1
// SPDX-License-Identifier: GPL-2.0
2
/*
3
 * handling kvm guest interrupts
4
 *
5
 * Copyright IBM Corp. 2008, 2020
6
 *
7
 *    Author(s): Carsten Otte <[email protected]>
8
 */
9

10
#define KMSG_COMPONENT "kvm-s390"
11
#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
12

13
#include <linux/cpufeature.h>
14
#include <linux/interrupt.h>
15
#include <linux/kvm_host.h>
16
#include <linux/hrtimer.h>
17
#include <linux/export.h>
18
#include <linux/mmu_context.h>
19
#include <linux/nospec.h>
20
#include <linux/signal.h>
21
#include <linux/slab.h>
22
#include <linux/bitmap.h>
23
#include <linux/vmalloc.h>
24
#include <asm/access-regs.h>
25
#include <asm/asm-offsets.h>
26
#include <asm/dis.h>
27
#include <linux/uaccess.h>
28
#include <asm/sclp.h>
29
#include <asm/isc.h>
30
#include <asm/gmap.h>
31
#include <asm/nmi.h>
32
#include <asm/airq.h>
33
#include <asm/tpi.h>
34
#include "kvm-s390.h"
35
#include "gaccess.h"
36
#include "trace-s390.h"
37
#include "pci.h"
38

39
#define PFAULT_INIT 0x0600
40
#define PFAULT_DONE 0x0680
41
#define VIRTIO_PARAM 0x0d00
42

43
static struct kvm_s390_gib *gib;
44

45
/* handle external calls via sigp interpretation facility */
46
static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
47
{
48
	int c, scn;
49

50
	if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND))
51
		return 0;
52

53
	BUG_ON(!kvm_s390_use_sca_entries());
54
	read_lock(&vcpu->kvm->arch.sca_lock);
55
	if (vcpu->kvm->arch.use_esca) {
56
		struct esca_block *sca = vcpu->kvm->arch.sca;
57
		union esca_sigp_ctrl sigp_ctrl =
58
			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
59

60
		c = sigp_ctrl.c;
61
		scn = sigp_ctrl.scn;
62
	} else {
63
		struct bsca_block *sca = vcpu->kvm->arch.sca;
64
		union bsca_sigp_ctrl sigp_ctrl =
65
			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
66

67
		c = sigp_ctrl.c;
68
		scn = sigp_ctrl.scn;
69
	}
70
	read_unlock(&vcpu->kvm->arch.sca_lock);
71

72
	if (src_id)
73
		*src_id = scn;
74

75
	return c;
76
}
77

78
static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
79
{
80
	int expect, rc;
81

82
	BUG_ON(!kvm_s390_use_sca_entries());
83
	read_lock(&vcpu->kvm->arch.sca_lock);
84
	if (vcpu->kvm->arch.use_esca) {
85
		struct esca_block *sca = vcpu->kvm->arch.sca;
86
		union esca_sigp_ctrl *sigp_ctrl =
87
			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
88
		union esca_sigp_ctrl new_val = {0}, old_val;
89

90
		old_val = READ_ONCE(*sigp_ctrl);
91
		new_val.scn = src_id;
92
		new_val.c = 1;
93
		old_val.c = 0;
94

95
		expect = old_val.value;
96
		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
97
	} else {
98
		struct bsca_block *sca = vcpu->kvm->arch.sca;
99
		union bsca_sigp_ctrl *sigp_ctrl =
100
			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
101
		union bsca_sigp_ctrl new_val = {0}, old_val;
102

103
		old_val = READ_ONCE(*sigp_ctrl);
104
		new_val.scn = src_id;
105
		new_val.c = 1;
106
		old_val.c = 0;
107

108
		expect = old_val.value;
109
		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
110
	}
111
	read_unlock(&vcpu->kvm->arch.sca_lock);
112

113
	if (rc != expect) {
114
		/* another external call is pending */
115
		return -EBUSY;
116
	}
117
	kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
118
	return 0;
119
}
120

121
static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
122
{
123
	if (!kvm_s390_use_sca_entries())
124
		return;
125
	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
126
	read_lock(&vcpu->kvm->arch.sca_lock);
127
	if (vcpu->kvm->arch.use_esca) {
128
		struct esca_block *sca = vcpu->kvm->arch.sca;
129
		union esca_sigp_ctrl *sigp_ctrl =
130
			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
131

132
		WRITE_ONCE(sigp_ctrl->value, 0);
133
	} else {
134
		struct bsca_block *sca = vcpu->kvm->arch.sca;
135
		union bsca_sigp_ctrl *sigp_ctrl =
136
			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
137

138
		WRITE_ONCE(sigp_ctrl->value, 0);
139
	}
140
	read_unlock(&vcpu->kvm->arch.sca_lock);
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
	word = READ_ONCE(gisa->u64.word[0]);
244
	do {
245
		if ((u64)gisa != word >> 32)
246
			return -EBUSY;
247
		_word = (word & ~0xffUL) | iam;
248
	} while (!try_cmpxchg(&gisa->u64.word[0], &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
	word = READ_ONCE(gisa->u64.word[0]);
267
	do {
268
		_word = word & ~(0xffUL << 24);
269
	} while (!try_cmpxchg(&gisa->u64.word[0], &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
	word = READ_ONCE(gi->origin->u64.word[0]);
288
	do {
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 (!try_cmpxchg(&gi->origin->u64.word[0], &word, _word));
295

296
	return 0;
297
}
298

299
static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
300
{
301
	set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
302
}
303

304
static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
305
{
306
	return READ_ONCE(gisa->ipm);
307
}
308

309
static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
310
{
311
	return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
312
}
313

314
static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
315
{
316
	unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs |
317
				vcpu->arch.local_int.pending_irqs;
318

319
	pending &= ~vcpu->kvm->arch.float_int.masked_irqs;
320
	return pending;
321
}
322

323
static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
324
{
325
	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
326
	unsigned long pending_mask;
327

328
	pending_mask = pending_irqs_no_gisa(vcpu);
329
	if (gi->origin)
330
		pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
331
	return pending_mask;
332
}
333

334
static inline int isc_to_irq_type(unsigned long isc)
335
{
336
	return IRQ_PEND_IO_ISC_0 - isc;
337
}
338

339
static inline int irq_type_to_isc(unsigned long irq_type)
340
{
341
	return IRQ_PEND_IO_ISC_0 - irq_type;
342
}
343

344
static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
345
				   unsigned long active_mask)
346
{
347
	int i;
348

349
	for (i = 0; i <= MAX_ISC; i++)
350
		if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
351
			active_mask &= ~(1UL << (isc_to_irq_type(i)));
352

353
	return active_mask;
354
}
355

356
static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
357
{
358
	unsigned long active_mask;
359

360
	active_mask = pending_irqs(vcpu);
361
	if (!active_mask)
362
		return 0;
363

364
	if (psw_extint_disabled(vcpu))
365
		active_mask &= ~IRQ_PEND_EXT_MASK;
366
	if (psw_ioint_disabled(vcpu))
367
		active_mask &= ~IRQ_PEND_IO_MASK;
368
	else
369
		active_mask = disable_iscs(vcpu, active_mask);
370
	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
371
		__clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
372
	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
373
		__clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
374
	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
375
		__clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
376
	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
377
		__clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
378
	if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) {
379
		__clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
380
		__clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask);
381
	}
382
	if (psw_mchk_disabled(vcpu))
383
		active_mask &= ~IRQ_PEND_MCHK_MASK;
384
	/* PV guest cpus can have a single interruption injected at a time. */
385
	if (kvm_s390_pv_cpu_get_handle(vcpu) &&
386
	    vcpu->arch.sie_block->iictl != IICTL_CODE_NONE)
387
		active_mask &= ~(IRQ_PEND_EXT_II_MASK |
388
				 IRQ_PEND_IO_MASK |
389
				 IRQ_PEND_MCHK_MASK);
390
	/*
391
	 * Check both floating and local interrupt's cr14 because
392
	 * bit IRQ_PEND_MCHK_REP could be set in both cases.
393
	 */
394
	if (!(vcpu->arch.sie_block->gcr[14] &
395
	   (vcpu->kvm->arch.float_int.mchk.cr14 |
396
	   vcpu->arch.local_int.irq.mchk.cr14)))
397
		__clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
398

399
	/*
400
	 * STOP irqs will never be actively delivered. They are triggered via
401
	 * intercept requests and cleared when the stop intercept is performed.
402
	 */
403
	__clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
404

405
	return active_mask;
406
}
407

408
static void __set_cpu_idle(struct kvm_vcpu *vcpu)
409
{
410
	kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
411
	set_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
412
}
413

414
static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
415
{
416
	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
417
	clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
418
}
419

420
static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
421
{
422
	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
423
				      CPUSTAT_STOP_INT);
424
	vcpu->arch.sie_block->lctl = 0x0000;
425
	vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
426

427
	if (guestdbg_enabled(vcpu)) {
428
		vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
429
					       LCTL_CR10 | LCTL_CR11);
430
		vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
431
	}
432
}
433

434
static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
435
{
436
	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
437
		return;
438
	if (psw_ioint_disabled(vcpu))
439
		kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
440
	else
441
		vcpu->arch.sie_block->lctl |= LCTL_CR6;
442
}
443

444
static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
445
{
446
	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
447
		return;
448
	if (psw_extint_disabled(vcpu))
449
		kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
450
	else
451
		vcpu->arch.sie_block->lctl |= LCTL_CR0;
452
}
453

454
static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
455
{
456
	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
457
		return;
458
	if (psw_mchk_disabled(vcpu))
459
		vcpu->arch.sie_block->ictl |= ICTL_LPSW;
460
	else
461
		vcpu->arch.sie_block->lctl |= LCTL_CR14;
462
}
463

464
static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
465
{
466
	if (kvm_s390_is_stop_irq_pending(vcpu))
467
		kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
468
}
469

470
/* Set interception request for non-deliverable interrupts */
471
static void set_intercept_indicators(struct kvm_vcpu *vcpu)
472
{
473
	set_intercept_indicators_io(vcpu);
474
	set_intercept_indicators_ext(vcpu);
475
	set_intercept_indicators_mchk(vcpu);
476
	set_intercept_indicators_stop(vcpu);
477
}
478

479
static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
480
{
481
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
482
	int rc = 0;
483

484
	vcpu->stat.deliver_cputm++;
485
	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
486
					 0, 0);
487
	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
488
		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
489
		vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER;
490
	} else {
491
		rc  = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
492
				   (u16 *)__LC_EXT_INT_CODE);
493
		rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
494
		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
495
				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
496
		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
497
				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
498
	}
499
	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
500
	return rc ? -EFAULT : 0;
501
}
502

503
static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
504
{
505
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
506
	int rc = 0;
507

508
	vcpu->stat.deliver_ckc++;
509
	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
510
					 0, 0);
511
	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
512
		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
513
		vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP;
514
	} else {
515
		rc  = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
516
				   (u16 __user *)__LC_EXT_INT_CODE);
517
		rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
518
		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
519
				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
520
		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
521
				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
522
	}
523
	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
524
	return rc ? -EFAULT : 0;
525
}
526

527
static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
528
{
529
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
530
	struct kvm_s390_ext_info ext;
531
	int rc;
532

533
	spin_lock(&li->lock);
534
	ext = li->irq.ext;
535
	clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
536
	li->irq.ext.ext_params2 = 0;
537
	spin_unlock(&li->lock);
538

539
	VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
540
		   ext.ext_params2);
541
	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
542
					 KVM_S390_INT_PFAULT_INIT,
543
					 0, ext.ext_params2);
544

545
	rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
546
	rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
547
	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
548
			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
549
	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
550
			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
551
	rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
552
	return rc ? -EFAULT : 0;
553
}
554

555
static int __write_machine_check(struct kvm_vcpu *vcpu,
556
				 struct kvm_s390_mchk_info *mchk)
557
{
558
	unsigned long ext_sa_addr;
559
	unsigned long lc;
560
	freg_t fprs[NUM_FPRS];
561
	union mci mci;
562
	int rc;
563

564
	/*
565
	 * All other possible payload for a machine check (e.g. the register
566
	 * contents in the save area) will be handled by the ultravisor, as
567
	 * the hypervisor does not not have the needed information for
568
	 * protected guests.
569
	 */
570
	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
571
		vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK;
572
		vcpu->arch.sie_block->mcic = mchk->mcic;
573
		vcpu->arch.sie_block->faddr = mchk->failing_storage_address;
574
		vcpu->arch.sie_block->edc = mchk->ext_damage_code;
575
		return 0;
576
	}
577

578
	mci.val = mchk->mcic;
579
	/* take care of lazy register loading */
580
	kvm_s390_fpu_store(vcpu->run);
581
	save_access_regs(vcpu->run->s.regs.acrs);
582
	if (cpu_has_gs() && vcpu->arch.gs_enabled)
583
		save_gs_cb(current->thread.gs_cb);
584

585
	/* Extended save area */
586
	rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
587
			   sizeof(unsigned long));
588
	/* Only bits 0 through 63-LC are used for address formation */
589
	lc = ext_sa_addr & MCESA_LC_MASK;
590
	if (test_kvm_facility(vcpu->kvm, 133)) {
591
		switch (lc) {
592
		case 0:
593
		case 10:
594
			ext_sa_addr &= ~0x3ffUL;
595
			break;
596
		case 11:
597
			ext_sa_addr &= ~0x7ffUL;
598
			break;
599
		case 12:
600
			ext_sa_addr &= ~0xfffUL;
601
			break;
602
		default:
603
			ext_sa_addr = 0;
604
			break;
605
		}
606
	} else {
607
		ext_sa_addr &= ~0x3ffUL;
608
	}
609

