT_PARAMS2);
975 		kfree(inti);
976 	}
977 	return rc ? -EFAULT : 0;
978 }
979 
980 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
981 {
982 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
983 	struct kvm_s390_interrupt_info *inti;
984 	int rc = 0;
985 
986 	spin_lock(&fi->lock);
987 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
988 					struct kvm_s390_interrupt_info,
989 					list);
990 	if (inti) {
991 		VCPU_EVENT(vcpu, 4,
992 			   "deliver: virtio parm: 0x%x,parm64: 0x%llx",
993 			   inti->ext.ext_params, inti->ext.ext_params2);
994 		vcpu->stat.deliver_virtio++;
995 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
996 				inti->type,
997 				inti->ext.ext_params,
998 				inti->ext.ext_params2);
999 		list_del(&inti->list);
1000 		fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1001 	}
1002 	if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1003 		clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1004 	spin_unlock(&fi->lock);
1005 
1006 	if (inti) {
1007 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1008 				(u16 *)__LC_EXT_INT_CODE);
1009 		rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1010 				(u16 *)__LC_EXT_CPU_ADDR);
1011 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1012 				&vcpu->arch.sie_block->gpsw,
1013 				sizeof(psw_t));
1014 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1015 				&vcpu->arch.sie_block->gpsw,
1016 				sizeof(psw_t));
1017 		rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1018 				(u32 *)__LC_EXT_PARAMS);
1019 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1020 				(u64 *)__LC_EXT_PARAMS2);
1021 		kfree(inti);
1022 	}
1023 	return rc ? -EFAULT : 0;
1024 }
1025 
1026 static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1027 {
1028 	int rc;
1029 
1030 	rc  = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1031 	rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1032 	rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1033 	rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1034 	rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1035 			     &vcpu->arch.sie_block->gpsw,
1036 			     sizeof(psw_t));
1037 	rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1038 			    &vcpu->arch.sie_block->gpsw,
1039 			    sizeof(psw_t));
1040 	return rc ? -EFAULT : 0;
1041 }
1042 
1043 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1044 				     unsigned long irq_type)
1045 {
1046 	struct list_head *isc_list;
1047 	struct kvm_s390_float_interrupt *fi;
1048 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1049 	struct kvm_s390_interrupt_info *inti = NULL;
1050 	struct kvm_s390_io_info io;
1051 	u32 isc;
1052 	int rc = 0;
1053 
1054 	fi = &vcpu->kvm->arch.float_int;
1055 
1056 	spin_lock(&fi->lock);
1057 	isc = irq_type_to_isc(irq_type);
1058 	isc_list = &fi->lists[isc];
1059 	inti = list_first_entry_or_null(isc_list,
1060 					struct kvm_s390_interrupt_info,
1061 					list);
1062 	if (inti) {
1063 		if (inti->type & KVM_S390_INT_IO_AI_MASK)
1064 			VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1065 		else
1066 			VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1067 			inti->io.subchannel_id >> 8,
1068 			inti->io.subchannel_id >> 1 & 0x3,
1069 			inti->io.subchannel_nr);
1070 
1071 		vcpu->stat.deliver_io++;
1072 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1073 				inti->type,
1074 				((__u32)inti->io.subchannel_id << 16) |
1075 				inti->io.subchannel_nr,
1076 				((__u64)inti->io.io_int_parm << 32) |
1077 				inti->io.io_int_word);
1078 		list_del(&inti->list);
1079 		fi->counters[FIRQ_CNTR_IO] -= 1;
1080 	}
1081 	if (list_empty(isc_list))
1082 		clear_bit(irq_type, &fi->pending_irqs);
1083 	spin_unlock(&fi->lock);
1084 
1085 	if (inti) {
1086 		rc = __do_deliver_io(vcpu, &(inti->io));
1087 		kfree(inti);
1088 		goto out;
1089 	}
1090 
1091 	if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1092 		/*
1093 		 * in case an adapter interrupt was not delivered
1094 		 * in SIE context KVM will handle the delivery
1095 		 */
1096 		VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1097 		memset(&io, 0, sizeof(io));
1098 		io.io_int_word = isc_to_int_word(isc);
1099 		vcpu->stat.deliver_io++;
1100 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1101 			KVM_S390_INT_IO(1, 0, 0, 0),
1102 			((__u32)io.subchannel_id << 16) |
1103 			io.subchannel_nr,
1104 			((__u64)io.io_int_parm << 32) |
1105 			io.io_int_word);
1106 		rc = __do_deliver_io(vcpu, &io);
1107 	}
1108 out:
1109 	return rc;
1110 }
1111 
1112 /* Check whether an external call is pending (deliverable or not) */
1113 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1114 {
1115 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1116 
1117 	if (!sclp.has_sigpif)
1118 		return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1119 
1120 	return sca_ext_call_pending(vcpu, NULL);
1121 }
1122 
1123 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1124 {
1125 	if (deliverable_irqs(vcpu))
1126 		return 1;
1127 
1128 	if (kvm_cpu_has_pending_timer(vcpu))
1129 		return 1;
1130 
1131 	/* external call pending and deliverable */
1132 	if (kvm_s390_ext_call_pending(vcpu) &&
1133 	    !psw_extint_disabled(vcpu) &&
1134 	    (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1135 		return 1;
1136 
1137 	if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1138 		return 1;
1139 	return 0;
1140 }
1141 
1142 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1143 {
1144 	return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1145 }
1146 
1147 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1148 {
1149 	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1150 	const u64 ckc = vcpu->arch.sie_block->ckc;
1151 	u64 cputm, sltime = 0;
1152 
1153 	if (ckc_interrupts_enabled(vcpu)) {
1154 		if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1155 			if ((s64)now < (s64)ckc)
1156 				sltime = tod_to_ns((s64)ckc - (s64)now);
1157 		} else if (now < ckc) {
1158 			sltime = tod_to_ns(ckc - now);
1159 		}
1160 		/* already expired */
1161 		if (!sltime)
1162 			return 0;
1163 		if (cpu_timer_interrupts_enabled(vcpu)) {
1164 			cputm = kvm_s390_get_cpu_timer(vcpu);
1165 			/* already expired? */
1166 			if (cputm >> 63)
1167 				return 0;
1168 			return min(sltime, tod_to_ns(cputm));
1169 		}
1170 	} else if (cpu_timer_interrupts_enabled(vcpu)) {
1171 		sltime = kvm_s390_get_cpu_timer(vcpu);
1172 		/* already expired? */
1173 		if (sltime >> 63)
1174 			return 0;
1175 	}
1176 	return sltime;
1177 }
1178 
1179 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1180 {
1181 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1182 	u64 sltime;
1183 
1184 	vcpu->stat.exit_wait_state++;
1185 
1186 	/* fast path */
1187 	if (kvm_arch_vcpu_runnable(vcpu))
1188 		return 0;
1189 
1190 	if (psw_interrupts_disabled(vcpu)) {
1191 		VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1192 		return -EOPNOTSUPP; /* disabled wait */
1193 	}
1194 
1195 	if (gi->origin &&
1196 	    (gisa_get_ipm_or_restore_iam(gi) &
1197 	     vcpu->arch.sie_block->gcr[6] >> 24))
1198 		return 0;
1199 
1200 	if (!ckc_interrupts_enabled(vcpu) &&
1201 	    !cpu_timer_interrupts_enabled(vcpu)) {
1202 		VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1203 		__set_cpu_idle(vcpu);
1204 		goto no_timer;
1205 	}
1206 
1207 	sltime = __calculate_sltime(vcpu);
1208 	if (!sltime)
1209 		return 0;
1210 
1211 	__set_cpu_idle(vcpu);
1212 	hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1213 	VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1214 no_timer:
1215 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1216 	kvm_vcpu_block(vcpu);
1217 	__unset_cpu_idle(vcpu);
1218 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1219 
1220 	hrtimer_cancel(&vcpu->arch.ckc_timer);
1221 	return 0;
1222 }
1223 
1224 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1225 {
1226 	/*
1227 	 * We cannot move this into the if, as the CPU might be already
1228 	 * in kvm_vcpu_block without having the waitqueue set (polling)
1229 	 */
1230 	vcpu->valid_wakeup = true;
1231 	/*
1232 	 * This is mostly to document, that the read in swait_active could
1233 	 * be moved before other stores, leading to subtle races.
1234 	 * All current users do not store or use an atomic like update
1235 	 */
1236 	smp_mb__after_atomic();
1237 	if (swait_active(&vcpu->wq)) {
1238 		/*
1239 		 * The vcpu gave up the cpu voluntarily, mark it as a good
1240 		 * yield-candidate.
1241 		 */
1242 		vcpu->preempted = true;
1243 		swake_up_one(&vcpu->wq);
1244 		vcpu->stat.halt_wakeup++;
1245 	}
1246 	/*
1247 	 * The VCPU might not be sleeping but is executing the VSIE. Let's
1248 	 * kick it, so it leaves the SIE to process the request.
1249 	 */
1250 	kvm_s390_vsie_kick(vcpu);
1251 }
1252 
1253 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1254 {
1255 	struct kvm_vcpu *vcpu;
1256 	u64 sltime;
1257 
1258 	vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1259 	sltime = __calculate_sltime(vcpu);
1260 
1261 	/*
1262 	 * If the monotonic clock runs faster than the tod clock we might be
1263 	 * woken up too early and have to go back to sleep to avoid deadlocks.
