xref: /linux/arch/s390/kvm/interrupt.c (revision bc46b7cbc58c4cb562b6a45a1fbc7b8e7b23df58)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * handling kvm guest interrupts
4  *
5  * Copyright IBM Corp. 2008, 2020
6  *
7  *    Author(s): Carsten Otte <cotte@de.ibm.com>
8  */
9 
10 #define KMSG_COMPONENT "kvm-s390"
11 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
12 
13 #include <linux/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 */
sca_ext_call_pending(struct kvm_vcpu * vcpu,int * src_id)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 
sca_inject_ext_call(struct kvm_vcpu * vcpu,int src_id)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 
sca_clear_ext_call(struct kvm_vcpu * vcpu)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 
psw_extint_disabled(struct kvm_vcpu * vcpu)143 int psw_extint_disabled(struct kvm_vcpu *vcpu)
144 {
145 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
146 }
147 
psw_ioint_disabled(struct kvm_vcpu * vcpu)148 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
149 {
150 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
151 }
152 
psw_mchk_disabled(struct kvm_vcpu * vcpu)153 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
154 {
155 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
156 }
157 
psw_interrupts_disabled(struct kvm_vcpu * vcpu)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 
ckc_interrupts_enabled(struct kvm_vcpu * vcpu)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 
ckc_irq_pending(struct kvm_vcpu * vcpu)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 
cpu_timer_interrupts_enabled(struct kvm_vcpu * vcpu)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 
cpu_timer_irq_pending(struct kvm_vcpu * vcpu)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 
isc_to_isc_bits(int isc)203 static uint64_t isc_to_isc_bits(int isc)
204 {
205 	return (0x80 >> isc) << 24;
206 }
207 
isc_to_int_word(u8 isc)208 static inline u32 isc_to_int_word(u8 isc)
209 {
210 	return ((u32)isc << 27) | 0x80000000;
211 }
212 
int_word_to_isc(u32 int_word)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  */
gisa_set_iam(struct kvm_s390_gisa * gisa,u8 iam)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  */
gisa_clear_ipm(struct kvm_s390_gisa * gisa)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  */
gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt * gi)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 
gisa_set_ipm_gisc(struct kvm_s390_gisa * gisa,u32 gisc)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 
gisa_get_ipm(struct kvm_s390_gisa * gisa)304 static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
305 {
306 	return READ_ONCE(gisa->ipm);
307 }
308 
gisa_tac_ipm_gisc(struct kvm_s390_gisa * gisa,u32 gisc)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 
pending_irqs_no_gisa(struct kvm_vcpu * vcpu)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 
pending_irqs(struct kvm_vcpu * vcpu)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 
isc_to_irq_type(unsigned long isc)334 static inline int isc_to_irq_type(unsigned long isc)
335 {
336 	return IRQ_PEND_IO_ISC_0 - isc;
337 }
338 
irq_type_to_isc(unsigned long irq_type)339 static inline int irq_type_to_isc(unsigned long irq_type)
340 {
341 	return IRQ_PEND_IO_ISC_0 - irq_type;
342 }
343 
disable_iscs(struct kvm_vcpu * vcpu,unsigned long active_mask)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 
deliverable_irqs(struct kvm_vcpu * vcpu)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 
__set_cpu_idle(struct kvm_vcpu * vcpu)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 
__unset_cpu_idle(struct kvm_vcpu * vcpu)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 
__reset_intercept_indicators(struct kvm_vcpu * vcpu)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 
set_intercept_indicators_io(struct kvm_vcpu * vcpu)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 
set_intercept_indicators_ext(struct kvm_vcpu * vcpu)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 
set_intercept_indicators_mchk(struct kvm_vcpu * vcpu)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 
set_intercept_indicators_stop(struct kvm_vcpu * vcpu)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 */
set_intercept_indicators(struct kvm_vcpu * vcpu)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 
__deliver_cpu_timer(struct kvm_vcpu * vcpu)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 
__deliver_ckc(struct kvm_vcpu * vcpu)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 
__deliver_pfault_init(struct kvm_vcpu * vcpu)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 
__write_machine_check(struct kvm_vcpu * vcpu,struct kvm_s390_mchk_info * mchk)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 
__deliver_machine_check(struct kvm_vcpu * vcpu)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 
__deliver_restart(struct kvm_vcpu * vcpu)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 
__deliver_set_prefix(struct kvm_vcpu * vcpu)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 
__deliver_emergency_signal(struct kvm_vcpu * vcpu)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 
__deliver_external_call(struct kvm_vcpu * vcpu)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 
__deliver_prog_pv(struct kvm_vcpu * vcpu,u16 code)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 
__deliver_prog(struct kvm_vcpu * vcpu)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 
write_sclp(struct kvm_vcpu * vcpu,u32 parm)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 
__deliver_service(struct kvm_vcpu * vcpu)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 
__deliver_service_ev(struct kvm_vcpu * vcpu)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 
__deliver_pfault_done(struct kvm_vcpu * vcpu)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 
__deliver_virtio(struct kvm_vcpu * vcpu)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 
__do_deliver_io(struct kvm_vcpu * vcpu,struct kvm_s390_io_info * io)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 
__deliver_io(struct kvm_vcpu * vcpu,unsigned long irq_type)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) */
kvm_s390_ext_call_pending(struct kvm_vcpu * vcpu)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 
kvm_s390_vcpu_has_irq(struct kvm_vcpu * vcpu,int exclude_stop)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 
kvm_cpu_has_pending_timer(struct kvm_vcpu * vcpu)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 
__calculate_sltime(struct kvm_vcpu * vcpu)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 
kvm_s390_handle_wait(struct kvm_vcpu * vcpu)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 	kvm_vcpu_halt(vcpu);
1327 	vcpu->valid_wakeup = false;
1328 	__unset_cpu_idle(vcpu);
1329 	kvm_vcpu_srcu_read_lock(vcpu);
1330 
1331 	hrtimer_cancel(&vcpu->arch.ckc_timer);
1332 	return 0;
1333 }
1334 
kvm_s390_vcpu_wakeup(struct kvm_vcpu * vcpu)1335 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1336 {
1337 	vcpu->valid_wakeup = true;
1338 	kvm_vcpu_wake_up(vcpu);
1339 
1340 	/*
1341 	 * The VCPU might not be sleeping but rather executing VSIE. Let's
1342 	 * kick it, so it leaves the SIE to process the request.
1343 	 */
1344 	kvm_s390_vsie_kick(vcpu);
1345 }
1346 
kvm_s390_idle_wakeup(struct hrtimer * timer)1347 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1348 {
1349 	struct kvm_vcpu *vcpu;
1350 	u64 sltime;
1351 
1352 	vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1353 	sltime = __calculate_sltime(vcpu);
1354 
1355 	/*
1356 	 * If the monotonic clock runs faster than the tod clock we might be
1357 	 * woken up too early and have to go back to sleep to avoid deadlocks.
1358 	 */
1359 	if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1360 		return HRTIMER_RESTART;
1361 	kvm_s390_vcpu_wakeup(vcpu);
1362 	return HRTIMER_NORESTART;
1363 }
1364 
kvm_s390_clear_local_irqs(struct kvm_vcpu * vcpu)1365 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1366 {
1367 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1368 
1369 	spin_lock(&li->lock);
1370 	li->pending_irqs = 0;
1371 	bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1372 	memset(&li->irq, 0, sizeof(li->irq));
1373 	spin_unlock(&li->lock);
1374 
1375 	sca_clear_ext_call(vcpu);
1376 }
1377 
kvm_s390_deliver_pending_interrupts(struct kvm_vcpu * vcpu)1378 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1379 {
1380 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1381 	int rc = 0;
1382 	bool delivered = false;
1383 	unsigned long irq_type;
1384 	unsigned long irqs;
1385 
1386 	__reset_intercept_indicators(vcpu);
1387 
1388 	/* pending ckc conditions might have been invalidated */
1389 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1390 	if (ckc_irq_pending(vcpu))
1391 		set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1392 
1393 	/* pending cpu timer conditions might have been invalidated */
1394 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1395 	if (cpu_timer_irq_pending(vcpu))
1396 		set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1397 
1398 	while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1399 		/* bits are in the reverse order of interrupt priority */
1400 		irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1401 		switch (irq_type) {
1402 		case IRQ_PEND_IO_ISC_0:
1403 		case IRQ_PEND_IO_ISC_1:
1404 		case IRQ_PEND_IO_ISC_2:
1405 		case IRQ_PEND_IO_ISC_3:
1406 		case IRQ_PEND_IO_ISC_4:
1407 		case IRQ_PEND_IO_ISC_5:
1408 		case IRQ_PEND_IO_ISC_6:
1409 		case IRQ_PEND_IO_ISC_7:
1410 			rc = __deliver_io(vcpu, irq_type);
1411 			break;
1412 		case IRQ_PEND_MCHK_EX:
1413 		case IRQ_PEND_MCHK_REP:
1414 			rc = __deliver_machine_check(vcpu);
1415 			break;
1416 		case IRQ_PEND_PROG:
1417 			rc = __deliver_prog(vcpu);
1418 			break;
1419 		case IRQ_PEND_EXT_EMERGENCY:
1420 			rc = __deliver_emergency_signal(vcpu);
1421 			break;
1422 		case IRQ_PEND_EXT_EXTERNAL:
1423 			rc = __deliver_external_call(vcpu);
1424 			break;
1425 		case IRQ_PEND_EXT_CLOCK_COMP:
1426 			rc = __deliver_ckc(vcpu);
1427 			break;
1428 		case IRQ_PEND_EXT_CPU_TIMER:
1429 			rc = __deliver_cpu_timer(vcpu);
1430 			break;
1431 		case IRQ_PEND_RESTART:
1432 			rc = __deliver_restart(vcpu);
1433 			break;
1434 		case IRQ_PEND_SET_PREFIX:
1435 			rc = __deliver_set_prefix(vcpu);
1436 			break;
1437 		case IRQ_PEND_PFAULT_INIT:
1438 			rc = __deliver_pfault_init(vcpu);
1439 			break;
1440 		case IRQ_PEND_EXT_SERVICE:
1441 			rc = __deliver_service(vcpu);
1442 			break;
1443 		case IRQ_PEND_EXT_SERVICE_EV:
1444 			rc = __deliver_service_ev(vcpu);
1445 			break;
1446 		case IRQ_PEND_PFAULT_DONE:
1447 			rc = __deliver_pfault_done(vcpu);
1448 			break;
1449 		case IRQ_PEND_VIRTIO:
1450 			rc = __deliver_virtio(vcpu);
1451 			break;
1452 		default:
1453 			WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1454 			clear_bit(irq_type, &li->pending_irqs);
1455 		}
1456 		delivered |= !rc;
1457 	}
1458 
1459 	/*
1460 	 * We delivered at least one interrupt and modified the PC. Force a
1461 	 * singlestep event now.