610
	if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
611
		if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
612
				    512))
613
			mci.vr = 0;
614
	} else {
615
		mci.vr = 0;
616
	}
617
	if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
618
	    && (lc == 11 || lc == 12)) {
619
		if (write_guest_abs(vcpu, ext_sa_addr + 1024,
620
				    &vcpu->run->s.regs.gscb, 32))
621
			mci.gs = 0;
622
	} else {
623
		mci.gs = 0;
624
	}
625

626
	/* General interruption information */
627
	rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
628
	rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
629
			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
630
	rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
631
			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
632
	rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
633

634
	/* Register-save areas */
635
	if (cpu_has_vx()) {
636
		convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
637
		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
638
	} else {
639
		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
640
				     vcpu->run->s.regs.fprs, 128);
641
	}
642
	rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
643
			     vcpu->run->s.regs.gprs, 128);
644
	rc |= put_guest_lc(vcpu, vcpu->run->s.regs.fpc,
645
			   (u32 __user *) __LC_FP_CREG_SAVE_AREA);
646
	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
647
			   (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
648
	rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
649
			   (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
650
	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
651
			   (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
652
	rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
653
			     &vcpu->run->s.regs.acrs, 64);
654
	rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
655
			     &vcpu->arch.sie_block->gcr, 128);
656

657
	/* Extended interruption information */
658
	rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
659
			   (u32 __user *) __LC_EXT_DAMAGE_CODE);
660
	rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
661
			   (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
662
	rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
663
			     sizeof(mchk->fixed_logout));
664
	return rc ? -EFAULT : 0;
665
}
666

667
static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
668
{
669
	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
670
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
671
	struct kvm_s390_mchk_info mchk = {};
672
	int deliver = 0;
673
	int rc = 0;
674

675
	spin_lock(&fi->lock);
676
	spin_lock(&li->lock);
677
	if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
678
	    test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
679
		/*
680
		 * If there was an exigent machine check pending, then any
681
		 * repressible machine checks that might have been pending
682
		 * are indicated along with it, so always clear bits for
683
		 * repressible and exigent interrupts
684
		 */
685
		mchk = li->irq.mchk;
686
		clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
687
		clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
688
		memset(&li->irq.mchk, 0, sizeof(mchk));
689
		deliver = 1;
690
	}
691
	/*
692
	 * We indicate floating repressible conditions along with
693
	 * other pending conditions. Channel Report Pending and Channel
694
	 * Subsystem damage are the only two and are indicated by
695
	 * bits in mcic and masked in cr14.
696
	 */
697
	if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
698
		mchk.mcic |= fi->mchk.mcic;
699
		mchk.cr14 |= fi->mchk.cr14;
700
		memset(&fi->mchk, 0, sizeof(mchk));
701
		deliver = 1;
702
	}
703
	spin_unlock(&li->lock);
704
	spin_unlock(&fi->lock);
705

706
	if (deliver) {
707
		VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
708
			   mchk.mcic);
709
		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
710
						 KVM_S390_MCHK,
711
						 mchk.cr14, mchk.mcic);
712
		vcpu->stat.deliver_machine_check++;
713
		rc = __write_machine_check(vcpu, &mchk);
714
	}
715
	return rc;
716
}
717

718
static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
719
{
720
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
721
	int rc = 0;
722

723
	VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
724
	vcpu->stat.deliver_restart_signal++;
725
	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
726

727
	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
728
		vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART;
729
	} else {
730
		rc  = write_guest_lc(vcpu,
731
				     offsetof(struct lowcore, restart_old_psw),
732
				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
733
		rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
734
				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
735
	}
736
	clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
737
	return rc ? -EFAULT : 0;
738
}
739

740
static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
741
{
742
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
743
	struct kvm_s390_prefix_info prefix;
744

745
	spin_lock(&li->lock);
746
	prefix = li->irq.prefix;
747
	li->irq.prefix.address = 0;
748
	clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
749
	spin_unlock(&li->lock);
750

751
	vcpu->stat.deliver_prefix_signal++;
752
	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
753
					 KVM_S390_SIGP_SET_PREFIX,
754
					 prefix.address, 0);
755

756
	kvm_s390_set_prefix(vcpu, prefix.address);
757
	return 0;
758
}
759

760
static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
761
{
762
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
763
	int rc;
764
	int cpu_addr;
765

766
	spin_lock(&li->lock);
767
	cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
768
	clear_bit(cpu_addr, li->sigp_emerg_pending);
769
	if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
770
		clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
771
	spin_unlock(&li->lock);
772

773
	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
774
	vcpu->stat.deliver_emergency_signal++;
775
	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
776
					 cpu_addr, 0);
777
	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
778
		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
779
		vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG;
780
		vcpu->arch.sie_block->extcpuaddr = cpu_addr;
781
		return 0;
782
	}
783

784
	rc  = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
785
			   (u16 *)__LC_EXT_INT_CODE);
786
	rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
787
	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
788
			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
789
	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
790
			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
791
	return rc ? -EFAULT : 0;
792
}
793

794
static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
795
{
796
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
797
	struct kvm_s390_extcall_info extcall;
798
	int rc;
799

800
	spin_lock(&li->lock);
801
	extcall = li->irq.extcall;
802
	li->irq.extcall.code = 0;
803
	clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
804
	spin_unlock(&li->lock);
805

806
	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
807
	vcpu->stat.deliver_external_call++;
808
	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
809
					 KVM_S390_INT_EXTERNAL_CALL,
810
					 extcall.code, 0);
811
	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
812
		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
813
		vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL;
814
		vcpu->arch.sie_block->extcpuaddr = extcall.code;
815
		return 0;
816
	}
817

818
	rc  = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
819
			   (u16 *)__LC_EXT_INT_CODE);
820
	rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
821
	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
822
			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
823
	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
824
			    sizeof(psw_t));
825
	return rc ? -EFAULT : 0;
826
}
827

828
static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code)
829
{
830
	switch (code) {
831
	case PGM_SPECIFICATION:
832
		vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION;
833
		break;
834
	case PGM_OPERAND:
835
		vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND;
836
		break;
837
	default:
838
		return -EINVAL;
839
	}
840
	return 0;
841
}
842

843
static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
844
{
845
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
846
	struct kvm_s390_pgm_info pgm_info;
847
	int rc = 0, nullifying = false;
848
	u16 ilen;
849

850
	spin_lock(&li->lock);
851
	pgm_info = li->irq.pgm;
852
	clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
853
	memset(&li->irq.pgm, 0, sizeof(pgm_info));
854
	spin_unlock(&li->lock);
855

856
	ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
857
	VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
858
		   pgm_info.code, ilen);
859
	vcpu->stat.deliver_program++;
860
	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
861
					 pgm_info.code, 0);
862

863
	/* PER is handled by the ultravisor */
864
	if (kvm_s390_pv_cpu_is_protected(vcpu))
865
		return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER);
866

867
	switch (pgm_info.code & ~PGM_PER) {
868
	case PGM_AFX_TRANSLATION:
869
	case PGM_ASX_TRANSLATION:
870
	case PGM_EX_TRANSLATION:
871
	case PGM_LFX_TRANSLATION:
872
	case PGM_LSTE_SEQUENCE:
873
	case PGM_LSX_TRANSLATION:
874
	case PGM_LX_TRANSLATION:
875
	case PGM_PRIMARY_AUTHORITY:
876
	case PGM_SECONDARY_AUTHORITY:
877
		nullifying = true;
878
		fallthrough;
879
	case PGM_SPACE_SWITCH:
880
		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
881
				  (u64 *)__LC_TRANS_EXC_CODE);
882
		break;
883
	case PGM_ALEN_TRANSLATION:
884
	case PGM_ALE_SEQUENCE:
885
	case PGM_ASTE_INSTANCE:
886
	case PGM_ASTE_SEQUENCE:
887
	case PGM_ASTE_VALIDITY:
888
	case PGM_EXTENDED_AUTHORITY:
889
		rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
890
				  (u8 *)__LC_EXC_ACCESS_ID);
891
		nullifying = true;
892
		break;
893
	case PGM_ASCE_TYPE:
894
	case PGM_PAGE_TRANSLATION:
895
	case PGM_REGION_FIRST_TRANS:
896
	case PGM_REGION_SECOND_TRANS:
897
	case PGM_REGION_THIRD_TRANS:
898
	case PGM_SEGMENT_TRANSLATION:
899
		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
900
				  (u64 *)__LC_TRANS_EXC_CODE);
901
		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
902
				   (u8 *)__LC_EXC_ACCESS_ID);
903
		rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
904
				   (u8 *)__LC_OP_ACCESS_ID);
905
		nullifying = true;
906
		break;
907
	case PGM_MONITOR:
908
		rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
909
				  (u16 *)__LC_MON_CLASS_NR);
910
		rc |= put_guest_lc(vcpu, pgm_info.mon_code,
911
				   (u64 *)__LC_MON_CODE);
912
		break;
913
	case PGM_VECTOR_PROCESSING:
914
	case PGM_DATA:
915
		rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
916
				  (u32 *)__LC_DATA_EXC_CODE);
917
		break;
918
	case PGM_PROTECTION:
919
		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
920
				  (u64 *)__LC_TRANS_EXC_CODE);
921
		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
922
				   (u8 *)__LC_EXC_ACCESS_ID);
923
		break;
924
	case PGM_STACK_FULL:
925
	case PGM_STACK_EMPTY:
926
	case PGM_STACK_SPECIFICATION:
927
	case PGM_STACK_TYPE:
928
	case PGM_STACK_OPERATION:
929
	case PGM_TRACE_TABEL:
930
	case PGM_CRYPTO_OPERATION:
931
		nullifying = true;
932
		break;
933
	}
934

935
	if (pgm_info.code & PGM_PER) {
936
		rc |= put_guest_lc(vcpu, pgm_info.per_code,
937
				   (u8 *) __LC_PER_CODE);
938
		rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
939
				   (u8 *)__LC_PER_ATMID);
940
		rc |= put_guest_lc(vcpu, pgm_info.per_address,
941
				   (u64 *) __LC_PER_ADDRESS);
942
		rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
943
				   (u8 *) __LC_PER_ACCESS_ID);
944
	}
945

946
	if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
947
		kvm_s390_rewind_psw(vcpu, ilen);
948

949
	/* bit 1+2 of the target are the ilc, so we can directly use ilen */
950
	rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
951
	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
952
				 (u64 *) __LC_PGM_LAST_BREAK);
953
	rc |= put_guest_lc(vcpu, pgm_info.code, (u16 *)__LC_PGM_CODE);
954
	rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
955
			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
956
	rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
957
			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
958
	return rc ? -EFAULT : 0;
959
}
960

961
#define SCCB_MASK 0xFFFFFFF8
962
#define SCCB_EVENT_PENDING 0x3
963

964
static int write_sclp(struct kvm_vcpu *vcpu, u32 parm)
965
{
966
	int rc;
967

968
	if (kvm_s390_pv_cpu_get_handle(vcpu)) {
969
		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
970
		vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG;
971
		vcpu->arch.sie_block->eiparams = parm;
972
		return 0;
973
	}
974

975
	rc  = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
976
	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
977
	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
978
			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
979
	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
980
			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
981
	rc |= put_guest_lc(vcpu, parm,
982
			   (u32 *)__LC_EXT_PARAMS);
983

984
	return rc ? -EFAULT : 0;
985
}
986

987
static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
988
{
989
	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
990
	struct kvm_s390_ext_info ext;
991

992
	spin_lock(&fi->lock);
993
	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) ||
994
	    !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
995
		spin_unlock(&fi->lock);
996
		return 0;
997
	}
998
	ext = fi->srv_signal;
999
	memset(&fi->srv_signal, 0, sizeof(ext));
1000
	clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1001
	clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1002
	if (kvm_s390_pv_cpu_is_protected(vcpu))
1003
		set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs);
1004
	spin_unlock(&fi->lock);
1005

1006
	VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
1007
		   ext.ext_params);
1008
	vcpu->stat.deliver_service_signal++;
1009
	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1010
					 ext.ext_params, 0);
1011

1012
	return write_sclp(vcpu, ext.ext_params);
1013
}
1014

1015
static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)
1016
{
1017
	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1018
	struct kvm_s390_ext_info ext;
1019

1020
	spin_lock(&fi->lock);
1021
	if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) {
1022
		spin_unlock(&fi->lock);
1023
		return 0;
1024
	}
1025
	ext = fi->srv_signal;
1026
	/* only clear the event bits */
1027
	fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING;
1028
	clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1029
	spin_unlock(&fi->lock);
1030

1031
	VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event");
1032
	vcpu->stat.deliver_service_signal++;
1033
	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1034
					 ext.ext_params, 0);
1035

1036
	return write_sclp(vcpu, ext.ext_params & SCCB_EVENT_PENDING);
1037
}
1038

1039
static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
1040
{
1041
	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1042
	struct kvm_s390_interrupt_info *inti;
1043
	int rc = 0;
1044

1045
	spin_lock(&fi->lock);
1046
	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
1047
					struct kvm_s390_interrupt_info,
1048
					list);
1049
	if (inti) {
1050
		list_del(&inti->list);
1051
		fi->counters[FIRQ_CNTR_PFAULT] -= 1;
1052
	}
1053
	if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
1054
		clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1055
	spin_unlock(&fi->lock);
1056