1264 	 */
1265 	if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1266 		return HRTIMER_RESTART;
1267 	kvm_s390_vcpu_wakeup(vcpu);
1268 	return HRTIMER_NORESTART;
1269 }
1270 
1271 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1272 {
1273 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1274 
1275 	spin_lock(&li->lock);
1276 	li->pending_irqs = 0;
1277 	bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1278 	memset(&li->irq, 0, sizeof(li->irq));
1279 	spin_unlock(&li->lock);
1280 
1281 	sca_clear_ext_call(vcpu);
1282 }
1283 
1284 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1285 {
1286 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1287 	int rc = 0;
1288 	unsigned long irq_type;
1289 	unsigned long irqs;
1290 
1291 	__reset_intercept_indicators(vcpu);
1292 
1293 	/* pending ckc conditions might have been invalidated */
1294 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1295 	if (ckc_irq_pending(vcpu))
1296 		set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1297 
1298 	/* pending cpu timer conditions might have been invalidated */
1299 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1300 	if (cpu_timer_irq_pending(vcpu))
1301 		set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1302 
1303 	while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1304 		/* bits are in the reverse order of interrupt priority */
1305 		irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1306 		switch (irq_type) {
1307 		case IRQ_PEND_IO_ISC_0:
1308 		case IRQ_PEND_IO_ISC_1:
1309 		case IRQ_PEND_IO_ISC_2:
1310 		case IRQ_PEND_IO_ISC_3:
1311 		case IRQ_PEND_IO_ISC_4:
1312 		case IRQ_PEND_IO_ISC_5:
1313 		case IRQ_PEND_IO_ISC_6:
1314 		case IRQ_PEND_IO_ISC_7:
1315 			rc = __deliver_io(vcpu, irq_type);
1316 			break;
1317 		case IRQ_PEND_MCHK_EX:
1318 		case IRQ_PEND_MCHK_REP:
1319 			rc = __deliver_machine_check(vcpu);
1320 			break;
1321 		case IRQ_PEND_PROG:
1322 			rc = __deliver_prog(vcpu);
1323 			break;
1324 		case IRQ_PEND_EXT_EMERGENCY:
1325 			rc = __deliver_emergency_signal(vcpu);
1326 			break;
1327 		case IRQ_PEND_EXT_EXTERNAL:
1328 			rc = __deliver_external_call(vcpu);
1329 			break;
1330 		case IRQ_PEND_EXT_CLOCK_COMP:
1331 			rc = __deliver_ckc(vcpu);
1332 			break;
1333 		case IRQ_PEND_EXT_CPU_TIMER:
1334 			rc = __deliver_cpu_timer(vcpu);
1335 			break;
1336 		case IRQ_PEND_RESTART:
1337 			rc = __deliver_restart(vcpu);
1338 			break;
1339 		case IRQ_PEND_SET_PREFIX:
1340 			rc = __deliver_set_prefix(vcpu);
1341 			break;
1342 		case IRQ_PEND_PFAULT_INIT:
1343 			rc = __deliver_pfault_init(vcpu);
1344 			break;
1345 		case IRQ_PEND_EXT_SERVICE:
1346 			rc = __deliver_service(vcpu);
1347 			break;
1348 		case IRQ_PEND_PFAULT_DONE:
1349 			rc = __deliver_pfault_done(vcpu);
1350 			break;
1351 		case IRQ_PEND_VIRTIO:
1352 			rc = __deliver_virtio(vcpu);
1353 			break;
1354 		default:
1355 			WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1356 			clear_bit(irq_type, &li->pending_irqs);
1357 		}
1358 	}
1359 
1360 	set_intercept_indicators(vcpu);
1361 
1362 	return rc;
1363 }
1364 
1365 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1366 {
1367 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1368 
1369 	vcpu->stat.inject_program++;
1370 	VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1371 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1372 				   irq->u.pgm.code, 0);
1373 
1374 	if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1375 		/* auto detection if no valid ILC was given */
1376 		irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1377 		irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1378 		irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1379 	}
1380 
1381 	if (irq->u.pgm.code == PGM_PER) {
1382 		li->irq.pgm.code |= PGM_PER;
1383 		li->irq.pgm.flags = irq->u.pgm.flags;
1384 		/* only modify PER related information */
1385 		li->irq.pgm.per_address = irq->u.pgm.per_address;
1386 		li->irq.pgm.per_code = irq->u.pgm.per_code;
1387 		li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1388 		li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1389 	} else if (!(irq->u.pgm.code & PGM_PER)) {
1390 		li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1391 				   irq->u.pgm.code;
1392 		li->irq.pgm.flags = irq->u.pgm.flags;
1393 		/* only modify non-PER information */
1394 		li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1395 		li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1396 		li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1397 		li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1398 		li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1399 		li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1400 	} else {
1401 		li->irq.pgm = irq->u.pgm;
1402 	}
1403 	set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1404 	return 0;
1405 }
1406 
1407 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1408 {
1409 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1410 
1411 	vcpu->stat.inject_pfault_init++;
1412 	VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1413 		   irq->u.ext.ext_params2);
1414 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1415 				   irq->u.ext.ext_params,
1416 				   irq->u.ext.ext_params2);
1417 
1418 	li->irq.ext = irq->u.ext;
1419 	set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1420 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1421 	return 0;
1422 }
1423 
1424 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1425 {
1426 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1427 	struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1428 	uint16_t src_id = irq->u.extcall.code;
1429 
1430 	vcpu->stat.inject_external_call++;
1431 	VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1432 		   src_id);
1433 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1434 				   src_id, 0);
1435 
1436 	/* sending vcpu invalid */
1437 	if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1438 		return -EINVAL;
1439 
1440 	if (sclp.has_sigpif)
1441 		return sca_inject_ext_call(vcpu, src_id);
1442 
1443 	if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1444 		return -EBUSY;
1445 	*extcall = irq->u.extcall;
1446 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1447 	return 0;
1448 }
1449 
1450 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1451 {
1452 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1453 	struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1454 
1455 	vcpu->stat.inject_set_prefix++;
1456 	VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1457 		   irq->u.prefix.address);
1458 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1459 				   irq->u.prefix.address, 0);
1460 
1461 	if (!is_vcpu_stopped(vcpu))
1462 		return -EBUSY;
1463 
1464 	*prefix = irq->u.prefix;
1465 	set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1466 	return 0;
1467 }
1468 
1469 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1470 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1471 {
1472 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1473 	struct kvm_s390_stop_info *stop = &li->irq.stop;
1474 	int rc = 0;
1475 
1476 	vcpu->stat.inject_stop_signal++;
1477 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1478 
1479 	if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1480 		return -EINVAL;
1481 
1482 	if (is_vcpu_stopped(vcpu)) {
1483 		if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1484 			rc = kvm_s390_store_status_unloaded(vcpu,
1485 						KVM_S390_STORE_STATUS_NOADDR);
1486 		return rc;
1487 	}
1488 
1489 	if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1490 		return -EBUSY;
1491 	stop->flags = irq->u.stop.flags;
1492 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1493 	return 0;
1494 }
1495 
1496 static int __inject_sigp_restart(struct kvm_vcpu *vcpu,
1497 				 struct kvm_s390_irq *irq)
1498 {
1499 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1500 
1501 	vcpu->stat.inject_restart++;
1502 	VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1503 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1504 
1505 	set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1506 	return 0;
1507 }
1508 
1509 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1510 				   struct kvm_s390_irq *irq)
1511 {
1512 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1513 
1514 	vcpu->stat.inject_emergency_signal++;
1515 	VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1516 		   irq->u.emerg.code);
1517 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1518 				   irq->u.emerg.code, 0);
1519 
1520 	/* sending vcpu invalid */
1521 	if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1522 		return -EINVAL;
1523 
1524 	set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1525 	set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1526 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1527 	return 0;
1528 }
1529 
1530 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1531 {
1532 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1533 	struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1534 
1535 	vcpu->stat.inject_mchk++;
1536 	VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1537 		   irq->u.mchk.mcic);
1538 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1539 				   irq->u.mchk.mcic);
1540 
1541 	/*
1542 	 * Because repressible machine checks can be indicated along with
1543 	 * exigent machine checks (PoP, Chapter 11, Interruption action)
1544 	 * we need to combine cr14, mcic and external damage code.