1462 	 */
1463 	if (delivered && guestdbg_sstep_enabled(vcpu)) {
1464 		struct kvm_debug_exit_arch *debug_exit = &vcpu->run->debug.arch;
1465 
1466 		debug_exit->addr = vcpu->arch.sie_block->gpsw.addr;
1467 		debug_exit->type = KVM_SINGLESTEP;
1468 		vcpu->guest_debug |= KVM_GUESTDBG_EXIT_PENDING;
1469 	}
1470 
1471 	set_intercept_indicators(vcpu);
1472 
1473 	return rc;
1474 }
1475 
__inject_prog(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1476 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1477 {
1478 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1479 
1480 	vcpu->stat.inject_program++;
1481 	VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1482 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1483 				   irq->u.pgm.code, 0);
1484 
1485 	if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1486 		/* auto detection if no valid ILC was given */
1487 		irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1488 		irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1489 		irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1490 	}
1491 
1492 	if (irq->u.pgm.code == PGM_PER) {
1493 		li->irq.pgm.code |= PGM_PER;
1494 		li->irq.pgm.flags = irq->u.pgm.flags;
1495 		/* only modify PER related information */
1496 		li->irq.pgm.per_address = irq->u.pgm.per_address;
1497 		li->irq.pgm.per_code = irq->u.pgm.per_code;
1498 		li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1499 		li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1500 	} else if (!(irq->u.pgm.code & PGM_PER)) {
1501 		li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1502 				   irq->u.pgm.code;
1503 		li->irq.pgm.flags = irq->u.pgm.flags;
1504 		/* only modify non-PER information */
1505 		li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1506 		li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1507 		li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1508 		li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1509 		li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1510 		li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1511 	} else {
1512 		li->irq.pgm = irq->u.pgm;
1513 	}
1514 	set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1515 	return 0;
1516 }
1517 
__inject_pfault_init(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1518 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1519 {
1520 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1521 
1522 	vcpu->stat.inject_pfault_init++;
1523 	VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1524 		   irq->u.ext.ext_params2);
1525 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1526 				   irq->u.ext.ext_params,
1527 				   irq->u.ext.ext_params2);
1528 
1529 	li->irq.ext = irq->u.ext;
1530 	set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1531 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1532 	return 0;
1533 }
1534 
__inject_extcall(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1535 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1536 {
1537 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1538 	struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1539 	uint16_t src_id = irq->u.extcall.code;
1540 
1541 	vcpu->stat.inject_external_call++;
1542 	VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1543 		   src_id);
1544 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1545 				   src_id, 0);
1546 
1547 	/* sending vcpu invalid */
1548 	if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1549 		return -EINVAL;
1550 
1551 	if (sclp.has_sigpif && !kvm_s390_pv_cpu_get_handle(vcpu))
1552 		return sca_inject_ext_call(vcpu, src_id);
1553 
1554 	if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1555 		return -EBUSY;
1556 	*extcall = irq->u.extcall;
1557 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1558 	return 0;
1559 }
1560 
__inject_set_prefix(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1561 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1562 {
1563 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1564 	struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1565 
1566 	vcpu->stat.inject_set_prefix++;
1567 	VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1568 		   irq->u.prefix.address);
1569 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1570 				   irq->u.prefix.address, 0);
1571 
1572 	if (!is_vcpu_stopped(vcpu))
1573 		return -EBUSY;
1574 
1575 	*prefix = irq->u.prefix;
1576 	set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1577 	return 0;
1578 }
1579 
1580 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
__inject_sigp_stop(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1581 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1582 {
1583 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1584 	struct kvm_s390_stop_info *stop = &li->irq.stop;
1585 	int rc = 0;
1586 
1587 	vcpu->stat.inject_stop_signal++;
1588 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1589 
1590 	if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1591 		return -EINVAL;
1592 
1593 	if (is_vcpu_stopped(vcpu)) {
1594 		if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1595 			rc = kvm_s390_store_status_unloaded(vcpu,
1596 						KVM_S390_STORE_STATUS_NOADDR);
1597 		return rc;
1598 	}
1599 
1600 	if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1601 		return -EBUSY;
1602 	stop->flags = irq->u.stop.flags;
1603 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1604 	return 0;
1605 }
1606 
__inject_sigp_restart(struct kvm_vcpu * vcpu)1607 static int __inject_sigp_restart(struct kvm_vcpu *vcpu)
1608 {
1609 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1610 
1611 	vcpu->stat.inject_restart++;
1612 	VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1613 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1614 
1615 	set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1616 	return 0;
1617 }
1618 
__inject_sigp_emergency(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1619 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1620 				   struct kvm_s390_irq *irq)
1621 {
1622 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1623 
1624 	vcpu->stat.inject_emergency_signal++;
1625 	VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1626 		   irq->u.emerg.code);
1627 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1628 				   irq->u.emerg.code, 0);
1629 
1630 	/* sending vcpu invalid */
1631 	if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1632 		return -EINVAL;
1633 
1634 	set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1635 	set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1636 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1637 	return 0;
1638 }
1639 
__inject_mchk(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1640 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1641 {
1642 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1643 	struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1644 
1645 	vcpu->stat.inject_mchk++;
1646 	VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1647 		   irq->u.mchk.mcic);
1648 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1649 				   irq->u.mchk.mcic);
1650 
1651 	/*
1652 	 * Because repressible machine checks can be indicated along with
1653 	 * exigent machine checks (PoP, Chapter 11, Interruption action)
1654 	 * we need to combine cr14, mcic and external damage code.
1655 	 * Failing storage address and the logout area should not be or'ed
1656 	 * together, we just indicate the last occurrence of the corresponding
1657 	 * machine check
1658 	 */
1659 	mchk->cr14 |= irq->u.mchk.cr14;
1660 	mchk->mcic |= irq->u.mchk.mcic;
1661 	mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1662 	mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1663 	memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1664 	       sizeof(mchk->fixed_logout));
1665 	if (mchk->mcic & MCHK_EX_MASK)
1666 		set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1667 	else if (mchk->mcic & MCHK_REP_MASK)
1668 		set_bit(IRQ_PEND_MCHK_REP,  &li->pending_irqs);
1669 	return 0;
1670 }
1671 
__inject_ckc(struct kvm_vcpu * vcpu)1672 static int __inject_ckc(struct kvm_vcpu *vcpu)
1673 {
1674 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1675 
1676 	vcpu->stat.inject_ckc++;
1677 	VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1678 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1679 				   0, 0);
1680 
1681 	set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1682 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1683 	return 0;
1684 }
1685 
__inject_cpu_timer(struct kvm_vcpu * vcpu)1686 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1687 {
1688 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1689 
1690 	vcpu->stat.inject_cputm++;
1691 	VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1692 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1693 				   0, 0);
1694 
1695 	set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1696 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1697 	return 0;
1698 }
1699 
get_io_int(struct kvm * kvm,int isc,u32 schid)1700 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1701 						  int isc, u32 schid)
1702 {
1703 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1704 	struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1705 	struct kvm_s390_interrupt_info *iter;
1706 	u16 id = (schid & 0xffff0000U) >> 16;
1707 	u16 nr = schid & 0x0000ffffU;
1708 
1709 	spin_lock(&fi->lock);
1710 	list_for_each_entry(iter, isc_list, list) {
1711 		if (schid && (id != iter->io.subchannel_id ||
1712 			      nr != iter->io.subchannel_nr))
1713 			continue;
1714 		/* found an appropriate entry */
1715 		list_del_init(&iter->list);
1716 		fi->counters[FIRQ_CNTR_IO] -= 1;
1717 		if (list_empty(isc_list))
1718 			clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1719 		spin_unlock(&fi->lock);
1720 		return iter;
1721 	}
1722 	spin_unlock(&fi->lock);
1723 	return NULL;
1724 }
1725 
get_top_io_int(struct kvm * kvm,u64 isc_mask,u32 schid)1726 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1727 						      u64 isc_mask, u32 schid)
1728 {
1729 	struct kvm_s390_interrupt_info *inti = NULL;
1730 	int isc;
1731 
1732 	for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1733 		if (isc_mask & isc_to_isc_bits(isc))
1734 			inti = get_io_int(kvm, isc, schid);
1735 	}
1736 	return inti;
1737 }
1738 
get_top_gisa_isc(struct kvm * kvm,u64 isc_mask,u32 schid)1739 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1740 {
1741 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1742 	unsigned long active_mask;
1743 	int isc;
1744 
1745 	if (schid)
1746 		goto out;
1747 	if (!gi->origin)
1748 		goto out;
1749 
1750 	active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1751 	while (active_mask) {
1752 		isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1753 		if (gisa_tac_ipm_gisc(gi->origin, isc))
1754 			return isc;
1755 		clear_bit_inv(isc, &active_mask);
1756 	}
1757 out:
1758 	return -EINVAL;
1759 }
1760 
1761 /*
1762  * Dequeue and return an I/O interrupt matching any of the interruption
1763  * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1764  * Take into account the interrupts pending in the interrupt list and in GISA.
1765  *
1766  * Note that for a guest that does not enable I/O interrupts
1767  * but relies on TPI, a flood of classic interrupts may starve
1768  * out adapter interrupts on the same isc. Linux does not do
1769  * that, and it is possible to work around the issue by configuring
1770  * different iscs for classic and adapter interrupts in the guest,
1771  * but we may want to revisit this in the future.