1057
	if (inti) {
1058
		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1059
						 KVM_S390_INT_PFAULT_DONE, 0,
1060
						 inti->ext.ext_params2);
1061
		VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
1062
			   inti->ext.ext_params2);
1063

1064
		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1065
				(u16 *)__LC_EXT_INT_CODE);
1066
		rc |= put_guest_lc(vcpu, PFAULT_DONE,
1067
				(u16 *)__LC_EXT_CPU_ADDR);
1068
		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1069
				&vcpu->arch.sie_block->gpsw,
1070
				sizeof(psw_t));
1071
		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1072
				&vcpu->arch.sie_block->gpsw,
1073
				sizeof(psw_t));
1074
		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1075
				(u64 *)__LC_EXT_PARAMS2);
1076
		kfree(inti);
1077
	}
1078
	return rc ? -EFAULT : 0;
1079
}
1080

1081
static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
1082
{
1083
	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1084
	struct kvm_s390_interrupt_info *inti;
1085
	int rc = 0;
1086

1087
	spin_lock(&fi->lock);
1088
	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
1089
					struct kvm_s390_interrupt_info,
1090
					list);
1091
	if (inti) {
1092
		VCPU_EVENT(vcpu, 4,
1093
			   "deliver: virtio parm: 0x%x,parm64: 0x%llx",
1094
			   inti->ext.ext_params, inti->ext.ext_params2);
1095
		vcpu->stat.deliver_virtio++;
1096
		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1097
				inti->type,
1098
				inti->ext.ext_params,
1099
				inti->ext.ext_params2);
1100
		list_del(&inti->list);
1101
		fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1102
	}
1103
	if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1104
		clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1105
	spin_unlock(&fi->lock);
1106

1107
	if (inti) {
1108
		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1109
				(u16 *)__LC_EXT_INT_CODE);
1110
		rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1111
				(u16 *)__LC_EXT_CPU_ADDR);
1112
		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1113
				&vcpu->arch.sie_block->gpsw,
1114
				sizeof(psw_t));
1115
		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1116
				&vcpu->arch.sie_block->gpsw,
1117
				sizeof(psw_t));
1118
		rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1119
				(u32 *)__LC_EXT_PARAMS);
1120
		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1121
				(u64 *)__LC_EXT_PARAMS2);
1122
		kfree(inti);
1123
	}
1124
	return rc ? -EFAULT : 0;
1125
}
1126

1127
static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1128
{
1129
	int rc;
1130

1131
	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
1132
		vcpu->arch.sie_block->iictl = IICTL_CODE_IO;
1133
		vcpu->arch.sie_block->subchannel_id = io->subchannel_id;
1134
		vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr;
1135
		vcpu->arch.sie_block->io_int_parm = io->io_int_parm;
1136
		vcpu->arch.sie_block->io_int_word = io->io_int_word;
1137
		return 0;
1138
	}
1139

1140
	rc  = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1141
	rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1142
	rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1143
	rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1144
	rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1145
			     &vcpu->arch.sie_block->gpsw,
1146
			     sizeof(psw_t));
1147
	rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1148
			    &vcpu->arch.sie_block->gpsw,
1149
			    sizeof(psw_t));
1150
	return rc ? -EFAULT : 0;
1151
}
1152

1153
static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1154
				     unsigned long irq_type)
1155
{
1156
	struct list_head *isc_list;
1157
	struct kvm_s390_float_interrupt *fi;
1158
	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1159
	struct kvm_s390_interrupt_info *inti = NULL;
1160
	struct kvm_s390_io_info io;
1161
	u32 isc;
1162
	int rc = 0;
1163

1164
	fi = &vcpu->kvm->arch.float_int;
1165

1166
	spin_lock(&fi->lock);
1167
	isc = irq_type_to_isc(irq_type);
1168
	isc_list = &fi->lists[isc];
1169
	inti = list_first_entry_or_null(isc_list,
1170
					struct kvm_s390_interrupt_info,
1171
					list);
1172
	if (inti) {
1173
		if (inti->type & KVM_S390_INT_IO_AI_MASK)
1174
			VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1175
		else
1176
			VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1177
			inti->io.subchannel_id >> 8,
1178
			inti->io.subchannel_id >> 1 & 0x3,
1179
			inti->io.subchannel_nr);
1180

1181
		vcpu->stat.deliver_io++;
1182
		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1183
				inti->type,
1184
				((__u32)inti->io.subchannel_id << 16) |
1185
				inti->io.subchannel_nr,
1186
				((__u64)inti->io.io_int_parm << 32) |
1187
				inti->io.io_int_word);
1188
		list_del(&inti->list);
1189
		fi->counters[FIRQ_CNTR_IO] -= 1;
1190
	}
1191
	if (list_empty(isc_list))
1192
		clear_bit(irq_type, &fi->pending_irqs);
1193
	spin_unlock(&fi->lock);
1194

1195
	if (inti) {
1196
		rc = __do_deliver_io(vcpu, &(inti->io));
1197
		kfree(inti);
1198
		goto out;
1199
	}
1200

1201
	if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1202
		/*
1203
		 * in case an adapter interrupt was not delivered
1204
		 * in SIE context KVM will handle the delivery
1205
		 */
1206
		VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1207
		memset(&io, 0, sizeof(io));
1208
		io.io_int_word = isc_to_int_word(isc);
1209
		vcpu->stat.deliver_io++;
1210
		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1211
			KVM_S390_INT_IO(1, 0, 0, 0),
1212
			((__u32)io.subchannel_id << 16) |
1213
			io.subchannel_nr,
1214
			((__u64)io.io_int_parm << 32) |
1215
			io.io_int_word);
1216
		rc = __do_deliver_io(vcpu, &io);
1217
	}
1218
out:
1219
	return rc;
1220
}
1221

1222
/* Check whether an external call is pending (deliverable or not) */
1223
int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1224
{
1225
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1226

1227
	if (!sclp.has_sigpif)
1228
		return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1229

1230
	return sca_ext_call_pending(vcpu, NULL);
1231
}
1232

1233
int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1234
{
1235
	if (deliverable_irqs(vcpu))
1236
		return 1;
1237

1238
	if (kvm_cpu_has_pending_timer(vcpu))
1239
		return 1;
1240

1241
	/* external call pending and deliverable */
1242
	if (kvm_s390_ext_call_pending(vcpu) &&
1243
	    !psw_extint_disabled(vcpu) &&
1244
	    (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1245
		return 1;
1246

1247
	if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1248
		return 1;
1249
	return 0;
1250
}
1251

1252
int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1253
{
1254
	return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1255
}
1256

1257
static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1258
{
1259
	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1260
	const u64 ckc = vcpu->arch.sie_block->ckc;
1261
	u64 cputm, sltime = 0;
1262

1263
	if (ckc_interrupts_enabled(vcpu)) {
1264
		if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1265
			if ((s64)now < (s64)ckc)
1266
				sltime = tod_to_ns((s64)ckc - (s64)now);
1267
		} else if (now < ckc) {
1268
			sltime = tod_to_ns(ckc - now);
1269
		}
1270
		/* already expired */
1271
		if (!sltime)
1272
			return 0;
1273
		if (cpu_timer_interrupts_enabled(vcpu)) {
1274
			cputm = kvm_s390_get_cpu_timer(vcpu);
1275
			/* already expired? */
1276
			if (cputm >> 63)
1277
				return 0;
1278
			return min_t(u64, sltime, tod_to_ns(cputm));
1279
		}
1280
	} else if (cpu_timer_interrupts_enabled(vcpu)) {
1281
		sltime = kvm_s390_get_cpu_timer(vcpu);
1282
		/* already expired? */
1283
		if (sltime >> 63)
1284
			return 0;
1285
	}
1286
	return sltime;
1287
}
1288

1289
int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1290
{
1291
	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1292
	u64 sltime;
1293

1294
	vcpu->stat.exit_wait_state++;
1295

1296
	/* fast path */
1297
	if (kvm_arch_vcpu_runnable(vcpu))
1298
		return 0;
1299

1300
	if (psw_interrupts_disabled(vcpu)) {
1301
		VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1302
		return -EOPNOTSUPP; /* disabled wait */
1303
	}
1304

1305
	if (gi->origin &&
1306
	    (gisa_get_ipm_or_restore_iam(gi) &
1307
	     vcpu->arch.sie_block->gcr[6] >> 24))
1308
		return 0;
1309

1310
	if (!ckc_interrupts_enabled(vcpu) &&
1311
	    !cpu_timer_interrupts_enabled(vcpu)) {
1312
		VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1313
		__set_cpu_idle(vcpu);
1314
		goto no_timer;
1315
	}
1316

1317
	sltime = __calculate_sltime(vcpu);
1318
	if (!sltime)
1319
		return 0;
1320

1321
	__set_cpu_idle(vcpu);
1322
	hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1323
	VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1324
no_timer:
1325
	kvm_vcpu_srcu_read_unlock(vcpu);
1326
	vcpu->kvm->arch.float_int.last_sleep_cpu = vcpu->vcpu_idx;
1327
	kvm_vcpu_halt(vcpu);
1328
	vcpu->valid_wakeup = false;
1329
	__unset_cpu_idle(vcpu);
1330
	kvm_vcpu_srcu_read_lock(vcpu);
1331

1332
	hrtimer_cancel(&vcpu->arch.ckc_timer);
1333
	return 0;
1334
}
1335

1336
void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1337
{
1338
	vcpu->valid_wakeup = true;
1339
	kvm_vcpu_wake_up(vcpu);
1340

1341
	/*
1342
	 * The VCPU might not be sleeping but rather executing VSIE. Let's
1343
	 * kick it, so it leaves the SIE to process the request.
1344
	 */
1345
	kvm_s390_vsie_kick(vcpu);
1346
}
1347

1348
enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1349
{
1350
	struct kvm_vcpu *vcpu;
1351
	u64 sltime;
1352

1353
	vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1354
	sltime = __calculate_sltime(vcpu);
1355

1356
	/*
1357
	 * If the monotonic clock runs faster than the tod clock we might be
1358
	 * woken up too early and have to go back to sleep to avoid deadlocks.
1359
	 */
1360
	if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1361
		return HRTIMER_RESTART;
1362
	kvm_s390_vcpu_wakeup(vcpu);
1363
	return HRTIMER_NORESTART;
1364
}
1365

1366
void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1367
{
1368
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1369

1370
	spin_lock(&li->lock);
1371
	li->pending_irqs = 0;
1372
	bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1373
	memset(&li->irq, 0, sizeof(li->irq));
1374
	spin_unlock(&li->lock);
1375

1376
	sca_clear_ext_call(vcpu);
1377
}
1378

1379
int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1380
{
1381
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1382
	int rc = 0;
1383
	bool delivered = false;
1384
	unsigned long irq_type;
1385
	unsigned long irqs;
1386

1387
	__reset_intercept_indicators(vcpu);
1388

1389
	/* pending ckc conditions might have been invalidated */
1390
	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1391
	if (ckc_irq_pending(vcpu))
1392
		set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1393

1394
	/* pending cpu timer conditions might have been invalidated */
1395
	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1396
	if (cpu_timer_irq_pending(vcpu))
1397
		set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1398

1399
	while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1400
		/* bits are in the reverse order of interrupt priority */
1401
		irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1402
		switch (irq_type) {
1403
		case IRQ_PEND_IO_ISC_0:
1404
		case IRQ_PEND_IO_ISC_1:
1405
		case IRQ_PEND_IO_ISC_2:
1406
		case IRQ_PEND_IO_ISC_3:
1407
		case IRQ_PEND_IO_ISC_4:
1408
		case IRQ_PEND_IO_ISC_5:
1409
		case IRQ_PEND_IO_ISC_6:
1410
		case IRQ_PEND_IO_ISC_7:
1411
			rc = __deliver_io(vcpu, irq_type);
1412
			break;
1413
		case IRQ_PEND_MCHK_EX:
1414
		case IRQ_PEND_MCHK_REP:
1415
			rc = __deliver_machine_check(vcpu);
1416
			break;
1417
		case IRQ_PEND_PROG:
1418
			rc = __deliver_prog(vcpu);
1419
			break;
1420
		case IRQ_PEND_EXT_EMERGENCY:
1421
			rc = __deliver_emergency_signal(vcpu);
1422
			break;
1423
		case IRQ_PEND_EXT_EXTERNAL:
1424
			rc = __deliver_external_call(vcpu);
1425
			break;
1426
		case IRQ_PEND_EXT_CLOCK_COMP:
1427
			rc = __deliver_ckc(vcpu);
1428
			break;
1429
		case IRQ_PEND_EXT_CPU_TIMER:
1430
			rc = __deliver_cpu_timer(vcpu);
1431
			break;
1432
		case IRQ_PEND_RESTART:
1433
			rc = __deliver_restart(vcpu);
1434
			break;
1435
		case IRQ_PEND_SET_PREFIX:
1436
			rc = __deliver_set_prefix(vcpu);
1437
			break;
1438
		case IRQ_PEND_PFAULT_INIT:
1439
			rc = __deliver_pfault_init(vcpu);
1440
			break;
1441
		case IRQ_PEND_EXT_SERVICE:
1442
			rc = __deliver_service(vcpu);
1443
			break;
1444
		case IRQ_PEND_EXT_SERVICE_EV:
1445
			rc = __deliver_service_ev(vcpu);
1446
			break;
1447
		case IRQ_PEND_PFAULT_DONE:
1448
			rc = __deliver_pfault_done(vcpu);
1449
			break;
1450
		case IRQ_PEND_VIRTIO:
1451
			rc = __deliver_virtio(vcpu);
1452
			break;
1453
		default:
1454
			WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1455
			clear_bit(irq_type, &li->pending_irqs);
1456
		}
1457
		delivered |= !rc;
1458
	}
1459