1545 	 * Failing storage address and the logout area should not be or'ed
1546 	 * together, we just indicate the last occurrence of the corresponding
1547 	 * machine check
1548 	 */
1549 	mchk->cr14 |= irq->u.mchk.cr14;
1550 	mchk->mcic |= irq->u.mchk.mcic;
1551 	mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1552 	mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1553 	memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1554 	       sizeof(mchk->fixed_logout));
1555 	if (mchk->mcic & MCHK_EX_MASK)
1556 		set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1557 	else if (mchk->mcic & MCHK_REP_MASK)
1558 		set_bit(IRQ_PEND_MCHK_REP,  &li->pending_irqs);
1559 	return 0;
1560 }
1561 
1562 static int __inject_ckc(struct kvm_vcpu *vcpu)
1563 {
1564 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1565 
1566 	vcpu->stat.inject_ckc++;
1567 	VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1568 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1569 				   0, 0);
1570 
1571 	set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1572 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1573 	return 0;
1574 }
1575 
1576 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1577 {
1578 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1579 
1580 	vcpu->stat.inject_cputm++;
1581 	VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1582 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1583 				   0, 0);
1584 
1585 	set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1586 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1587 	return 0;
1588 }
1589 
1590 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1591 						  int isc, u32 schid)
1592 {
1593 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1594 	struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1595 	struct kvm_s390_interrupt_info *iter;
1596 	u16 id = (schid & 0xffff0000U) >> 16;
1597 	u16 nr = schid & 0x0000ffffU;
1598 
1599 	spin_lock(&fi->lock);
1600 	list_for_each_entry(iter, isc_list, list) {
1601 		if (schid && (id != iter->io.subchannel_id ||
1602 			      nr != iter->io.subchannel_nr))
1603 			continue;
1604 		/* found an appropriate entry */
1605 		list_del_init(&iter->list);
1606 		fi->counters[FIRQ_CNTR_IO] -= 1;
1607 		if (list_empty(isc_list))
1608 			clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1609 		spin_unlock(&fi->lock);
1610 		return iter;
1611 	}
1612 	spin_unlock(&fi->lock);
1613 	return NULL;
1614 }
1615 
1616 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1617 						      u64 isc_mask, u32 schid)
1618 {
1619 	struct kvm_s390_interrupt_info *inti = NULL;
1620 	int isc;
1621 
1622 	for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1623 		if (isc_mask & isc_to_isc_bits(isc))
1624 			inti = get_io_int(kvm, isc, schid);
1625 	}
1626 	return inti;
1627 }
1628 
1629 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1630 {
1631 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1632 	unsigned long active_mask;
1633 	int isc;
1634 
1635 	if (schid)
1636 		goto out;
1637 	if (!gi->origin)
1638 		goto out;
1639 
1640 	active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1641 	while (active_mask) {
1642 		isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1643 		if (gisa_tac_ipm_gisc(gi->origin, isc))
1644 			return isc;
1645 		clear_bit_inv(isc, &active_mask);
1646 	}
1647 out:
1648 	return -EINVAL;
1649 }
1650 
1651 /*
1652  * Dequeue and return an I/O interrupt matching any of the interruption
1653  * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1654  * Take into account the interrupts pending in the interrupt list and in GISA.
1655  *
1656  * Note that for a guest that does not enable I/O interrupts
1657  * but relies on TPI, a flood of classic interrupts may starve
1658  * out adapter interrupts on the same isc. Linux does not do
1659  * that, and it is possible to work around the issue by configuring
1660  * different iscs for classic and adapter interrupts in the guest,
1661  * but we may want to revisit this in the future.
1662  */
1663 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1664 						    u64 isc_mask, u32 schid)
1665 {
1666 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1667 	struct kvm_s390_interrupt_info *inti, *tmp_inti;
1668 	int isc;
1669 
1670 	inti = get_top_io_int(kvm, isc_mask, schid);
1671 
1672 	isc = get_top_gisa_isc(kvm, isc_mask, schid);
1673 	if (isc < 0)
1674 		/* no AI in GISA */
1675 		goto out;
1676 
1677 	if (!inti)
1678 		/* AI in GISA but no classical IO int */
1679 		goto gisa_out;
1680 
1681 	/* both types of interrupts present */
1682 	if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1683 		/* classical IO int with higher priority */
1684 		gisa_set_ipm_gisc(gi->origin, isc);
1685 		goto out;
1686 	}
1687 gisa_out:
1688 	tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1689 	if (tmp_inti) {
1690 		tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1691 		tmp_inti->io.io_int_word = isc_to_int_word(isc);
1692 		if (inti)
1693 			kvm_s390_reinject_io_int(kvm, inti);
1694 		inti = tmp_inti;
1695 	} else
1696 		gisa_set_ipm_gisc(gi->origin, isc);
1697 out:
1698 	return inti;
1699 }
1700 
1701 #define SCCB_MASK 0xFFFFFFF8
1702 #define SCCB_EVENT_PENDING 0x3
1703 
1704 static int __inject_service(struct kvm *kvm,
1705 			     struct kvm_s390_interrupt_info *inti)
1706 {
1707 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1708 
1709 	kvm->stat.inject_service_signal++;
1710 	spin_lock(&fi->lock);
1711 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1712 	/*
1713 	 * Early versions of the QEMU s390 bios will inject several
1714 	 * service interrupts after another without handling a
1715 	 * condition code indicating busy.
1716 	 * We will silently ignore those superfluous sccb values.
1717 	 * A future version of QEMU will take care of serialization
1718 	 * of servc requests
1719 	 */
1720 	if (fi->srv_signal.ext_params & SCCB_MASK)
1721 		goto out;
1722 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1723 	set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1724 out:
1725 	spin_unlock(&fi->lock);
1726 	kfree(inti);
1727 	return 0;
1728 }
1729 
1730 static int __inject_virtio(struct kvm *kvm,
1731 			    struct kvm_s390_interrupt_info *inti)
1732 {
1733 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1734 
1735 	kvm->stat.inject_virtio++;
1736 	spin_lock(&fi->lock);
1737 	if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1738 		spin_unlock(&fi->lock);
1739 		return -EBUSY;
1740 	}
1741 	fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1742 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1743 	set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1744 	spin_unlock(&fi->lock);
1745 	return 0;
1746 }
1747 
1748 static int __inject_pfault_done(struct kvm *kvm,
1749 				 struct kvm_s390_interrupt_info *inti)
1750 {
1751 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1752 
1753 	kvm->stat.inject_pfault_done++;
1754 	spin_lock(&fi->lock);
1755 	if (fi->counters[FIRQ_CNTR_PFAULT] >=
1756 		(ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1757 		spin_unlock(&fi->lock);
1758 		return -EBUSY;
1759 	}
1760 	fi->counters[FIRQ_CNTR_PFAULT] += 1;
1761 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1762 	set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1763 	spin_unlock(&fi->lock);
1764 	return 0;
1765 }
1766 
1767 #define CR_PENDING_SUBCLASS 28
1768 static int __inject_float_mchk(struct kvm *kvm,
1769 				struct kvm_s390_interrupt_info *inti)
1770 {
1771 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1772 
1773 	kvm->stat.inject_float_mchk++;
1774 	spin_lock(&fi->lock);
1775 	fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1776 	fi->mchk.mcic |= inti->mchk.mcic;
1777 	set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1778 	spin_unlock(&fi->lock);
1779 	kfree(inti);
1780 	return 0;
1781 }
1782 
1783 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1784 {
1785 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1786 	struct kvm_s390_float_interrupt *fi;
1787 	struct list_head *list;
1788 	int isc;
1789 
1790 	kvm->stat.inject_io++;
1791 	isc = int_word_to_isc(inti->io.io_int_word);
1792 
1793 	if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1794 		VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1795 		gisa_set_ipm_gisc(gi->origin, isc);
1796 		kfree(inti);
1797 		return 0;
1798 	}
1799 
1800 	fi = &kvm->arch.float_int;
1801 	spin_lock(&fi->lock);
1802 	if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1803 		spin_unlock(&fi->lock);
1804 		return -EBUSY;
1805 	}
1806 	fi->counters[FIRQ_CNTR_IO] += 1;
1807 
1808 	if (inti->type & KVM_S390_INT_IO_AI_MASK)
1809 		VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1810 	else
1811 		VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1812 			inti->io.subchannel_id >> 8,
1813 			inti->io.subchannel_id >> 1 & 0x3,
1814 			inti->io.subchannel_nr);
1815 	list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1816 	list_add_tail(&inti->list, list);
1817 	set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1818 	spin_unlock(&fi->lock);
1819 	return 0;
1820 }
1821 
1822 /*
1823  * Find a destination VCPU for a floating irq and kick it.