1772  */
kvm_s390_get_io_int(struct kvm * kvm,u64 isc_mask,u32 schid)1773 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1774 						    u64 isc_mask, u32 schid)
1775 {
1776 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1777 	struct kvm_s390_interrupt_info *inti, *tmp_inti;
1778 	int isc;
1779 
1780 	inti = get_top_io_int(kvm, isc_mask, schid);
1781 
1782 	isc = get_top_gisa_isc(kvm, isc_mask, schid);
1783 	if (isc < 0)
1784 		/* no AI in GISA */
1785 		goto out;
1786 
1787 	if (!inti)
1788 		/* AI in GISA but no classical IO int */
1789 		goto gisa_out;
1790 
1791 	/* both types of interrupts present */
1792 	if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1793 		/* classical IO int with higher priority */
1794 		gisa_set_ipm_gisc(gi->origin, isc);
1795 		goto out;
1796 	}
1797 gisa_out:
1798 	tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
1799 	if (tmp_inti) {
1800 		tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1801 		tmp_inti->io.io_int_word = isc_to_int_word(isc);
1802 		if (inti)
1803 			kvm_s390_reinject_io_int(kvm, inti);
1804 		inti = tmp_inti;
1805 	} else
1806 		gisa_set_ipm_gisc(gi->origin, isc);
1807 out:
1808 	return inti;
1809 }
1810 
__inject_service(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1811 static int __inject_service(struct kvm *kvm,
1812 			     struct kvm_s390_interrupt_info *inti)
1813 {
1814 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1815 
1816 	kvm->stat.inject_service_signal++;
1817 	spin_lock(&fi->lock);
1818 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1819 
1820 	/* We always allow events, track them separately from the sccb ints */
1821 	if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING)
1822 		set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1823 
1824 	/*
1825 	 * Early versions of the QEMU s390 bios will inject several
1826 	 * service interrupts after another without handling a
1827 	 * condition code indicating busy.
1828 	 * We will silently ignore those superfluous sccb values.
1829 	 * A future version of QEMU will take care of serialization
1830 	 * of servc requests
1831 	 */
1832 	if (fi->srv_signal.ext_params & SCCB_MASK)
1833 		goto out;
1834 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1835 	set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1836 out:
1837 	spin_unlock(&fi->lock);
1838 	kfree(inti);
1839 	return 0;
1840 }
1841 
__inject_virtio(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1842 static int __inject_virtio(struct kvm *kvm,
1843 			    struct kvm_s390_interrupt_info *inti)
1844 {
1845 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1846 
1847 	kvm->stat.inject_virtio++;
1848 	spin_lock(&fi->lock);
1849 	if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1850 		spin_unlock(&fi->lock);
1851 		return -EBUSY;
1852 	}
1853 	fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1854 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1855 	set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1856 	spin_unlock(&fi->lock);
1857 	return 0;
1858 }
1859 
__inject_pfault_done(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1860 static int __inject_pfault_done(struct kvm *kvm,
1861 				 struct kvm_s390_interrupt_info *inti)
1862 {
1863 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1864 
1865 	kvm->stat.inject_pfault_done++;
1866 	spin_lock(&fi->lock);
1867 	if (fi->counters[FIRQ_CNTR_PFAULT] >=
1868 		(ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1869 		spin_unlock(&fi->lock);
1870 		return -EBUSY;
1871 	}
1872 	fi->counters[FIRQ_CNTR_PFAULT] += 1;
1873 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1874 	set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1875 	spin_unlock(&fi->lock);
1876 	return 0;
1877 }
1878 
1879 #define CR_PENDING_SUBCLASS 28
__inject_float_mchk(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1880 static int __inject_float_mchk(struct kvm *kvm,
1881 				struct kvm_s390_interrupt_info *inti)
1882 {
1883 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1884 
1885 	kvm->stat.inject_float_mchk++;
1886 	spin_lock(&fi->lock);
1887 	fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1888 	fi->mchk.mcic |= inti->mchk.mcic;
1889 	set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1890 	spin_unlock(&fi->lock);
1891 	kfree(inti);
1892 	return 0;
1893 }
1894 
__inject_io(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1895 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1896 {
1897 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1898 	struct kvm_s390_float_interrupt *fi;
1899 	struct list_head *list;
1900 	int isc;
1901 
1902 	kvm->stat.inject_io++;
1903 	isc = int_word_to_isc(inti->io.io_int_word);
1904 
1905 	/*
1906 	 * We do not use the lock checking variant as this is just a
1907 	 * performance optimization and we do not hold the lock here.
1908 	 * This is ok as the code will pick interrupts from both "lists"
1909 	 * for delivery.
1910 	 */
1911 	if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1912 		VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1913 		gisa_set_ipm_gisc(gi->origin, isc);
1914 		kfree(inti);
1915 		return 0;
1916 	}
1917 
1918 	fi = &kvm->arch.float_int;
1919 	spin_lock(&fi->lock);
1920 	if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1921 		spin_unlock(&fi->lock);
1922 		return -EBUSY;
1923 	}
1924 	fi->counters[FIRQ_CNTR_IO] += 1;
1925 
1926 	if (inti->type & KVM_S390_INT_IO_AI_MASK)
1927 		VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1928 	else
1929 		VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1930 			inti->io.subchannel_id >> 8,
1931 			inti->io.subchannel_id >> 1 & 0x3,
1932 			inti->io.subchannel_nr);
1933 	list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1934 	list_add_tail(&inti->list, list);
1935 	set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1936 	spin_unlock(&fi->lock);
1937 	return 0;
1938 }
1939 
1940 /*
1941  * Find a destination VCPU for a floating irq and kick it.
1942  */
__floating_irq_kick(struct kvm * kvm,u64 type)1943 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1944 {
1945 	struct kvm_vcpu *dst_vcpu;
1946 	int sigcpu, online_vcpus, nr_tries = 0;
1947 
1948 	online_vcpus = atomic_read(&kvm->online_vcpus);
1949 	if (!online_vcpus)
1950 		return;
1951 
1952 	/* find idle VCPUs first, then round robin */
1953 	sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1954 	if (sigcpu == online_vcpus) {
1955 		do {
1956 			sigcpu = kvm->arch.float_int.next_rr_cpu++;
1957 			kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1958 			/* avoid endless loops if all vcpus are stopped */
1959 			if (nr_tries++ >= online_vcpus)
1960 				return;
1961 		} while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1962 	}
1963 	dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1964 
1965 	/* make the VCPU drop out of the SIE, or wake it up if sleeping */
1966 	switch (type) {
1967 	case KVM_S390_MCHK:
1968 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1969 		break;
1970 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1971 		if (!(type & KVM_S390_INT_IO_AI_MASK &&
1972 		      kvm->arch.gisa_int.origin) ||
1973 		      kvm_s390_pv_cpu_get_handle(dst_vcpu))
1974 			kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1975 		break;
1976 	default:
1977 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1978 		break;
1979 	}
1980 	kvm_s390_vcpu_wakeup(dst_vcpu);
1981 }
1982 
__inject_vm(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1983 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1984 {
1985 	u64 type = READ_ONCE(inti->type);
1986 	int rc;
1987 
1988 	switch (type) {
1989 	case KVM_S390_MCHK:
1990 		rc = __inject_float_mchk(kvm, inti);
1991 		break;
1992 	case KVM_S390_INT_VIRTIO:
1993 		rc = __inject_virtio(kvm, inti);
1994 		break;
1995 	case KVM_S390_INT_SERVICE:
1996 		rc = __inject_service(kvm, inti);
1997 		break;
1998 	case KVM_S390_INT_PFAULT_DONE:
1999 		rc = __inject_pfault_done(kvm, inti);
2000 		break;
2001 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2002 		rc = __inject_io(kvm, inti);
2003 		break;
2004 	default:
2005 		rc = -EINVAL;
2006 	}
2007 	if (rc)
2008 		return rc;
2009 
2010 	__floating_irq_kick(kvm, type);
2011 	return 0;
2012 }
2013 
kvm_s390_inject_vm(struct kvm * kvm,struct kvm_s390_interrupt * s390int)2014 int kvm_s390_inject_vm(struct kvm *kvm,
2015 		       struct kvm_s390_interrupt *s390int)
2016 {
2017 	struct kvm_s390_interrupt_info *inti;
2018 	int rc;
2019 
2020 	inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2021 	if (!inti)
2022 		return -ENOMEM;
2023 
2024 	inti->type = s390int->type;
2025 	switch (inti->type) {
2026 	case KVM_S390_INT_VIRTIO:
2027 		VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
2028 			 s390int->parm, s390int->parm64);
2029 		inti->ext.ext_params = s390int->parm;
2030 		inti->ext.ext_params2 = s390int->parm64;
2031 		break;
2032 	case KVM_S390_INT_SERVICE:
2033 		VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
2034 		inti->ext.ext_params = s390int->parm;
2035 		break;
2036 	case KVM_S390_INT_PFAULT_DONE:
2037 		inti->ext.ext_params2 = s390int->parm64;
2038 		break;
2039 	case KVM_S390_MCHK:
2040 		VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
2041 			 s390int->parm64);
2042 		inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
2043 		inti->mchk.mcic = s390int->parm64;
2044 		break;
2045 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2046 		inti->io.subchannel_id = s390int->parm >> 16;
2047 		inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
2048 		inti->io.io_int_parm = s390int->parm64 >> 32;
2049 		inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
2050 		break;
2051 	default:
2052 		kfree(inti);
2053 		return -EINVAL;