1460
	/*
1461
	 * We delivered at least one interrupt and modified the PC. Force a
1462
	 * singlestep event now.
1463
	 */
1464
	if (delivered && guestdbg_sstep_enabled(vcpu)) {
1465
		struct kvm_debug_exit_arch *debug_exit = &vcpu->run->debug.arch;
1466

1467
		debug_exit->addr = vcpu->arch.sie_block->gpsw.addr;
1468
		debug_exit->type = KVM_SINGLESTEP;
1469
		vcpu->guest_debug |= KVM_GUESTDBG_EXIT_PENDING;
1470
	}
1471

1472
	set_intercept_indicators(vcpu);
1473

1474
	return rc;
1475
}
1476

1477
static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1478
{
1479
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1480

1481
	vcpu->stat.inject_program++;
1482
	VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1483
	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1484
				   irq->u.pgm.code, 0);
1485

1486
	if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1487
		/* auto detection if no valid ILC was given */
1488
		irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1489
		irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1490
		irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1491
	}
1492

1493
	if (irq->u.pgm.code == PGM_PER) {
1494
		li->irq.pgm.code |= PGM_PER;
1495
		li->irq.pgm.flags = irq->u.pgm.flags;
1496
		/* only modify PER related information */
1497
		li->irq.pgm.per_address = irq->u.pgm.per_address;
1498
		li->irq.pgm.per_code = irq->u.pgm.per_code;
1499
		li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1500
		li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1501
	} else if (!(irq->u.pgm.code & PGM_PER)) {
1502
		li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1503
				   irq->u.pgm.code;
1504
		li->irq.pgm.flags = irq->u.pgm.flags;
1505
		/* only modify non-PER information */
1506
		li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1507
		li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1508
		li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1509
		li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1510
		li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1511
		li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1512
	} else {
1513
		li->irq.pgm = irq->u.pgm;
1514
	}
1515
	set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1516
	return 0;
1517
}
1518

1519
static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1520
{
1521
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1522

1523
	vcpu->stat.inject_pfault_init++;
1524
	VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1525
		   irq->u.ext.ext_params2);
1526
	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1527
				   irq->u.ext.ext_params,
1528
				   irq->u.ext.ext_params2);
1529

1530
	li->irq.ext = irq->u.ext;
1531
	set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1532
	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1533
	return 0;
1534
}
1535

1536
static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1537
{
1538
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1539
	struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1540
	uint16_t src_id = irq->u.extcall.code;
1541

1542
	vcpu->stat.inject_external_call++;
1543
	VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1544
		   src_id);
1545
	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1546
				   src_id, 0);
1547

1548
	/* sending vcpu invalid */
1549
	if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1550
		return -EINVAL;
1551

1552
	if (sclp.has_sigpif && !kvm_s390_pv_cpu_get_handle(vcpu))
1553
		return sca_inject_ext_call(vcpu, src_id);
1554

1555
	if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1556
		return -EBUSY;
1557
	*extcall = irq->u.extcall;
1558
	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1559
	return 0;
1560
}
1561

1562
static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1563
{
1564
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1565
	struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1566

1567
	vcpu->stat.inject_set_prefix++;
1568
	VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1569
		   irq->u.prefix.address);
1570
	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1571
				   irq->u.prefix.address, 0);
1572

1573
	if (!is_vcpu_stopped(vcpu))
1574
		return -EBUSY;
1575

1576
	*prefix = irq->u.prefix;
1577
	set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1578
	return 0;
1579
}
1580

1581
#define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1582
static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1583
{
1584
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1585
	struct kvm_s390_stop_info *stop = &li->irq.stop;
1586
	int rc = 0;
1587

1588
	vcpu->stat.inject_stop_signal++;
1589
	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1590

1591
	if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1592
		return -EINVAL;
1593

1594
	if (is_vcpu_stopped(vcpu)) {
1595
		if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1596
			rc = kvm_s390_store_status_unloaded(vcpu,
1597
						KVM_S390_STORE_STATUS_NOADDR);
1598
		return rc;
1599
	}
1600

1601
	if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1602
		return -EBUSY;
1603
	stop->flags = irq->u.stop.flags;
1604
	kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1605
	return 0;
1606
}
1607

1608
static int __inject_sigp_restart(struct kvm_vcpu *vcpu)
1609
{
1610
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1611

1612
	vcpu->stat.inject_restart++;
1613
	VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1614
	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1615

1616
	set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1617
	return 0;
1618
}
1619

1620
static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1621
				   struct kvm_s390_irq *irq)
1622
{
1623
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1624

1625
	vcpu->stat.inject_emergency_signal++;
1626
	VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1627
		   irq->u.emerg.code);
1628
	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1629
				   irq->u.emerg.code, 0);
1630

1631
	/* sending vcpu invalid */
1632
	if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1633
		return -EINVAL;
1634

1635
	set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1636
	set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1637
	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1638
	return 0;
1639
}
1640

1641
static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1642
{
1643
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1644
	struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1645

1646
	vcpu->stat.inject_mchk++;
1647
	VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1648
		   irq->u.mchk.mcic);
1649
	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1650
				   irq->u.mchk.mcic);
1651

1652
	/*
1653
	 * Because repressible machine checks can be indicated along with
1654
	 * exigent machine checks (PoP, Chapter 11, Interruption action)
1655
	 * we need to combine cr14, mcic and external damage code.
1656
	 * Failing storage address and the logout area should not be or'ed
1657
	 * together, we just indicate the last occurrence of the corresponding
1658
	 * machine check
1659
	 */
1660
	mchk->cr14 |= irq->u.mchk.cr14;
1661
	mchk->mcic |= irq->u.mchk.mcic;
1662
	mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1663
	mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1664
	memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1665
	       sizeof(mchk->fixed_logout));
1666
	if (mchk->mcic & MCHK_EX_MASK)
1667
		set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1668
	else if (mchk->mcic & MCHK_REP_MASK)
1669
		set_bit(IRQ_PEND_MCHK_REP,  &li->pending_irqs);
1670
	return 0;
1671
}
1672

1673
static int __inject_ckc(struct kvm_vcpu *vcpu)
1674
{
1675
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1676

1677
	vcpu->stat.inject_ckc++;
1678
	VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1679
	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1680
				   0, 0);
1681

1682
	set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1683
	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1684
	return 0;
1685
}
1686

1687
static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1688
{
1689
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1690

1691
	vcpu->stat.inject_cputm++;
1692
	VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1693
	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1694
				   0, 0);
1695

1696
	set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1697
	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1698
	return 0;
1699
}
1700

1701
static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1702
						  int isc, u32 schid)
1703
{
1704
	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1705
	struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1706
	struct kvm_s390_interrupt_info *iter;
1707
	u16 id = (schid & 0xffff0000U) >> 16;
1708
	u16 nr = schid & 0x0000ffffU;
1709

1710
	spin_lock(&fi->lock);
1711
	list_for_each_entry(iter, isc_list, list) {
1712
		if (schid && (id != iter->io.subchannel_id ||
1713
			      nr != iter->io.subchannel_nr))
1714
			continue;
1715
		/* found an appropriate entry */
1716
		list_del_init(&iter->list);
1717
		fi->counters[FIRQ_CNTR_IO] -= 1;
1718
		if (list_empty(isc_list))
1719
			clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1720
		spin_unlock(&fi->lock);
1721
		return iter;
1722
	}
1723
	spin_unlock(&fi->lock);
1724
	return NULL;
1725
}
1726

1727
static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1728
						      u64 isc_mask, u32 schid)
1729
{
1730
	struct kvm_s390_interrupt_info *inti = NULL;
1731
	int isc;
1732

1733
	for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1734
		if (isc_mask & isc_to_isc_bits(isc))
1735
			inti = get_io_int(kvm, isc, schid);
1736
	}
1737
	return inti;
1738
}
1739

1740
static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1741
{
1742
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1743
	unsigned long active_mask;
1744
	int isc;
1745

1746
	if (schid)
1747
		goto out;
1748
	if (!gi->origin)
1749
		goto out;
1750

1751
	active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1752
	while (active_mask) {
1753
		isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1754
		if (gisa_tac_ipm_gisc(gi->origin, isc))
1755
			return isc;
1756
		clear_bit_inv(isc, &active_mask);
1757
	}
1758
out:
1759
	return -EINVAL;
1760
}
1761

1762
/*
1763
 * Dequeue and return an I/O interrupt matching any of the interruption
1764
 * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1765
 * Take into account the interrupts pending in the interrupt list and in GISA.
1766
 *
1767
 * Note that for a guest that does not enable I/O interrupts
1768
 * but relies on TPI, a flood of classic interrupts may starve
1769
 * out adapter interrupts on the same isc. Linux does not do
1770
 * that, and it is possible to work around the issue by configuring
1771
 * different iscs for classic and adapter interrupts in the guest,
1772
 * but we may want to revisit this in the future.
1773
 */
1774
struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1775
						    u64 isc_mask, u32 schid)
1776
{
1777
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1778
	struct kvm_s390_interrupt_info *inti, *tmp_inti;
1779
	int isc;
1780

1781
	inti = get_top_io_int(kvm, isc_mask, schid);
1782

1783
	isc = get_top_gisa_isc(kvm, isc_mask, schid);
1784
	if (isc < 0)
1785
		/* no AI in GISA */
1786
		goto out;
1787

1788
	if (!inti)
1789
		/* AI in GISA but no classical IO int */
1790
		goto gisa_out;
1791

1792
	/* both types of interrupts present */
1793
	if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1794
		/* classical IO int with higher priority */
1795
		gisa_set_ipm_gisc(gi->origin, isc);
1796
		goto out;
1797
	}
1798
gisa_out:
1799
	tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
1800
	if (tmp_inti) {
1801
		tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1802
		tmp_inti->io.io_int_word = isc_to_int_word(isc);
1803
		if (inti)
1804
			kvm_s390_reinject_io_int(kvm, inti);
1805
		inti = tmp_inti;
1806
	} else
1807
		gisa_set_ipm_gisc(gi->origin, isc);
1808
out:
1809
	return inti;
1810
}
1811

1812
static int __inject_service(struct kvm *kvm,
1813
			     struct kvm_s390_interrupt_info *inti)
1814
{
1815
	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1816

1817
	kvm->stat.inject_service_signal++;
1818
	spin_lock(&fi->lock);
1819
	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1820

1821
	/* We always allow events, track them separately from the sccb ints */
1822
	if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING)
1823
		set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1824

1825
	/*
1826
	 * Early versions of the QEMU s390 bios will inject several
1827
	 * service interrupts after another without handling a
1828
	 * condition code indicating busy.
1829
	 * We will silently ignore those superfluous sccb values.
1830
	 * A future version of QEMU will take care of serialization
1831
	 * of servc requests
1832
	 */
1833
	if (fi->srv_signal.ext_params & SCCB_MASK)
1834
		goto out;
1835
	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1836
	set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1837
out:
1838
	spin_unlock(&fi->lock);
1839
	kfree(inti);
1840
	return 0;
1841
}
1842

1843
static int __inject_virtio(struct kvm *kvm,
1844
			    struct kvm_s390_interrupt_info *inti)
1845
{
1846
	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1847

1848
	kvm->stat.inject_virtio++;
1849
	spin_lock(&fi->lock);
1850
	if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1851
		spin_unlock(&fi->lock);
1852
		return -EBUSY;
1853
	}
1854
	fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1855
	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1856
	set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1857
	spin_unlock(&fi->lock);
1858
	return 0;
1859
}
1860

1861
static int __inject_pfault_done(struct kvm *kvm,
1862
				 struct kvm_s390_interrupt_info *inti)
1863
{
1864
	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1865

1866
	kvm->stat.inject_pfault_done++;
1867
	spin_lock(&fi->lock);
1868
	if (fi->counters[FIRQ_CNTR_PFAULT] >=
1869
		(ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1870
		spin_unlock(&fi->lock);
1871
		return -EBUSY;
1872
	}
1873
	fi->counters[FIRQ_CNTR_PFAULT] += 1;
1874
	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1875
	set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1876
	spin_unlock(&fi->lock);
1877
	return 0;
1878
}
1879

1880
#define CR_PENDING_SUBCLASS 28
1881
static int __inject_float_mchk(struct kvm *kvm,
1882
				struct kvm_s390_interrupt_info *inti)
1883
{
1884
	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1885

1886
	kvm->stat.inject_float_mchk++;
1887
	spin_lock(&fi->lock);
1888
	fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1889
	fi->mchk.mcic |= inti->mchk.mcic;
1890
	set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1891
	spin_unlock(&fi->lock);
1892
	kfree(inti);
1893
	return 0;
1894
}
1895

1896
static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1897
{
1898
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1899
	struct kvm_s390_float_interrupt *fi;
1900
	struct list_head *list;
1901
	int isc;
1902

1903
	kvm->stat.inject_io++;
1904
	isc = int_word_to_isc(inti->io.io_int_word);
1905

1906
	/*
1907
	 * We do not use the lock checking variant as this is just a
1908
	 * performance optimization and we do not hold the lock here.
1909
	 * This is ok as the code will pick interrupts from both "lists"
1910
	 * for delivery.
1911
	 */
1912
	if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1913
		VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1914
		gisa_set_ipm_gisc(gi->origin, isc);
1915
		kfree(inti);
1916
		return 0;
1917
	}
1918

1919
	fi = &kvm->arch.float_int;
1920
	spin_lock(&fi->lock);
1921
	if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1922
		spin_unlock(&fi->lock);
1923
		return -EBUSY;
1924
	}
1925
	fi->counters[FIRQ_CNTR_IO] += 1;
1926

