xref: /linux/arch/s390/kvm/interrupt.c (revision ef69f8d2ff09518657c3ecaf2db8408c16549829)
1 /*
2  * handling kvm guest interrupts
3  *
4  * Copyright IBM Corp. 2008, 2015
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License (version 2 only)
8  * as published by the Free Software Foundation.
9  *
10  *    Author(s): Carsten Otte <cotte@de.ibm.com>
11  */
12 
13 #include <linux/interrupt.h>
14 #include <linux/kvm_host.h>
15 #include <linux/hrtimer.h>
16 #include <linux/mmu_context.h>
17 #include <linux/signal.h>
18 #include <linux/slab.h>
19 #include <linux/bitmap.h>
20 #include <linux/vmalloc.h>
21 #include <asm/asm-offsets.h>
22 #include <asm/dis.h>
23 #include <linux/uaccess.h>
24 #include <asm/sclp.h>
25 #include <asm/isc.h>
26 #include <asm/gmap.h>
27 #include <asm/switch_to.h>
28 #include <asm/nmi.h>
29 #include "kvm-s390.h"
30 #include "gaccess.h"
31 #include "trace-s390.h"
32 
33 #define PFAULT_INIT 0x0600
34 #define PFAULT_DONE 0x0680
35 #define VIRTIO_PARAM 0x0d00
36 
37 /* handle external calls via sigp interpretation facility */
38 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
39 {
40 	int c, scn;
41 
42 	if (!(atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_ECALL_PEND))
43 		return 0;
44 
45 	BUG_ON(!kvm_s390_use_sca_entries());
46 	read_lock(&vcpu->kvm->arch.sca_lock);
47 	if (vcpu->kvm->arch.use_esca) {
48 		struct esca_block *sca = vcpu->kvm->arch.sca;
49 		union esca_sigp_ctrl sigp_ctrl =
50 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
51 
52 		c = sigp_ctrl.c;
53 		scn = sigp_ctrl.scn;
54 	} else {
55 		struct bsca_block *sca = vcpu->kvm->arch.sca;
56 		union bsca_sigp_ctrl sigp_ctrl =
57 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
58 
59 		c = sigp_ctrl.c;
60 		scn = sigp_ctrl.scn;
61 	}
62 	read_unlock(&vcpu->kvm->arch.sca_lock);
63 
64 	if (src_id)
65 		*src_id = scn;
66 
67 	return c;
68 }
69 
70 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
71 {
72 	int expect, rc;
73 
74 	BUG_ON(!kvm_s390_use_sca_entries());
75 	read_lock(&vcpu->kvm->arch.sca_lock);
76 	if (vcpu->kvm->arch.use_esca) {
77 		struct esca_block *sca = vcpu->kvm->arch.sca;
78 		union esca_sigp_ctrl *sigp_ctrl =
79 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
80 		union esca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
81 
82 		new_val.scn = src_id;
83 		new_val.c = 1;
84 		old_val.c = 0;
85 
86 		expect = old_val.value;
87 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
88 	} else {
89 		struct bsca_block *sca = vcpu->kvm->arch.sca;
90 		union bsca_sigp_ctrl *sigp_ctrl =
91 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
92 		union bsca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
93 
94 		new_val.scn = src_id;
95 		new_val.c = 1;
96 		old_val.c = 0;
97 
98 		expect = old_val.value;
99 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
100 	}
101 	read_unlock(&vcpu->kvm->arch.sca_lock);
102 
103 	if (rc != expect) {
104 		/* another external call is pending */
105 		return -EBUSY;
106 	}
107 	atomic_or(CPUSTAT_ECALL_PEND, &vcpu->arch.sie_block->cpuflags);
108 	return 0;
109 }
110 
111 static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
112 {
113 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
114 	int rc, expect;
115 
116 	if (!kvm_s390_use_sca_entries())
117 		return;
118 	atomic_andnot(CPUSTAT_ECALL_PEND, li->cpuflags);
119 	read_lock(&vcpu->kvm->arch.sca_lock);
120 	if (vcpu->kvm->arch.use_esca) {
121 		struct esca_block *sca = vcpu->kvm->arch.sca;
122 		union esca_sigp_ctrl *sigp_ctrl =
123 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
124 		union esca_sigp_ctrl old = *sigp_ctrl;
125 
126 		expect = old.value;
127 		rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
128 	} else {
129 		struct bsca_block *sca = vcpu->kvm->arch.sca;
130 		union bsca_sigp_ctrl *sigp_ctrl =
131 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
132 		union bsca_sigp_ctrl old = *sigp_ctrl;
133 
134 		expect = old.value;
135 		rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
136 	}
137 	read_unlock(&vcpu->kvm->arch.sca_lock);
138 	WARN_ON(rc != expect); /* cannot clear? */
139 }
140 
141 int psw_extint_disabled(struct kvm_vcpu *vcpu)
142 {
143 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
144 }
145 
146 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
147 {
148 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
149 }
150 
151 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
152 {
153 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
154 }
155 
156 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
157 {
158 	return psw_extint_disabled(vcpu) &&
159 	       psw_ioint_disabled(vcpu) &&
160 	       psw_mchk_disabled(vcpu);
161 }
162 
163 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
164 {
165 	if (psw_extint_disabled(vcpu) ||
166 	    !(vcpu->arch.sie_block->gcr[0] & 0x800ul))
167 		return 0;
168 	if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
169 		/* No timer interrupts when single stepping */
170 		return 0;
171 	return 1;
172 }
173 
174 static int ckc_irq_pending(struct kvm_vcpu *vcpu)
175 {
176 	if (vcpu->arch.sie_block->ckc >= kvm_s390_get_tod_clock_fast(vcpu->kvm))
177 		return 0;
178 	return ckc_interrupts_enabled(vcpu);
179 }
180 
181 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
182 {
183 	return !psw_extint_disabled(vcpu) &&
184 	       (vcpu->arch.sie_block->gcr[0] & 0x400ul);
185 }
186 
187 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
188 {
189 	if (!cpu_timer_interrupts_enabled(vcpu))
190 		return 0;
191 	return kvm_s390_get_cpu_timer(vcpu) >> 63;
192 }
193 
194 static inline int is_ioirq(unsigned long irq_type)
195 {
196 	return ((irq_type >= IRQ_PEND_IO_ISC_0) &&
197 		(irq_type <= IRQ_PEND_IO_ISC_7));
198 }
199 
200 static uint64_t isc_to_isc_bits(int isc)
201 {
202 	return (0x80 >> isc) << 24;
203 }
204 
205 static inline u8 int_word_to_isc(u32 int_word)
206 {
207 	return (int_word & 0x38000000) >> 27;
208 }
209 
210 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
211 {
212 	return vcpu->kvm->arch.float_int.pending_irqs |
213 	       vcpu->arch.local_int.pending_irqs;
214 }
215 
216 static inline int isc_to_irq_type(unsigned long isc)
217 {
218 	return IRQ_PEND_IO_ISC_0 + isc;
219 }
220 
221 static inline int irq_type_to_isc(unsigned long irq_type)
222 {
223 	return irq_type - IRQ_PEND_IO_ISC_0;
224 }
225 
226 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
227 				   unsigned long active_mask)
228 {
229 	int i;
230 
231 	for (i = 0; i <= MAX_ISC; i++)
232 		if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
233 			active_mask &= ~(1UL << (isc_to_irq_type(i)));
234 
235 	return active_mask;
236 }
237 
238 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
239 {
240 	unsigned long active_mask;
241 
242 	active_mask = pending_irqs(vcpu);
243 	if (!active_mask)
244 		return 0;
245 
246 	if (psw_extint_disabled(vcpu))
247 		active_mask &= ~IRQ_PEND_EXT_MASK;
248 	if (psw_ioint_disabled(vcpu))
249 		active_mask &= ~IRQ_PEND_IO_MASK;
250 	else
251 		active_mask = disable_iscs(vcpu, active_mask);
252 	if (!(vcpu->arch.sie_block->gcr[0] & 0x2000ul))
253 		__clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
254 	if (!(vcpu->arch.sie_block->gcr[0] & 0x4000ul))
255 		__clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
256 	if (!(vcpu->arch.sie_block->gcr[0] & 0x800ul))
257 		__clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
258 	if (!(vcpu->arch.sie_block->gcr[0] & 0x400ul))
259 		__clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
260 	if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
261 		__clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
262 	if (psw_mchk_disabled(vcpu))
263 		active_mask &= ~IRQ_PEND_MCHK_MASK;
264 	/*
265 	 * Check both floating and local interrupt's cr14 because
266 	 * bit IRQ_PEND_MCHK_REP could be set in both cases.
267 	 */
268 	if (!(vcpu->arch.sie_block->gcr[14] &
269 	   (vcpu->kvm->arch.float_int.mchk.cr14 |
270 	   vcpu->arch.local_int.irq.mchk.cr14)))
271 		__clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
272 
273 	/*
274 	 * STOP irqs will never be actively delivered. They are triggered via
275 	 * intercept requests and cleared when the stop intercept is performed.