2054 	}
2055 	trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
2056 				 2);
2057 
2058 	rc = __inject_vm(kvm, inti);
2059 	if (rc)
2060 		kfree(inti);
2061 	return rc;
2062 }
2063 
kvm_s390_reinject_io_int(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)2064 int kvm_s390_reinject_io_int(struct kvm *kvm,
2065 			      struct kvm_s390_interrupt_info *inti)
2066 {
2067 	return __inject_vm(kvm, inti);
2068 }
2069 
s390int_to_s390irq(struct kvm_s390_interrupt * s390int,struct kvm_s390_irq * irq)2070 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
2071 		       struct kvm_s390_irq *irq)
2072 {
2073 	irq->type = s390int->type;
2074 	switch (irq->type) {
2075 	case KVM_S390_PROGRAM_INT:
2076 		if (s390int->parm & 0xffff0000)
2077 			return -EINVAL;
2078 		irq->u.pgm.code = s390int->parm;
2079 		break;
2080 	case KVM_S390_SIGP_SET_PREFIX:
2081 		irq->u.prefix.address = s390int->parm;
2082 		break;
2083 	case KVM_S390_SIGP_STOP:
2084 		irq->u.stop.flags = s390int->parm;
2085 		break;
2086 	case KVM_S390_INT_EXTERNAL_CALL:
2087 		if (s390int->parm & 0xffff0000)
2088 			return -EINVAL;
2089 		irq->u.extcall.code = s390int->parm;
2090 		break;
2091 	case KVM_S390_INT_EMERGENCY:
2092 		if (s390int->parm & 0xffff0000)
2093 			return -EINVAL;
2094 		irq->u.emerg.code = s390int->parm;
2095 		break;
2096 	case KVM_S390_MCHK:
2097 		irq->u.mchk.mcic = s390int->parm64;
2098 		break;
2099 	case KVM_S390_INT_PFAULT_INIT:
2100 		irq->u.ext.ext_params = s390int->parm;
2101 		irq->u.ext.ext_params2 = s390int->parm64;
2102 		break;
2103 	case KVM_S390_RESTART:
2104 	case KVM_S390_INT_CLOCK_COMP:
2105 	case KVM_S390_INT_CPU_TIMER:
2106 		break;
2107 	default:
2108 		return -EINVAL;
2109 	}
2110 	return 0;
2111 }
2112 
kvm_s390_is_stop_irq_pending(struct kvm_vcpu * vcpu)2113 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
2114 {
2115 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2116 
2117 	return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2118 }
2119 
kvm_s390_is_restart_irq_pending(struct kvm_vcpu * vcpu)2120 int kvm_s390_is_restart_irq_pending(struct kvm_vcpu *vcpu)
2121 {
2122 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2123 
2124 	return test_bit(IRQ_PEND_RESTART, &li->pending_irqs);
2125 }
2126 
kvm_s390_clear_stop_irq(struct kvm_vcpu * vcpu)2127 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
2128 {
2129 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2130 
2131 	spin_lock(&li->lock);
2132 	li->irq.stop.flags = 0;
2133 	clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2134 	spin_unlock(&li->lock);
2135 }
2136 
do_inject_vcpu(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)2137 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2138 {
2139 	int rc;
2140 
2141 	switch (irq->type) {
2142 	case KVM_S390_PROGRAM_INT:
2143 		rc = __inject_prog(vcpu, irq);
2144 		break;
2145 	case KVM_S390_SIGP_SET_PREFIX:
2146 		rc = __inject_set_prefix(vcpu, irq);
2147 		break;
2148 	case KVM_S390_SIGP_STOP:
2149 		rc = __inject_sigp_stop(vcpu, irq);
2150 		break;
2151 	case KVM_S390_RESTART:
2152 		rc = __inject_sigp_restart(vcpu);
2153 		break;
2154 	case KVM_S390_INT_CLOCK_COMP:
2155 		rc = __inject_ckc(vcpu);
2156 		break;
2157 	case KVM_S390_INT_CPU_TIMER:
2158 		rc = __inject_cpu_timer(vcpu);
2159 		break;
2160 	case KVM_S390_INT_EXTERNAL_CALL:
2161 		rc = __inject_extcall(vcpu, irq);
2162 		break;
2163 	case KVM_S390_INT_EMERGENCY:
2164 		rc = __inject_sigp_emergency(vcpu, irq);
2165 		break;
2166 	case KVM_S390_MCHK:
2167 		rc = __inject_mchk(vcpu, irq);
2168 		break;
2169 	case KVM_S390_INT_PFAULT_INIT:
2170 		rc = __inject_pfault_init(vcpu, irq);
2171 		break;
2172 	case KVM_S390_INT_VIRTIO:
2173 	case KVM_S390_INT_SERVICE:
2174 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2175 	default:
2176 		rc = -EINVAL;
2177 	}
2178 
2179 	return rc;
2180 }
2181 
kvm_s390_inject_vcpu(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)2182 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2183 {
2184 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2185 	int rc;
2186 
2187 	spin_lock(&li->lock);
2188 	rc = do_inject_vcpu(vcpu, irq);
2189 	spin_unlock(&li->lock);
2190 	if (!rc)
2191 		kvm_s390_vcpu_wakeup(vcpu);
2192 	return rc;
2193 }
2194 
clear_irq_list(struct list_head * _list)2195 static inline void clear_irq_list(struct list_head *_list)
2196 {
2197 	struct kvm_s390_interrupt_info *inti, *n;
2198 
2199 	list_for_each_entry_safe(inti, n, _list, list) {
2200 		list_del(&inti->list);
2201 		kfree(inti);
2202 	}
2203 }
2204 
inti_to_irq(struct kvm_s390_interrupt_info * inti,struct kvm_s390_irq * irq)2205 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2206 		       struct kvm_s390_irq *irq)
2207 {
2208 	irq->type = inti->type;
2209 	switch (inti->type) {
2210 	case KVM_S390_INT_PFAULT_INIT:
2211 	case KVM_S390_INT_PFAULT_DONE:
2212 	case KVM_S390_INT_VIRTIO:
2213 		irq->u.ext = inti->ext;
2214 		break;
2215 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2216 		irq->u.io = inti->io;
2217 		break;
2218 	}
2219 }
2220 
kvm_s390_clear_float_irqs(struct kvm * kvm)2221 void kvm_s390_clear_float_irqs(struct kvm *kvm)
2222 {
2223 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2224 	int i;
2225 
2226 	mutex_lock(&kvm->lock);
2227 	if (!kvm_s390_pv_is_protected(kvm))
2228 		fi->masked_irqs = 0;
2229 	mutex_unlock(&kvm->lock);
2230 	spin_lock(&fi->lock);
2231 	fi->pending_irqs = 0;
2232 	memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2233 	memset(&fi->mchk, 0, sizeof(fi->mchk));
2234 	for (i = 0; i < FIRQ_LIST_COUNT; i++)
2235 		clear_irq_list(&fi->lists[i]);
2236 	for (i = 0; i < FIRQ_MAX_COUNT; i++)
2237 		fi->counters[i] = 0;
2238 	spin_unlock(&fi->lock);
2239 	kvm_s390_gisa_clear(kvm);
2240 };
2241 
get_all_floating_irqs(struct kvm * kvm,u8 __user * usrbuf,u64 len)2242 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2243 {
2244 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2245 	struct kvm_s390_interrupt_info *inti;
2246 	struct kvm_s390_float_interrupt *fi;
2247 	struct kvm_s390_irq *buf;
2248 	struct kvm_s390_irq *irq;
2249 	int max_irqs;
2250 	int ret = 0;
2251 	int n = 0;
2252 	int i;
2253 
2254 	if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2255 		return -EINVAL;
2256 
2257 	/*
2258 	 * We are already using -ENOMEM to signal
2259 	 * userspace it may retry with a bigger buffer,
2260 	 * so we need to use something else for this case
2261 	 */
2262 	buf = vzalloc(len);
2263 	if (!buf)
2264 		return -ENOBUFS;
2265 
2266 	max_irqs = len / sizeof(struct kvm_s390_irq);
2267 
2268 	if (gi->origin && gisa_get_ipm(gi->origin)) {
2269 		for (i = 0; i <= MAX_ISC; i++) {
2270 			if (n == max_irqs) {
2271 				/* signal userspace to try again */
2272 				ret = -ENOMEM;
2273 				goto out_nolock;
2274 			}
2275 			if (gisa_tac_ipm_gisc(gi->origin, i)) {
2276 				irq = (struct kvm_s390_irq *) &buf[n];
2277 				irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2278 				irq->u.io.io_int_word = isc_to_int_word(i);
2279 				n++;
2280 			}
2281 		}
2282 	}
2283 	fi = &kvm->arch.float_int;
2284 	spin_lock(&fi->lock);
2285 	for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2286 		list_for_each_entry(inti, &fi->lists[i], list) {
2287 			if (n == max_irqs) {
2288 				/* signal userspace to try again */
2289 				ret = -ENOMEM;
2290 				goto out;
2291 			}
2292 			inti_to_irq(inti, &buf[n]);
2293 			n++;
2294 		}
2295 	}
2296 	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) ||
2297 	    test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) {
2298 		if (n == max_irqs) {
2299 			/* signal userspace to try again */
2300 			ret = -ENOMEM;
2301 			goto out;
2302 		}
2303 		irq = (struct kvm_s390_irq *) &buf[n];
2304 		irq->type = KVM_S390_INT_SERVICE;
2305 		irq->u.ext = fi->srv_signal;
2306 		n++;
2307 	}
2308 	if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2309 		if (n == max_irqs) {
2310 				/* signal userspace to try again */
2311 				ret = -ENOMEM;
2312 				goto out;
2313 		}
2314 		irq = (struct kvm_s390_irq *) &buf[n];
2315 		irq->type = KVM_S390_MCHK;
2316 		irq->u.mchk = fi->mchk;
2317 		n++;
2318 }
2319 
2320 out:
2321 	spin_unlock(&fi->lock);
2322 out_nolock:
2323 	if (!ret && n > 0) {
2324 		if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2325 			ret = -EFAULT;
2326 	}
2327 	vfree(buf);
2328 
2329 	return ret < 0 ? ret : n;
2330 }
2331 
flic_ais_mode_get_all(struct kvm * kvm,struct kvm_device_attr * attr)2332 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2333 {
2334 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2335 	struct kvm_s390_ais_all ais;
2336 
2337 	if (attr->attr < sizeof(ais))
2338 		return -EINVAL;
2339 
2340 	if (!test_kvm_facility(kvm, 72))
2341 		return -EOPNOTSUPP;
2342 
2343 	mutex_lock(&fi->ais_lock);
2344 	ais.simm = fi->simm;
2345 	ais.nimm = fi->nimm;
2346 	mutex_unlock(&fi->ais_lock);
2347 
2348 	if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2349 		return -EFAULT;
2350 
2351 	return 0;
2352 }
2353 
flic_get_attr(struct kvm_device * dev,struct kvm_device_attr * attr)2354 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2355 {
2356 	int r;
2357 
2358 	switch (attr->group) {
2359 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2360 		r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2361 					  attr->attr);
2362 		break;
2363 	case KVM_DEV_FLIC_AISM_ALL:
2364 		r = flic_ais_mode_get_all(dev->kvm, attr);
2365 		break;
2366 	default:
2367 		r = -EINVAL;
2368 	}
2369 
2370 	return r;
2371 }
2372 
copy_irq_from_user(struct kvm_s390_interrupt_info * inti,u64 addr)2373 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2374 				     u64 addr)
2375 {
2376 	struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2377 	void *target = NULL;