1927
	if (inti->type & KVM_S390_INT_IO_AI_MASK)
1928
		VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1929
	else
1930
		VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1931
			inti->io.subchannel_id >> 8,
1932
			inti->io.subchannel_id >> 1 & 0x3,
1933
			inti->io.subchannel_nr);
1934
	list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1935
	list_add_tail(&inti->list, list);
1936
	set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1937
	spin_unlock(&fi->lock);
1938
	return 0;
1939
}
1940

1941
/*
1942
 * Find a destination VCPU for a floating irq and kick it.
1943
 */
1944
static void __floating_irq_kick(struct kvm *kvm, u64 type)
1945
{
1946
	struct kvm_vcpu *dst_vcpu;
1947
	int sigcpu, online_vcpus, nr_tries = 0;
1948

1949
	online_vcpus = atomic_read(&kvm->online_vcpus);
1950
	if (!online_vcpus)
1951
		return;
1952

1953
	for (sigcpu = kvm->arch.float_int.last_sleep_cpu; ; sigcpu++) {
1954
		sigcpu %= online_vcpus;
1955
		dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1956
		if (!is_vcpu_stopped(dst_vcpu))
1957
			break;
1958
		/* avoid endless loops if all vcpus are stopped */
1959
		if (nr_tries++ >= online_vcpus)
1960
			return;
1961
	}
1962

1963
	/* make the VCPU drop out of the SIE, or wake it up if sleeping */
1964
	switch (type) {
1965
	case KVM_S390_MCHK:
1966
		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1967
		break;
1968
	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1969
		if (!(type & KVM_S390_INT_IO_AI_MASK &&
1970
		      kvm->arch.gisa_int.origin) ||
1971
		      kvm_s390_pv_cpu_get_handle(dst_vcpu))
1972
			kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1973
		break;
1974
	default:
1975
		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1976
		break;
1977
	}
1978
	kvm_s390_vcpu_wakeup(dst_vcpu);
1979
}
1980

1981
static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1982
{
1983
	u64 type = READ_ONCE(inti->type);
1984
	int rc;
1985

1986
	switch (type) {
1987
	case KVM_S390_MCHK:
1988
		rc = __inject_float_mchk(kvm, inti);
1989
		break;
1990
	case KVM_S390_INT_VIRTIO:
1991
		rc = __inject_virtio(kvm, inti);
1992
		break;
1993
	case KVM_S390_INT_SERVICE:
1994
		rc = __inject_service(kvm, inti);
1995
		break;
1996
	case KVM_S390_INT_PFAULT_DONE:
1997
		rc = __inject_pfault_done(kvm, inti);
1998
		break;
1999
	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2000
		rc = __inject_io(kvm, inti);
2001
		break;
2002
	default:
2003
		rc = -EINVAL;
2004
	}
2005
	if (rc)
2006
		return rc;
2007

2008
	__floating_irq_kick(kvm, type);
2009
	return 0;
2010
}
2011

2012
int kvm_s390_inject_vm(struct kvm *kvm,
2013
		       struct kvm_s390_interrupt *s390int)
2014
{
2015
	struct kvm_s390_interrupt_info *inti;
2016
	int rc;
2017

2018
	inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2019
	if (!inti)
2020
		return -ENOMEM;
2021

2022
	inti->type = s390int->type;
2023
	switch (inti->type) {
2024
	case KVM_S390_INT_VIRTIO:
2025
		VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
2026
			 s390int->parm, s390int->parm64);
2027
		inti->ext.ext_params = s390int->parm;
2028
		inti->ext.ext_params2 = s390int->parm64;
2029
		break;
2030
	case KVM_S390_INT_SERVICE:
2031
		VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
2032
		inti->ext.ext_params = s390int->parm;
2033
		break;
2034
	case KVM_S390_INT_PFAULT_DONE:
2035
		inti->ext.ext_params2 = s390int->parm64;
2036
		break;
2037
	case KVM_S390_MCHK:
2038
		VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
2039
			 s390int->parm64);
2040
		inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
2041
		inti->mchk.mcic = s390int->parm64;
2042
		break;
2043
	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2044
		inti->io.subchannel_id = s390int->parm >> 16;
2045
		inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
2046
		inti->io.io_int_parm = s390int->parm64 >> 32;
2047
		inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
2048
		break;
2049
	default:
2050
		kfree(inti);
2051
		return -EINVAL;
2052
	}
2053
	trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
2054
				 2);
2055

2056
	rc = __inject_vm(kvm, inti);
2057
	if (rc)
2058
		kfree(inti);
2059
	return rc;
2060
}
2061

2062
int kvm_s390_reinject_io_int(struct kvm *kvm,
2063
			      struct kvm_s390_interrupt_info *inti)
2064
{
2065
	return __inject_vm(kvm, inti);
2066
}
2067

2068
int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
2069
		       struct kvm_s390_irq *irq)
2070
{
2071
	irq->type = s390int->type;
2072
	switch (irq->type) {
2073
	case KVM_S390_PROGRAM_INT:
2074
		if (s390int->parm & 0xffff0000)
2075
			return -EINVAL;
2076
		irq->u.pgm.code = s390int->parm;
2077
		break;
2078
	case KVM_S390_SIGP_SET_PREFIX:
2079
		irq->u.prefix.address = s390int->parm;
2080
		break;
2081
	case KVM_S390_SIGP_STOP:
2082
		irq->u.stop.flags = s390int->parm;
2083
		break;
2084
	case KVM_S390_INT_EXTERNAL_CALL:
2085
		if (s390int->parm & 0xffff0000)
2086
			return -EINVAL;
2087
		irq->u.extcall.code = s390int->parm;
2088
		break;
2089
	case KVM_S390_INT_EMERGENCY:
2090
		if (s390int->parm & 0xffff0000)
2091
			return -EINVAL;
2092
		irq->u.emerg.code = s390int->parm;
2093
		break;
2094
	case KVM_S390_MCHK:
2095
		irq->u.mchk.mcic = s390int->parm64;
2096
		break;
2097
	case KVM_S390_INT_PFAULT_INIT:
2098
		irq->u.ext.ext_params = s390int->parm;
2099
		irq->u.ext.ext_params2 = s390int->parm64;
2100
		break;
2101
	case KVM_S390_RESTART:
2102
	case KVM_S390_INT_CLOCK_COMP:
2103
	case KVM_S390_INT_CPU_TIMER:
2104
		break;
2105
	default:
2106
		return -EINVAL;
2107
	}
2108
	return 0;
2109
}
2110

2111
int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
2112
{
2113
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2114

2115
	return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2116
}
2117

2118
int kvm_s390_is_restart_irq_pending(struct kvm_vcpu *vcpu)
2119
{
2120
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2121

2122
	return test_bit(IRQ_PEND_RESTART, &li->pending_irqs);
2123
}
2124

2125
void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
2126
{
2127
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2128

2129
	spin_lock(&li->lock);
2130
	li->irq.stop.flags = 0;
2131
	clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2132
	spin_unlock(&li->lock);
2133
}
2134

2135
static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2136
{
2137
	int rc;
2138

2139
	switch (irq->type) {
2140
	case KVM_S390_PROGRAM_INT:
2141
		rc = __inject_prog(vcpu, irq);
2142
		break;
2143
	case KVM_S390_SIGP_SET_PREFIX:
2144
		rc = __inject_set_prefix(vcpu, irq);
2145
		break;
2146
	case KVM_S390_SIGP_STOP:
2147
		rc = __inject_sigp_stop(vcpu, irq);
2148
		break;
2149
	case KVM_S390_RESTART:
2150
		rc = __inject_sigp_restart(vcpu);
2151
		break;
2152
	case KVM_S390_INT_CLOCK_COMP:
2153
		rc = __inject_ckc(vcpu);
2154
		break;
2155
	case KVM_S390_INT_CPU_TIMER:
2156
		rc = __inject_cpu_timer(vcpu);
2157
		break;
2158
	case KVM_S390_INT_EXTERNAL_CALL:
2159
		rc = __inject_extcall(vcpu, irq);
2160
		break;
2161
	case KVM_S390_INT_EMERGENCY:
2162
		rc = __inject_sigp_emergency(vcpu, irq);
2163
		break;
2164
	case KVM_S390_MCHK:
2165
		rc = __inject_mchk(vcpu, irq);
2166
		break;
2167
	case KVM_S390_INT_PFAULT_INIT:
2168
		rc = __inject_pfault_init(vcpu, irq);
2169
		break;
2170
	case KVM_S390_INT_VIRTIO:
2171
	case KVM_S390_INT_SERVICE:
2172
	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2173
	default:
2174
		rc = -EINVAL;
2175
	}
2176

2177
	return rc;
2178
}
2179

2180
int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2181
{
2182
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2183
	int rc;
2184

2185
	spin_lock(&li->lock);
2186
	rc = do_inject_vcpu(vcpu, irq);
2187
	spin_unlock(&li->lock);
2188
	if (!rc)
2189
		kvm_s390_vcpu_wakeup(vcpu);
2190
	return rc;
2191
}
2192

2193
static inline void clear_irq_list(struct list_head *_list)
2194
{
2195
	struct kvm_s390_interrupt_info *inti, *n;
2196

2197
	list_for_each_entry_safe(inti, n, _list, list) {
2198
		list_del(&inti->list);
2199
		kfree(inti);
2200
	}
2201
}
2202

2203
static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2204
		       struct kvm_s390_irq *irq)
2205
{
2206
	irq->type = inti->type;
2207
	switch (inti->type) {
2208
	case KVM_S390_INT_PFAULT_INIT:
2209
	case KVM_S390_INT_PFAULT_DONE:
2210
	case KVM_S390_INT_VIRTIO:
2211
		irq->u.ext = inti->ext;
2212
		break;
2213
	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2214
		irq->u.io = inti->io;
2215
		break;
2216
	}
2217
}
2218

2219
void kvm_s390_clear_float_irqs(struct kvm *kvm)
2220
{
2221
	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2222
	int i;
2223

2224
	mutex_lock(&kvm->lock);
2225
	if (!kvm_s390_pv_is_protected(kvm))
2226
		fi->masked_irqs = 0;
2227
	mutex_unlock(&kvm->lock);
2228
	spin_lock(&fi->lock);
2229
	fi->pending_irqs = 0;
2230
	memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2231
	memset(&fi->mchk, 0, sizeof(fi->mchk));
2232
	for (i = 0; i < FIRQ_LIST_COUNT; i++)
2233
		clear_irq_list(&fi->lists[i]);
2234
	for (i = 0; i < FIRQ_MAX_COUNT; i++)
2235
		fi->counters[i] = 0;
2236
	spin_unlock(&fi->lock);
2237
	kvm_s390_gisa_clear(kvm);
2238
};
2239

2240
static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2241
{
2242
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2243
	struct kvm_s390_interrupt_info *inti;
2244
	struct kvm_s390_float_interrupt *fi;
2245
	struct kvm_s390_irq *buf;
2246
	struct kvm_s390_irq *irq;
2247
	int max_irqs;
2248
	int ret = 0;
2249
	int n = 0;
2250
	int i;
2251

2252
	if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2253
		return -EINVAL;
2254

2255
	/*
2256
	 * We are already using -ENOMEM to signal
2257
	 * userspace it may retry with a bigger buffer,
2258
	 * so we need to use something else for this case
2259
	 */
2260
	buf = vzalloc(len);
2261
	if (!buf)
2262
		return -ENOBUFS;
2263

2264
	max_irqs = len / sizeof(struct kvm_s390_irq);
2265

2266
	if (gi->origin && gisa_get_ipm(gi->origin)) {
2267
		for (i = 0; i <= MAX_ISC; i++) {
2268
			if (n == max_irqs) {
2269
				/* signal userspace to try again */
2270
				ret = -ENOMEM;
2271
				goto out_nolock;
2272
			}
2273
			if (gisa_tac_ipm_gisc(gi->origin, i)) {
2274
				irq = (struct kvm_s390_irq *) &buf[n];
2275
				irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2276
				irq->u.io.io_int_word = isc_to_int_word(i);
2277
				n++;
2278
			}
2279
		}
2280
	}
2281
	fi = &kvm->arch.float_int;
2282
	spin_lock(&fi->lock);
2283
	for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2284
		list_for_each_entry(inti, &fi->lists[i], list) {
2285
			if (n == max_irqs) {
2286
				/* signal userspace to try again */
2287
				ret = -ENOMEM;
2288
				goto out;
2289
			}
2290
			inti_to_irq(inti, &buf[n]);
2291
			n++;
2292
		}
2293
	}
2294
	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) ||
2295
	    test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) {
2296
		if (n == max_irqs) {
2297
			/* signal userspace to try again */
2298
			ret = -ENOMEM;
2299
			goto out;
2300
		}
2301
		irq = (struct kvm_s390_irq *) &buf[n];
2302
		irq->type = KVM_S390_INT_SERVICE;
2303
		irq->u.ext = fi->srv_signal;
2304
		n++;
2305
	}
2306
	if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2307
		if (n == max_irqs) {
2308
				/* signal userspace to try again */
2309
				ret = -ENOMEM;
2310
				goto out;
2311
		}
2312
		irq = (struct kvm_s390_irq *) &buf[n];
2313
		irq->type = KVM_S390_MCHK;
2314
		irq->u.mchk = fi->mchk;
2315
		n++;
2316
}
2317

2318
out:
2319
	spin_unlock(&fi->lock);
2320
out_nolock:
2321
	if (!ret && n > 0) {
2322
		if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2323
			ret = -EFAULT;
2324
	}
2325
	vfree(buf);
2326