276 	 */
277 	__clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
278 
279 	return active_mask;
280 }
281 
282 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
283 {
284 	atomic_or(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
285 	set_bit(vcpu->vcpu_id, vcpu->arch.local_int.float_int->idle_mask);
286 }
287 
288 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
289 {
290 	atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
291 	clear_bit(vcpu->vcpu_id, vcpu->arch.local_int.float_int->idle_mask);
292 }
293 
294 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
295 {
296 	atomic_andnot(CPUSTAT_IO_INT | CPUSTAT_EXT_INT | CPUSTAT_STOP_INT,
297 		    &vcpu->arch.sie_block->cpuflags);
298 	vcpu->arch.sie_block->lctl = 0x0000;
299 	vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
300 
301 	if (guestdbg_enabled(vcpu)) {
302 		vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
303 					       LCTL_CR10 | LCTL_CR11);
304 		vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
305 	}
306 }
307 
308 static void __set_cpuflag(struct kvm_vcpu *vcpu, u32 flag)
309 {
310 	atomic_or(flag, &vcpu->arch.sie_block->cpuflags);
311 }
312 
313 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
314 {
315 	if (!(pending_irqs(vcpu) & IRQ_PEND_IO_MASK))
316 		return;
317 	else if (psw_ioint_disabled(vcpu))
318 		__set_cpuflag(vcpu, CPUSTAT_IO_INT);
319 	else
320 		vcpu->arch.sie_block->lctl |= LCTL_CR6;
321 }
322 
323 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
324 {
325 	if (!(pending_irqs(vcpu) & IRQ_PEND_EXT_MASK))
326 		return;
327 	if (psw_extint_disabled(vcpu))
328 		__set_cpuflag(vcpu, CPUSTAT_EXT_INT);
329 	else
330 		vcpu->arch.sie_block->lctl |= LCTL_CR0;
331 }
332 
333 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
334 {
335 	if (!(pending_irqs(vcpu) & IRQ_PEND_MCHK_MASK))
336 		return;
337 	if (psw_mchk_disabled(vcpu))
338 		vcpu->arch.sie_block->ictl |= ICTL_LPSW;
339 	else
340 		vcpu->arch.sie_block->lctl |= LCTL_CR14;
341 }
342 
343 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
344 {
345 	if (kvm_s390_is_stop_irq_pending(vcpu))
346 		__set_cpuflag(vcpu, CPUSTAT_STOP_INT);
347 }
348 
349 /* Set interception request for non-deliverable interrupts */
350 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
351 {
352 	set_intercept_indicators_io(vcpu);
353 	set_intercept_indicators_ext(vcpu);
354 	set_intercept_indicators_mchk(vcpu);
355 	set_intercept_indicators_stop(vcpu);
356 }
357 
358 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
359 {
360 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
361 	int rc;
362 
363 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
364 					 0, 0);
365 
366 	rc  = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
367 			   (u16 *)__LC_EXT_INT_CODE);
368 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
369 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
370 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
371 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
372 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
373 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
374 	return rc ? -EFAULT : 0;
375 }
376 
377 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
378 {
379 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
380 	int rc;
381 
382 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
383 					 0, 0);
384 
385 	rc  = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
386 			   (u16 __user *)__LC_EXT_INT_CODE);
387 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
388 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
389 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
390 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
391 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
392 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
393 	return rc ? -EFAULT : 0;
394 }
395 
396 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
397 {
398 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
399 	struct kvm_s390_ext_info ext;
400 	int rc;
401 
402 	spin_lock(&li->lock);
403 	ext = li->irq.ext;
404 	clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
405 	li->irq.ext.ext_params2 = 0;
406 	spin_unlock(&li->lock);
407 
408 	VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
409 		   ext.ext_params2);
410 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
411 					 KVM_S390_INT_PFAULT_INIT,
412 					 0, ext.ext_params2);
413 
414 	rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
415 	rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
416 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
417 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
418 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
419 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
420 	rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
421 	return rc ? -EFAULT : 0;
422 }
423 
424 static int __write_machine_check(struct kvm_vcpu *vcpu,
425 				 struct kvm_s390_mchk_info *mchk)
426 {
427 	unsigned long ext_sa_addr;
428 	unsigned long lc;
429 	freg_t fprs[NUM_FPRS];
430 	union mci mci;
431 	int rc;
432 
433 	mci.val = mchk->mcic;
434 	/* take care of lazy register loading */
435 	save_fpu_regs();
436 	save_access_regs(vcpu->run->s.regs.acrs);
437 	if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
438 		save_gs_cb(current->thread.gs_cb);
439 
440 	/* Extended save area */
441 	rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
442 			   sizeof(unsigned long));
443 	/* Only bits 0 through 63-LC are used for address formation */
444 	lc = ext_sa_addr & MCESA_LC_MASK;
445 	if (test_kvm_facility(vcpu->kvm, 133)) {
446 		switch (lc) {
447 		case 0:
448 		case 10:
449 			ext_sa_addr &= ~0x3ffUL;
450 			break;
451 		case 11:
452 			ext_sa_addr &= ~0x7ffUL;
453 			break;
454 		case 12:
455 			ext_sa_addr &= ~0xfffUL;
456 			break;
457 		default:
458 			ext_sa_addr = 0;
459 			break;
460 		}
461 	} else {
462 		ext_sa_addr &= ~0x3ffUL;
463 	}
464 
465 	if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
466 		if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
467 				    512))
468 			mci.vr = 0;
469 	} else {
470 		mci.vr = 0;
471 	}
472 	if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
473 	    && (lc == 11 || lc == 12)) {
474 		if (write_guest_abs(vcpu, ext_sa_addr + 1024,
475 				    &vcpu->run->s.regs.gscb, 32))
476 			mci.gs = 0;
477 	} else {
478 		mci.gs = 0;
479 	}
480 
481 	/* General interruption information */
482 	rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
483 	rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
484 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
485 	rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
486 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
487 	rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
488 
489 	/* Register-save areas */
490 	if (MACHINE_HAS_VX) {
491 		convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
492 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
493 	} else {
494 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
495 				     vcpu->run->s.regs.fprs, 128);
496 	}
497 	rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
498 			     vcpu->run->s.regs.gprs, 128);
499 	rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
500 			   (u32 __user *) __LC_FP_CREG_SAVE_AREA);
501 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
502 			   (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
503 	rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
504 			   (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
505 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
506 			   (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
507 	rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
508 			     &vcpu->run->s.regs.acrs, 64);
509 	rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
510 			     &vcpu->arch.sie_block->gcr, 128);
511 
512 	/* Extended interruption information */
513 	rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
514 			   (u32 __user *) __LC_EXT_DAMAGE_CODE);
515 	rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
516 			   (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
517 	rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
518 			     sizeof(mchk->fixed_logout));
519 	return rc ? -EFAULT : 0;
520 }
521 
522 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
523 {
524 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
525 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
526 	struct kvm_s390_mchk_info mchk = {};
527 	int deliver = 0;
528 	int rc = 0;
529 
530 	spin_lock(&fi->lock);
531 	spin_lock(&li->lock);
532 	if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
533 	    test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
534 		/*
535 		 * If there was an exigent machine check pending, then any
536 		 * repressible machine checks that might have been pending
537 		 * are indicated along with it, so always clear bits for
538 		 * repressible and exigent interrupts
539 		 */
540 		mchk = li->irq.mchk;
541 		clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
542 		clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
543 		memset(&li->irq.mchk, 0, sizeof(mchk));
544 		deliver = 1;
545 	}
546 	/*
547 	 * We indicate floating repressible conditions along with
548 	 * other pending conditions. Channel Report Pending and Channel
549 	 * Subsystem damage are the only two and and are indicated by
550 	 * bits in mcic and masked in cr14.
551 	 */
552 	if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
553 		mchk.mcic |= fi->mchk.mcic;
554 		mchk.cr14 |= fi->mchk.cr14;
555 		memset(&fi->mchk, 0, sizeof(mchk));
556 		deliver = 1;
557 	}
558 	spin_unlock(&li->lock);
559 	spin_unlock(&fi->lock);
560 
561 	if (deliver) {
562 		VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
563 			   mchk.mcic);
564 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
565 						 KVM_S390_MCHK,
566 						 mchk.cr14, mchk.mcic);
567 		rc = __write_machine_check(vcpu, &mchk);
568 	}
569 	return rc;
570 }
571 
572 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
573 {
574 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
575 	int rc;
576 
577 	VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
578 	vcpu->stat.deliver_restart_signal++;
579 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
580 
581 	rc  = write_guest_lc(vcpu,
582 			     offsetof(struct lowcore, restart_old_psw),
583 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
584 	rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
585 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
586 	clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
587 	return rc ? -EFAULT : 0;
588 }
589 
590 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
591 {
592 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
593 	struct kvm_s390_prefix_info prefix;
594 
595 	spin_lock(&li->lock);
596 	prefix = li->irq.prefix;
597 	li->irq.prefix.address = 0;
598 	clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
599 	spin_unlock(&li->lock);
600 
601 	vcpu->stat.deliver_prefix_signal++;
602 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
603 					 KVM_S390_SIGP_SET_PREFIX,
604 					 prefix.address, 0);
605 
606 	kvm_s390_set_prefix(vcpu, prefix.address);
607 	return 0;
608 }
609 
610 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
611 {
612 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
613 	int rc;
614 	int cpu_addr;
615 
616 	spin_lock(&li->lock);
617 	cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
618 	clear_bit(cpu_addr, li->sigp_emerg_pending);
619 	if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
620 		clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
621 	spin_unlock(&li->lock);
622 
623 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
624 	vcpu->stat.deliver_emergency_signal++;
625 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
626 					 cpu_addr, 0);
627 
628 	rc  = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
629 			   (u16 *)__LC_EXT_INT_CODE);
630 	rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
631 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
632 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
633 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
634 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
635 	return rc ? -EFAULT : 0;
636 }
637 
638 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
639 {
640 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
641 	struct kvm_s390_extcall_info extcall;
642 	int rc;
643 
644 	spin_lock(&li->lock);
645 	extcall = li->irq.extcall;
646 	li->irq.extcall.code = 0;
647 	clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
648 	spin_unlock(&li->lock);
649 
650 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
651 	vcpu->stat.deliver_external_call++;
652 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
653 					 KVM_S390_INT_EXTERNAL_CALL,
654 					 extcall.code, 0);
655 
656 	rc  = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
657 			   (u16 *)__LC_EXT_INT_CODE);
658 	rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
659 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
660 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
661 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
662 			    sizeof(psw_t));
663 	return rc ? -EFAULT : 0;
664 }
665 
666 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
667 {
668 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
669 	struct kvm_s390_pgm_info pgm_info;
670 	int rc = 0, nullifying = false;
671 	u16 ilen;
672 
673 	spin_lock(&li->lock);
674 	pgm_info = li->irq.pgm;
675 	clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
676 	memset(&li->irq.pgm, 0, sizeof(pgm_info));
677 	spin_unlock(&li->lock);
678 