1824  */
1825 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1826 {
1827 	struct kvm_vcpu *dst_vcpu;
1828 	int sigcpu, online_vcpus, nr_tries = 0;
1829 
1830 	online_vcpus = atomic_read(&kvm->online_vcpus);
1831 	if (!online_vcpus)
1832 		return;
1833 
1834 	/* find idle VCPUs first, then round robin */
1835 	sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1836 	if (sigcpu == online_vcpus) {
1837 		do {
1838 			sigcpu = kvm->arch.float_int.next_rr_cpu++;
1839 			kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1840 			/* avoid endless loops if all vcpus are stopped */
1841 			if (nr_tries++ >= online_vcpus)
1842 				return;
1843 		} while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1844 	}
1845 	dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1846 
1847 	/* make the VCPU drop out of the SIE, or wake it up if sleeping */
1848 	switch (type) {
1849 	case KVM_S390_MCHK:
1850 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1851 		break;
1852 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1853 		if (!(type & KVM_S390_INT_IO_AI_MASK &&
1854 		      kvm->arch.gisa_int.origin))
1855 			kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1856 		break;
1857 	default:
1858 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1859 		break;
1860 	}
1861 	kvm_s390_vcpu_wakeup(dst_vcpu);
1862 }
1863 
1864 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1865 {
1866 	u64 type = READ_ONCE(inti->type);
1867 	int rc;
1868 
1869 	switch (type) {
1870 	case KVM_S390_MCHK:
1871 		rc = __inject_float_mchk(kvm, inti);
1872 		break;
1873 	case KVM_S390_INT_VIRTIO:
1874 		rc = __inject_virtio(kvm, inti);
1875 		break;
1876 	case KVM_S390_INT_SERVICE:
1877 		rc = __inject_service(kvm, inti);
1878 		break;
1879 	case KVM_S390_INT_PFAULT_DONE:
1880 		rc = __inject_pfault_done(kvm, inti);
1881 		break;
1882 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1883 		rc = __inject_io(kvm, inti);
1884 		break;
1885 	default:
1886 		rc = -EINVAL;
1887 	}
1888 	if (rc)
1889 		return rc;
1890 
1891 	__floating_irq_kick(kvm, type);
1892 	return 0;
1893 }
1894 
1895 int kvm_s390_inject_vm(struct kvm *kvm,
1896 		       struct kvm_s390_interrupt *s390int)
1897 {
1898 	struct kvm_s390_interrupt_info *inti;
1899 	int rc;
1900 
1901 	inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1902 	if (!inti)
1903 		return -ENOMEM;
1904 
1905 	inti->type = s390int->type;
1906 	switch (inti->type) {
1907 	case KVM_S390_INT_VIRTIO:
1908 		VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
1909 			 s390int->parm, s390int->parm64);
1910 		inti->ext.ext_params = s390int->parm;
1911 		inti->ext.ext_params2 = s390int->parm64;
1912 		break;
1913 	case KVM_S390_INT_SERVICE:
1914 		VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
1915 		inti->ext.ext_params = s390int->parm;
1916 		break;
1917 	case KVM_S390_INT_PFAULT_DONE:
1918 		inti->ext.ext_params2 = s390int->parm64;
1919 		break;
1920 	case KVM_S390_MCHK:
1921 		VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
1922 			 s390int->parm64);
1923 		inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
1924 		inti->mchk.mcic = s390int->parm64;
1925 		break;
1926 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1927 		inti->io.subchannel_id = s390int->parm >> 16;
1928 		inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
1929 		inti->io.io_int_parm = s390int->parm64 >> 32;
1930 		inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
1931 		break;
1932 	default:
1933 		kfree(inti);
1934 		return -EINVAL;
1935 	}
1936 	trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
1937 				 2);
1938 
1939 	rc = __inject_vm(kvm, inti);
1940 	if (rc)
1941 		kfree(inti);
1942 	return rc;
1943 }
1944 
1945 int kvm_s390_reinject_io_int(struct kvm *kvm,
1946 			      struct kvm_s390_interrupt_info *inti)
1947 {
1948 	return __inject_vm(kvm, inti);
1949 }
1950 
1951 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
1952 		       struct kvm_s390_irq *irq)
1953 {
1954 	irq->type = s390int->type;
1955 	switch (irq->type) {
1956 	case KVM_S390_PROGRAM_INT:
1957 		if (s390int->parm & 0xffff0000)
1958 			return -EINVAL;
1959 		irq->u.pgm.code = s390int->parm;
1960 		break;
1961 	case KVM_S390_SIGP_SET_PREFIX:
1962 		irq->u.prefix.address = s390int->parm;
1963 		break;
1964 	case KVM_S390_SIGP_STOP:
1965 		irq->u.stop.flags = s390int->parm;
1966 		break;
1967 	case KVM_S390_INT_EXTERNAL_CALL:
1968 		if (s390int->parm & 0xffff0000)
1969 			return -EINVAL;
1970 		irq->u.extcall.code = s390int->parm;
1971 		break;
1972 	case KVM_S390_INT_EMERGENCY:
1973 		if (s390int->parm & 0xffff0000)
1974 			return -EINVAL;
1975 		irq->u.emerg.code = s390int->parm;
1976 		break;
1977 	case KVM_S390_MCHK:
1978 		irq->u.mchk.mcic = s390int->parm64;
1979 		break;
1980 	}
1981 	return 0;
1982 }
1983 
1984 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
1985 {
1986 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1987 
1988 	return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1989 }
1990 
1991 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
1992 {
1993 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1994 
1995 	spin_lock(&li->lock);
1996 	li->irq.stop.flags = 0;
1997 	clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1998 	spin_unlock(&li->lock);
1999 }
2000 
2001 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2002 {
2003 	int rc;
2004 
2005 	switch (irq->type) {
2006 	case KVM_S390_PROGRAM_INT:
2007 		rc = __inject_prog(vcpu, irq);
2008 		break;
2009 	case KVM_S390_SIGP_SET_PREFIX:
2010 		rc = __inject_set_prefix(vcpu, irq);
2011 		break;
2012 	case KVM_S390_SIGP_STOP:
2013 		rc = __inject_sigp_stop(vcpu, irq);
2014 		break;
2015 	case KVM_S390_RESTART:
2016 		rc = __inject_sigp_restart(vcpu, irq);
2017 		break;
2018 	case KVM_S390_INT_CLOCK_COMP:
2019 		rc = __inject_ckc(vcpu);
2020 		break;
2021 	case KVM_S390_INT_CPU_TIMER:
2022 		rc = __inject_cpu_timer(vcpu);
2023 		break;
2024 	case KVM_S390_INT_EXTERNAL_CALL:
2025 		rc = __inject_extcall(vcpu, irq);
2026 		break;
2027 	case KVM_S390_INT_EMERGENCY:
2028 		rc = __inject_sigp_emergency(vcpu, irq);
2029 		break;
2030 	case KVM_S390_MCHK:
2031 		rc = __inject_mchk(vcpu, irq);
2032 		break;
2033 	case KVM_S390_INT_PFAULT_INIT:
2034 		rc = __inject_pfault_init(vcpu, irq);
2035 		break;
2036 	case KVM_S390_INT_VIRTIO:
2037 	case KVM_S390_INT_SERVICE:
2038 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2039 	default:
2040 		rc = -EINVAL;
2041 	}
2042 
2043 	return rc;
2044 }
2045 
2046 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2047 {
2048 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2049 	int rc;
2050 
2051 	spin_lock(&li->lock);
2052 	rc = do_inject_vcpu(vcpu, irq);
2053 	spin_unlock(&li->lock);
2054 	if (!rc)
2055 		kvm_s390_vcpu_wakeup(vcpu);
2056 	return rc;
2057 }
2058 
2059 static inline void clear_irq_list(struct list_head *_list)
2060 {
2061 	struct kvm_s390_interrupt_info *inti, *n;
2062 
2063 	list_for_each_entry_safe(inti, n, _list, list) {
2064 		list_del(&inti->list);
2065 		kfree(inti);
2066 	}
2067 }
2068 
2069 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2070 		       struct kvm_s390_irq *irq)
2071 {
2072 	irq->type = inti->type;
2073 	switch (inti->type) {
2074 	case KVM_S390_INT_PFAULT_INIT:
2075 	case KVM_S390_INT_PFAULT_DONE:
2076 	case KVM_S390_INT_VIRTIO:
2077 		irq->u.ext = inti->ext;
2078 		break;
2079 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2080 		irq->u.io = inti->io;
2081 		break;
2082 	}
2083 }
2084 
2085 void kvm_s390_clear_float_irqs(struct kvm *kvm)
2086 {
2087 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2088 	int i;
2089 
2090 	spin_lock(&fi->lock);
2091 	fi->pending_irqs = 0;
2092 	memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2093 	memset(&fi->mchk, 0, sizeof(fi->mchk));
2094 	for (i = 0; i < FIRQ_LIST_COUNT; i++)
2095 		clear_irq_list(&fi->lists[i]);
2096 	for (i = 0; i < FIRQ_MAX_COUNT; i++)
2097 		fi->counters[i] = 0;
2098 	spin_unlock(&fi->lock);
2099 	kvm_s390_gisa_clear(kvm);
2100 };
2101 
2102 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2103 {
2104 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2105 	struct kvm_s390_interrupt_info *inti;
2106 	struct kvm_s390_float_interrupt *fi;
2107 	struct kvm_s390_irq *buf;
2108 	struct kvm_s390_irq *irq;
2109 	int max_irqs;
2110 	int ret = 0;
2111 	int n = 0;
2112 	int i;
2113 
2114 	if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2115 		return -EINVAL;
2116 
2117 	/*
2118 	 * We are already using -ENOMEM to signal
2119 	 * userspace it may retry with a bigger buffer,
2120 	 * so we need to use something else for this case
2121 	 */
2122 	buf = vzalloc(len);
2123 	if (!buf)
2124 		return -ENOBUFS;
2125 
2126 	max_irqs = len / sizeof(struct kvm_s390_irq);
2127 
2128 	if (gi->origin && gisa_get_ipm(gi->origin)) {
2129 		for (i = 0; i <= MAX_ISC; i++) {
2130 			if (n == max_irqs) {
2131 				/* signal userspace to try again */
2132 				ret = -ENOMEM;
2133 				goto out_nolock;
2134 			}
2135 			if (gisa_tac_ipm_gisc(gi->origin, i)) {
2136 				irq = (struct kvm_s390_irq *) &buf[n];
2137 				irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2138 				irq->u.io.io_int_word = isc_to_int_word(i);
2139 				n++;
2140 			}
2141 		}
2142 	}
2143 	fi = &kvm->arch.float_int;
2144 	spin_lock(&fi->lock);