2327
	return ret < 0 ? ret : n;
2328
}
2329

2330
static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2331
{
2332
	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2333
	struct kvm_s390_ais_all ais;
2334

2335
	if (attr->attr < sizeof(ais))
2336
		return -EINVAL;
2337

2338
	if (!test_kvm_facility(kvm, 72))
2339
		return -EOPNOTSUPP;
2340

2341
	mutex_lock(&fi->ais_lock);
2342
	ais.simm = fi->simm;
2343
	ais.nimm = fi->nimm;
2344
	mutex_unlock(&fi->ais_lock);
2345

2346
	if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2347
		return -EFAULT;
2348

2349
	return 0;
2350
}
2351

2352
static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2353
{
2354
	int r;
2355

2356
	switch (attr->group) {
2357
	case KVM_DEV_FLIC_GET_ALL_IRQS:
2358
		r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2359
					  attr->attr);
2360
		break;
2361
	case KVM_DEV_FLIC_AISM_ALL:
2362
		r = flic_ais_mode_get_all(dev->kvm, attr);
2363
		break;
2364
	default:
2365
		r = -EINVAL;
2366
	}
2367

2368
	return r;
2369
}
2370

2371
static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2372
				     u64 addr)
2373
{
2374
	struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2375
	void *target = NULL;
2376
	void __user *source;
2377
	u64 size;
2378

2379
	if (get_user(inti->type, (u64 __user *)addr))
2380
		return -EFAULT;
2381

2382
	switch (inti->type) {
2383
	case KVM_S390_INT_PFAULT_INIT:
2384
	case KVM_S390_INT_PFAULT_DONE:
2385
	case KVM_S390_INT_VIRTIO:
2386
	case KVM_S390_INT_SERVICE:
2387
		target = (void *) &inti->ext;
2388
		source = &uptr->u.ext;
2389
		size = sizeof(inti->ext);
2390
		break;
2391
	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2392
		target = (void *) &inti->io;
2393
		source = &uptr->u.io;
2394
		size = sizeof(inti->io);
2395
		break;
2396
	case KVM_S390_MCHK:
2397
		target = (void *) &inti->mchk;
2398
		source = &uptr->u.mchk;
2399
		size = sizeof(inti->mchk);
2400
		break;
2401
	default:
2402
		return -EINVAL;
2403
	}
2404

2405
	if (copy_from_user(target, source, size))
2406
		return -EFAULT;
2407

2408
	return 0;
2409
}
2410

2411
static int enqueue_floating_irq(struct kvm_device *dev,
2412
				struct kvm_device_attr *attr)
2413
{
2414
	struct kvm_s390_interrupt_info *inti = NULL;
2415
	int r = 0;
2416
	int len = attr->attr;
2417

2418
	if (len % sizeof(struct kvm_s390_irq) != 0)
2419
		return -EINVAL;
2420
	else if (len > KVM_S390_FLIC_MAX_BUFFER)
2421
		return -EINVAL;
2422

2423
	while (len >= sizeof(struct kvm_s390_irq)) {
2424
		inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2425
		if (!inti)
2426
			return -ENOMEM;
2427

2428
		r = copy_irq_from_user(inti, attr->addr);
2429
		if (r) {
2430
			kfree(inti);
2431
			return r;
2432
		}
2433
		r = __inject_vm(dev->kvm, inti);
2434
		if (r) {
2435
			kfree(inti);
2436
			return r;
2437
		}
2438
		len -= sizeof(struct kvm_s390_irq);
2439
		attr->addr += sizeof(struct kvm_s390_irq);
2440
	}
2441

2442
	return r;
2443
}
2444

2445
static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2446
{
2447
	if (id >= MAX_S390_IO_ADAPTERS)
2448
		return NULL;
2449
	id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2450
	return kvm->arch.adapters[id];
2451
}
2452

2453
static int register_io_adapter(struct kvm_device *dev,
2454
			       struct kvm_device_attr *attr)
2455
{
2456
	struct s390_io_adapter *adapter;
2457
	struct kvm_s390_io_adapter adapter_info;
2458

2459
	if (copy_from_user(&adapter_info,
2460
			   (void __user *)attr->addr, sizeof(adapter_info)))
2461
		return -EFAULT;
2462

2463
	if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2464
		return -EINVAL;
2465

2466
	adapter_info.id = array_index_nospec(adapter_info.id,
2467
					     MAX_S390_IO_ADAPTERS);
2468

2469
	if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2470
		return -EINVAL;
2471

2472
	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL_ACCOUNT);
2473
	if (!adapter)
2474
		return -ENOMEM;
2475

2476
	adapter->id = adapter_info.id;
2477
	adapter->isc = adapter_info.isc;
2478
	adapter->maskable = adapter_info.maskable;
2479
	adapter->masked = false;
2480
	adapter->swap = adapter_info.swap;
2481
	adapter->suppressible = (adapter_info.flags) &
2482
				KVM_S390_ADAPTER_SUPPRESSIBLE;
2483
	dev->kvm->arch.adapters[adapter->id] = adapter;
2484

2485
	return 0;
2486
}
2487

2488
int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2489
{
2490
	int ret;
2491
	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2492

2493
	if (!adapter || !adapter->maskable)
2494
		return -EINVAL;
2495
	ret = adapter->masked;
2496
	adapter->masked = masked;
2497
	return ret;
2498
}
2499

2500
void kvm_s390_destroy_adapters(struct kvm *kvm)
2501
{
2502
	int i;
2503

2504
	for (i = 0; i < MAX_S390_IO_ADAPTERS; i++)
2505
		kfree(kvm->arch.adapters[i]);
2506
}
2507

2508
static int modify_io_adapter(struct kvm_device *dev,
2509
			     struct kvm_device_attr *attr)
2510
{
2511
	struct kvm_s390_io_adapter_req req;
2512
	struct s390_io_adapter *adapter;
2513
	int ret;
2514

2515
	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2516
		return -EFAULT;
2517

2518
	adapter = get_io_adapter(dev->kvm, req.id);
2519
	if (!adapter)
2520
		return -EINVAL;
2521
	switch (req.type) {
2522
	case KVM_S390_IO_ADAPTER_MASK:
2523
		ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2524
		if (ret > 0)
2525
			ret = 0;
2526
		break;
2527
	/*
2528
	 * The following operations are no longer needed and therefore no-ops.
2529
	 * The gpa to hva translation is done when an IRQ route is set up. The
2530
	 * set_irq code uses get_user_pages_remote() to do the actual write.
2531
	 */
2532
	case KVM_S390_IO_ADAPTER_MAP:
2533
	case KVM_S390_IO_ADAPTER_UNMAP:
2534
		ret = 0;
2535
		break;
2536
	default:
2537
		ret = -EINVAL;
2538
	}
2539

2540
	return ret;
2541
}
2542

2543
static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2544

2545
{
2546
	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2547
	u32 schid;
2548

2549
	if (attr->flags)
2550
		return -EINVAL;
2551
	if (attr->attr != sizeof(schid))
2552
		return -EINVAL;
2553
	if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2554
		return -EFAULT;
2555
	if (!schid)
2556
		return -EINVAL;
2557
	kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2558
	/*
2559
	 * If userspace is conforming to the architecture, we can have at most
2560
	 * one pending I/O interrupt per subchannel, so this is effectively a
2561
	 * clear all.
2562
	 */
2563
	return 0;
2564
}
2565

2566
static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2567
{
2568
	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2569
	struct kvm_s390_ais_req req;
2570
	int ret = 0;
2571

2572
	if (!test_kvm_facility(kvm, 72))
2573
		return -EOPNOTSUPP;
2574

2575
	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2576
		return -EFAULT;
2577

2578
	if (req.isc > MAX_ISC)
2579
		return -EINVAL;
2580

2581
	trace_kvm_s390_modify_ais_mode(req.isc,
2582
				       (fi->simm & AIS_MODE_MASK(req.isc)) ?
2583
				       (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2584
				       2 : KVM_S390_AIS_MODE_SINGLE :
2585
				       KVM_S390_AIS_MODE_ALL, req.mode);
2586

2587
	mutex_lock(&fi->ais_lock);
2588
	switch (req.mode) {
2589
	case KVM_S390_AIS_MODE_ALL:
2590
		fi->simm &= ~AIS_MODE_MASK(req.isc);
2591
		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2592
		break;
2593
	case KVM_S390_AIS_MODE_SINGLE:
2594
		fi->simm |= AIS_MODE_MASK(req.isc);
2595
		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2596
		break;
2597
	default:
2598
		ret = -EINVAL;
2599
	}
2600
	mutex_unlock(&fi->ais_lock);
2601

2602
	return ret;
2603
}
2604

2605
static int kvm_s390_inject_airq(struct kvm *kvm,
2606
				struct s390_io_adapter *adapter)
2607
{
2608
	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2609
	struct kvm_s390_interrupt s390int = {
2610
		.type = KVM_S390_INT_IO(1, 0, 0, 0),
2611
		.parm = 0,
2612
		.parm64 = isc_to_int_word(adapter->isc),
2613
	};
2614
	int ret = 0;
2615

2616
	if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2617
		return kvm_s390_inject_vm(kvm, &s390int);
2618

2619
	mutex_lock(&fi->ais_lock);
2620
	if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2621
		trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2622
		goto out;
2623
	}
2624

2625
	ret = kvm_s390_inject_vm(kvm, &s390int);
2626
	if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2627
		fi->nimm |= AIS_MODE_MASK(adapter->isc);
2628
		trace_kvm_s390_modify_ais_mode(adapter->isc,
2629
					       KVM_S390_AIS_MODE_SINGLE, 2);
2630
	}
2631
out:
2632
	mutex_unlock(&fi->ais_lock);
2633
	return ret;
2634
}
2635

2636
static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2637
{
2638
	unsigned int id = attr->attr;
2639
	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2640

2641
	if (!adapter)
2642
		return -EINVAL;
2643

2644
	return kvm_s390_inject_airq(kvm, adapter);
2645
}
2646

2647
static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2648
{
2649
	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2650
	struct kvm_s390_ais_all ais;
2651

2652
	if (!test_kvm_facility(kvm, 72))
2653
		return -EOPNOTSUPP;
2654

2655
	if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2656
		return -EFAULT;
2657

2658
	mutex_lock(&fi->ais_lock);
2659
	fi->simm = ais.simm;
2660
	fi->nimm = ais.nimm;
2661
	mutex_unlock(&fi->ais_lock);
2662

2663
	return 0;
2664
}
2665

2666
static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2667
{
2668
	int r = 0;
2669
	unsigned long i;
2670
	struct kvm_vcpu *vcpu;
2671

2672
	switch (attr->group) {
2673
	case KVM_DEV_FLIC_ENQUEUE:
2674
		r = enqueue_floating_irq(dev, attr);
2675
		break;
2676
	case KVM_DEV_FLIC_CLEAR_IRQS:
2677
		kvm_s390_clear_float_irqs(dev->kvm);
2678
		break;
2679
	case KVM_DEV_FLIC_APF_ENABLE:
2680
		if (kvm_is_ucontrol(dev->kvm))
2681
			return -EINVAL;
2682
		dev->kvm->arch.gmap->pfault_enabled = 1;
2683
		break;
2684
	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2685
		if (kvm_is_ucontrol(dev->kvm))
2686
			return -EINVAL;
2687
		dev->kvm->arch.gmap->pfault_enabled = 0;
2688
		/*
2689
		 * Make sure no async faults are in transition when
2690
		 * clearing the queues. So we don't need to worry
2691
		 * about late coming workers.
2692
		 */
2693
		synchronize_srcu(&dev->kvm->srcu);
2694
		kvm_for_each_vcpu(i, vcpu, dev->kvm)
2695
			kvm_clear_async_pf_completion_queue(vcpu);
2696
		break;
2697
	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2698
		r = register_io_adapter(dev, attr);
2699
		break;
2700
	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2701
		r = modify_io_adapter(dev, attr);
2702
		break;
2703
	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2704
		r = clear_io_irq(dev->kvm, attr);
2705
		break;
2706
	case KVM_DEV_FLIC_AISM:
2707
		r = modify_ais_mode(dev->kvm, attr);
2708
		break;
2709
	case KVM_DEV_FLIC_AIRQ_INJECT:
2710
		r = flic_inject_airq(dev->kvm, attr);
2711
		break;
2712
	case KVM_DEV_FLIC_AISM_ALL:
2713
		r = flic_ais_mode_set_all(dev->kvm, attr);
2714
		break;
2715
	default:
2716
		r = -EINVAL;
2717
	}
2718

2719
	return r;
2720
}
2721

2722
static int flic_has_attr(struct kvm_device *dev,
2723
			     struct kvm_device_attr *attr)
2724
{
2725
	switch (attr->group) {
2726
	case KVM_DEV_FLIC_GET_ALL_IRQS:
2727
	case KVM_DEV_FLIC_ENQUEUE:
2728
	case KVM_DEV_FLIC_CLEAR_IRQS:
2729
	case KVM_DEV_FLIC_APF_ENABLE:
2730
	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2731
	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2732
	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2733
	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2734
	case KVM_DEV_FLIC_AISM:
2735
	case KVM_DEV_FLIC_AIRQ_INJECT:
2736
	case KVM_DEV_FLIC_AISM_ALL:
2737
		return 0;
2738
	}
2739
	return -ENXIO;
2740
}
2741