679 	ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
680 	VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
681 		   pgm_info.code, ilen);
682 	vcpu->stat.deliver_program_int++;
683 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
684 					 pgm_info.code, 0);
685 
686 	switch (pgm_info.code & ~PGM_PER) {
687 	case PGM_AFX_TRANSLATION:
688 	case PGM_ASX_TRANSLATION:
689 	case PGM_EX_TRANSLATION:
690 	case PGM_LFX_TRANSLATION:
691 	case PGM_LSTE_SEQUENCE:
692 	case PGM_LSX_TRANSLATION:
693 	case PGM_LX_TRANSLATION:
694 	case PGM_PRIMARY_AUTHORITY:
695 	case PGM_SECONDARY_AUTHORITY:
696 		nullifying = true;
697 		/* fall through */
698 	case PGM_SPACE_SWITCH:
699 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
700 				  (u64 *)__LC_TRANS_EXC_CODE);
701 		break;
702 	case PGM_ALEN_TRANSLATION:
703 	case PGM_ALE_SEQUENCE:
704 	case PGM_ASTE_INSTANCE:
705 	case PGM_ASTE_SEQUENCE:
706 	case PGM_ASTE_VALIDITY:
707 	case PGM_EXTENDED_AUTHORITY:
708 		rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
709 				  (u8 *)__LC_EXC_ACCESS_ID);
710 		nullifying = true;
711 		break;
712 	case PGM_ASCE_TYPE:
713 	case PGM_PAGE_TRANSLATION:
714 	case PGM_REGION_FIRST_TRANS:
715 	case PGM_REGION_SECOND_TRANS:
716 	case PGM_REGION_THIRD_TRANS:
717 	case PGM_SEGMENT_TRANSLATION:
718 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
719 				  (u64 *)__LC_TRANS_EXC_CODE);
720 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
721 				   (u8 *)__LC_EXC_ACCESS_ID);
722 		rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
723 				   (u8 *)__LC_OP_ACCESS_ID);
724 		nullifying = true;
725 		break;
726 	case PGM_MONITOR:
727 		rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
728 				  (u16 *)__LC_MON_CLASS_NR);
729 		rc |= put_guest_lc(vcpu, pgm_info.mon_code,
730 				   (u64 *)__LC_MON_CODE);
731 		break;
732 	case PGM_VECTOR_PROCESSING:
733 	case PGM_DATA:
734 		rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
735 				  (u32 *)__LC_DATA_EXC_CODE);
736 		break;
737 	case PGM_PROTECTION:
738 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
739 				  (u64 *)__LC_TRANS_EXC_CODE);
740 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
741 				   (u8 *)__LC_EXC_ACCESS_ID);
742 		break;
743 	case PGM_STACK_FULL:
744 	case PGM_STACK_EMPTY:
745 	case PGM_STACK_SPECIFICATION:
746 	case PGM_STACK_TYPE:
747 	case PGM_STACK_OPERATION:
748 	case PGM_TRACE_TABEL:
749 	case PGM_CRYPTO_OPERATION:
750 		nullifying = true;
751 		break;
752 	}
753 
754 	if (pgm_info.code & PGM_PER) {
755 		rc |= put_guest_lc(vcpu, pgm_info.per_code,
756 				   (u8 *) __LC_PER_CODE);
757 		rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
758 				   (u8 *)__LC_PER_ATMID);
759 		rc |= put_guest_lc(vcpu, pgm_info.per_address,
760 				   (u64 *) __LC_PER_ADDRESS);
761 		rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
762 				   (u8 *) __LC_PER_ACCESS_ID);
763 	}
764 
765 	if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
766 		kvm_s390_rewind_psw(vcpu, ilen);
767 
768 	/* bit 1+2 of the target are the ilc, so we can directly use ilen */
769 	rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
770 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
771 				 (u64 *) __LC_LAST_BREAK);
772 	rc |= put_guest_lc(vcpu, pgm_info.code,
773 			   (u16 *)__LC_PGM_INT_CODE);
774 	rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
775 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
776 	rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
777 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
778 	return rc ? -EFAULT : 0;
779 }
780 
781 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
782 {
783 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
784 	struct kvm_s390_ext_info ext;
785 	int rc = 0;
786 
787 	spin_lock(&fi->lock);
788 	if (!(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
789 		spin_unlock(&fi->lock);
790 		return 0;
791 	}
792 	ext = fi->srv_signal;
793 	memset(&fi->srv_signal, 0, sizeof(ext));
794 	clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
795 	spin_unlock(&fi->lock);
796 
797 	VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
798 		   ext.ext_params);
799 	vcpu->stat.deliver_service_signal++;
800 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
801 					 ext.ext_params, 0);
802 
803 	rc  = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
804 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
805 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
806 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
807 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
808 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
809 	rc |= put_guest_lc(vcpu, ext.ext_params,
810 			   (u32 *)__LC_EXT_PARAMS);
811 
812 	return rc ? -EFAULT : 0;
813 }
814 
815 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
816 {
817 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
818 	struct kvm_s390_interrupt_info *inti;
819 	int rc = 0;
820 
821 	spin_lock(&fi->lock);
822 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
823 					struct kvm_s390_interrupt_info,
824 					list);
825 	if (inti) {
826 		list_del(&inti->list);
827 		fi->counters[FIRQ_CNTR_PFAULT] -= 1;
828 	}
829 	if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
830 		clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
831 	spin_unlock(&fi->lock);
832 
833 	if (inti) {
834 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
835 						 KVM_S390_INT_PFAULT_DONE, 0,
836 						 inti->ext.ext_params2);
837 		VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
838 			   inti->ext.ext_params2);
839 
840 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
841 				(u16 *)__LC_EXT_INT_CODE);
842 		rc |= put_guest_lc(vcpu, PFAULT_DONE,
843 				(u16 *)__LC_EXT_CPU_ADDR);
844 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
845 				&vcpu->arch.sie_block->gpsw,
846 				sizeof(psw_t));
847 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
848 				&vcpu->arch.sie_block->gpsw,
849 				sizeof(psw_t));
850 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
851 				(u64 *)__LC_EXT_PARAMS2);
852 		kfree(inti);
853 	}
854 	return rc ? -EFAULT : 0;
855 }
856 
857 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
858 {
859 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
860 	struct kvm_s390_interrupt_info *inti;
861 	int rc = 0;
862 
863 	spin_lock(&fi->lock);
864 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
865 					struct kvm_s390_interrupt_info,
866 					list);
867 	if (inti) {
868 		VCPU_EVENT(vcpu, 4,
869 			   "deliver: virtio parm: 0x%x,parm64: 0x%llx",
870 			   inti->ext.ext_params, inti->ext.ext_params2);
871 		vcpu->stat.deliver_virtio_interrupt++;
872 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
873 				inti->type,
874 				inti->ext.ext_params,
875 				inti->ext.ext_params2);
876 		list_del(&inti->list);
877 		fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
878 	}
879 	if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
880 		clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
881 	spin_unlock(&fi->lock);
882 
883 	if (inti) {
884 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
885 				(u16 *)__LC_EXT_INT_CODE);
886 		rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
887 				(u16 *)__LC_EXT_CPU_ADDR);
888 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
889 				&vcpu->arch.sie_block->gpsw,
890 				sizeof(psw_t));
891 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
892 				&vcpu->arch.sie_block->gpsw,
893 				sizeof(psw_t));
894 		rc |= put_guest_lc(vcpu, inti->ext.ext_params,
895 				(u32 *)__LC_EXT_PARAMS);
896 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
897 				(u64 *)__LC_EXT_PARAMS2);
898 		kfree(inti);
899 	}
900 	return rc ? -EFAULT : 0;
901 }
902 
903 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
904 				     unsigned long irq_type)
905 {
906 	struct list_head *isc_list;
907 	struct kvm_s390_float_interrupt *fi;
908 	struct kvm_s390_interrupt_info *inti = NULL;
909 	int rc = 0;
910 
911 	fi = &vcpu->kvm->arch.float_int;
912 
913 	spin_lock(&fi->lock);
914 	isc_list = &fi->lists[irq_type_to_isc(irq_type)];
915 	inti = list_first_entry_or_null(isc_list,
916 					struct kvm_s390_interrupt_info,
917 					list);
918 	if (inti) {
919 		if (inti->type & KVM_S390_INT_IO_AI_MASK)
920 			VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
921 		else
922 			VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
923 			inti->io.subchannel_id >> 8,
924 			inti->io.subchannel_id >> 1 & 0x3,
925 			inti->io.subchannel_nr);
926 
927 		vcpu->stat.deliver_io_int++;
928 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
929 				inti->type,
930 				((__u32)inti->io.subchannel_id << 16) |
931 				inti->io.subchannel_nr,
932 				((__u64)inti->io.io_int_parm << 32) |
933 				inti->io.io_int_word);
934 		list_del(&inti->list);
935 		fi->counters[FIRQ_CNTR_IO] -= 1;
936 	}
937 	if (list_empty(isc_list))
938 		clear_bit(irq_type, &fi->pending_irqs);
939 	spin_unlock(&fi->lock);
940 
941 	if (inti) {
942 		rc  = put_guest_lc(vcpu, inti->io.subchannel_id,
943 				(u16 *)__LC_SUBCHANNEL_ID);
944 		rc |= put_guest_lc(vcpu, inti->io.subchannel_nr,
945 				(u16 *)__LC_SUBCHANNEL_NR);
946 		rc |= put_guest_lc(vcpu, inti->io.io_int_parm,
947 				(u32 *)__LC_IO_INT_PARM);
948 		rc |= put_guest_lc(vcpu, inti->io.io_int_word,
949 				(u32 *)__LC_IO_INT_WORD);
950 		rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
951 				&vcpu->arch.sie_block->gpsw,
952 				sizeof(psw_t));
953 		rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
954 				&vcpu->arch.sie_block->gpsw,
955 				sizeof(psw_t));
956 		kfree(inti);
957 	}
958 
959 	return rc ? -EFAULT : 0;
960 }
961 
962 typedef int (*deliver_irq_t)(struct kvm_vcpu *vcpu);
963 
964 static const deliver_irq_t deliver_irq_funcs[] = {
965 	[IRQ_PEND_MCHK_EX]        = __deliver_machine_check,
966 	[IRQ_PEND_MCHK_REP]       = __deliver_machine_check,
967 	[IRQ_PEND_PROG]           = __deliver_prog,
968 	[IRQ_PEND_EXT_EMERGENCY]  = __deliver_emergency_signal,
969 	[IRQ_PEND_EXT_EXTERNAL]   = __deliver_external_call,
970 	[IRQ_PEND_EXT_CLOCK_COMP] = __deliver_ckc,
971 	[IRQ_PEND_EXT_CPU_TIMER]  = __deliver_cpu_timer,
972 	[IRQ_PEND_RESTART]        = __deliver_restart,
973 	[IRQ_PEND_SET_PREFIX]     = __deliver_set_prefix,
974 	[IRQ_PEND_PFAULT_INIT]    = __deliver_pfault_init,
975 	[IRQ_PEND_EXT_SERVICE]    = __deliver_service,
976 	[IRQ_PEND_PFAULT_DONE]    = __deliver_pfault_done,
977 	[IRQ_PEND_VIRTIO]         = __deliver_virtio,
978 };
979 
980 /* Check whether an external call is pending (deliverable or not) */
981 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
982 {
983 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
984 
985 	if (!sclp.has_sigpif)
986 		return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
987 
988 	return sca_ext_call_pending(vcpu, NULL);
989 }
990 
991 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
992 {
993 	if (deliverable_irqs(vcpu))
994 		return 1;
995 
996 	if (kvm_cpu_has_pending_timer(vcpu))
997 		return 1;
998 
999 	/* external call pending and deliverable */
1000 	if (kvm_s390_ext_call_pending(vcpu) &&
1001 	    !psw_extint_disabled(vcpu) &&
1002 	    (vcpu->arch.sie_block->gcr[0] & 0x2000ul))
1003 		return 1;
1004 
1005 	if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1006 		return 1;
1007 	return 0;
1008 }
1009 
1010 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1011 {
1012 	return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1013 }
1014 
1015 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1016 {
1017 	u64 now, cputm, sltime = 0;
1018 
1019 	if (ckc_interrupts_enabled(vcpu)) {
1020 		now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1021 		sltime = tod_to_ns(vcpu->arch.sie_block->ckc - now);
1022 		/* already expired or overflow? */
1023 		if (!sltime || vcpu->arch.sie_block->ckc <= now)
1024 			return 0;
1025 		if (cpu_timer_interrupts_enabled(vcpu)) {
1026 			cputm = kvm_s390_get_cpu_timer(vcpu);
1027 			/* already expired? */
1028 			if (cputm >> 63)
1029 				return 0;
1030 			return min(sltime, tod_to_ns(cputm));
1031 		}
1032 	} else if (cpu_timer_interrupts_enabled(vcpu)) {
1033 		sltime = kvm_s390_get_cpu_timer(vcpu);
1034 		/* already expired? */
1035 		if (sltime >> 63)
1036 			return 0;
1037 	}
1038 	return sltime;
1039 }
1040 
1041 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1042 {
1043 	u64 sltime;
1044 
1045 	vcpu->stat.exit_wait_state++;
1046 
1047 	/* fast path */
1048 	if (kvm_arch_vcpu_runnable(vcpu))
1049 		return 0;
1050 
1051 	if (psw_interrupts_disabled(vcpu)) {
1052 		VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1053 		return -EOPNOTSUPP; /* disabled wait */
1054 	}
1055 
1056 	if (!ckc_interrupts_enabled(vcpu) &&
1057 	    !cpu_timer_interrupts_enabled(vcpu)) {
1058 		VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1059 		__set_cpu_idle(vcpu);
1060 		goto no_timer;
1061 	}
1062 
1063 	sltime = __calculate_sltime(vcpu);
1064 	if (!sltime)
1065 		return 0;
1066 
1067 	__set_cpu_idle(vcpu);
1068 	hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1069 	VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1070 no_timer:
1071 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1072 	kvm_vcpu_block(vcpu);
1073 	__unset_cpu_idle(vcpu);
1074 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1075 
1076 	hrtimer_cancel(&vcpu->arch.ckc_timer);
1077 	return 0;
1078 }
1079 
1080 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1081 {
1082 	/*
1083 	 * We cannot move this into the if, as the CPU might be already
1084 	 * in kvm_vcpu_block without having the waitqueue set (polling)
1085 	 */
1086 	vcpu->valid_wakeup = true;
1087 	/*
1088 	 * This is mostly to document, that the read in swait_active could
1089 	 * be moved before other stores, leading to subtle races.