2378 	void __user *source;
2379 	u64 size;
2380 
2381 	if (get_user(inti->type, (u64 __user *)addr))
2382 		return -EFAULT;
2383 
2384 	switch (inti->type) {
2385 	case KVM_S390_INT_PFAULT_INIT:
2386 	case KVM_S390_INT_PFAULT_DONE:
2387 	case KVM_S390_INT_VIRTIO:
2388 	case KVM_S390_INT_SERVICE:
2389 		target = (void *) &inti->ext;
2390 		source = &uptr->u.ext;
2391 		size = sizeof(inti->ext);
2392 		break;
2393 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2394 		target = (void *) &inti->io;
2395 		source = &uptr->u.io;
2396 		size = sizeof(inti->io);
2397 		break;
2398 	case KVM_S390_MCHK:
2399 		target = (void *) &inti->mchk;
2400 		source = &uptr->u.mchk;
2401 		size = sizeof(inti->mchk);
2402 		break;
2403 	default:
2404 		return -EINVAL;
2405 	}
2406 
2407 	if (copy_from_user(target, source, size))
2408 		return -EFAULT;
2409 
2410 	return 0;
2411 }
2412 
enqueue_floating_irq(struct kvm_device * dev,struct kvm_device_attr * attr)2413 static int enqueue_floating_irq(struct kvm_device *dev,
2414 				struct kvm_device_attr *attr)
2415 {
2416 	struct kvm_s390_interrupt_info *inti = NULL;
2417 	int r = 0;
2418 	int len = attr->attr;
2419 
2420 	if (len % sizeof(struct kvm_s390_irq) != 0)
2421 		return -EINVAL;
2422 	else if (len > KVM_S390_FLIC_MAX_BUFFER)
2423 		return -EINVAL;
2424 
2425 	while (len >= sizeof(struct kvm_s390_irq)) {
2426 		inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2427 		if (!inti)
2428 			return -ENOMEM;
2429 
2430 		r = copy_irq_from_user(inti, attr->addr);
2431 		if (r) {
2432 			kfree(inti);
2433 			return r;
2434 		}
2435 		r = __inject_vm(dev->kvm, inti);
2436 		if (r) {
2437 			kfree(inti);
2438 			return r;
2439 		}
2440 		len -= sizeof(struct kvm_s390_irq);
2441 		attr->addr += sizeof(struct kvm_s390_irq);
2442 	}
2443 
2444 	return r;
2445 }
2446 
get_io_adapter(struct kvm * kvm,unsigned int id)2447 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2448 {
2449 	if (id >= MAX_S390_IO_ADAPTERS)
2450 		return NULL;
2451 	id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2452 	return kvm->arch.adapters[id];
2453 }
2454 
register_io_adapter(struct kvm_device * dev,struct kvm_device_attr * attr)2455 static int register_io_adapter(struct kvm_device *dev,
2456 			       struct kvm_device_attr *attr)
2457 {
2458 	struct s390_io_adapter *adapter;
2459 	struct kvm_s390_io_adapter adapter_info;
2460 
2461 	if (copy_from_user(&adapter_info,
2462 			   (void __user *)attr->addr, sizeof(adapter_info)))
2463 		return -EFAULT;
2464 
2465 	if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2466 		return -EINVAL;
2467 
2468 	adapter_info.id = array_index_nospec(adapter_info.id,
2469 					     MAX_S390_IO_ADAPTERS);
2470 
2471 	if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2472 		return -EINVAL;
2473 
2474 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL_ACCOUNT);
2475 	if (!adapter)
2476 		return -ENOMEM;
2477 
2478 	adapter->id = adapter_info.id;
2479 	adapter->isc = adapter_info.isc;
2480 	adapter->maskable = adapter_info.maskable;
2481 	adapter->masked = false;
2482 	adapter->swap = adapter_info.swap;
2483 	adapter->suppressible = (adapter_info.flags) &
2484 				KVM_S390_ADAPTER_SUPPRESSIBLE;
2485 	dev->kvm->arch.adapters[adapter->id] = adapter;
2486 
2487 	return 0;
2488 }
2489 
kvm_s390_mask_adapter(struct kvm * kvm,unsigned int id,bool masked)2490 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2491 {
2492 	int ret;
2493 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2494 
2495 	if (!adapter || !adapter->maskable)
2496 		return -EINVAL;
2497 	ret = adapter->masked;
2498 	adapter->masked = masked;
2499 	return ret;
2500 }
2501 
kvm_s390_destroy_adapters(struct kvm * kvm)2502 void kvm_s390_destroy_adapters(struct kvm *kvm)
2503 {
2504 	int i;
2505 
2506 	for (i = 0; i < MAX_S390_IO_ADAPTERS; i++)
2507 		kfree(kvm->arch.adapters[i]);
2508 }
2509 
modify_io_adapter(struct kvm_device * dev,struct kvm_device_attr * attr)2510 static int modify_io_adapter(struct kvm_device *dev,
2511 			     struct kvm_device_attr *attr)
2512 {
2513 	struct kvm_s390_io_adapter_req req;
2514 	struct s390_io_adapter *adapter;
2515 	int ret;
2516 
2517 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2518 		return -EFAULT;
2519 
2520 	adapter = get_io_adapter(dev->kvm, req.id);
2521 	if (!adapter)
2522 		return -EINVAL;
2523 	switch (req.type) {
2524 	case KVM_S390_IO_ADAPTER_MASK:
2525 		ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2526 		if (ret > 0)
2527 			ret = 0;
2528 		break;
2529 	/*
2530 	 * The following operations are no longer needed and therefore no-ops.
2531 	 * The gpa to hva translation is done when an IRQ route is set up. The
2532 	 * set_irq code uses get_user_pages_remote() to do the actual write.
2533 	 */
2534 	case KVM_S390_IO_ADAPTER_MAP:
2535 	case KVM_S390_IO_ADAPTER_UNMAP:
2536 		ret = 0;
2537 		break;
2538 	default:
2539 		ret = -EINVAL;
2540 	}
2541 
2542 	return ret;
2543 }
2544 
clear_io_irq(struct kvm * kvm,struct kvm_device_attr * attr)2545 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2546 
2547 {
2548 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2549 	u32 schid;
2550 
2551 	if (attr->flags)
2552 		return -EINVAL;
2553 	if (attr->attr != sizeof(schid))
2554 		return -EINVAL;
2555 	if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2556 		return -EFAULT;
2557 	if (!schid)
2558 		return -EINVAL;
2559 	kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2560 	/*
2561 	 * If userspace is conforming to the architecture, we can have at most
2562 	 * one pending I/O interrupt per subchannel, so this is effectively a
2563 	 * clear all.
2564 	 */
2565 	return 0;
2566 }
2567 
modify_ais_mode(struct kvm * kvm,struct kvm_device_attr * attr)2568 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2569 {
2570 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2571 	struct kvm_s390_ais_req req;
2572 	int ret = 0;
2573 
2574 	if (!test_kvm_facility(kvm, 72))
2575 		return -EOPNOTSUPP;
2576 
2577 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2578 		return -EFAULT;
2579 
2580 	if (req.isc > MAX_ISC)
2581 		return -EINVAL;
2582 
2583 	trace_kvm_s390_modify_ais_mode(req.isc,
2584 				       (fi->simm & AIS_MODE_MASK(req.isc)) ?
2585 				       (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2586 				       2 : KVM_S390_AIS_MODE_SINGLE :
2587 				       KVM_S390_AIS_MODE_ALL, req.mode);
2588 
2589 	mutex_lock(&fi->ais_lock);
2590 	switch (req.mode) {
2591 	case KVM_S390_AIS_MODE_ALL:
2592 		fi->simm &= ~AIS_MODE_MASK(req.isc);
2593 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2594 		break;
2595 	case KVM_S390_AIS_MODE_SINGLE:
2596 		fi->simm |= AIS_MODE_MASK(req.isc);
2597 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2598 		break;
2599 	default:
2600 		ret = -EINVAL;
2601 	}
2602 	mutex_unlock(&fi->ais_lock);
2603 
2604 	return ret;
2605 }
2606 
kvm_s390_inject_airq(struct kvm * kvm,struct s390_io_adapter * adapter)2607 static int kvm_s390_inject_airq(struct kvm *kvm,
2608 				struct s390_io_adapter *adapter)
2609 {
2610 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2611 	struct kvm_s390_interrupt s390int = {
2612 		.type = KVM_S390_INT_IO(1, 0, 0, 0),
2613 		.parm = 0,
2614 		.parm64 = isc_to_int_word(adapter->isc),
2615 	};
2616 	int ret = 0;
2617 
2618 	if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2619 		return kvm_s390_inject_vm(kvm, &s390int);
2620 
2621 	mutex_lock(&fi->ais_lock);
2622 	if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2623 		trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2624 		goto out;
2625 	}
2626 
2627 	ret = kvm_s390_inject_vm(kvm, &s390int);
2628 	if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2629 		fi->nimm |= AIS_MODE_MASK(adapter->isc);
2630 		trace_kvm_s390_modify_ais_mode(adapter->isc,
2631 					       KVM_S390_AIS_MODE_SINGLE, 2);
2632 	}
2633 out:
2634 	mutex_unlock(&fi->ais_lock);
2635 	return ret;
2636 }
2637 
flic_inject_airq(struct kvm * kvm,struct kvm_device_attr * attr)2638 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2639 {
2640 	unsigned int id = attr->attr;
2641 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2642 
2643 	if (!adapter)
2644 		return -EINVAL;
2645 
2646 	return kvm_s390_inject_airq(kvm, adapter);
2647 }
2648 
flic_ais_mode_set_all(struct kvm * kvm,struct kvm_device_attr * attr)2649 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2650 {
2651 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2652 	struct kvm_s390_ais_all ais;
2653 
2654 	if (!test_kvm_facility(kvm, 72))
2655 		return -EOPNOTSUPP;
2656 
2657 	if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2658 		return -EFAULT;
2659 
2660 	mutex_lock(&fi->ais_lock);
2661 	fi->simm = ais.simm;
2662 	fi->nimm = ais.nimm;
2663 	mutex_unlock(&fi->ais_lock);
2664 
2665 	return 0;
2666 }
2667 
flic_set_attr(struct kvm_device * dev,struct kvm_device_attr * attr)2668 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2669 {
2670 	int r = 0;
2671 	unsigned long i;
2672 	struct kvm_vcpu *vcpu;
2673 
2674 	switch (attr->group) {
2675 	case KVM_DEV_FLIC_ENQUEUE:
2676 		r = enqueue_floating_irq(dev, attr);
2677 		break;
2678 	case KVM_DEV_FLIC_CLEAR_IRQS:
2679 		kvm_s390_clear_float_irqs(dev->kvm);
2680 		break;
2681 	case KVM_DEV_FLIC_APF_ENABLE:
2682 		if (kvm_is_ucontrol(dev->kvm))
2683 			return -EINVAL;
2684 		dev->kvm->arch.gmap->pfault_enabled = 1;
2685 		break;
2686 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2687 		if (kvm_is_ucontrol(dev->kvm))
2688 			return -EINVAL;
2689 		dev->kvm->arch.gmap->pfault_enabled = 0;
2690 		/*
2691 		 * Make sure no async faults are in transition when
2692 		 * clearing the queues. So we don't need to worry
2693 		 * about late coming workers.