2145 	for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2146 		list_for_each_entry(inti, &fi->lists[i], list) {
2147 			if (n == max_irqs) {
2148 				/* signal userspace to try again */
2149 				ret = -ENOMEM;
2150 				goto out;
2151 			}
2152 			inti_to_irq(inti, &buf[n]);
2153 			n++;
2154 		}
2155 	}
2156 	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs)) {
2157 		if (n == max_irqs) {
2158 			/* signal userspace to try again */
2159 			ret = -ENOMEM;
2160 			goto out;
2161 		}
2162 		irq = (struct kvm_s390_irq *) &buf[n];
2163 		irq->type = KVM_S390_INT_SERVICE;
2164 		irq->u.ext = fi->srv_signal;
2165 		n++;
2166 	}
2167 	if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2168 		if (n == max_irqs) {
2169 				/* signal userspace to try again */
2170 				ret = -ENOMEM;
2171 				goto out;
2172 		}
2173 		irq = (struct kvm_s390_irq *) &buf[n];
2174 		irq->type = KVM_S390_MCHK;
2175 		irq->u.mchk = fi->mchk;
2176 		n++;
2177 }
2178 
2179 out:
2180 	spin_unlock(&fi->lock);
2181 out_nolock:
2182 	if (!ret && n > 0) {
2183 		if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2184 			ret = -EFAULT;
2185 	}
2186 	vfree(buf);
2187 
2188 	return ret < 0 ? ret : n;
2189 }
2190 
2191 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2192 {
2193 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2194 	struct kvm_s390_ais_all ais;
2195 
2196 	if (attr->attr < sizeof(ais))
2197 		return -EINVAL;
2198 
2199 	if (!test_kvm_facility(kvm, 72))
2200 		return -ENOTSUPP;
2201 
2202 	mutex_lock(&fi->ais_lock);
2203 	ais.simm = fi->simm;
2204 	ais.nimm = fi->nimm;
2205 	mutex_unlock(&fi->ais_lock);
2206 
2207 	if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2208 		return -EFAULT;
2209 
2210 	return 0;
2211 }
2212 
2213 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2214 {
2215 	int r;
2216 
2217 	switch (attr->group) {
2218 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2219 		r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2220 					  attr->attr);
2221 		break;
2222 	case KVM_DEV_FLIC_AISM_ALL:
2223 		r = flic_ais_mode_get_all(dev->kvm, attr);
2224 		break;
2225 	default:
2226 		r = -EINVAL;
2227 	}
2228 
2229 	return r;
2230 }
2231 
2232 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2233 				     u64 addr)
2234 {
2235 	struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2236 	void *target = NULL;
2237 	void __user *source;
2238 	u64 size;
2239 
2240 	if (get_user(inti->type, (u64 __user *)addr))
2241 		return -EFAULT;
2242 
2243 	switch (inti->type) {
2244 	case KVM_S390_INT_PFAULT_INIT:
2245 	case KVM_S390_INT_PFAULT_DONE:
2246 	case KVM_S390_INT_VIRTIO:
2247 	case KVM_S390_INT_SERVICE:
2248 		target = (void *) &inti->ext;
2249 		source = &uptr->u.ext;
2250 		size = sizeof(inti->ext);
2251 		break;
2252 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2253 		target = (void *) &inti->io;
2254 		source = &uptr->u.io;
2255 		size = sizeof(inti->io);
2256 		break;
2257 	case KVM_S390_MCHK:
2258 		target = (void *) &inti->mchk;
2259 		source = &uptr->u.mchk;
2260 		size = sizeof(inti->mchk);
2261 		break;
2262 	default:
2263 		return -EINVAL;
2264 	}
2265 
2266 	if (copy_from_user(target, source, size))
2267 		return -EFAULT;
2268 
2269 	return 0;
2270 }
2271 
2272 static int enqueue_floating_irq(struct kvm_device *dev,
2273 				struct kvm_device_attr *attr)
2274 {
2275 	struct kvm_s390_interrupt_info *inti = NULL;
2276 	int r = 0;
2277 	int len = attr->attr;
2278 
2279 	if (len % sizeof(struct kvm_s390_irq) != 0)
2280 		return -EINVAL;
2281 	else if (len > KVM_S390_FLIC_MAX_BUFFER)
2282 		return -EINVAL;
2283 
2284 	while (len >= sizeof(struct kvm_s390_irq)) {
2285 		inti = kzalloc(sizeof(*inti), GFP_KERNEL);
2286 		if (!inti)
2287 			return -ENOMEM;
2288 
2289 		r = copy_irq_from_user(inti, attr->addr);
2290 		if (r) {
2291 			kfree(inti);
2292 			return r;
2293 		}
2294 		r = __inject_vm(dev->kvm, inti);
2295 		if (r) {
2296 			kfree(inti);
2297 			return r;
2298 		}
2299 		len -= sizeof(struct kvm_s390_irq);
2300 		attr->addr += sizeof(struct kvm_s390_irq);
2301 	}
2302 
2303 	return r;
2304 }
2305 
2306 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2307 {
2308 	if (id >= MAX_S390_IO_ADAPTERS)
2309 		return NULL;
2310 	return kvm->arch.adapters[id];
2311 }
2312 
2313 static int register_io_adapter(struct kvm_device *dev,
2314 			       struct kvm_device_attr *attr)
2315 {
2316 	struct s390_io_adapter *adapter;
2317 	struct kvm_s390_io_adapter adapter_info;
2318 
2319 	if (copy_from_user(&adapter_info,
2320 			   (void __user *)attr->addr, sizeof(adapter_info)))
2321 		return -EFAULT;
2322 
2323 	if ((adapter_info.id >= MAX_S390_IO_ADAPTERS) ||
2324 	    (dev->kvm->arch.adapters[adapter_info.id] != NULL))
2325 		return -EINVAL;
2326 
2327 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2328 	if (!adapter)
2329 		return -ENOMEM;
2330 
2331 	INIT_LIST_HEAD(&adapter->maps);
2332 	init_rwsem(&adapter->maps_lock);
2333 	atomic_set(&adapter->nr_maps, 0);
2334 	adapter->id = adapter_info.id;
2335 	adapter->isc = adapter_info.isc;
2336 	adapter->maskable = adapter_info.maskable;
2337 	adapter->masked = false;
2338 	adapter->swap = adapter_info.swap;
2339 	adapter->suppressible = (adapter_info.flags) &
2340 				KVM_S390_ADAPTER_SUPPRESSIBLE;
2341 	dev->kvm->arch.adapters[adapter->id] = adapter;
2342 
2343 	return 0;
2344 }
2345 
2346 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2347 {
2348 	int ret;
2349 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2350 
2351 	if (!adapter || !adapter->maskable)
2352 		return -EINVAL;
2353 	ret = adapter->masked;
2354 	adapter->masked = masked;
2355 	return ret;
2356 }
2357 
2358 static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr)
2359 {
2360 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2361 	struct s390_map_info *map;
2362 	int ret;
2363 
2364 	if (!adapter || !addr)
2365 		return -EINVAL;
2366 
2367 	map = kzalloc(sizeof(*map), GFP_KERNEL);
2368 	if (!map) {
2369 		ret = -ENOMEM;
2370 		goto out;
2371 	}
2372 	INIT_LIST_HEAD(&map->list);
2373 	map->guest_addr = addr;
2374 	map->addr = gmap_translate(kvm->arch.gmap, addr);
2375 	if (map->addr == -EFAULT) {
2376 		ret = -EFAULT;
2377 		goto out;
2378 	}
2379 	ret = get_user_pages_fast(map->addr, 1, 1, &map->page);
2380 	if (ret < 0)
2381 		goto out;
2382 	BUG_ON(ret != 1);
2383 	down_write(&adapter->maps_lock);
2384 	if (atomic_inc_return(&adapter->nr_maps) < MAX_S390_ADAPTER_MAPS) {
2385 		list_add_tail(&map->list, &adapter->maps);
2386 		ret = 0;
2387 	} else {
2388 		put_page(map->page);
2389 		ret = -EINVAL;
2390 	}
2391 	up_write(&adapter->maps_lock);
2392 out:
2393 	if (ret)
2394 		kfree(map);
2395 	return ret;
2396 }
2397 
2398 static int kvm_s390_adapter_unmap(struct kvm *kvm, unsigned int id, __u64 addr)
2399 {
2400 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2401 	struct s390_map_info *map, *tmp;
2402 	int found = 0;
2403 
2404 	if (!adapter || !addr)
2405 		return -EINVAL;
2406 
2407 	down_write(&adapter->maps_lock);
2408 	list_for_each_entry_safe(map, tmp, &adapter->maps, list) {
2409 		if (map->guest_addr == addr) {
2410 			found = 1;
2411 			atomic_dec(&adapter->nr_maps);
2412 			list_del(&map->list);
2413 			put_page(map->page);
2414 			kfree(map);
2415 			break;
2416 		}
2417 	}
2418 	up_write(&adapter->maps_lock);
2419 
2420 	return found ? 0 : -EINVAL;
2421 }
2422 
2423 void kvm_s390_destroy_adapters(struct kvm *kvm)
2424 {
2425 	int i;
2426 	struct s390_map_info *map, *tmp;
2427 
2428 	for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) {
2429 		if (!kvm->arch.adapters[i])
2430 			continue;
2431 		list_for_each_entry_safe(map, tmp,
2432 					 &kvm->arch.adapters[i]->maps, list) {
2433 			list_del(&map->list);
2434 			put_page(map->page);
2435 			kfree(map);
2436 		}
2437 		kfree(kvm->arch.adapters[i]);
2438 	}
2439 }
2440 
2441 static int modify_io_adapter(struct kvm_device *dev,
2442 			     struct kvm_device_attr *attr)
2443 {
2444 	struct kvm_s390_io_adapter_req req;
2445 	struct s390_io_adapter *adapter;
2446 	int ret;
2447 
2448 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2449 		return -EFAULT;
2450 
2451 	adapter = get_io_adapter(dev->kvm, req.id);
2452 	if (!adapter)
2453 		return -EINVAL;
2454 	switch (req.type) {
2455 	case KVM_S390_IO_ADAPTER_MASK:
2456 		ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2457 		if (ret > 0)
2458 			ret = 0;
2459 		break;
2460 	case KVM_S390_IO_ADAPTER_MAP:
2461 		ret = kvm_s390_adapter_map(dev->kvm, req.id, req.addr);
2462 		break;
2463 	case KVM_S390_IO_ADAPTER_UNMAP:
2464 		ret = kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr);
2465 		break;
2466 	default:
2467 		ret = -EINVAL;
2468 	}
2469 
2470 	return ret;
2471 }
2472 
2473 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2474 
2475 {
2476 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2477 	u32 schid;
2478 
2479 	if (attr->flags)
2480 		return -EINVAL;
2481 	if (attr->attr != sizeof(schid))
2482 		return -EINVAL;
2483 	if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2484 		return -EFAULT;
2485 	if (!schid)
2486 		return -EINVAL;
2487 	kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2488 	/*
2489 	 * If userspace is conforming to the architecture, we can have at most
2490 	 * one pending I/O interrupt per subchannel, so this is effectively a
2491 	 * clear all.