2742
static int flic_create(struct kvm_device *dev, u32 type)
2743
{
2744
	if (!dev)
2745
		return -EINVAL;
2746
	if (dev->kvm->arch.flic)
2747
		return -EINVAL;
2748
	dev->kvm->arch.flic = dev;
2749
	return 0;
2750
}
2751

2752
static void flic_destroy(struct kvm_device *dev)
2753
{
2754
	dev->kvm->arch.flic = NULL;
2755
	kfree(dev);
2756
}
2757

2758
/* s390 floating irq controller (flic) */
2759
struct kvm_device_ops kvm_flic_ops = {
2760
	.name = "kvm-flic",
2761
	.get_attr = flic_get_attr,
2762
	.set_attr = flic_set_attr,
2763
	.has_attr = flic_has_attr,
2764
	.create = flic_create,
2765
	.destroy = flic_destroy,
2766
};
2767

2768
static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2769
{
2770
	unsigned long bit;
2771

2772
	bit = bit_nr + (addr % PAGE_SIZE) * 8;
2773

2774
	return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2775
}
2776

2777
static struct page *get_map_page(struct kvm *kvm, u64 uaddr)
2778
{
2779
	struct mm_struct *mm = kvm->mm;
2780
	struct page *page = NULL;
2781
	int locked = 1;
2782

2783
	if (mmget_not_zero(mm)) {
2784
		mmap_read_lock(mm);
2785
		get_user_pages_remote(mm, uaddr, 1, FOLL_WRITE,
2786
				      &page, &locked);
2787
		if (locked)
2788
			mmap_read_unlock(mm);
2789
		mmput(mm);
2790
	}
2791

2792
	return page;
2793
}
2794

2795
static int adapter_indicators_set(struct kvm *kvm,
2796
				  struct s390_io_adapter *adapter,
2797
				  struct kvm_s390_adapter_int *adapter_int)
2798
{
2799
	unsigned long bit;
2800
	int summary_set, idx;
2801
	struct page *ind_page, *summary_page;
2802
	void *map;
2803

2804
	ind_page = get_map_page(kvm, adapter_int->ind_addr);
2805
	if (!ind_page)
2806
		return -1;
2807
	summary_page = get_map_page(kvm, adapter_int->summary_addr);
2808
	if (!summary_page) {
2809
		put_page(ind_page);
2810
		return -1;
2811
	}
2812

2813
	idx = srcu_read_lock(&kvm->srcu);
2814
	map = page_address(ind_page);
2815
	bit = get_ind_bit(adapter_int->ind_addr,
2816
			  adapter_int->ind_offset, adapter->swap);
2817
	set_bit(bit, map);
2818
	mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT);
2819
	set_page_dirty_lock(ind_page);
2820
	map = page_address(summary_page);
2821
	bit = get_ind_bit(adapter_int->summary_addr,
2822
			  adapter_int->summary_offset, adapter->swap);
2823
	summary_set = test_and_set_bit(bit, map);
2824
	mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT);
2825
	set_page_dirty_lock(summary_page);
2826
	srcu_read_unlock(&kvm->srcu, idx);
2827

2828
	put_page(ind_page);
2829
	put_page(summary_page);
2830
	return summary_set ? 0 : 1;
2831
}
2832

2833
/*
2834
 * < 0 - not injected due to error
2835
 * = 0 - coalesced, summary indicator already active
2836
 * > 0 - injected interrupt
2837
 */
2838
static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2839
			   struct kvm *kvm, int irq_source_id, int level,
2840
			   bool line_status)
2841
{
2842
	int ret;
2843
	struct s390_io_adapter *adapter;
2844

2845
	/* We're only interested in the 0->1 transition. */
2846
	if (!level)
2847
		return 0;
2848
	adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2849
	if (!adapter)
2850
		return -1;
2851
	ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2852
	if ((ret > 0) && !adapter->masked) {
2853
		ret = kvm_s390_inject_airq(kvm, adapter);
2854
		if (ret == 0)
2855
			ret = 1;
2856
	}
2857
	return ret;
2858
}
2859

2860
/*
2861
 * Inject the machine check to the guest.
2862
 */
2863
void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2864
				     struct mcck_volatile_info *mcck_info)
2865
{
2866
	struct kvm_s390_interrupt_info inti;
2867
	struct kvm_s390_irq irq;
2868
	struct kvm_s390_mchk_info *mchk;
2869
	union mci mci;
2870
	__u64 cr14 = 0;         /* upper bits are not used */
2871
	int rc;
2872

2873
	mci.val = mcck_info->mcic;
2874
	if (mci.sr)
2875
		cr14 |= CR14_RECOVERY_SUBMASK;
2876
	if (mci.dg)
2877
		cr14 |= CR14_DEGRADATION_SUBMASK;
2878
	if (mci.w)
2879
		cr14 |= CR14_WARNING_SUBMASK;
2880

2881
	mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2882
	mchk->cr14 = cr14;
2883
	mchk->mcic = mcck_info->mcic;
2884
	mchk->ext_damage_code = mcck_info->ext_damage_code;
2885
	mchk->failing_storage_address = mcck_info->failing_storage_address;
2886
	if (mci.ck) {
2887
		/* Inject the floating machine check */
2888
		inti.type = KVM_S390_MCHK;
2889
		rc = __inject_vm(vcpu->kvm, &inti);
2890
	} else {
2891
		/* Inject the machine check to specified vcpu */
2892
		irq.type = KVM_S390_MCHK;
2893
		rc = kvm_s390_inject_vcpu(vcpu, &irq);
2894
	}
2895
	WARN_ON_ONCE(rc);
2896
}
2897

2898
int kvm_set_routing_entry(struct kvm *kvm,
2899
			  struct kvm_kernel_irq_routing_entry *e,
2900
			  const struct kvm_irq_routing_entry *ue)
2901
{
2902
	u64 uaddr_s, uaddr_i;
2903
	int idx;
2904

2905
	switch (ue->type) {
2906
	/* we store the userspace addresses instead of the guest addresses */
2907
	case KVM_IRQ_ROUTING_S390_ADAPTER:
2908
		if (kvm_is_ucontrol(kvm))
2909
			return -EINVAL;
2910
		e->set = set_adapter_int;
2911

2912
		idx = srcu_read_lock(&kvm->srcu);
2913
		uaddr_s = gpa_to_hva(kvm, ue->u.adapter.summary_addr);
2914
		uaddr_i = gpa_to_hva(kvm, ue->u.adapter.ind_addr);
2915
		srcu_read_unlock(&kvm->srcu, idx);
2916

2917
		if (kvm_is_error_hva(uaddr_s) || kvm_is_error_hva(uaddr_i))
2918
			return -EFAULT;
2919
		e->adapter.summary_addr = uaddr_s;
2920
		e->adapter.ind_addr = uaddr_i;
2921
		e->adapter.summary_offset = ue->u.adapter.summary_offset;
2922
		e->adapter.ind_offset = ue->u.adapter.ind_offset;
2923
		e->adapter.adapter_id = ue->u.adapter.adapter_id;
2924
		return 0;
2925
	default:
2926
		return -EINVAL;
2927
	}
2928
}
2929

2930
int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2931
		int irq_source_id, int level, bool line_status)
2932
{
2933
	return -EINVAL;
2934
}
2935

2936
int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2937
{
2938
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2939
	struct kvm_s390_irq *buf;
2940
	int r = 0;
2941
	int n;
2942

2943
	buf = vmalloc(len);
2944
	if (!buf)
2945
		return -ENOMEM;
2946

2947
	if (copy_from_user((void *) buf, irqstate, len)) {
2948
		r = -EFAULT;
2949
		goto out_free;
2950
	}
2951

2952
	/*
2953
	 * Don't allow setting the interrupt state
2954
	 * when there are already interrupts pending
2955
	 */
2956
	spin_lock(&li->lock);
2957
	if (li->pending_irqs) {
2958
		r = -EBUSY;
2959
		goto out_unlock;
2960
	}
2961

2962
	for (n = 0; n < len / sizeof(*buf); n++) {
2963
		r = do_inject_vcpu(vcpu, &buf[n]);
2964
		if (r)
2965
			break;
2966
	}
2967

2968
out_unlock:
2969
	spin_unlock(&li->lock);
2970
out_free:
2971
	vfree(buf);
2972

2973
	return r;
2974
}
2975

2976
static void store_local_irq(struct kvm_s390_local_interrupt *li,
2977
			    struct kvm_s390_irq *irq,
2978
			    unsigned long irq_type)
2979
{
2980
	switch (irq_type) {
2981
	case IRQ_PEND_MCHK_EX:
2982
	case IRQ_PEND_MCHK_REP:
2983
		irq->type = KVM_S390_MCHK;
2984
		irq->u.mchk = li->irq.mchk;
2985
		break;
2986
	case IRQ_PEND_PROG:
2987
		irq->type = KVM_S390_PROGRAM_INT;
2988
		irq->u.pgm = li->irq.pgm;
2989
		break;
2990
	case IRQ_PEND_PFAULT_INIT:
2991
		irq->type = KVM_S390_INT_PFAULT_INIT;
2992
		irq->u.ext = li->irq.ext;
2993
		break;
2994
	case IRQ_PEND_EXT_EXTERNAL:
2995
		irq->type = KVM_S390_INT_EXTERNAL_CALL;
2996
		irq->u.extcall = li->irq.extcall;
2997
		break;
2998
	case IRQ_PEND_EXT_CLOCK_COMP:
2999
		irq->type = KVM_S390_INT_CLOCK_COMP;
3000
		break;
3001
	case IRQ_PEND_EXT_CPU_TIMER:
3002
		irq->type = KVM_S390_INT_CPU_TIMER;
3003
		break;
3004
	case IRQ_PEND_SIGP_STOP:
3005
		irq->type = KVM_S390_SIGP_STOP;
3006
		irq->u.stop = li->irq.stop;
3007
		break;
3008
	case IRQ_PEND_RESTART:
3009
		irq->type = KVM_S390_RESTART;
3010
		break;
3011
	case IRQ_PEND_SET_PREFIX:
3012
		irq->type = KVM_S390_SIGP_SET_PREFIX;
3013
		irq->u.prefix = li->irq.prefix;
3014
		break;
3015
	}
3016
}
3017

3018
int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
3019
{
3020
	int scn;
3021
	DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
3022
	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
3023
	unsigned long pending_irqs;
3024
	struct kvm_s390_irq irq;
3025
	unsigned long irq_type;
3026
	int cpuaddr;
3027
	int n = 0;
3028

3029
	spin_lock(&li->lock);
3030
	pending_irqs = li->pending_irqs;
3031
	memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
3032
	       sizeof(sigp_emerg_pending));
3033
	spin_unlock(&li->lock);
3034

3035
	for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
3036
		memset(&irq, 0, sizeof(irq));
3037
		if (irq_type == IRQ_PEND_EXT_EMERGENCY)
3038
			continue;
3039
		if (n + sizeof(irq) > len)
3040
			return -ENOBUFS;
3041
		store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
3042
		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3043
			return -EFAULT;
3044
		n += sizeof(irq);
3045
	}
3046

3047
	if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
3048
		for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
3049
			memset(&irq, 0, sizeof(irq));
3050
			if (n + sizeof(irq) > len)
3051
				return -ENOBUFS;
3052
			irq.type = KVM_S390_INT_EMERGENCY;
3053
			irq.u.emerg.code = cpuaddr;
3054
			if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3055
				return -EFAULT;
3056
			n += sizeof(irq);
3057
		}
3058
	}
3059

3060
	if (sca_ext_call_pending(vcpu, &scn)) {
3061
		if (n + sizeof(irq) > len)
3062
			return -ENOBUFS;
3063
		memset(&irq, 0, sizeof(irq));
3064
		irq.type = KVM_S390_INT_EXTERNAL_CALL;
3065
		irq.u.extcall.code = scn;
3066
		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3067
			return -EFAULT;
3068
		n += sizeof(irq);
3069
	}
3070

3071
	return n;
3072
}
3073

3074
static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
3075
{
3076
	int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus);
3077
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3078
	struct kvm_vcpu *vcpu;
3079
	u8 vcpu_isc_mask;
3080

3081
	for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) {
3082
		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
3083
		if (psw_ioint_disabled(vcpu))
3084
			continue;
3085
		vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
3086
		if (deliverable_mask & vcpu_isc_mask) {
3087
			/* lately kicked but not yet running */
3088
			if (test_and_set_bit(vcpu_idx, gi->kicked_mask))
3089
				return;
3090
			kvm_s390_vcpu_wakeup(vcpu);
3091
			return;
3092
		}
3093
	}
3094
}
3095

3096
static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3097
{
3098
	struct kvm_s390_gisa_interrupt *gi =
3099
		container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3100
	struct kvm *kvm =
3101
		container_of(gi->origin, struct sie_page2, gisa)->kvm;
3102
	u8 pending_mask;
3103

3104
	pending_mask = gisa_get_ipm_or_restore_iam(gi);
3105
	if (pending_mask) {
3106
		__airqs_kick_single_vcpu(kvm, pending_mask);
3107
		hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3108
		return HRTIMER_RESTART;
3109
	}
3110

3111
	return HRTIMER_NORESTART;
3112
}
3113

3114
#define NULL_GISA_ADDR 0x00000000UL
3115
#define NONE_GISA_ADDR 0x00000001UL
3116
#define GISA_ADDR_MASK 0xfffff000UL
3117

3118
static void process_gib_alert_list(void)
3119
{
3120
	struct kvm_s390_gisa_interrupt *gi;
3121
	u32 final, gisa_phys, origin = 0UL;
3122
	struct kvm_s390_gisa *gisa;
3123
	struct kvm *kvm;
3124