1090 	 * All current users do not store or use an atomic like update
1091 	 */
1092 	smp_mb__after_atomic();
1093 	if (swait_active(&vcpu->wq)) {
1094 		/*
1095 		 * The vcpu gave up the cpu voluntarily, mark it as a good
1096 		 * yield-candidate.
1097 		 */
1098 		vcpu->preempted = true;
1099 		swake_up(&vcpu->wq);
1100 		vcpu->stat.halt_wakeup++;
1101 	}
1102 	/*
1103 	 * The VCPU might not be sleeping but is executing the VSIE. Let's
1104 	 * kick it, so it leaves the SIE to process the request.
1105 	 */
1106 	kvm_s390_vsie_kick(vcpu);
1107 }
1108 
1109 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1110 {
1111 	struct kvm_vcpu *vcpu;
1112 	u64 sltime;
1113 
1114 	vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1115 	sltime = __calculate_sltime(vcpu);
1116 
1117 	/*
1118 	 * If the monotonic clock runs faster than the tod clock we might be
1119 	 * woken up too early and have to go back to sleep to avoid deadlocks.
1120 	 */
1121 	if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1122 		return HRTIMER_RESTART;
1123 	kvm_s390_vcpu_wakeup(vcpu);
1124 	return HRTIMER_NORESTART;
1125 }
1126 
1127 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1128 {
1129 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1130 
1131 	spin_lock(&li->lock);
1132 	li->pending_irqs = 0;
1133 	bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1134 	memset(&li->irq, 0, sizeof(li->irq));
1135 	spin_unlock(&li->lock);
1136 
1137 	sca_clear_ext_call(vcpu);
1138 }
1139 
1140 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1141 {
1142 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1143 	deliver_irq_t func;
1144 	int rc = 0;
1145 	unsigned long irq_type;
1146 	unsigned long irqs;
1147 
1148 	__reset_intercept_indicators(vcpu);
1149 
1150 	/* pending ckc conditions might have been invalidated */
1151 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1152 	if (ckc_irq_pending(vcpu))
1153 		set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1154 
1155 	/* pending cpu timer conditions might have been invalidated */
1156 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1157 	if (cpu_timer_irq_pending(vcpu))
1158 		set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1159 
1160 	while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1161 		/* bits are in the order of interrupt priority */
1162 		irq_type = find_first_bit(&irqs, IRQ_PEND_COUNT);
1163 		if (is_ioirq(irq_type)) {
1164 			rc = __deliver_io(vcpu, irq_type);
1165 		} else {
1166 			func = deliver_irq_funcs[irq_type];
1167 			if (!func) {
1168 				WARN_ON_ONCE(func == NULL);
1169 				clear_bit(irq_type, &li->pending_irqs);
1170 				continue;
1171 			}
1172 			rc = func(vcpu);
1173 		}
1174 	}
1175 
1176 	set_intercept_indicators(vcpu);
1177 
1178 	return rc;
1179 }
1180 
1181 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1182 {
1183 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1184 
1185 	VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1186 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1187 				   irq->u.pgm.code, 0);
1188 
1189 	if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1190 		/* auto detection if no valid ILC was given */
1191 		irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1192 		irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1193 		irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1194 	}
1195 
1196 	if (irq->u.pgm.code == PGM_PER) {
1197 		li->irq.pgm.code |= PGM_PER;
1198 		li->irq.pgm.flags = irq->u.pgm.flags;
1199 		/* only modify PER related information */
1200 		li->irq.pgm.per_address = irq->u.pgm.per_address;
1201 		li->irq.pgm.per_code = irq->u.pgm.per_code;
1202 		li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1203 		li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1204 	} else if (!(irq->u.pgm.code & PGM_PER)) {
1205 		li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1206 				   irq->u.pgm.code;
1207 		li->irq.pgm.flags = irq->u.pgm.flags;
1208 		/* only modify non-PER information */
1209 		li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1210 		li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1211 		li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1212 		li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1213 		li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1214 		li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1215 	} else {
1216 		li->irq.pgm = irq->u.pgm;
1217 	}
1218 	set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1219 	return 0;
1220 }
1221 
1222 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1223 {
1224 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1225 
1226 	VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1227 		   irq->u.ext.ext_params2);
1228 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1229 				   irq->u.ext.ext_params,
1230 				   irq->u.ext.ext_params2);
1231 
1232 	li->irq.ext = irq->u.ext;
1233 	set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1234 	atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1235 	return 0;
1236 }
1237 
1238 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1239 {
1240 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1241 	struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1242 	uint16_t src_id = irq->u.extcall.code;
1243 
1244 	VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1245 		   src_id);
1246 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1247 				   src_id, 0);
1248 
1249 	/* sending vcpu invalid */
1250 	if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1251 		return -EINVAL;
1252 
1253 	if (sclp.has_sigpif)
1254 		return sca_inject_ext_call(vcpu, src_id);
1255 
1256 	if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1257 		return -EBUSY;
1258 	*extcall = irq->u.extcall;
1259 	atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1260 	return 0;
1261 }
1262 
1263 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1264 {
1265 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1266 	struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1267 
1268 	VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1269 		   irq->u.prefix.address);
1270 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1271 				   irq->u.prefix.address, 0);
1272 
1273 	if (!is_vcpu_stopped(vcpu))
1274 		return -EBUSY;
1275 
1276 	*prefix = irq->u.prefix;
1277 	set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1278 	return 0;
1279 }
1280 
1281 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1282 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1283 {
1284 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1285 	struct kvm_s390_stop_info *stop = &li->irq.stop;
1286 	int rc = 0;
1287 
1288 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1289 
1290 	if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1291 		return -EINVAL;
1292 
1293 	if (is_vcpu_stopped(vcpu)) {
1294 		if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1295 			rc = kvm_s390_store_status_unloaded(vcpu,
1296 						KVM_S390_STORE_STATUS_NOADDR);
1297 		return rc;
1298 	}
1299 
1300 	if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1301 		return -EBUSY;
1302 	stop->flags = irq->u.stop.flags;
1303 	__set_cpuflag(vcpu, CPUSTAT_STOP_INT);
1304 	return 0;
1305 }
1306 
1307 static int __inject_sigp_restart(struct kvm_vcpu *vcpu,
1308 				 struct kvm_s390_irq *irq)
1309 {
1310 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1311 
1312 	VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1313 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1314 
1315 	set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1316 	return 0;
1317 }
1318 
1319 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1320 				   struct kvm_s390_irq *irq)
1321 {
1322 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1323 
1324 	VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1325 		   irq->u.emerg.code);
1326 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1327 				   irq->u.emerg.code, 0);
1328 
1329 	/* sending vcpu invalid */
1330 	if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1331 		return -EINVAL;
1332 
1333 	set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1334 	set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1335 	atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1336 	return 0;
1337 }
1338 
1339 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1340 {
1341 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1342 	struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1343 
1344 	VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1345 		   irq->u.mchk.mcic);
1346 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1347 				   irq->u.mchk.mcic);
1348 
1349 	/*
1350 	 * Because repressible machine checks can be indicated along with
1351 	 * exigent machine checks (PoP, Chapter 11, Interruption action)
1352 	 * we need to combine cr14, mcic and external damage code.