2492 	 */
2493 	return 0;
2494 }
2495 
2496 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2497 {
2498 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2499 	struct kvm_s390_ais_req req;
2500 	int ret = 0;
2501 
2502 	if (!test_kvm_facility(kvm, 72))
2503 		return -ENOTSUPP;
2504 
2505 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2506 		return -EFAULT;
2507 
2508 	if (req.isc > MAX_ISC)
2509 		return -EINVAL;
2510 
2511 	trace_kvm_s390_modify_ais_mode(req.isc,
2512 				       (fi->simm & AIS_MODE_MASK(req.isc)) ?
2513 				       (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2514 				       2 : KVM_S390_AIS_MODE_SINGLE :
2515 				       KVM_S390_AIS_MODE_ALL, req.mode);
2516 
2517 	mutex_lock(&fi->ais_lock);
2518 	switch (req.mode) {
2519 	case KVM_S390_AIS_MODE_ALL:
2520 		fi->simm &= ~AIS_MODE_MASK(req.isc);
2521 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2522 		break;
2523 	case KVM_S390_AIS_MODE_SINGLE:
2524 		fi->simm |= AIS_MODE_MASK(req.isc);
2525 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2526 		break;
2527 	default:
2528 		ret = -EINVAL;
2529 	}
2530 	mutex_unlock(&fi->ais_lock);
2531 
2532 	return ret;
2533 }
2534 
2535 static int kvm_s390_inject_airq(struct kvm *kvm,
2536 				struct s390_io_adapter *adapter)
2537 {
2538 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2539 	struct kvm_s390_interrupt s390int = {
2540 		.type = KVM_S390_INT_IO(1, 0, 0, 0),
2541 		.parm = 0,
2542 		.parm64 = isc_to_int_word(adapter->isc),
2543 	};
2544 	int ret = 0;
2545 
2546 	if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2547 		return kvm_s390_inject_vm(kvm, &s390int);
2548 
2549 	mutex_lock(&fi->ais_lock);
2550 	if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2551 		trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2552 		goto out;
2553 	}
2554 
2555 	ret = kvm_s390_inject_vm(kvm, &s390int);
2556 	if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2557 		fi->nimm |= AIS_MODE_MASK(adapter->isc);
2558 		trace_kvm_s390_modify_ais_mode(adapter->isc,
2559 					       KVM_S390_AIS_MODE_SINGLE, 2);
2560 	}
2561 out:
2562 	mutex_unlock(&fi->ais_lock);
2563 	return ret;
2564 }
2565 
2566 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2567 {
2568 	unsigned int id = attr->attr;
2569 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2570 
2571 	if (!adapter)
2572 		return -EINVAL;
2573 
2574 	return kvm_s390_inject_airq(kvm, adapter);
2575 }
2576 
2577 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2578 {
2579 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2580 	struct kvm_s390_ais_all ais;
2581 
2582 	if (!test_kvm_facility(kvm, 72))
2583 		return -ENOTSUPP;
2584 
2585 	if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2586 		return -EFAULT;
2587 
2588 	mutex_lock(&fi->ais_lock);
2589 	fi->simm = ais.simm;
2590 	fi->nimm = ais.nimm;
2591 	mutex_unlock(&fi->ais_lock);
2592 
2593 	return 0;
2594 }
2595 
2596 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2597 {
2598 	int r = 0;
2599 	unsigned int i;
2600 	struct kvm_vcpu *vcpu;
2601 
2602 	switch (attr->group) {
2603 	case KVM_DEV_FLIC_ENQUEUE:
2604 		r = enqueue_floating_irq(dev, attr);
2605 		break;
2606 	case KVM_DEV_FLIC_CLEAR_IRQS:
2607 		kvm_s390_clear_float_irqs(dev->kvm);
2608 		break;
2609 	case KVM_DEV_FLIC_APF_ENABLE:
2610 		dev->kvm->arch.gmap->pfault_enabled = 1;
2611 		break;
2612 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2613 		dev->kvm->arch.gmap->pfault_enabled = 0;
2614 		/*
2615 		 * Make sure no async faults are in transition when
2616 		 * clearing the queues. So we don't need to worry
2617 		 * about late coming workers.
2618 		 */
2619 		synchronize_srcu(&dev->kvm->srcu);
2620 		kvm_for_each_vcpu(i, vcpu, dev->kvm)
2621 			kvm_clear_async_pf_completion_queue(vcpu);
2622 		break;
2623 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2624 		r = register_io_adapter(dev, attr);
2625 		break;
2626 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2627 		r = modify_io_adapter(dev, attr);
2628 		break;
2629 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2630 		r = clear_io_irq(dev->kvm, attr);
2631 		break;
2632 	case KVM_DEV_FLIC_AISM:
2633 		r = modify_ais_mode(dev->kvm, attr);
2634 		break;
2635 	case KVM_DEV_FLIC_AIRQ_INJECT:
2636 		r = flic_inject_airq(dev->kvm, attr);
2637 		break;
2638 	case KVM_DEV_FLIC_AISM_ALL:
2639 		r = flic_ais_mode_set_all(dev->kvm, attr);
2640 		break;
2641 	default:
2642 		r = -EINVAL;
2643 	}
2644 
2645 	return r;
2646 }
2647 
2648 static int flic_has_attr(struct kvm_device *dev,
2649 			     struct kvm_device_attr *attr)
2650 {
2651 	switch (attr->group) {
2652 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2653 	case KVM_DEV_FLIC_ENQUEUE:
2654 	case KVM_DEV_FLIC_CLEAR_IRQS:
2655 	case KVM_DEV_FLIC_APF_ENABLE:
2656 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2657 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2658 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2659 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2660 	case KVM_DEV_FLIC_AISM:
2661 	case KVM_DEV_FLIC_AIRQ_INJECT:
2662 	case KVM_DEV_FLIC_AISM_ALL:
2663 		return 0;
2664 	}
2665 	return -ENXIO;
2666 }
2667 
2668 static int flic_create(struct kvm_device *dev, u32 type)
2669 {
2670 	if (!dev)
2671 		return -EINVAL;
2672 	if (dev->kvm->arch.flic)
2673 		return -EINVAL;
2674 	dev->kvm->arch.flic = dev;
2675 	return 0;
2676 }
2677 
2678 static void flic_destroy(struct kvm_device *dev)
2679 {
2680 	dev->kvm->arch.flic = NULL;
2681 	kfree(dev);
2682 }
2683 
2684 /* s390 floating irq controller (flic) */
2685 struct kvm_device_ops kvm_flic_ops = {
2686 	.name = "kvm-flic",
2687 	.get_attr = flic_get_attr,
2688 	.set_attr = flic_set_attr,
2689 	.has_attr = flic_has_attr,
2690 	.create = flic_create,
2691 	.destroy = flic_destroy,
2692 };
2693 
2694 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2695 {
2696 	unsigned long bit;
2697 
2698 	bit = bit_nr + (addr % PAGE_SIZE) * 8;
2699 
2700 	return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2701 }
2702 
2703 static struct s390_map_info *get_map_info(struct s390_io_adapter *adapter,
2704 					  u64 addr)
2705 {
2706 	struct s390_map_info *map;
2707 
2708 	if (!adapter)
2709 		return NULL;
2710 
2711 	list_for_each_entry(map, &adapter->maps, list) {
2712 		if (map->guest_addr == addr)
2713 			return map;
2714 	}
2715 	return NULL;
2716 }
2717 
2718 static int adapter_indicators_set(struct kvm *kvm,
2719 				  struct s390_io_adapter *adapter,
2720 				  struct kvm_s390_adapter_int *adapter_int)
2721 {
2722 	unsigned long bit;
2723 	int summary_set, idx;
2724 	struct s390_map_info *info;
2725 	void *map;
2726 
2727 	info = get_map_info(adapter, adapter_int->ind_addr);
2728 	if (!info)
2729 		return -1;
2730 	map = page_address(info->page);
2731 	bit = get_ind_bit(info->addr, adapter_int->ind_offset, adapter->swap);
2732 	set_bit(bit, map);
2733 	idx = srcu_read_lock(&kvm->srcu);
2734 	mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2735 	set_page_dirty_lock(info->page);
2736 	info = get_map_info(adapter, adapter_int->summary_addr);
2737 	if (!info) {
2738 		srcu_read_unlock(&kvm->srcu, idx);
2739 		return -1;
2740 	}
2741 	map = page_address(info->page);
2742 	bit = get_ind_bit(info->addr, adapter_int->summary_offset,
2743 			  adapter->swap);
2744 	summary_set = test_and_set_bit(bit, map);
2745 	mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2746 	set_page_dirty_lock(info->page);
2747 	srcu_read_unlock(&kvm->srcu, idx);
2748 	return summary_set ? 0 : 1;
2749 }
2750 
2751 /*
2752  * < 0 - not injected due to error
2753  * = 0 - coalesced, summary indicator already active
2754  * > 0 - injected interrupt
2755  */
2756 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2757 			   struct kvm *kvm, int irq_source_id, int level,
2758 			   bool line_status)
2759 {
2760 	int ret;
2761 	struct s390_io_adapter *adapter;
2762 
2763 	/* We're only interested in the 0->1 transition. */
2764 	if (!level)
2765 		return 0;
2766 	adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2767 	if (!adapter)
2768 		return -1;
2769 	down_read(&adapter->maps_lock);
2770 	ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2771 	up_read(&adapter->maps_lock);
2772 	if ((ret > 0) && !adapter->masked) {
2773 		ret = kvm_s390_inject_airq(kvm, adapter);
2774 		if (ret == 0)
2775 			ret = 1;
2776 	}
2777 	return ret;
2778 }
2779 
2780 /*
2781  * Inject the machine check to the guest.