2694 		 */
2695 		synchronize_srcu(&dev->kvm->srcu);
2696 		kvm_for_each_vcpu(i, vcpu, dev->kvm)
2697 			kvm_clear_async_pf_completion_queue(vcpu);
2698 		break;
2699 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2700 		r = register_io_adapter(dev, attr);
2701 		break;
2702 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2703 		r = modify_io_adapter(dev, attr);
2704 		break;
2705 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2706 		r = clear_io_irq(dev->kvm, attr);
2707 		break;
2708 	case KVM_DEV_FLIC_AISM:
2709 		r = modify_ais_mode(dev->kvm, attr);
2710 		break;
2711 	case KVM_DEV_FLIC_AIRQ_INJECT:
2712 		r = flic_inject_airq(dev->kvm, attr);
2713 		break;
2714 	case KVM_DEV_FLIC_AISM_ALL:
2715 		r = flic_ais_mode_set_all(dev->kvm, attr);
2716 		break;
2717 	default:
2718 		r = -EINVAL;
2719 	}
2720 
2721 	return r;
2722 }
2723 
flic_has_attr(struct kvm_device * dev,struct kvm_device_attr * attr)2724 static int flic_has_attr(struct kvm_device *dev,
2725 			     struct kvm_device_attr *attr)
2726 {
2727 	switch (attr->group) {
2728 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2729 	case KVM_DEV_FLIC_ENQUEUE:
2730 	case KVM_DEV_FLIC_CLEAR_IRQS:
2731 	case KVM_DEV_FLIC_APF_ENABLE:
2732 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2733 	case KVM_DEV_FLIC_ADAPTER_REGISTER:
2734 	case KVM_DEV_FLIC_ADAPTER_MODIFY:
2735 	case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2736 	case KVM_DEV_FLIC_AISM:
2737 	case KVM_DEV_FLIC_AIRQ_INJECT:
2738 	case KVM_DEV_FLIC_AISM_ALL:
2739 		return 0;
2740 	}
2741 	return -ENXIO;
2742 }
2743 
flic_create(struct kvm_device * dev,u32 type)2744 static int flic_create(struct kvm_device *dev, u32 type)
2745 {
2746 	if (!dev)
2747 		return -EINVAL;
2748 	if (dev->kvm->arch.flic)
2749 		return -EINVAL;
2750 	dev->kvm->arch.flic = dev;
2751 	return 0;
2752 }
2753 
flic_destroy(struct kvm_device * dev)2754 static void flic_destroy(struct kvm_device *dev)
2755 {
2756 	dev->kvm->arch.flic = NULL;
2757 	kfree(dev);
2758 }
2759 
2760 /* s390 floating irq controller (flic) */
2761 struct kvm_device_ops kvm_flic_ops = {
2762 	.name = "kvm-flic",
2763 	.get_attr = flic_get_attr,
2764 	.set_attr = flic_set_attr,
2765 	.has_attr = flic_has_attr,
2766 	.create = flic_create,
2767 	.destroy = flic_destroy,
2768 };
2769 
get_ind_bit(__u64 addr,unsigned long bit_nr,bool swap)2770 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2771 {
2772 	unsigned long bit;
2773 
2774 	bit = bit_nr + (addr % PAGE_SIZE) * 8;
2775 
2776 	return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2777 }
2778 
get_map_page(struct kvm * kvm,u64 uaddr)2779 static struct page *get_map_page(struct kvm *kvm, u64 uaddr)
2780 {
2781 	struct page *page = NULL;
2782 
2783 	mmap_read_lock(kvm->mm);
2784 	get_user_pages_remote(kvm->mm, uaddr, 1, FOLL_WRITE,
2785 			      &page, NULL);
2786 	mmap_read_unlock(kvm->mm);
2787 	return page;
2788 }
2789 
adapter_indicators_set(struct kvm * kvm,struct s390_io_adapter * adapter,struct kvm_s390_adapter_int * adapter_int)2790 static int adapter_indicators_set(struct kvm *kvm,
2791 				  struct s390_io_adapter *adapter,
2792 				  struct kvm_s390_adapter_int *adapter_int)
2793 {
2794 	unsigned long bit;
2795 	int summary_set, idx;
2796 	struct page *ind_page, *summary_page;
2797 	void *map;
2798 
2799 	ind_page = get_map_page(kvm, adapter_int->ind_addr);
2800 	if (!ind_page)
2801 		return -1;
2802 	summary_page = get_map_page(kvm, adapter_int->summary_addr);
2803 	if (!summary_page) {
2804 		put_page(ind_page);
2805 		return -1;
2806 	}
2807 
2808 	idx = srcu_read_lock(&kvm->srcu);
2809 	map = page_address(ind_page);
2810 	bit = get_ind_bit(adapter_int->ind_addr,
2811 			  adapter_int->ind_offset, adapter->swap);
2812 	set_bit(bit, map);
2813 	mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT);
2814 	set_page_dirty_lock(ind_page);
2815 	map = page_address(summary_page);
2816 	bit = get_ind_bit(adapter_int->summary_addr,
2817 			  adapter_int->summary_offset, adapter->swap);
2818 	summary_set = test_and_set_bit(bit, map);
2819 	mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT);
2820 	set_page_dirty_lock(summary_page);
2821 	srcu_read_unlock(&kvm->srcu, idx);
2822 
2823 	put_page(ind_page);
2824 	put_page(summary_page);
2825 	return summary_set ? 0 : 1;
2826 }
2827 
2828 /*
2829  * < 0 - not injected due to error
2830  * = 0 - coalesced, summary indicator already active
2831  * > 0 - injected interrupt
2832  */
set_adapter_int(struct kvm_kernel_irq_routing_entry * e,struct kvm * kvm,int irq_source_id,int level,bool line_status)2833 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2834 			   struct kvm *kvm, int irq_source_id, int level,
2835 			   bool line_status)
2836 {
2837 	int ret;
2838 	struct s390_io_adapter *adapter;
2839 
2840 	/* We're only interested in the 0->1 transition. */
2841 	if (!level)
2842 		return 0;
2843 	adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2844 	if (!adapter)
2845 		return -1;
2846 	ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2847 	if ((ret > 0) && !adapter->masked) {
2848 		ret = kvm_s390_inject_airq(kvm, adapter);
2849 		if (ret == 0)
2850 			ret = 1;
2851 	}
2852 	return ret;
2853 }
2854 
2855 /*
2856  * Inject the machine check to the guest.
2857  */
kvm_s390_reinject_machine_check(struct kvm_vcpu * vcpu,struct mcck_volatile_info * mcck_info)2858 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2859 				     struct mcck_volatile_info *mcck_info)
2860 {
2861 	struct kvm_s390_interrupt_info inti;
2862 	struct kvm_s390_irq irq;
2863 	struct kvm_s390_mchk_info *mchk;
2864 	union mci mci;
2865 	__u64 cr14 = 0;         /* upper bits are not used */
2866 	int rc;
2867 
2868 	mci.val = mcck_info->mcic;
2869 	if (mci.sr)
2870 		cr14 |= CR14_RECOVERY_SUBMASK;
2871 	if (mci.dg)
2872 		cr14 |= CR14_DEGRADATION_SUBMASK;
2873 	if (mci.w)
2874 		cr14 |= CR14_WARNING_SUBMASK;
2875 
2876 	mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2877 	mchk->cr14 = cr14;
2878 	mchk->mcic = mcck_info->mcic;
2879 	mchk->ext_damage_code = mcck_info->ext_damage_code;
2880 	mchk->failing_storage_address = mcck_info->failing_storage_address;
2881 	if (mci.ck) {
2882 		/* Inject the floating machine check */
2883 		inti.type = KVM_S390_MCHK;
2884 		rc = __inject_vm(vcpu->kvm, &inti);
2885 	} else {
2886 		/* Inject the machine check to specified vcpu */
2887 		irq.type = KVM_S390_MCHK;
2888 		rc = kvm_s390_inject_vcpu(vcpu, &irq);
2889 	}
2890 	WARN_ON_ONCE(rc);
2891 }
2892 
kvm_set_routing_entry(struct kvm * kvm,struct kvm_kernel_irq_routing_entry * e,const struct kvm_irq_routing_entry * ue)2893 int kvm_set_routing_entry(struct kvm *kvm,
2894 			  struct kvm_kernel_irq_routing_entry *e,
2895 			  const struct kvm_irq_routing_entry *ue)
2896 {
2897 	u64 uaddr_s, uaddr_i;
2898 	int idx;
2899 
2900 	switch (ue->type) {
2901 	/* we store the userspace addresses instead of the guest addresses */
2902 	case KVM_IRQ_ROUTING_S390_ADAPTER:
2903 		if (kvm_is_ucontrol(kvm))
2904 			return -EINVAL;
2905 		e->set = set_adapter_int;
2906 
2907 		idx = srcu_read_lock(&kvm->srcu);
2908 		uaddr_s = gpa_to_hva(kvm, ue->u.adapter.summary_addr);
2909 		uaddr_i = gpa_to_hva(kvm, ue->u.adapter.ind_addr);
2910 		srcu_read_unlock(&kvm->srcu, idx);
2911 
2912 		if (kvm_is_error_hva(uaddr_s) || kvm_is_error_hva(uaddr_i))
2913 			return -EFAULT;
2914 		e->adapter.summary_addr = uaddr_s;
2915 		e->adapter.ind_addr = uaddr_i;
2916 		e->adapter.summary_offset = ue->u.adapter.summary_offset;
2917 		e->adapter.ind_offset = ue->u.adapter.ind_offset;
2918 		e->adapter.adapter_id = ue->u.adapter.adapter_id;
2919 		return 0;
2920 	default:
2921 		return -EINVAL;
2922 	}
2923 }
2924 
kvm_set_msi(struct kvm_kernel_irq_routing_entry * e,struct kvm * kvm,int irq_source_id,int level,bool line_status)2925 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2926 		int irq_source_id, int level, bool line_status)
2927 {
2928 	return -EINVAL;
2929 }
2930 
kvm_s390_set_irq_state(struct kvm_vcpu * vcpu,void __user * irqstate,int len)2931 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2932 {
2933 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2934 	struct kvm_s390_irq *buf;
2935 	int r = 0;
2936 	int n;
2937 
2938 	buf = vmalloc(len);
2939 	if (!buf)
2940 		return -ENOMEM;
2941 
2942 	if (copy_from_user((void *) buf, irqstate, len)) {
2943 		r = -EFAULT;
2944 		goto out_free;
2945 	}
2946 
2947 	/*
2948 	 * Don't allow setting the interrupt state
2949 	 * when there are already interrupts pending
2950 	 */
2951 	spin_lock(&li->lock);
2952 	if (li->pending_irqs) {
2953 		r = -EBUSY;
2954 		goto out_unlock;
2955 	}
2956 
2957 	for (n = 0; n < len / sizeof(*buf); n++) {
2958 		r = do_inject_vcpu(vcpu, &buf[n]);
2959 		if (r)
2960 			break;
2961 	}
2962 
2963 out_unlock:
2964 	spin_unlock(&li->lock);
2965 out_free:
2966 	vfree(buf);
2967 
2968 	return r;
2969 }
2970 
store_local_irq(struct kvm_s390_local_interrupt * li,struct kvm_s390_irq * irq,unsigned long irq_type)2971 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2972 			    struct kvm_s390_irq *irq,
2973 			    unsigned long irq_type)
2974 {
2975 	switch (irq_type) {
2976 	case IRQ_PEND_MCHK_EX:
2977 	case IRQ_PEND_MCHK_REP:
2978 		irq->type = KVM_S390_MCHK;
2979 		irq->u.mchk = li->irq.mchk;
2980 		break;
2981 	case IRQ_PEND_PROG:
2982 		irq->type = KVM_S390_PROGRAM_INT;
2983 		irq->u.pgm = li->irq.pgm;
2984 		break;
2985 	case IRQ_PEND_PFAULT_INIT:
2986 		irq->type = KVM_S390_INT_PFAULT_INIT;
2987 		irq->u.ext = li->irq.ext;
2988 		break;
2989 	case IRQ_PEND_EXT_EXTERNAL:
2990 		irq->type = KVM_S390_INT_EXTERNAL_CALL;
2991 		irq->u.extcall = li->irq.extcall;
2992 		break;
2993 	case IRQ_PEND_EXT_CLOCK_COMP:
2994 		irq->type = KVM_S390_INT_CLOCK_COMP;
2995 		break;
2996 	case IRQ_PEND_EXT_CPU_TIMER:
2997 		irq->type = KVM_S390_INT_CPU_TIMER;
2998 		break;
2999 	case IRQ_PEND_SIGP_STOP:
3000 		irq->type = KVM_S390_SIGP_STOP;
3001 		irq->u.stop = li->irq.stop;
3002 		break;
3003 	case IRQ_PEND_RESTART:
3004 		irq->type = KVM_S390_RESTART;
3005 		break;
3006 	case IRQ_PEND_SET_PREFIX:
3007 		irq->type = KVM_S390_SIGP_SET_PREFIX;
3008 		irq->u.prefix = li->irq.prefix;
3009 		break;
3010 	}
3011 }
3012 
kvm_s390_get_irq_state(struct kvm_vcpu * vcpu,__u8 __user * buf,int len)3013 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
3014 {
3015 	int scn;
3016 	DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
3017 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
3018 	unsigned long pending_irqs;
3019 	struct kvm_s390_irq irq;
3020 	unsigned long irq_type;
3021 	int cpuaddr;
3022 	int n = 0;
3023 
3024 	spin_lock(&li->lock);
3025 	pending_irqs = li->pending_irqs;
3026 	memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
3027 	       sizeof(sigp_emerg_pending));
3028 	spin_unlock(&li->lock);
3029 
3030 	for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
3031 		memset(&irq, 0, sizeof(irq));
3032 		if (irq_type == IRQ_PEND_EXT_EMERGENCY)
3033 			continue;
3034 		if (n + sizeof(irq) > len)
3035 			return -ENOBUFS;
3036 		store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
3037 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3038 			return -EFAULT;
3039 		n += sizeof(irq);
3040 	}
3041 
3042 	if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
3043 		for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
3044 			memset(&irq, 0, sizeof(irq));
3045 			if (n + sizeof(irq) > len)
3046 				return -ENOBUFS;
3047 			irq.type = KVM_S390_INT_EMERGENCY;
3048 			irq.u.emerg.code = cpuaddr;
3049 			if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3050 				return -EFAULT;
3051 			n += sizeof(irq);
3052 		}
3053 	}
3054 
3055 	if (sca_ext_call_pending(vcpu, &scn)) {
3056 		if (n + sizeof(irq) > len)
3057 			return -ENOBUFS;
3058 		memset(&irq, 0, sizeof(irq));
3059 		irq.type = KVM_S390_INT_EXTERNAL_CALL;
3060 		irq.u.extcall.code = scn;
3061 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3062 			return -EFAULT;
3063 		n += sizeof(irq);
3064 	}
3065 
3066 	return n;
3067 }
3068 
__airqs_kick_single_vcpu(struct kvm * kvm,u8 deliverable_mask)3069 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
3070 {
3071 	int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus);
3072 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3073 	struct kvm_vcpu *vcpu;
3074 	u8 vcpu_isc_mask;
3075 
3076 	for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) {
3077 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
3078 		if (psw_ioint_disabled(vcpu))
3079 			continue;
3080 		vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
3081 		if (deliverable_mask & vcpu_isc_mask) {
3082 			/* lately kicked but not yet running */
3083 			if (test_and_set_bit(vcpu_idx, gi->kicked_mask))
3084 				return;
3085 			kvm_s390_vcpu_wakeup(vcpu);
3086 			return;
3087 		}
3088 	}
3089 }
3090 
gisa_vcpu_kicker(struct hrtimer * timer)3091 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3092 {
3093 	struct kvm_s390_gisa_interrupt *gi =
3094 		container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3095 	struct kvm *kvm =
3096 		container_of(gi->origin, struct sie_page2, gisa)->kvm;
3097 	u8 pending_mask;
3098 
3099 	pending_mask = gisa_get_ipm_or_restore_iam(gi);
3100 	if (pending_mask) {
3101 		__airqs_kick_single_vcpu(kvm, pending_mask);
3102 		hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3103 		return HRTIMER_RESTART;
3104 	}
3105 
3106 	return HRTIMER_NORESTART;
3107 }
3108 
3109 #define NULL_GISA_ADDR 0x00000000UL
3110 #define NONE_GISA_ADDR 0x00000001UL
3111 #define GISA_ADDR_MASK 0xfffff000UL
3112 
process_gib_alert_list(void)3113 static void process_gib_alert_list(void)
3114 {
3115 	struct kvm_s390_gisa_interrupt *gi;
3116 	u32 final, gisa_phys, origin = 0UL;
3117 	struct kvm_s390_gisa *gisa;
3118 	struct kvm *kvm;
3119 
3120 	do {
3121 		/*
3122 		 * If the NONE_GISA_ADDR is still stored in the alert list
3123 		 * origin, we will leave the outer loop. No further GISA has
3124 		 * been added to the alert list by millicode while processing
3125 		 * the current alert list.
3126 		 */
3127 		final = (origin & NONE_GISA_ADDR);
3128 		/*
3129 		 * Cut off the alert list and store the NONE_GISA_ADDR in the
3130 		 * alert list origin to avoid further GAL interruptions.
3131 		 * A new alert list can be build up by millicode in parallel
3132 		 * for guests not in the yet cut-off alert list. When in the
3133 		 * final loop, store the NULL_GISA_ADDR instead. This will re-
3134 		 * enable GAL interruptions on the host again.
3135 		 */
3136 		origin = xchg(&gib->alert_list_origin,
3137 			      (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3138 		/*
3139 		 * Loop through the just cut-off alert list and start the
3140 		 * gisa timers to kick idle vcpus to consume the pending
3141 		 * interruptions asap.
3142 		 */
3143 		while (origin & GISA_ADDR_MASK) {
3144 			gisa_phys = origin;
3145 			gisa = phys_to_virt(gisa_phys);
3146 			origin = gisa->next_alert;
3147 			gisa->next_alert = gisa_phys;
3148 			kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3149 			gi = &kvm->arch.gisa_int;
3150 			if (hrtimer_active(&gi->timer))
3151 				hrtimer_cancel(&gi->timer);
3152 			hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3153 		}
3154 	} while (!final);
3155 
3156 }
3157 
kvm_s390_gisa_clear(struct kvm * kvm)3158 void kvm_s390_gisa_clear(struct kvm *kvm)
3159 {
3160 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3161 
3162 	if (!gi->origin)
3163 		return;
3164 	gisa_clear_ipm(gi->origin);
3165 	VM_EVENT(kvm, 3, "gisa 0x%p cleared", gi->origin);
3166 }
3167 
kvm_s390_gisa_init(struct kvm * kvm)3168 void kvm_s390_gisa_init(struct kvm *kvm)
3169 {
3170 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3171 
3172 	if (!css_general_characteristics.aiv)
3173 		return;
3174 	gi->origin = &kvm->arch.sie_page2->gisa;
3175 	gi->alert.mask = 0;
3176 	spin_lock_init(&gi->alert.ref_lock);
3177 	gi->expires = 50 * 1000; /* 50 usec */
3178 	hrtimer_setup(&gi->timer, gisa_vcpu_kicker, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3179 	memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3180 	gi->origin->next_alert = (u32)virt_to_phys(gi->origin);
3181 	VM_EVENT(kvm, 3, "gisa 0x%p initialized", gi->origin);
3182 }
3183 
kvm_s390_gisa_enable(struct kvm * kvm)3184 void kvm_s390_gisa_enable(struct kvm *kvm)
3185 {
3186 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3187 	struct kvm_vcpu *vcpu;
3188 	unsigned long i;
3189 	u32 gisa_desc;
3190 
3191 	if (gi->origin)
3192 		return;
3193 	kvm_s390_gisa_init(kvm);
3194 	gisa_desc = kvm_s390_get_gisa_desc(kvm);
3195 	if (!gisa_desc)
3196 		return;
3197 	kvm_for_each_vcpu(i, vcpu, kvm) {
3198 		mutex_lock(&vcpu->mutex);
3199 		vcpu->arch.sie_block->gd = gisa_desc;
3200 		vcpu->arch.sie_block->eca |= ECA_AIV;
3201 		VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3202 			   vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3203 		mutex_unlock(&vcpu->mutex);
3204 	}
3205 }
3206 
kvm_s390_gisa_destroy(struct kvm * kvm)3207 void kvm_s390_gisa_destroy(struct kvm *kvm)
3208 {
3209 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3210 	struct kvm_s390_gisa *gisa = gi->origin;
3211 
3212 	if (!gi->origin)
3213 		return;
3214 	WARN(gi->alert.mask != 0x00,
3215 	     "unexpected non zero alert.mask 0x%02x",
3216 	     gi->alert.mask);
3217 	gi->alert.mask = 0x00;
3218 	if (gisa_set_iam(gi->origin, gi->alert.mask))
3219 		process_gib_alert_list();
3220 	hrtimer_cancel(&gi->timer);
3221 	gi->origin = NULL;
3222 	VM_EVENT(kvm, 3, "gisa 0x%p destroyed", gisa);
3223 }
3224 
kvm_s390_gisa_disable(struct kvm * kvm)3225 void kvm_s390_gisa_disable(struct kvm *kvm)
3226 {
3227 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3228 	struct kvm_vcpu *vcpu;
3229 	unsigned long i;
3230 
3231 	if (!gi->origin)
3232 		return;
3233 	kvm_for_each_vcpu(i, vcpu, kvm) {
3234 		mutex_lock(&vcpu->mutex);
3235 		vcpu->arch.sie_block->eca &= ~ECA_AIV;
3236 		vcpu->arch.sie_block->gd = 0U;
3237 		mutex_unlock(&vcpu->mutex);
3238 		VCPU_EVENT(vcpu, 3, "AIV disabled for cpu %03u", vcpu->vcpu_id);
3239 	}
3240 	kvm_s390_gisa_destroy(kvm);
3241 }
3242 
3243 /**
3244  * kvm_s390_gisc_register - register a guest ISC
3245  *
3246  * @kvm:  the kernel vm to work with
3247  * @gisc: the guest interruption sub class to register
3248  *
3249  * The function extends the vm specific alert mask to use.