1353 	 * Failing storage address and the logout area should not be or'ed
1354 	 * together, we just indicate the last occurrence of the corresponding
1355 	 * machine check
1356 	 */
1357 	mchk->cr14 |= irq->u.mchk.cr14;
1358 	mchk->mcic |= irq->u.mchk.mcic;
1359 	mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1360 	mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1361 	memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1362 	       sizeof(mchk->fixed_logout));
1363 	if (mchk->mcic & MCHK_EX_MASK)
1364 		set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1365 	else if (mchk->mcic & MCHK_REP_MASK)
1366 		set_bit(IRQ_PEND_MCHK_REP,  &li->pending_irqs);
1367 	return 0;
1368 }
1369 
1370 static int __inject_ckc(struct kvm_vcpu *vcpu)
1371 {
1372 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1373 
1374 	VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1375 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1376 				   0, 0);
1377 
1378 	set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1379 	atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1380 	return 0;
1381 }
1382 
1383 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1384 {
1385 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1386 
1387 	VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1388 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1389 				   0, 0);
1390 
1391 	set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1392 	atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1393 	return 0;
1394 }
1395 
1396 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1397 						  int isc, u32 schid)
1398 {
1399 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1400 	struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1401 	struct kvm_s390_interrupt_info *iter;
1402 	u16 id = (schid & 0xffff0000U) >> 16;
1403 	u16 nr = schid & 0x0000ffffU;
1404 
1405 	spin_lock(&fi->lock);
1406 	list_for_each_entry(iter, isc_list, list) {
1407 		if (schid && (id != iter->io.subchannel_id ||
1408 			      nr != iter->io.subchannel_nr))
1409 			continue;
1410 		/* found an appropriate entry */
1411 		list_del_init(&iter->list);
1412 		fi->counters[FIRQ_CNTR_IO] -= 1;
1413 		if (list_empty(isc_list))
1414 			clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1415 		spin_unlock(&fi->lock);
1416 		return iter;
1417 	}
1418 	spin_unlock(&fi->lock);
1419 	return NULL;
1420 }
1421 
1422 /*
1423  * Dequeue and return an I/O interrupt matching any of the interruption
1424  * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1425  */
1426 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1427 						    u64 isc_mask, u32 schid)
1428 {
1429 	struct kvm_s390_interrupt_info *inti = NULL;
1430 	int isc;
1431 
1432 	for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1433 		if (isc_mask & isc_to_isc_bits(isc))
1434 			inti = get_io_int(kvm, isc, schid);
1435 	}
1436 	return inti;
1437 }
1438 
1439 #define SCCB_MASK 0xFFFFFFF8
1440 #define SCCB_EVENT_PENDING 0x3
1441 
1442 static int __inject_service(struct kvm *kvm,
1443 			     struct kvm_s390_interrupt_info *inti)
1444 {
1445 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1446 
1447 	spin_lock(&fi->lock);
1448 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1449 	/*
1450 	 * Early versions of the QEMU s390 bios will inject several
1451 	 * service interrupts after another without handling a
1452 	 * condition code indicating busy.
1453 	 * We will silently ignore those superfluous sccb values.
1454 	 * A future version of QEMU will take care of serialization
1455 	 * of servc requests
1456 	 */
1457 	if (fi->srv_signal.ext_params & SCCB_MASK)
1458 		goto out;
1459 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1460 	set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1461 out:
1462 	spin_unlock(&fi->lock);
1463 	kfree(inti);
1464 	return 0;
1465 }
1466 
1467 static int __inject_virtio(struct kvm *kvm,
1468 			    struct kvm_s390_interrupt_info *inti)
1469 {
1470 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1471 
1472 	spin_lock(&fi->lock);
1473 	if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1474 		spin_unlock(&fi->lock);
1475 		return -EBUSY;
1476 	}
1477 	fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1478 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1479 	set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1480 	spin_unlock(&fi->lock);
1481 	return 0;
1482 }
1483 
1484 static int __inject_pfault_done(struct kvm *kvm,
1485 				 struct kvm_s390_interrupt_info *inti)
1486 {
1487 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1488 
1489 	spin_lock(&fi->lock);
1490 	if (fi->counters[FIRQ_CNTR_PFAULT] >=
1491 		(ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1492 		spin_unlock(&fi->lock);
1493 		return -EBUSY;
1494 	}
1495 	fi->counters[FIRQ_CNTR_PFAULT] += 1;
1496 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1497 	set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1498 	spin_unlock(&fi->lock);
1499 	return 0;
1500 }
1501 
1502 #define CR_PENDING_SUBCLASS 28
1503 static int __inject_float_mchk(struct kvm *kvm,
1504 				struct kvm_s390_interrupt_info *inti)
1505 {
1506 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1507 
1508 	spin_lock(&fi->lock);
1509 	fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1510 	fi->mchk.mcic |= inti->mchk.mcic;
1511 	set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1512 	spin_unlock(&fi->lock);
1513 	kfree(inti);
1514 	return 0;
1515 }
1516 
1517 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1518 {
1519 	struct kvm_s390_float_interrupt *fi;
1520 	struct list_head *list;
1521 	int isc;
1522 
1523 	fi = &kvm->arch.float_int;
1524 	spin_lock(&fi->lock);
1525 	if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1526 		spin_unlock(&fi->lock);
1527 		return -EBUSY;
1528 	}
1529 	fi->counters[FIRQ_CNTR_IO] += 1;
1530 
1531 	if (inti->type & KVM_S390_INT_IO_AI_MASK)
1532 		VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1533 	else
1534 		VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1535 			inti->io.subchannel_id >> 8,
1536 			inti->io.subchannel_id >> 1 & 0x3,
1537 			inti->io.subchannel_nr);
1538 	isc = int_word_to_isc(inti->io.io_int_word);
1539 	list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1540 	list_add_tail(&inti->list, list);
1541 	set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1542 	spin_unlock(&fi->lock);
1543 	return 0;
1544 }
1545 
1546 /*
1547  * Find a destination VCPU for a floating irq and kick it.
1548  */
1549 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1550 {
1551 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1552 	struct kvm_s390_local_interrupt *li;
1553 	struct kvm_vcpu *dst_vcpu;
1554 	int sigcpu, online_vcpus, nr_tries = 0;
1555 
1556 	online_vcpus = atomic_read(&kvm->online_vcpus);
1557 	if (!online_vcpus)
1558 		return;
1559 
1560 	/* find idle VCPUs first, then round robin */
1561 	sigcpu = find_first_bit(fi->idle_mask, online_vcpus);
1562 	if (sigcpu == online_vcpus) {
1563 		do {
1564 			sigcpu = fi->next_rr_cpu;
1565 			fi->next_rr_cpu = (fi->next_rr_cpu + 1) % online_vcpus;
1566 			/* avoid endless loops if all vcpus are stopped */
1567 			if (nr_tries++ >= online_vcpus)
1568 				return;
1569 		} while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1570 	}
1571 	dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1572 
1573 	/* make the VCPU drop out of the SIE, or wake it up if sleeping */
1574 	li = &dst_vcpu->arch.local_int;
1575 	spin_lock(&li->lock);
1576 	switch (type) {
1577 	case KVM_S390_MCHK:
1578 		atomic_or(CPUSTAT_STOP_INT, li->cpuflags);
1579 		break;
1580 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1581 		atomic_or(CPUSTAT_IO_INT, li->cpuflags);
1582 		break;
1583 	default:
1584 		atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1585 		break;
1586 	}
1587 	spin_unlock(&li->lock);
1588 	kvm_s390_vcpu_wakeup(dst_vcpu);
1589 }
1590 
1591 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1592 {
1593 	u64 type = READ_ONCE(inti->type);
1594 	int rc;
1595 
1596 	switch (type) {
1597 	case KVM_S390_MCHK:
1598 		rc = __inject_float_mchk(kvm, inti);
1599 		break;
1600 	case KVM_S390_INT_VIRTIO:
1601 		rc = __inject_virtio(kvm, inti);
1602 		break;
1603 	case KVM_S390_INT_SERVICE:
1604 		rc = __inject_service(kvm, inti);
1605 		break;
1606 	case KVM_S390_INT_PFAULT_DONE:
1607 		rc = __inject_pfault_done(kvm, inti);
1608 		break;
1609 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1610 		rc = __inject_io(kvm, inti);
1611 		break;
1612 	default:
1613 		rc = -EINVAL;
1614 	}
1615 	if (rc)
1616 		return rc;
1617 
1618 	__floating_irq_kick(kvm, type);
1619 	return 0;
1620 }
1621 
1622 int kvm_s390_inject_vm(struct kvm *kvm,
1623 		       struct kvm_s390_interrupt *s390int)
1624 {
1625 	struct kvm_s390_interrupt_info *inti;
1626 	int rc;
1627 
1628 	inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1629 	if (!inti)
1630 		return -ENOMEM;
1631 
1632 	inti->type = s390int->type;
1633 	switch (inti->type) {
1634 	case KVM_S390_INT_VIRTIO:
1635 		VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
1636 			 s390int->parm, s390int->parm64);
1637 		inti->ext.ext_params = s390int->parm;
1638 		inti->ext.ext_params2 = s390int->parm64;
1639 		break;
1640 	case KVM_S390_INT_SERVICE:
1641 		VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
1642 		inti->ext.ext_params = s390int->parm;
1643 		break;