2782  */
2783 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2784 				     struct mcck_volatile_info *mcck_info)
2785 {
2786 	struct kvm_s390_interrupt_info inti;
2787 	struct kvm_s390_irq irq;
2788 	struct kvm_s390_mchk_info *mchk;
2789 	union mci mci;
2790 	__u64 cr14 = 0;         /* upper bits are not used */
2791 	int rc;
2792 
2793 	mci.val = mcck_info->mcic;
2794 	if (mci.sr)
2795 		cr14 |= CR14_RECOVERY_SUBMASK;
2796 	if (mci.dg)
2797 		cr14 |= CR14_DEGRADATION_SUBMASK;
2798 	if (mci.w)
2799 		cr14 |= CR14_WARNING_SUBMASK;
2800 
2801 	mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2802 	mchk->cr14 = cr14;
2803 	mchk->mcic = mcck_info->mcic;
2804 	mchk->ext_damage_code = mcck_info->ext_damage_code;
2805 	mchk->failing_storage_address = mcck_info->failing_storage_address;
2806 	if (mci.ck) {
2807 		/* Inject the floating machine check */
2808 		inti.type = KVM_S390_MCHK;
2809 		rc = __inject_vm(vcpu->kvm, &inti);
2810 	} else {
2811 		/* Inject the machine check to specified vcpu */
2812 		irq.type = KVM_S390_MCHK;
2813 		rc = kvm_s390_inject_vcpu(vcpu, &irq);
2814 	}
2815 	WARN_ON_ONCE(rc);
2816 }
2817 
2818 int kvm_set_routing_entry(struct kvm *kvm,
2819 			  struct kvm_kernel_irq_routing_entry *e,
2820 			  const struct kvm_irq_routing_entry *ue)
2821 {
2822 	int ret;
2823 
2824 	switch (ue->type) {
2825 	case KVM_IRQ_ROUTING_S390_ADAPTER:
2826 		e->set = set_adapter_int;
2827 		e->adapter.summary_addr = ue->u.adapter.summary_addr;
2828 		e->adapter.ind_addr = ue->u.adapter.ind_addr;
2829 		e->adapter.summary_offset = ue->u.adapter.summary_offset;
2830 		e->adapter.ind_offset = ue->u.adapter.ind_offset;
2831 		e->adapter.adapter_id = ue->u.adapter.adapter_id;
2832 		ret = 0;
2833 		break;
2834 	default:
2835 		ret = -EINVAL;
2836 	}
2837 
2838 	return ret;
2839 }
2840 
2841 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2842 		int irq_source_id, int level, bool line_status)
2843 {
2844 	return -EINVAL;
2845 }
2846 
2847 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2848 {
2849 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2850 	struct kvm_s390_irq *buf;
2851 	int r = 0;
2852 	int n;
2853 
2854 	buf = vmalloc(len);
2855 	if (!buf)
2856 		return -ENOMEM;
2857 
2858 	if (copy_from_user((void *) buf, irqstate, len)) {
2859 		r = -EFAULT;
2860 		goto out_free;
2861 	}
2862 
2863 	/*
2864 	 * Don't allow setting the interrupt state
2865 	 * when there are already interrupts pending
2866 	 */
2867 	spin_lock(&li->lock);
2868 	if (li->pending_irqs) {
2869 		r = -EBUSY;
2870 		goto out_unlock;
2871 	}
2872 
2873 	for (n = 0; n < len / sizeof(*buf); n++) {
2874 		r = do_inject_vcpu(vcpu, &buf[n]);
2875 		if (r)
2876 			break;
2877 	}
2878 
2879 out_unlock:
2880 	spin_unlock(&li->lock);
2881 out_free:
2882 	vfree(buf);
2883 
2884 	return r;
2885 }
2886 
2887 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2888 			    struct kvm_s390_irq *irq,
2889 			    unsigned long irq_type)
2890 {
2891 	switch (irq_type) {
2892 	case IRQ_PEND_MCHK_EX:
2893 	case IRQ_PEND_MCHK_REP:
2894 		irq->type = KVM_S390_MCHK;
2895 		irq->u.mchk = li->irq.mchk;
2896 		break;
2897 	case IRQ_PEND_PROG:
2898 		irq->type = KVM_S390_PROGRAM_INT;
2899 		irq->u.pgm = li->irq.pgm;
2900 		break;
2901 	case IRQ_PEND_PFAULT_INIT:
2902 		irq->type = KVM_S390_INT_PFAULT_INIT;
2903 		irq->u.ext = li->irq.ext;
2904 		break;
2905 	case IRQ_PEND_EXT_EXTERNAL:
2906 		irq->type = KVM_S390_INT_EXTERNAL_CALL;
2907 		irq->u.extcall = li->irq.extcall;
2908 		break;
2909 	case IRQ_PEND_EXT_CLOCK_COMP:
2910 		irq->type = KVM_S390_INT_CLOCK_COMP;
2911 		break;
2912 	case IRQ_PEND_EXT_CPU_TIMER:
2913 		irq->type = KVM_S390_INT_CPU_TIMER;
2914 		break;
2915 	case IRQ_PEND_SIGP_STOP:
2916 		irq->type = KVM_S390_SIGP_STOP;
2917 		irq->u.stop = li->irq.stop;
2918 		break;
2919 	case IRQ_PEND_RESTART:
2920 		irq->type = KVM_S390_RESTART;
2921 		break;
2922 	case IRQ_PEND_SET_PREFIX:
2923 		irq->type = KVM_S390_SIGP_SET_PREFIX;
2924 		irq->u.prefix = li->irq.prefix;
2925 		break;
2926 	}
2927 }
2928 
2929 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
2930 {
2931 	int scn;
2932 	DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
2933 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2934 	unsigned long pending_irqs;
2935 	struct kvm_s390_irq irq;
2936 	unsigned long irq_type;
2937 	int cpuaddr;
2938 	int n = 0;
2939 
2940 	spin_lock(&li->lock);
2941 	pending_irqs = li->pending_irqs;
2942 	memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
2943 	       sizeof(sigp_emerg_pending));
2944 	spin_unlock(&li->lock);
2945 
2946 	for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
2947 		memset(&irq, 0, sizeof(irq));
2948 		if (irq_type == IRQ_PEND_EXT_EMERGENCY)
2949 			continue;
2950 		if (n + sizeof(irq) > len)
2951 			return -ENOBUFS;
2952 		store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
2953 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2954 			return -EFAULT;
2955 		n += sizeof(irq);
2956 	}
2957 
2958 	if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
2959 		for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
2960 			memset(&irq, 0, sizeof(irq));
2961 			if (n + sizeof(irq) > len)
2962 				return -ENOBUFS;
2963 			irq.type = KVM_S390_INT_EMERGENCY;
2964 			irq.u.emerg.code = cpuaddr;
2965 			if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2966 				return -EFAULT;
2967 			n += sizeof(irq);
2968 		}
2969 	}
2970 
2971 	if (sca_ext_call_pending(vcpu, &scn)) {
2972 		if (n + sizeof(irq) > len)
2973 			return -ENOBUFS;
2974 		memset(&irq, 0, sizeof(irq));
2975 		irq.type = KVM_S390_INT_EXTERNAL_CALL;
2976 		irq.u.extcall.code = scn;
2977 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2978 			return -EFAULT;
2979 		n += sizeof(irq);
2980 	}
2981 
2982 	return n;
2983 }
2984 
2985 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
2986 {
2987 	int vcpu_id, online_vcpus = atomic_read(&kvm->online_vcpus);
2988 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2989 	struct kvm_vcpu *vcpu;
2990 
2991 	for_each_set_bit(vcpu_id, kvm->arch.idle_mask, online_vcpus) {
2992 		vcpu = kvm_get_vcpu(kvm, vcpu_id);
2993 		if (psw_ioint_disabled(vcpu))
2994 			continue;
2995 		deliverable_mask &= (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
2996 		if (deliverable_mask) {
2997 			/* lately kicked but not yet running */
2998 			if (test_and_set_bit(vcpu_id, gi->kicked_mask))
2999 				return;
3000 			kvm_s390_vcpu_wakeup(vcpu);
3001 			return;
3002 		}
3003 	}
3004 }
3005 
3006 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3007 {
3008 	struct kvm_s390_gisa_interrupt *gi =
3009 		container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3010 	struct kvm *kvm =
3011 		container_of(gi->origin, struct sie_page2, gisa)->kvm;
3012 	u8 pending_mask;
3013 
3014 	pending_mask = gisa_get_ipm_or_restore_iam(gi);
3015 	if (pending_mask) {
3016 		__airqs_kick_single_vcpu(kvm, pending_mask);
3017 		hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3018 		return HRTIMER_RESTART;
3019 	};
3020 
3021 	return HRTIMER_NORESTART;
3022 }
3023 
3024 #define NULL_GISA_ADDR 0x00000000UL
3025 #define NONE_GISA_ADDR 0x00000001UL
3026 #define GISA_ADDR_MASK 0xfffff000UL
3027 
3028 static void process_gib_alert_list(void)
3029 {
3030 	struct kvm_s390_gisa_interrupt *gi;
3031 	struct kvm_s390_gisa *gisa;
3032 	struct kvm *kvm;
3033 	u32 final, origin = 0UL;
3034 
3035 	do {
3036 		/*
3037 		 * If the NONE_GISA_ADDR is still stored in the alert list
3038 		 * origin, we will leave the outer loop. No further GISA has
3039 		 * been added to the alert list by millicode while processing
3040 		 * the current alert list.