3250  * The effective IAM mask in the GISA is updated as well
3251  * in case the GISA is not part of the GIB alert list.
3252  * It will be updated latest when the IAM gets restored
3253  * by gisa_get_ipm_or_restore_iam().
3254  *
3255  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3256  *          has registered with the channel subsystem.
3257  *          -ENODEV in case the vm uses no GISA
3258  *          -ERANGE in case the guest ISC is invalid
3259  */
kvm_s390_gisc_register(struct kvm * kvm,u32 gisc)3260 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3261 {
3262 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3263 
3264 	if (!gi->origin)
3265 		return -ENODEV;
3266 	if (gisc > MAX_ISC)
3267 		return -ERANGE;
3268 
3269 	spin_lock(&gi->alert.ref_lock);
3270 	gi->alert.ref_count[gisc]++;
3271 	if (gi->alert.ref_count[gisc] == 1) {
3272 		gi->alert.mask |= 0x80 >> gisc;
3273 		gisa_set_iam(gi->origin, gi->alert.mask);
3274 	}
3275 	spin_unlock(&gi->alert.ref_lock);
3276 
3277 	return gib->nisc;
3278 }
3279 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3280 
3281 /**
3282  * kvm_s390_gisc_unregister - unregister a guest ISC
3283  *
3284  * @kvm:  the kernel vm to work with
3285  * @gisc: the guest interruption sub class to register
3286  *
3287  * The function reduces the vm specific alert mask to use.
3288  * The effective IAM mask in the GISA is updated as well
3289  * in case the GISA is not part of the GIB alert list.
3290  * It will be updated latest when the IAM gets restored
3291  * by gisa_get_ipm_or_restore_iam().
3292  *
3293  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3294  *          has registered with the channel subsystem.
3295  *          -ENODEV in case the vm uses no GISA
3296  *          -ERANGE in case the guest ISC is invalid
3297  *          -EINVAL in case the guest ISC is not registered
3298  */
kvm_s390_gisc_unregister(struct kvm * kvm,u32 gisc)3299 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3300 {
3301 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3302 	int rc = 0;
3303 
3304 	if (!gi->origin)
3305 		return -ENODEV;
3306 	if (gisc > MAX_ISC)
3307 		return -ERANGE;
3308 
3309 	spin_lock(&gi->alert.ref_lock);
3310 	if (gi->alert.ref_count[gisc] == 0) {
3311 		rc = -EINVAL;
3312 		goto out;
3313 	}
3314 	gi->alert.ref_count[gisc]--;
3315 	if (gi->alert.ref_count[gisc] == 0) {
3316 		gi->alert.mask &= ~(0x80 >> gisc);
3317 		gisa_set_iam(gi->origin, gi->alert.mask);
3318 	}
3319 out:
3320 	spin_unlock(&gi->alert.ref_lock);
3321 
3322 	return rc;
3323 }
3324 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3325 
aen_host_forward(unsigned long si)3326 static void aen_host_forward(unsigned long si)
3327 {
3328 	struct kvm_s390_gisa_interrupt *gi;
3329 	struct zpci_gaite *gaite;
3330 	struct kvm *kvm;
3331 
3332 	gaite = (struct zpci_gaite *)aift->gait +
3333 		(si * sizeof(struct zpci_gaite));
3334 	if (gaite->count == 0)
3335 		return;
3336 	if (gaite->aisb != 0)
3337 		set_bit_inv(gaite->aisbo, phys_to_virt(gaite->aisb));
3338 
3339 	kvm = kvm_s390_pci_si_to_kvm(aift, si);
3340 	if (!kvm)
3341 		return;
3342 	gi = &kvm->arch.gisa_int;
3343 
3344 	if (!(gi->origin->g1.simm & AIS_MODE_MASK(gaite->gisc)) ||
3345 	    !(gi->origin->g1.nimm & AIS_MODE_MASK(gaite->gisc))) {
3346 		gisa_set_ipm_gisc(gi->origin, gaite->gisc);
3347 		if (hrtimer_active(&gi->timer))
3348 			hrtimer_cancel(&gi->timer);
3349 		hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3350 		kvm->stat.aen_forward++;
3351 	}
3352 }
3353 
aen_process_gait(u8 isc)3354 static void aen_process_gait(u8 isc)
3355 {
3356 	bool found = false, first = true;
3357 	union zpci_sic_iib iib = {{0}};
3358 	unsigned long si, flags;
3359 
3360 	spin_lock_irqsave(&aift->gait_lock, flags);
3361 
3362 	if (!aift->gait) {
3363 		spin_unlock_irqrestore(&aift->gait_lock, flags);
3364 		return;
3365 	}
3366 
3367 	for (si = 0;;) {
3368 		/* Scan adapter summary indicator bit vector */
3369 		si = airq_iv_scan(aift->sbv, si, airq_iv_end(aift->sbv));
3370 		if (si == -1UL) {
3371 			if (first || found) {
3372 				/* Re-enable interrupts. */
3373 				zpci_set_irq_ctrl(SIC_IRQ_MODE_SINGLE, isc,
3374 						  &iib);
3375 				first = found = false;
3376 			} else {
3377 				/* Interrupts on and all bits processed */
3378 				break;
3379 			}
3380 			found = false;
3381 			si = 0;
3382 			/* Scan again after re-enabling interrupts */
3383 			continue;
3384 		}
3385 		found = true;
3386 		aen_host_forward(si);
3387 	}
3388 
3389 	spin_unlock_irqrestore(&aift->gait_lock, flags);
3390 }
3391 
gib_alert_irq_handler(struct airq_struct * airq,struct tpi_info * tpi_info)3392 static void gib_alert_irq_handler(struct airq_struct *airq,
3393 				  struct tpi_info *tpi_info)
3394 {
3395 	struct tpi_adapter_info *info = (struct tpi_adapter_info *)tpi_info;
3396 
3397 	inc_irq_stat(IRQIO_GAL);
3398 
3399 	if ((info->forward || info->error) &&
3400 	    IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3401 		aen_process_gait(info->isc);
3402 		if (info->aism != 0)
3403 			process_gib_alert_list();
3404 	} else {
3405 		process_gib_alert_list();
3406 	}
3407 }
3408 
3409 static struct airq_struct gib_alert_irq = {
3410 	.handler = gib_alert_irq_handler,
3411 };
3412 
kvm_s390_gib_destroy(void)3413 void kvm_s390_gib_destroy(void)
3414 {
3415 	if (!gib)
3416 		return;
3417 	if (kvm_s390_pci_interp_allowed() && aift) {
3418 		mutex_lock(&aift->aift_lock);
3419 		kvm_s390_pci_aen_exit();
3420 		mutex_unlock(&aift->aift_lock);
3421 	}
3422 	chsc_sgib(0);
3423 	unregister_adapter_interrupt(&gib_alert_irq);
3424 	free_page((unsigned long)gib);
3425 	gib = NULL;
3426 }
3427 
kvm_s390_gib_init(u8 nisc)3428 int __init kvm_s390_gib_init(u8 nisc)
3429 {
3430 	u32 gib_origin;
3431 	int rc = 0;
3432 
3433 	if (!css_general_characteristics.aiv) {
3434 		KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3435 		goto out;
3436 	}
3437 
3438 	gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3439 	if (!gib) {
3440 		rc = -ENOMEM;
3441 		goto out;
3442 	}
3443 
3444 	gib_alert_irq.isc = nisc;
3445 	if (register_adapter_interrupt(&gib_alert_irq)) {
3446 		pr_err("Registering the GIB alert interruption handler failed\n");
3447 		rc = -EIO;
3448 		goto out_free_gib;
3449 	}
3450 	/* adapter interrupts used for AP (applicable here) don't use the LSI */
3451 	*gib_alert_irq.lsi_ptr = 0xff;
3452 
3453 	gib->nisc = nisc;
3454 	gib_origin = virt_to_phys(gib);
3455 	if (chsc_sgib(gib_origin)) {
3456 		pr_err("Associating the GIB with the AIV facility failed\n");
3457 		free_page((unsigned long)gib);
3458 		gib = NULL;
3459 		rc = -EIO;
3460 		goto out_unreg_gal;
3461 	}
3462 
3463 	if (kvm_s390_pci_interp_allowed()) {
3464 		if (kvm_s390_pci_aen_init(nisc)) {
3465 			pr_err("Initializing AEN for PCI failed\n");
3466 			rc = -EIO;
3467 			goto out_unreg_gal;
3468 		}
3469 	}
3470 
3471 	KVM_EVENT(3, "gib 0x%p (nisc=%d) initialized", gib, gib->nisc);
3472 	goto out;
3473 
3474 out_unreg_gal:
3475 	unregister_adapter_interrupt(&gib_alert_irq);
3476 out_free_gib:
3477 	free_page((unsigned long)gib);
3478 	gib = NULL;
3479 out:
3480 	return rc;
3481 }
3482