1644 	case KVM_S390_INT_PFAULT_DONE:
1645 		inti->ext.ext_params2 = s390int->parm64;
1646 		break;
1647 	case KVM_S390_MCHK:
1648 		VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
1649 			 s390int->parm64);
1650 		inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
1651 		inti->mchk.mcic = s390int->parm64;
1652 		break;
1653 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1654 		inti->io.subchannel_id = s390int->parm >> 16;
1655 		inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
1656 		inti->io.io_int_parm = s390int->parm64 >> 32;
1657 		inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
1658 		break;
1659 	default:
1660 		kfree(inti);
1661 		return -EINVAL;
1662 	}
1663 	trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
1664 				 2);
1665 
1666 	rc = __inject_vm(kvm, inti);
1667 	if (rc)
1668 		kfree(inti);
1669 	return rc;
1670 }
1671 
1672 int kvm_s390_reinject_io_int(struct kvm *kvm,
1673 			      struct kvm_s390_interrupt_info *inti)
1674 {
1675 	return __inject_vm(kvm, inti);
1676 }
1677 
1678 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
1679 		       struct kvm_s390_irq *irq)
1680 {
1681 	irq->type = s390int->type;
1682 	switch (irq->type) {
1683 	case KVM_S390_PROGRAM_INT:
1684 		if (s390int->parm & 0xffff0000)
1685 			return -EINVAL;
1686 		irq->u.pgm.code = s390int->parm;
1687 		break;
1688 	case KVM_S390_SIGP_SET_PREFIX:
1689 		irq->u.prefix.address = s390int->parm;
1690 		break;
1691 	case KVM_S390_SIGP_STOP:
1692 		irq->u.stop.flags = s390int->parm;
1693 		break;
1694 	case KVM_S390_INT_EXTERNAL_CALL:
1695 		if (s390int->parm & 0xffff0000)
1696 			return -EINVAL;
1697 		irq->u.extcall.code = s390int->parm;
1698 		break;
1699 	case KVM_S390_INT_EMERGENCY:
1700 		if (s390int->parm & 0xffff0000)
1701 			return -EINVAL;
1702 		irq->u.emerg.code = s390int->parm;
1703 		break;
1704 	case KVM_S390_MCHK:
1705 		irq->u.mchk.mcic = s390int->parm64;
1706 		break;
1707 	}
1708 	return 0;
1709 }
1710 
1711 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
1712 {
1713 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1714 
1715 	return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1716 }
1717 
1718 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
1719 {
1720 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1721 
1722 	spin_lock(&li->lock);
1723 	li->irq.stop.flags = 0;
1724 	clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1725 	spin_unlock(&li->lock);
1726 }
1727 
1728 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1729 {
1730 	int rc;
1731 
1732 	switch (irq->type) {
1733 	case KVM_S390_PROGRAM_INT:
1734 		rc = __inject_prog(vcpu, irq);
1735 		break;
1736 	case KVM_S390_SIGP_SET_PREFIX:
1737 		rc = __inject_set_prefix(vcpu, irq);
1738 		break;
1739 	case KVM_S390_SIGP_STOP:
1740 		rc = __inject_sigp_stop(vcpu, irq);
1741 		break;
1742 	case KVM_S390_RESTART:
1743 		rc = __inject_sigp_restart(vcpu, irq);
1744 		break;
1745 	case KVM_S390_INT_CLOCK_COMP:
1746 		rc = __inject_ckc(vcpu);
1747 		break;
1748 	case KVM_S390_INT_CPU_TIMER:
1749 		rc = __inject_cpu_timer(vcpu);
1750 		break;
1751 	case KVM_S390_INT_EXTERNAL_CALL:
1752 		rc = __inject_extcall(vcpu, irq);
1753 		break;
1754 	case KVM_S390_INT_EMERGENCY:
1755 		rc = __inject_sigp_emergency(vcpu, irq);
1756 		break;
1757 	case KVM_S390_MCHK:
1758 		rc = __inject_mchk(vcpu, irq);
1759 		break;
1760 	case KVM_S390_INT_PFAULT_INIT:
1761 		rc = __inject_pfault_init(vcpu, irq);
1762 		break;
1763 	case KVM_S390_INT_VIRTIO:
1764 	case KVM_S390_INT_SERVICE:
1765 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1766 	default:
1767 		rc = -EINVAL;
1768 	}
1769 
1770 	return rc;
1771 }
1772 
1773 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1774 {
1775 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1776 	int rc;
1777 
1778 	spin_lock(&li->lock);
1779 	rc = do_inject_vcpu(vcpu, irq);
1780 	spin_unlock(&li->lock);
1781 	if (!rc)
1782 		kvm_s390_vcpu_wakeup(vcpu);
1783 	return rc;
1784 }
1785 
1786 static inline void clear_irq_list(struct list_head *_list)
1787 {
1788 	struct kvm_s390_interrupt_info *inti, *n;
1789 
1790 	list_for_each_entry_safe(inti, n, _list, list) {
1791 		list_del(&inti->list);
1792 		kfree(inti);
1793 	}
1794 }
1795 
1796 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
1797 		       struct kvm_s390_irq *irq)
1798 {
1799 	irq->type = inti->type;
1800 	switch (inti->type) {
1801 	case KVM_S390_INT_PFAULT_INIT:
1802 	case KVM_S390_INT_PFAULT_DONE:
1803 	case KVM_S390_INT_VIRTIO:
1804 		irq->u.ext = inti->ext;
1805 		break;
1806 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1807 		irq->u.io = inti->io;
1808 		break;
1809 	}
1810 }
1811 
1812 void kvm_s390_clear_float_irqs(struct kvm *kvm)
1813 {
1814 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1815 	int i;
1816 
1817 	spin_lock(&fi->lock);
1818 	fi->pending_irqs = 0;
1819 	memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
1820 	memset(&fi->mchk, 0, sizeof(fi->mchk));
1821 	for (i = 0; i < FIRQ_LIST_COUNT; i++)
1822 		clear_irq_list(&fi->lists[i]);
1823 	for (i = 0; i < FIRQ_MAX_COUNT; i++)
1824 		fi->counters[i] = 0;
1825 	spin_unlock(&fi->lock);
1826 };
1827 
1828 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
1829 {
1830 	struct kvm_s390_interrupt_info *inti;
1831 	struct kvm_s390_float_interrupt *fi;
1832 	struct kvm_s390_irq *buf;
1833 	struct kvm_s390_irq *irq;
1834 	int max_irqs;
1835 	int ret = 0;
1836 	int n = 0;
1837 	int i;
1838 
1839 	if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
1840 		return -EINVAL;
1841 
1842 	/*
1843 	 * We are already using -ENOMEM to signal
1844 	 * userspace it may retry with a bigger buffer,
1845 	 * so we need to use something else for this case
1846 	 */
1847 	buf = vzalloc(len);
1848 	if (!buf)
1849 		return -ENOBUFS;
1850 
1851 	max_irqs = len / sizeof(struct kvm_s390_irq);
1852 
1853 	fi = &kvm->arch.float_int;
1854 	spin_lock(&fi->lock);
1855 	for (i = 0; i < FIRQ_LIST_COUNT; i++) {
1856 		list_for_each_entry(inti, &fi->lists[i], list) {
1857 			if (n == max_irqs) {
1858 				/* signal userspace to try again */
1859 				ret = -ENOMEM;
1860 				goto out;
1861 			}
1862 			inti_to_irq(inti, &buf[n]);
1863 			n++;
1864 		}
1865 	}
1866 	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs)) {
1867 		if (n == max_irqs) {
1868 			/* signal userspace to try again */
1869 			ret = -ENOMEM;
1870 			goto out;
1871 		}
1872 		irq = (struct kvm_s390_irq *) &buf[n];
1873 		irq->type = KVM_S390_INT_SERVICE;
1874 		irq->u.ext = fi->srv_signal;
1875 		n++;
1876 	}
1877 	if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
1878 		if (n == max_irqs) {
1879 				/* signal userspace to try again */
1880 				ret = -ENOMEM;
1881 				goto out;
1882 		}
1883 		irq = (struct kvm_s390_irq *) &buf[n];
1884 		irq->type = KVM_S390_MCHK;
1885 		irq->u.mchk = fi->mchk;
1886 		n++;
1887 }
1888 
1889 out:
1890 	spin_unlock(&fi->lock);
1891 	if (!ret && n > 0) {
1892 		if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
1893 			ret = -EFAULT;
1894 	}
1895 	vfree(buf);
1896 
1897 	return ret < 0 ? ret : n;
1898 }
1899 
1900 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
1901 {
1902 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1903 	struct kvm_s390_ais_all ais;
1904 
1905 	if (attr->attr < sizeof(ais))
1906 		return -EINVAL;
1907 
1908 	if (!test_kvm_facility(kvm, 72))
1909 		return -ENOTSUPP;
1910 
1911 	mutex_lock(&fi->ais_lock);
1912 	ais.simm = fi->simm;
1913 	ais.nimm = fi->nimm;
1914 	mutex_unlock(&fi->ais_lock);
1915 
1916 	if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
1917 		return -EFAULT;
1918 
1919 	return 0;
1920 }
1921 
1922 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1923 {
1924 	int r;
1925 
1926 	switch (attr->group) {
1927 	case KVM_DEV_FLIC_GET_ALL_IRQS:
1928 		r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
1929 					  attr->attr);
1930 		break;
1931 	case KVM_DEV_FLIC_AISM_ALL:
1932 		r = flic_ais_mode_get_all(dev->kvm, attr);
1933 		break;
1934 	default:
1935 		r = -EINVAL;
1936 	}
1937 
1938 	return r;
1939 }
1940 
1941 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
1942 				     u64 addr)
1943 {
1944 	struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
1945 	void *target = NULL;
1946 	void __user *source;
1947 	u64 size;
1948 
1949 	if (get_user(inti->type, (u64 __user *)addr))
1950 		return -EFAULT;
1951 
1952 	switch (inti->type) {
1953 	case KVM_S390_INT_PFAULT_INIT:
1954 	case KVM_S390_INT_PFAULT_DONE:
1955 	case KVM_S390_INT_VIRTIO:
1956 	case KVM_S390_INT_SERVICE:
1957 		target = (void *) &inti->ext;
1958 		source = &uptr->u.ext;
1959 		size = sizeof(inti->ext);
1960 		break;
1961 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1962 		target = (void *) &inti->io;
1963 		source = &uptr->u.io;
1964 		size = sizeof(inti->io);
1965 		break;
1966 	case KVM_S390_MCHK:
1967 		target = (void *) &inti->mchk;
1968 		source = &uptr->u.mchk;
1969 		size = sizeof(inti->mchk);
1970 		break;
1971 	default:
1972 		return -EINVAL;
1973 	}
1974 
1975 	if (copy_from_user(target, source, size))
1976 		return -EFAULT;
1977 
1978 	return 0;
1979 }
1980 
1981 static int enqueue_floating_irq(struct kvm_device *dev,
1982 				struct kvm_device_attr *attr)
1983 {
1984 	struct kvm_s390_interrupt_info *inti = NULL;
1985 	int r = 0;
1986 	int len = attr->attr;
1987 
1988 	if (len % sizeof(struct kvm_s390_irq) != 0)
1989 		return -EINVAL;
1990 	else if (len > KVM_S390_FLIC_MAX_BUFFER)
