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/interrupt.h>
14 #include <linux/kvm_host.h>
15 #include <linux/hrtimer.h>
16 #include <linux/mmu_context.h>
17 #include <linux/nospec.h>
18 #include <linux/signal.h>
19 #include <linux/slab.h>
20 #include <linux/bitmap.h>
21 #include <linux/vmalloc.h>
22 #include <asm/asm-offsets.h>
23 #include <asm/dis.h>
24 #include <linux/uaccess.h>
25 #include <asm/sclp.h>
26 #include <asm/isc.h>
27 #include <asm/gmap.h>
28 #include <asm/switch_to.h>
29 #include <asm/nmi.h>
30 #include <asm/airq.h>
31 #include <asm/tpi.h>
32 #include "kvm-s390.h"
33 #include "gaccess.h"
34 #include "trace-s390.h"
35 #include "pci.h"
36 
37 #define PFAULT_INIT 0x0600
38 #define PFAULT_DONE 0x0680
39 #define VIRTIO_PARAM 0x0d00
40 
41 static struct kvm_s390_gib *gib;
42 
43 /* handle external calls via sigp interpretation facility */
sca_ext_call_pending(struct kvm_vcpu * vcpu,int * src_id)44 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
45 {
46 	int c, scn;
47 
48 	if (!kvm_s390_test_cpuflags(vcpu, CPUSTAT_ECALL_PEND))
49 		return 0;
50 
51 	BUG_ON(!kvm_s390_use_sca_entries());
52 	read_lock(&vcpu->kvm->arch.sca_lock);
53 	if (vcpu->kvm->arch.use_esca) {
54 		struct esca_block *sca = vcpu->kvm->arch.sca;
55 		union esca_sigp_ctrl sigp_ctrl =
56 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
57 
58 		c = sigp_ctrl.c;
59 		scn = sigp_ctrl.scn;
60 	} else {
61 		struct bsca_block *sca = vcpu->kvm->arch.sca;
62 		union bsca_sigp_ctrl sigp_ctrl =
63 			sca->cpu[vcpu->vcpu_id].sigp_ctrl;
64 
65 		c = sigp_ctrl.c;
66 		scn = sigp_ctrl.scn;
67 	}
68 	read_unlock(&vcpu->kvm->arch.sca_lock);
69 
70 	if (src_id)
71 		*src_id = scn;
72 
73 	return c;
74 }
75 
sca_inject_ext_call(struct kvm_vcpu * vcpu,int src_id)76 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
77 {
78 	int expect, rc;
79 
80 	BUG_ON(!kvm_s390_use_sca_entries());
81 	read_lock(&vcpu->kvm->arch.sca_lock);
82 	if (vcpu->kvm->arch.use_esca) {
83 		struct esca_block *sca = vcpu->kvm->arch.sca;
84 		union esca_sigp_ctrl *sigp_ctrl =
85 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
86 		union esca_sigp_ctrl new_val = {0}, old_val;
87 
88 		old_val = READ_ONCE(*sigp_ctrl);
89 		new_val.scn = src_id;
90 		new_val.c = 1;
91 		old_val.c = 0;
92 
93 		expect = old_val.value;
94 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
95 	} else {
96 		struct bsca_block *sca = vcpu->kvm->arch.sca;
97 		union bsca_sigp_ctrl *sigp_ctrl =
98 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
99 		union bsca_sigp_ctrl new_val = {0}, old_val;
100 
101 		old_val = READ_ONCE(*sigp_ctrl);
102 		new_val.scn = src_id;
103 		new_val.c = 1;
104 		old_val.c = 0;
105 
106 		expect = old_val.value;
107 		rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
108 	}
109 	read_unlock(&vcpu->kvm->arch.sca_lock);
110 
111 	if (rc != expect) {
112 		/* another external call is pending */
113 		return -EBUSY;
114 	}
115 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
116 	return 0;
117 }
118 
sca_clear_ext_call(struct kvm_vcpu * vcpu)119 static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
120 {
121 	int rc, expect;
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 		union esca_sigp_ctrl old;
132 
133 		old = READ_ONCE(*sigp_ctrl);
134 		expect = old.value;
135 		rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
136 	} else {
137 		struct bsca_block *sca = vcpu->kvm->arch.sca;
138 		union bsca_sigp_ctrl *sigp_ctrl =
139 			&(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
140 		union bsca_sigp_ctrl old;
141 
142 		old = READ_ONCE(*sigp_ctrl);
143 		expect = old.value;
144 		rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
145 	}
146 	read_unlock(&vcpu->kvm->arch.sca_lock);
147 	WARN_ON(rc != expect); /* cannot clear? */
148 }
149 
psw_extint_disabled(struct kvm_vcpu * vcpu)150 int psw_extint_disabled(struct kvm_vcpu *vcpu)
151 {
152 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
153 }
154 