3125
	do {
3126
		/*
3127
		 * If the NONE_GISA_ADDR is still stored in the alert list
3128
		 * origin, we will leave the outer loop. No further GISA has
3129
		 * been added to the alert list by millicode while processing
3130
		 * the current alert list.
3131
		 */
3132
		final = (origin & NONE_GISA_ADDR);
3133
		/*
3134
		 * Cut off the alert list and store the NONE_GISA_ADDR in the
3135
		 * alert list origin to avoid further GAL interruptions.
3136
		 * A new alert list can be build up by millicode in parallel
3137
		 * for guests not in the yet cut-off alert list. When in the
3138
		 * final loop, store the NULL_GISA_ADDR instead. This will re-
3139
		 * enable GAL interruptions on the host again.
3140
		 */
3141
		origin = xchg(&gib->alert_list_origin,
3142
			      (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3143
		/*
3144
		 * Loop through the just cut-off alert list and start the
3145
		 * gisa timers to kick idle vcpus to consume the pending
3146
		 * interruptions asap.
3147
		 */
3148
		while (origin & GISA_ADDR_MASK) {
3149
			gisa_phys = origin;
3150
			gisa = phys_to_virt(gisa_phys);
3151
			origin = gisa->next_alert;
3152
			gisa->next_alert = gisa_phys;
3153
			kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3154
			gi = &kvm->arch.gisa_int;
3155
			if (hrtimer_active(&gi->timer))
3156
				hrtimer_cancel(&gi->timer);
3157
			hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3158
		}
3159
	} while (!final);
3160

3161
}
3162

3163
void kvm_s390_gisa_clear(struct kvm *kvm)
3164
{
3165
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3166

3167
	if (!gi->origin)
3168
		return;
3169
	gisa_clear_ipm(gi->origin);
3170
	VM_EVENT(kvm, 3, "gisa 0x%p cleared", gi->origin);
3171
}
3172

3173
void kvm_s390_gisa_init(struct kvm *kvm)
3174
{
3175
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3176

3177
	if (!css_general_characteristics.aiv)
3178
		return;
3179
	gi->origin = &kvm->arch.sie_page2->gisa;
3180
	gi->alert.mask = 0;
3181
	spin_lock_init(&gi->alert.ref_lock);
3182
	gi->expires = 50 * 1000; /* 50 usec */
3183
	hrtimer_setup(&gi->timer, gisa_vcpu_kicker, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3184
	memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3185
	gi->origin->next_alert = (u32)virt_to_phys(gi->origin);
3186
	VM_EVENT(kvm, 3, "gisa 0x%p initialized", gi->origin);
3187
}
3188

3189
void kvm_s390_gisa_enable(struct kvm *kvm)
3190
{
3191
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3192
	struct kvm_vcpu *vcpu;
3193
	unsigned long i;
3194
	u32 gisa_desc;
3195

3196
	if (gi->origin)
3197
		return;
3198
	kvm_s390_gisa_init(kvm);
3199
	gisa_desc = kvm_s390_get_gisa_desc(kvm);
3200
	if (!gisa_desc)
3201
		return;
3202
	kvm_for_each_vcpu(i, vcpu, kvm) {
3203
		mutex_lock(&vcpu->mutex);
3204
		vcpu->arch.sie_block->gd = gisa_desc;
3205
		vcpu->arch.sie_block->eca |= ECA_AIV;
3206
		VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3207
			   vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3208
		mutex_unlock(&vcpu->mutex);
3209
	}
3210
}
3211

3212
void kvm_s390_gisa_destroy(struct kvm *kvm)
3213
{
3214
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3215
	struct kvm_s390_gisa *gisa = gi->origin;
3216

3217
	if (!gi->origin)
3218
		return;
3219
	WARN(gi->alert.mask != 0x00,
3220
	     "unexpected non zero alert.mask 0x%02x",
3221
	     gi->alert.mask);
3222
	gi->alert.mask = 0x00;
3223
	if (gisa_set_iam(gi->origin, gi->alert.mask))
3224
		process_gib_alert_list();
3225
	hrtimer_cancel(&gi->timer);
3226
	gi->origin = NULL;
3227
	VM_EVENT(kvm, 3, "gisa 0x%p destroyed", gisa);
3228
}
3229

3230
void kvm_s390_gisa_disable(struct kvm *kvm)
3231
{
3232
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3233
	struct kvm_vcpu *vcpu;
3234
	unsigned long i;
3235

3236
	if (!gi->origin)
3237
		return;
3238
	kvm_for_each_vcpu(i, vcpu, kvm) {
3239
		mutex_lock(&vcpu->mutex);
3240
		vcpu->arch.sie_block->eca &= ~ECA_AIV;
3241
		vcpu->arch.sie_block->gd = 0U;
3242
		mutex_unlock(&vcpu->mutex);
3243
		VCPU_EVENT(vcpu, 3, "AIV disabled for cpu %03u", vcpu->vcpu_id);
3244
	}
3245
	kvm_s390_gisa_destroy(kvm);
3246
}
3247

3248
/**
3249
 * kvm_s390_gisc_register - register a guest ISC
3250
 *
3251
 * @kvm:  the kernel vm to work with
3252
 * @gisc: the guest interruption sub class to register
3253
 *
3254
 * The function extends the vm specific alert mask to use.
3255
 * The effective IAM mask in the GISA is updated as well
3256
 * in case the GISA is not part of the GIB alert list.
3257
 * It will be updated latest when the IAM gets restored
3258
 * by gisa_get_ipm_or_restore_iam().
3259
 *
3260
 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3261
 *          has registered with the channel subsystem.
3262
 *          -ENODEV in case the vm uses no GISA
3263
 *          -ERANGE in case the guest ISC is invalid
3264
 */
3265
int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3266
{
3267
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3268

3269
	if (!gi->origin)
3270
		return -ENODEV;
3271
	if (gisc > MAX_ISC)
3272
		return -ERANGE;
3273

3274
	spin_lock(&gi->alert.ref_lock);
3275
	gi->alert.ref_count[gisc]++;
3276
	if (gi->alert.ref_count[gisc] == 1) {
3277
		gi->alert.mask |= 0x80 >> gisc;
3278
		gisa_set_iam(gi->origin, gi->alert.mask);
3279
	}
3280
	spin_unlock(&gi->alert.ref_lock);
3281

3282
	return gib->nisc;
3283
}
3284
EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3285

3286
/**
3287
 * kvm_s390_gisc_unregister - unregister a guest ISC
3288
 *
3289
 * @kvm:  the kernel vm to work with
3290
 * @gisc: the guest interruption sub class to register
3291
 *
3292
 * The function reduces the vm specific alert mask to use.
3293
 * The effective IAM mask in the GISA is updated as well
3294
 * in case the GISA is not part of the GIB alert list.
3295
 * It will be updated latest when the IAM gets restored
3296
 * by gisa_get_ipm_or_restore_iam().
3297
 *
3298
 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3299
 *          has registered with the channel subsystem.
3300
 *          -ENODEV in case the vm uses no GISA
3301
 *          -ERANGE in case the guest ISC is invalid
3302
 *          -EINVAL in case the guest ISC is not registered
3303
 */
3304
int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3305
{
3306
	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3307
	int rc = 0;
3308

3309
	if (!gi->origin)
3310
		return -ENODEV;
3311
	if (gisc > MAX_ISC)
3312
		return -ERANGE;
3313

3314
	spin_lock(&gi->alert.ref_lock);
3315
	if (gi->alert.ref_count[gisc] == 0) {
3316
		rc = -EINVAL;
3317
		goto out;
3318
	}
3319
	gi->alert.ref_count[gisc]--;
3320
	if (gi->alert.ref_count[gisc] == 0) {
3321
		gi->alert.mask &= ~(0x80 >> gisc);
3322
		gisa_set_iam(gi->origin, gi->alert.mask);
3323
	}
3324
out:
3325
	spin_unlock(&gi->alert.ref_lock);
3326

3327
	return rc;
3328
}
3329
EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3330

3331
static void aen_host_forward(unsigned long si)
3332
{
3333
	struct kvm_s390_gisa_interrupt *gi;
3334
	struct zpci_gaite *gaite;
3335
	struct kvm *kvm;
3336

3337
	gaite = (struct zpci_gaite *)aift->gait +
3338
		(si * sizeof(struct zpci_gaite));
3339
	if (gaite->count == 0)
3340
		return;
3341
	if (gaite->aisb != 0)
3342
		set_bit_inv(gaite->aisbo, phys_to_virt(gaite->aisb));
3343

3344
	kvm = kvm_s390_pci_si_to_kvm(aift, si);
3345
	if (!kvm)
3346
		return;
3347
	gi = &kvm->arch.gisa_int;
3348

3349
	if (!(gi->origin->g1.simm & AIS_MODE_MASK(gaite->gisc)) ||
3350
	    !(gi->origin->g1.nimm & AIS_MODE_MASK(gaite->gisc))) {
3351
		gisa_set_ipm_gisc(gi->origin, gaite->gisc);
3352
		if (hrtimer_active(&gi->timer))
3353
			hrtimer_cancel(&gi->timer);
3354
		hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3355
		kvm->stat.aen_forward++;
3356
	}
3357
}
3358

3359
static void aen_process_gait(u8 isc)
3360
{
3361
	bool found = false, first = true;
3362
	union zpci_sic_iib iib = {{0}};
3363
	unsigned long si, flags;
3364

3365
	spin_lock_irqsave(&aift->gait_lock, flags);
3366

3367
	if (!aift->gait) {
3368
		spin_unlock_irqrestore(&aift->gait_lock, flags);
3369
		return;
3370
	}
3371

3372
	for (si = 0;;) {
3373
		/* Scan adapter summary indicator bit vector */
3374
		si = airq_iv_scan(aift->sbv, si, airq_iv_end(aift->sbv));
3375
		if (si == -1UL) {
3376
			if (first || found) {
3377
				/* Re-enable interrupts. */
3378
				zpci_set_irq_ctrl(SIC_IRQ_MODE_SINGLE, isc,
3379
						  &iib);
3380
				first = found = false;
3381
			} else {
3382
				/* Interrupts on and all bits processed */
3383
				break;
3384
			}
3385
			found = false;
3386
			si = 0;
3387
			/* Scan again after re-enabling interrupts */
3388
			continue;
3389
		}
3390
		found = true;
3391
		aen_host_forward(si);
3392
	}
3393

3394
	spin_unlock_irqrestore(&aift->gait_lock, flags);
3395
}
3396

3397
static void gib_alert_irq_handler(struct airq_struct *airq,
3398
				  struct tpi_info *tpi_info)
3399
{
3400
	struct tpi_adapter_info *info = (struct tpi_adapter_info *)tpi_info;
3401

3402
	inc_irq_stat(IRQIO_GAL);
3403

3404
	if ((info->forward || info->error) &&
3405
	    IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3406
		aen_process_gait(info->isc);
3407
		if (info->aism != 0)
3408
			process_gib_alert_list();
3409
	} else {
3410
		process_gib_alert_list();
3411
	}
3412
}
3413

3414
static struct airq_struct gib_alert_irq = {
3415
	.handler = gib_alert_irq_handler,
3416
};
3417

3418
void kvm_s390_gib_destroy(void)
3419
{
3420
	if (!gib)
3421
		return;
3422
	if (kvm_s390_pci_interp_allowed() && aift) {
3423
		mutex_lock(&aift->aift_lock);
3424
		kvm_s390_pci_aen_exit();
3425
		mutex_unlock(&aift->aift_lock);
3426
	}
3427
	chsc_sgib(0);
3428
	unregister_adapter_interrupt(&gib_alert_irq);
3429
	free_page((unsigned long)gib);
3430
	gib = NULL;
3431
}
3432

3433
int __init kvm_s390_gib_init(u8 nisc)
3434
{
3435
	u32 gib_origin;
3436
	int rc = 0;
3437

3438
	if (!css_general_characteristics.aiv) {
3439
		KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3440
		goto out;
3441
	}
3442

3443
	gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3444
	if (!gib) {
3445
		rc = -ENOMEM;
3446
		goto out;
3447
	}
3448

3449
	gib_alert_irq.isc = nisc;
3450
	if (register_adapter_interrupt(&gib_alert_irq)) {
3451
		pr_err("Registering the GIB alert interruption handler failed\n");
3452
		rc = -EIO;
3453
		goto out_free_gib;
3454
	}
3455
	/* adapter interrupts used for AP (applicable here) don't use the LSI */
3456
	*gib_alert_irq.lsi_ptr = 0xff;
3457

3458
	gib->nisc = nisc;
3459
	gib_origin = virt_to_phys(gib);
3460
	if (chsc_sgib(gib_origin)) {
3461
		pr_err("Associating the GIB with the AIV facility failed\n");
3462
		free_page((unsigned long)gib);
3463
		gib = NULL;
3464
		rc = -EIO;
3465
		goto out_unreg_gal;
3466
	}
3467

3468
	if (kvm_s390_pci_interp_allowed()) {
3469
		if (kvm_s390_pci_aen_init(nisc)) {
3470
			pr_err("Initializing AEN for PCI failed\n");
3471
			rc = -EIO;
3472
			goto out_unreg_gal;
3473
		}
3474
	}
3475

3476
	KVM_EVENT(3, "gib 0x%p (nisc=%d) initialized", gib, gib->nisc);
3477
	goto out;
3478

3479
out_unreg_gal:
3480
	unregister_adapter_interrupt(&gib_alert_irq);
3481
out_free_gib:
3482
	free_page((unsigned long)gib);
3483
	gib = NULL;
3484
out:
3485
	return rc;
3486
}
3487

3488