1991 		return -EINVAL;
1992 
1993 	while (len >= sizeof(struct kvm_s390_irq)) {
1994 		inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1995 		if (!inti)
1996 			return -ENOMEM;
1997 
1998 		r = copy_irq_from_user(inti, attr->addr);
1999 		if (r) {
2000 			kfree(inti);
2001 			return r;
2002 		}
2003 		r = __inject_vm(dev->kvm, inti);
2004 		if (r) {
2005 			kfree(inti);
2006 			return r;
2007 		}
2008 		len -= sizeof(struct kvm_s390_irq);
2009 		attr->addr += sizeof(struct kvm_s390_irq);
2010 	}
2011 
2012 	return r;
2013 }
2014 
2015 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2016 {
2017 	if (id >= MAX_S390_IO_ADAPTERS)
2018 		return NULL;
2019 	return kvm->arch.adapters[id];
2020 }
2021 
2022 static int register_io_adapter(struct kvm_device *dev,
2023 			       struct kvm_device_attr *attr)
2024 {
2025 	struct s390_io_adapter *adapter;
2026 	struct kvm_s390_io_adapter adapter_info;
2027 
2028 	if (copy_from_user(&adapter_info,
2029 			   (void __user *)attr->addr, sizeof(adapter_info)))
2030 		return -EFAULT;
2031 
2032 	if ((adapter_info.id >= MAX_S390_IO_ADAPTERS) ||
2033 	    (dev->kvm->arch.adapters[adapter_info.id] != NULL))
2034 		return -EINVAL;
2035 
2036 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2037 	if (!adapter)
2038 		return -ENOMEM;
2039 
2040 	INIT_LIST_HEAD(&adapter->maps);
2041 	init_rwsem(&adapter->maps_lock);
2042 	atomic_set(&adapter->nr_maps, 0);
2043 	adapter->id = adapter_info.id;
2044 	adapter->isc = adapter_info.isc;
2045 	adapter->maskable = adapter_info.maskable;
2046 	adapter->masked = false;
2047 	adapter->swap = adapter_info.swap;
2048 	adapter->suppressible = (adapter_info.flags) &
2049 				KVM_S390_ADAPTER_SUPPRESSIBLE;
2050 	dev->kvm->arch.adapters[adapter->id] = adapter;
2051 
2052 	return 0;
2053 }
2054 
2055 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2056 {
2057 	int ret;
2058 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2059 
2060 	if (!adapter || !adapter->maskable)
2061 		return -EINVAL;
2062 	ret = adapter->masked;
2063 	adapter->masked = masked;
2064 	return ret;
2065 }
2066 
2067 static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr)
2068 {
2069 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2070 	struct s390_map_info *map;
2071 	int ret;
2072 
2073 	if (!adapter || !addr)
2074 		return -EINVAL;
2075 
2076 	map = kzalloc(sizeof(*map), GFP_KERNEL);
2077 	if (!map) {
2078 		ret = -ENOMEM;
2079 		goto out;
2080 	}
2081 	INIT_LIST_HEAD(&map->list);
2082 	map->guest_addr = addr;
2083 	map->addr = gmap_translate(kvm->arch.gmap, addr);
2084 	if (map->addr == -EFAULT) {
2085 		ret = -EFAULT;
2086 		goto out;
2087 	}
2088 	ret = get_user_pages_fast(map->addr, 1, 1, &map->page);
2089 	if (ret < 0)
2090 		goto out;
2091 	BUG_ON(ret != 1);
2092 	down_write(&adapter->maps_lock);
2093 	if (atomic_inc_return(&adapter->nr_maps) < MAX_S390_ADAPTER_MAPS) {
2094 		list_add_tail(&map->list, &adapter->maps);
2095 		ret = 0;
2096 	} else {
2097 		put_page(map->page);
2098 		ret = -EINVAL;
2099 	}
2100 	up_write(&adapter->maps_lock);
2101 out:
2102 	if (ret)
2103 		kfree(map);
2104 	return ret;
2105 }
2106 
2107 static int kvm_s390_adapter_unmap(struct kvm *kvm, unsigned int id, __u64 addr)
2108 {
2109 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2110 	struct s390_map_info *map, *tmp;
2111 	int found = 0;
2112 
2113 	if (!adapter || !addr)
2114 		return -EINVAL;
2115 
2116 	down_write(&adapter->maps_lock);
2117 	list_for_each_entry_safe(map, tmp, &adapter->maps, list) {
2118 		if (map->guest_addr == addr) {
2119 			found = 1;
2120 			atomic_dec(&adapter->nr_maps);
2121 			list_del(&map->list);
2122 			put_page(map->page);
2123 			kfree(map);
2124 			break;
2125 		}
2126 	}
2127 	up_write(&adapter->maps_lock);
2128 
2129 	return found ? 0 : -EINVAL;
2130 }
2131 
2132 void kvm_s390_destroy_adapters(struct kvm *kvm)
2133 {
2134 	int i;
2135 	struct s390_map_info *map, *tmp;
2136 
2137 	for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) {
2138 		if (!kvm->arch.adapters[i])
2139 			continue;
2140 		list_for_each_entry_safe(map, tmp,
2141 					 &kvm->arch.adapters[i]->maps, list) {
2142 			list_del(&map->list);
2143 			put_page(map->page);
2144 			kfree(map);
2145 		}
2146 		kfree(kvm->arch.adapters[i]);
2147 	}
2148 }
2149 
2150 static int modify_io_adapter(struct kvm_device *dev,
2151 			     struct kvm_device_attr *attr)
2152 {
2153 	struct kvm_s390_io_adapter_req req;
2154 	struct s390_io_adapter *adapter;
2155 	int ret;
2156 
2157 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2158 		return -EFAULT;
2159 
2160 	adapter = get_io_adapter(dev->kvm, req.id);
2161 	if (!adapter)
2162 		return -EINVAL;
2163 	switch (req.type) {
2164 	case KVM_S390_IO_ADAPTER_MASK:
2165 		ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2166 		if (ret > 0)
2167 			ret = 0;
2168 		break;
2169 	case KVM_S390_IO_ADAPTER_MAP:
2170 		ret = kvm_s390_adapter_map(dev->kvm, req.id, req.addr);
2171 		break;
2172 	case KVM_S390_IO_ADAPTER_UNMAP:
2173 		ret = kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr);
2174 		break;
2175 	default:
2176 		ret = -EINVAL;
2177 	}
2178 
2179 	return ret;
2180 }
2181 
2182 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2183 
2184 {
2185 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2186 	u32 schid;
2187 
2188 	if (attr->flags)
2189 		return -EINVAL;
2190 	if (attr->attr != sizeof(schid))
2191 		return -EINVAL;
2192 	if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2193 		return -EFAULT;
2194 	if (!schid)
2195 		return -EINVAL;
2196 	kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2197 	/*
2198 	 * If userspace is conforming to the architecture, we can have at most
2199 	 * one pending I/O interrupt per subchannel, so this is effectively a
2200 	 * clear all.
2201 	 */
2202 	return 0;
2203 }
2204 
2205 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2206 {
2207 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2208 	struct kvm_s390_ais_req req;
2209 	int ret = 0;
2210 
2211 	if (!test_kvm_facility(kvm, 72))
2212 		return -ENOTSUPP;
2213 
2214 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2215 		return -EFAULT;
2216 
2217 	if (req.isc > MAX_ISC)
2218 		return -EINVAL;
2219 
2220 	trace_kvm_s390_modify_ais_mode(req.isc,
2221 				       (fi->simm & AIS_MODE_MASK(req.isc)) ?
2222 				       (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2223 				       2 : KVM_S390_AIS_MODE_SINGLE :
2224 				       KVM_S390_AIS_MODE_ALL, req.mode);
2225 
2226 	mutex_lock(&fi->ais_lock);
2227 	switch (req.mode) {
2228 	case KVM_S390_AIS_MODE_ALL:
2229 		fi->simm &= ~AIS_MODE_MASK(req.isc);
2230 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2231 		break;
2232 	case KVM_S390_AIS_MODE_SINGLE:
2233 		fi->simm |= AIS_MODE_MASK(req.isc);
2234 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2235 		break;
2236 	default:
2237 		ret = -EINVAL;
2238 	}
2239 	mutex_unlock(&fi->ais_lock);
2240 
2241 	return ret;
2242 }
2243 
2244 static int kvm_s390_inject_airq(struct kvm *kvm,
2245 				struct s390_io_adapter *adapter)
2246 {
2247 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2248 	struct kvm_s390_interrupt s390int = {
2249 		.type = KVM_S390_INT_IO(1, 0, 0, 0),
2250 		.parm = 0,
2251 		.parm64 = (adapter->isc << 27) | 0x80000000,
2252 	};
2253 	int ret = 0;
2254 
2255 	if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2256 		return kvm_s390_inject_vm(kvm, &s390int);
2257 
2258 	mutex_lock(&fi->ais_lock);
2259 	if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2260 		trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2261 		goto out;
2262 	}
2263 
2264 	ret = kvm_s390_inject_vm(kvm, &s390int);
2265 	if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2266 		fi->nimm |= AIS_MODE_MASK(adapter->isc);
2267 		trace_kvm_s390_modify_ais_mode(adapter->isc,
2268 					       KVM_S390_AIS_MODE_SINGLE, 2);
2269 	}
2270 out:
2271 	mutex_unlock(&fi->ais_lock);
2272 	return ret;
2273 }
2274 
2275 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2276 {
2277 	unsigned int id = attr->attr;
2278 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2279 
2280 	if (!adapter)
2281 		return -EINVAL;
2282 
2283 	return kvm_s390_inject_airq(kvm, adapter);
2284 }
2285 
2286 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2287 {
2288 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2289 	struct kvm_s390_ais_all ais;
2290 
2291 	if (!test_kvm_facility(kvm, 72))
2292 		return -ENOTSUPP;
2293 
2294 	if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2295 		return -EFAULT;
2296 
2297 	mutex_lock(&fi->ais_lock);
2298 	fi->simm = ais.simm;
2299 	fi->nimm = ais.nimm;
2300 	mutex_unlock(&fi->ais_lock);
2301 
2302 	return 0;
2303 }
2304 
2305 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2306 {
2307 	int r = 0;
2308 	unsigned int i;
2309 	struct kvm_vcpu *vcpu;
2310 
2311 	switch (attr->group) {
2312 	case KVM_DEV_FLIC_ENQUEUE:
2313 		r = enqueue_floating_irq(dev, attr);
2314 		break;
2315 	case KVM_DEV_FLIC_CLEAR_IRQS:
2316 		kvm_s390_clear_float_irqs(dev->kvm);
2317 		break;
2318 	case KVM_DEV_FLIC_APF_ENABLE:
2319 		dev->kvm->arch.gmap->pfault_enabled = 1;
2320 		break;
2321 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2322 		dev->kvm->arch.gmap->pfault_enabled = 0;
2323 		/*
2324 		 * Make sure no async faults are in transition when
2325 		 * clearing the queues. So we don't need to worry
2326 		 * about late coming workers.