3041 		 */
3042 		final = (origin & NONE_GISA_ADDR);
3043 		/*
3044 		 * Cut off the alert list and store the NONE_GISA_ADDR in the
3045 		 * alert list origin to avoid further GAL interruptions.
3046 		 * A new alert list can be build up by millicode in parallel
3047 		 * for guests not in the yet cut-off alert list. When in the
3048 		 * final loop, store the NULL_GISA_ADDR instead. This will re-
3049 		 * enable GAL interruptions on the host again.
3050 		 */
3051 		origin = xchg(&gib->alert_list_origin,
3052 			      (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3053 		/*
3054 		 * Loop through the just cut-off alert list and start the
3055 		 * gisa timers to kick idle vcpus to consume the pending
3056 		 * interruptions asap.
3057 		 */
3058 		while (origin & GISA_ADDR_MASK) {
3059 			gisa = (struct kvm_s390_gisa *)(u64)origin;
3060 			origin = gisa->next_alert;
3061 			gisa->next_alert = (u32)(u64)gisa;
3062 			kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3063 			gi = &kvm->arch.gisa_int;
3064 			if (hrtimer_active(&gi->timer))
3065 				hrtimer_cancel(&gi->timer);
3066 			hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3067 		}
3068 	} while (!final);
3069 
3070 }
3071 
3072 void kvm_s390_gisa_clear(struct kvm *kvm)
3073 {
3074 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3075 
3076 	if (!gi->origin)
3077 		return;
3078 	gisa_clear_ipm(gi->origin);
3079 	VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3080 }
3081 
3082 void kvm_s390_gisa_init(struct kvm *kvm)
3083 {
3084 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3085 
3086 	if (!css_general_characteristics.aiv)
3087 		return;
3088 	gi->origin = &kvm->arch.sie_page2->gisa;
3089 	gi->alert.mask = 0;
3090 	spin_lock_init(&gi->alert.ref_lock);
3091 	gi->expires = 50 * 1000; /* 50 usec */
3092 	hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3093 	gi->timer.function = gisa_vcpu_kicker;
3094 	memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3095 	gi->origin->next_alert = (u32)(u64)gi->origin;
3096 	VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3097 }
3098 
3099 void kvm_s390_gisa_destroy(struct kvm *kvm)
3100 {
3101 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3102 
3103 	if (!gi->origin)
3104 		return;
3105 	if (gi->alert.mask)
3106 		KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
3107 			  kvm, gi->alert.mask);
3108 	while (gisa_in_alert_list(gi->origin))
3109 		cpu_relax();
3110 	hrtimer_cancel(&gi->timer);
3111 	gi->origin = NULL;
3112 }
3113 
3114 /**
3115  * kvm_s390_gisc_register - register a guest ISC
3116  *
3117  * @kvm:  the kernel vm to work with
3118  * @gisc: the guest interruption sub class to register
3119  *
3120  * The function extends the vm specific alert mask to use.
3121  * The effective IAM mask in the GISA is updated as well
3122  * in case the GISA is not part of the GIB alert list.
3123  * It will be updated latest when the IAM gets restored
3124  * by gisa_get_ipm_or_restore_iam().
3125  *
3126  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3127  *          has registered with the channel subsystem.
3128  *          -ENODEV in case the vm uses no GISA
3129  *          -ERANGE in case the guest ISC is invalid
3130  */
3131 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3132 {
3133 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3134 
3135 	if (!gi->origin)
3136 		return -ENODEV;
3137 	if (gisc > MAX_ISC)
3138 		return -ERANGE;
3139 
3140 	spin_lock(&gi->alert.ref_lock);
3141 	gi->alert.ref_count[gisc]++;
3142 	if (gi->alert.ref_count[gisc] == 1) {
3143 		gi->alert.mask |= 0x80 >> gisc;
3144 		gisa_set_iam(gi->origin, gi->alert.mask);
3145 	}
3146 	spin_unlock(&gi->alert.ref_lock);
3147 
3148 	return gib->nisc;
3149 }
3150 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3151 
3152 /**
3153  * kvm_s390_gisc_unregister - unregister a guest ISC
3154  *
3155  * @kvm:  the kernel vm to work with
3156  * @gisc: the guest interruption sub class to register
3157  *
3158  * The function reduces the vm specific alert mask to use.
3159  * The effective IAM mask in the GISA is updated as well
3160  * in case the GISA is not part of the GIB alert list.
3161  * It will be updated latest when the IAM gets restored
3162  * by gisa_get_ipm_or_restore_iam().
3163  *
3164  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3165  *          has registered with the channel subsystem.
3166  *          -ENODEV in case the vm uses no GISA
3167  *          -ERANGE in case the guest ISC is invalid
3168  *          -EINVAL in case the guest ISC is not registered
3169  */
3170 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3171 {
3172 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3173 	int rc = 0;
3174 
3175 	if (!gi->origin)
3176 		return -ENODEV;
3177 	if (gisc > MAX_ISC)
3178 		return -ERANGE;
3179 
3180 	spin_lock(&gi->alert.ref_lock);
3181 	if (gi->alert.ref_count[gisc] == 0) {
3182 		rc = -EINVAL;
3183 		goto out;
3184 	}
3185 	gi->alert.ref_count[gisc]--;
3186 	if (gi->alert.ref_count[gisc] == 0) {
3187 		gi->alert.mask &= ~(0x80 >> gisc);
3188 		gisa_set_iam(gi->origin, gi->alert.mask);
3189 	}
3190 out:
3191 	spin_unlock(&gi->alert.ref_lock);
3192 
3193 	return rc;
3194 }
3195 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3196 
3197 static void gib_alert_irq_handler(struct airq_struct *airq)
3198 {
3199 	inc_irq_stat(IRQIO_GAL);
3200 	process_gib_alert_list();
3201 }
3202 
3203 static struct airq_struct gib_alert_irq = {
3204 	.handler = gib_alert_irq_handler,
3205 	.lsi_ptr = &gib_alert_irq.lsi_mask,
3206 };
3207 
3208 void kvm_s390_gib_destroy(void)
3209 {
3210 	if (!gib)
3211 		return;
3212 	chsc_sgib(0);
3213 	unregister_adapter_interrupt(&gib_alert_irq);
3214 	free_page((unsigned long)gib);
3215 	gib = NULL;
3216 }
3217 
3218 int kvm_s390_gib_init(u8 nisc)
3219 {
3220 	int rc = 0;
3221 
3222 	if (!css_general_characteristics.aiv) {
3223 		KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3224 		goto out;
3225 	}
3226 
3227 	gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL | GFP_DMA);
3228 	if (!gib) {
3229 		rc = -ENOMEM;
3230 		goto out;
3231 	}
3232 
3233 	gib_alert_irq.isc = nisc;
3234 	if (register_adapter_interrupt(&gib_alert_irq)) {
3235 		pr_err("Registering the GIB alert interruption handler failed\n");
3236 		rc = -EIO;
3237 		goto out_free_gib;
3238 	}
3239 
3240 	gib->nisc = nisc;
3241 	if (chsc_sgib((u32)(u64)gib)) {
3242 		pr_err("Associating the GIB with the AIV facility failed\n");
3243 		free_page((unsigned long)gib);
3244 		gib = NULL;
3245 		rc = -EIO;
3246 		goto out_unreg_gal;
3247 	}
3248 
3249 	KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3250 	goto out;
3251 
3252 out_unreg_gal:
3253 	unregister_adapter_interrupt(&gib_alert_irq);
3254 out_free_gib:
3255 	free_page((unsigned long)gib);
3256 	gib = NULL;
3257 out:
3258 	return rc;
3259 }
3260