2327 		 */
2328 		synchronize_srcu(&dev->kvm->srcu);
2329 		kvm_for_each_vcpu(i, vcpu, dev->kvm)
2330 			kvm_clear_async_pf_completion_queue(vcpu);
2331 		break;
2332 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2333 		r = register_io_adapter(dev, attr);
2334 		break;
2335 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2336 		r = modify_io_adapter(dev, attr);
2337 		break;
2338 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2339 		r = clear_io_irq(dev->kvm, attr);
2340 		break;
2341 	case KVM_DEV_FLIC_AISM:
2342 		r = modify_ais_mode(dev->kvm, attr);
2343 		break;
2344 	case KVM_DEV_FLIC_AIRQ_INJECT:
2345 		r = flic_inject_airq(dev->kvm, attr);
2346 		break;
2347 	case KVM_DEV_FLIC_AISM_ALL:
2348 		r = flic_ais_mode_set_all(dev->kvm, attr);
2349 		break;
2350 	default:
2351 		r = -EINVAL;
2352 	}
2353 
2354 	return r;
2355 }
2356 
2357 static int flic_has_attr(struct kvm_device *dev,
2358 			     struct kvm_device_attr *attr)
2359 {
2360 	switch (attr->group) {
2361 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2362 	case KVM_DEV_FLIC_ENQUEUE:
2363 	case KVM_DEV_FLIC_CLEAR_IRQS:
2364 	case KVM_DEV_FLIC_APF_ENABLE:
2365 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2366 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2367 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2368 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2369 	case KVM_DEV_FLIC_AISM:
2370 	case KVM_DEV_FLIC_AIRQ_INJECT:
2371 	case KVM_DEV_FLIC_AISM_ALL:
2372 		return 0;
2373 	}
2374 	return -ENXIO;
2375 }
2376 
2377 static int flic_create(struct kvm_device *dev, u32 type)
2378 {
2379 	if (!dev)
2380 		return -EINVAL;
2381 	if (dev->kvm->arch.flic)
2382 		return -EINVAL;
2383 	dev->kvm->arch.flic = dev;
2384 	return 0;
2385 }
2386 
2387 static void flic_destroy(struct kvm_device *dev)
2388 {
2389 	dev->kvm->arch.flic = NULL;
2390 	kfree(dev);
2391 }
2392 
2393 /* s390 floating irq controller (flic) */
2394 struct kvm_device_ops kvm_flic_ops = {
2395 	.name = "kvm-flic",
2396 	.get_attr = flic_get_attr,
2397 	.set_attr = flic_set_attr,
2398 	.has_attr = flic_has_attr,
2399 	.create = flic_create,
2400 	.destroy = flic_destroy,
2401 };
2402 
2403 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2404 {
2405 	unsigned long bit;
2406 
2407 	bit = bit_nr + (addr % PAGE_SIZE) * 8;
2408 
2409 	return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2410 }
2411 
2412 static struct s390_map_info *get_map_info(struct s390_io_adapter *adapter,
2413 					  u64 addr)
2414 {
2415 	struct s390_map_info *map;
2416 
2417 	if (!adapter)
2418 		return NULL;
2419 
2420 	list_for_each_entry(map, &adapter->maps, list) {
2421 		if (map->guest_addr == addr)
2422 			return map;
2423 	}
2424 	return NULL;
2425 }
2426 
2427 static int adapter_indicators_set(struct kvm *kvm,
2428 				  struct s390_io_adapter *adapter,
2429 				  struct kvm_s390_adapter_int *adapter_int)
2430 {
2431 	unsigned long bit;
2432 	int summary_set, idx;
2433 	struct s390_map_info *info;
2434 	void *map;
2435 
2436 	info = get_map_info(adapter, adapter_int->ind_addr);
2437 	if (!info)
2438 		return -1;
2439 	map = page_address(info->page);
2440 	bit = get_ind_bit(info->addr, adapter_int->ind_offset, adapter->swap);
2441 	set_bit(bit, map);
2442 	idx = srcu_read_lock(&kvm->srcu);
2443 	mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2444 	set_page_dirty_lock(info->page);
2445 	info = get_map_info(adapter, adapter_int->summary_addr);
2446 	if (!info) {
2447 		srcu_read_unlock(&kvm->srcu, idx);
2448 		return -1;
2449 	}
2450 	map = page_address(info->page);
2451 	bit = get_ind_bit(info->addr, adapter_int->summary_offset,
2452 			  adapter->swap);
2453 	summary_set = test_and_set_bit(bit, map);
2454 	mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2455 	set_page_dirty_lock(info->page);
2456 	srcu_read_unlock(&kvm->srcu, idx);
2457 	return summary_set ? 0 : 1;
2458 }
2459 
2460 /*
2461  * < 0 - not injected due to error
2462  * = 0 - coalesced, summary indicator already active
2463  * > 0 - injected interrupt
2464  */
2465 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2466 			   struct kvm *kvm, int irq_source_id, int level,
2467 			   bool line_status)
2468 {
2469 	int ret;
2470 	struct s390_io_adapter *adapter;
2471 
2472 	/* We're only interested in the 0->1 transition. */
2473 	if (!level)
2474 		return 0;
2475 	adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2476 	if (!adapter)
2477 		return -1;
2478 	down_read(&adapter->maps_lock);
2479 	ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2480 	up_read(&adapter->maps_lock);
2481 	if ((ret > 0) && !adapter->masked) {
2482 		ret = kvm_s390_inject_airq(kvm, adapter);
2483 		if (ret == 0)
2484 			ret = 1;
2485 	}
2486 	return ret;
2487 }
2488 
2489 /*
2490  * Inject the machine check to the guest.
2491  */
2492 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2493 				     struct mcck_volatile_info *mcck_info)
2494 {
2495 	struct kvm_s390_interrupt_info inti;
2496 	struct kvm_s390_irq irq;
2497 	struct kvm_s390_mchk_info *mchk;
2498 	union mci mci;
2499 	__u64 cr14 = 0;         /* upper bits are not used */
2500 	int rc;
2501 
2502 	mci.val = mcck_info->mcic;
2503 	if (mci.sr)
2504 		cr14 |= CR14_RECOVERY_SUBMASK;
2505 	if (mci.dg)
2506 		cr14 |= CR14_DEGRADATION_SUBMASK;
2507 	if (mci.w)
2508 		cr14 |= CR14_WARNING_SUBMASK;
2509 
2510 	mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2511 	mchk->cr14 = cr14;
2512 	mchk->mcic = mcck_info->mcic;
2513 	mchk->ext_damage_code = mcck_info->ext_damage_code;
2514 	mchk->failing_storage_address = mcck_info->failing_storage_address;
2515 	if (mci.ck) {
2516 		/* Inject the floating machine check */
2517 		inti.type = KVM_S390_MCHK;
2518 		rc = __inject_vm(vcpu->kvm, &inti);
2519 	} else {
2520 		/* Inject the machine check to specified vcpu */
2521 		irq.type = KVM_S390_MCHK;
2522 		rc = kvm_s390_inject_vcpu(vcpu, &irq);
2523 	}
2524 	WARN_ON_ONCE(rc);
2525 }
2526 
2527 int kvm_set_routing_entry(struct kvm *kvm,
2528 			  struct kvm_kernel_irq_routing_entry *e,
2529 			  const struct kvm_irq_routing_entry *ue)
2530 {
2531 	int ret;
2532 
2533 	switch (ue->type) {
2534 	case KVM_IRQ_ROUTING_S390_ADAPTER:
2535 		e->set = set_adapter_int;
2536 		e->adapter.summary_addr = ue->u.adapter.summary_addr;
2537 		e->adapter.ind_addr = ue->u.adapter.ind_addr;
2538 		e->adapter.summary_offset = ue->u.adapter.summary_offset;
2539 		e->adapter.ind_offset = ue->u.adapter.ind_offset;
2540 		e->adapter.adapter_id = ue->u.adapter.adapter_id;
2541 		ret = 0;
2542 		break;
2543 	default:
2544 		ret = -EINVAL;
2545 	}
2546 
2547 	return ret;
2548 }
2549 
2550 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2551 		int irq_source_id, int level, bool line_status)
2552 {
2553 	return -EINVAL;
2554 }
2555 
2556 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2557 {
2558 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2559 	struct kvm_s390_irq *buf;
2560 	int r = 0;
2561 	int n;
2562 
2563 	buf = vmalloc(len);
2564 	if (!buf)
2565 		return -ENOMEM;
2566 
2567 	if (copy_from_user((void *) buf, irqstate, len)) {
2568 		r = -EFAULT;
2569 		goto out_free;
2570 	}
2571 
2572 	/*
2573 	 * Don't allow setting the interrupt state
2574 	 * when there are already interrupts pending
2575 	 */
2576 	spin_lock(&li->lock);
2577 	if (li->pending_irqs) {
2578 		r = -EBUSY;
2579 		goto out_unlock;
2580 	}
2581 
2582 	for (n = 0; n < len / sizeof(*buf); n++) {
2583 		r = do_inject_vcpu(vcpu, &buf[n]);
2584 		if (r)
2585 			break;
2586 	}
2587 
2588 out_unlock:
2589 	spin_unlock(&li->lock);
2590 out_free:
2591 	vfree(buf);
2592 
2593 	return r;
2594 }
2595 
2596 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2597 			    struct kvm_s390_irq *irq,
2598 			    unsigned long irq_type)
2599 {
2600 	switch (irq_type) {
2601 	case IRQ_PEND_MCHK_EX:
2602 	case IRQ_PEND_MCHK_REP:
2603 		irq->type = KVM_S390_MCHK;
2604 		irq->u.mchk = li->irq.mchk;
2605 		break;
2606 	case IRQ_PEND_PROG:
2607 		irq->type = KVM_S390_PROGRAM_INT;
2608 		irq->u.pgm = li->irq.pgm;
2609 		break;
2610 	case IRQ_PEND_PFAULT_INIT:
2611 		irq->type = KVM_S390_INT_PFAULT_INIT;
2612 		irq->u.ext = li->irq.ext;
2613 		break;
2614 	case IRQ_PEND_EXT_EXTERNAL:
2615 		irq->type = KVM_S390_INT_EXTERNAL_CALL;
2616 		irq->u.extcall = li->irq.extcall;
2617 		break;
2618 	case IRQ_PEND_EXT_CLOCK_COMP:
2619 		irq->type = KVM_S390_INT_CLOCK_COMP;
2620 		break;
2621 	case IRQ_PEND_EXT_CPU_TIMER:
2622 		irq->type = KVM_S390_INT_CPU_TIMER;
2623 		break;
2624 	case IRQ_PEND_SIGP_STOP:
2625 		irq->type = KVM_S390_SIGP_STOP;
2626 		irq->u.stop = li->irq.stop;
2627 		break;
2628 	case IRQ_PEND_RESTART:
2629 		irq->type = KVM_S390_RESTART;
2630 		break;
2631 	case IRQ_PEND_SET_PREFIX:
2632 		irq->type = KVM_S390_SIGP_SET_PREFIX;
2633 		irq->u.prefix = li->irq.prefix;
2634 		break;
2635 	}
2636 }
2637 
2638 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
2639 {
2640 	int scn;
2641 	unsigned long sigp_emerg_pending[BITS_TO_LONGS(KVM_MAX_VCPUS)];
2642 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2643 	unsigned long pending_irqs;
2644 	struct kvm_s390_irq irq;
2645 	unsigned long irq_type;
2646 	int cpuaddr;
2647 	int n = 0;
2648 
2649 	spin_lock(&li->lock);
2650 	pending_irqs = li->pending_irqs;
2651 	memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
2652 	       sizeof(sigp_emerg_pending));
2653 	spin_unlock(&li->lock);
2654 
2655 	for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
2656 		memset(&irq, 0, sizeof(irq));
2657 		if (irq_type == IRQ_PEND_EXT_EMERGENCY)
2658 			continue;
2659 		if (n + sizeof(irq) > len)
2660 			return -ENOBUFS;
2661 		store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
2662 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2663 			return -EFAULT;
2664 		n += sizeof(irq);
2665 	}
2666 
2667 	if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
2668 		for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
2669 			memset(&irq, 0, sizeof(irq));
2670 			if (n + sizeof(irq) > len)
2671 				return -ENOBUFS;
2672 			irq.type = KVM_S390_INT_EMERGENCY;
2673 			irq.u.emerg.code = cpuaddr;
2674 			if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2675 				return -EFAULT;
2676 			n += sizeof(irq);
2677 		}
2678 	}
2679 
2680 	if (sca_ext_call_pending(vcpu, &scn)) {
2681 		if (n + sizeof(irq) > len)
2682 			return -ENOBUFS;
2683 		memset(&irq, 0, sizeof(irq));
2684 		irq.type = KVM_S390_INT_EXTERNAL_CALL;
2685 		irq.u.extcall.code = scn;
2686 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2687 			return -EFAULT;
2688 		n += sizeof(irq);
2689 	}
2690 
2691 	return n;
2692 }
2693