psw_ioint_disabled(struct kvm_vcpu * vcpu)155 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
156 {
157 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
158 }
159 
psw_mchk_disabled(struct kvm_vcpu * vcpu)160 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
161 {
162 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
163 }
164 
psw_interrupts_disabled(struct kvm_vcpu * vcpu)165 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
166 {
167 	return psw_extint_disabled(vcpu) &&
168 	       psw_ioint_disabled(vcpu) &&
169 	       psw_mchk_disabled(vcpu);
170 }
171 
ckc_interrupts_enabled(struct kvm_vcpu * vcpu)172 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
173 {
174 	if (psw_extint_disabled(vcpu) ||
175 	    !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
176 		return 0;
177 	if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
178 		/* No timer interrupts when single stepping */
179 		return 0;
180 	return 1;
181 }
182 
ckc_irq_pending(struct kvm_vcpu * vcpu)183 static int ckc_irq_pending(struct kvm_vcpu *vcpu)
184 {
185 	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
186 	const u64 ckc = vcpu->arch.sie_block->ckc;
187 
188 	if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
189 		if ((s64)ckc >= (s64)now)
190 			return 0;
191 	} else if (ckc >= now) {
192 		return 0;
193 	}
194 	return ckc_interrupts_enabled(vcpu);
195 }
196 
cpu_timer_interrupts_enabled(struct kvm_vcpu * vcpu)197 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
198 {
199 	return !psw_extint_disabled(vcpu) &&
200 	       (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK);
201 }
202 
cpu_timer_irq_pending(struct kvm_vcpu * vcpu)203 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
204 {
205 	if (!cpu_timer_interrupts_enabled(vcpu))
206 		return 0;
207 	return kvm_s390_get_cpu_timer(vcpu) >> 63;
208 }
209 
isc_to_isc_bits(int isc)210 static uint64_t isc_to_isc_bits(int isc)
211 {
212 	return (0x80 >> isc) << 24;
213 }
214 
isc_to_int_word(u8 isc)215 static inline u32 isc_to_int_word(u8 isc)
216 {
217 	return ((u32)isc << 27) | 0x80000000;
218 }
219 
int_word_to_isc(u32 int_word)220 static inline u8 int_word_to_isc(u32 int_word)
221 {
222 	return (int_word & 0x38000000) >> 27;
223 }
224 
225 /*
226  * To use atomic bitmap functions, we have to provide a bitmap address
227  * that is u64 aligned. However, the ipm might be u32 aligned.
228  * Therefore, we logically start the bitmap at the very beginning of the
229  * struct and fixup the bit number.
230  */
231 #define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE)
232 
233 /**
234  * gisa_set_iam - change the GISA interruption alert mask
235  *
236  * @gisa: gisa to operate on
237  * @iam: new IAM value to use
238  *
239  * Change the IAM atomically with the next alert address and the IPM
240  * of the GISA if the GISA is not part of the GIB alert list. All three
241  * fields are located in the first long word of the GISA.
242  *
243  * Returns: 0 on success
244  *          -EBUSY in case the gisa is part of the alert list
245  */
gisa_set_iam(struct kvm_s390_gisa * gisa,u8 iam)246 static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam)
247 {
248 	u64 word, _word;
249 
250 	do {
251 		word = READ_ONCE(gisa->u64.word[0]);
252 		if ((u64)gisa != word >> 32)
253 			return -EBUSY;
254 		_word = (word & ~0xffUL) | iam;
255 	} while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
256 
257 	return 0;
258 }
259 
260 /**
261  * gisa_clear_ipm - clear the GISA interruption pending mask
262  *
263  * @gisa: gisa to operate on
264  *
265  * Clear the IPM atomically with the next alert address and the IAM
266  * of the GISA unconditionally. All three fields are located in the
267  * first long word of the GISA.
268  */
gisa_clear_ipm(struct kvm_s390_gisa * gisa)269 static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa)
270 {
271 	u64 word, _word;
272 
273 	do {
274 		word = READ_ONCE(gisa->u64.word[0]);
275 		_word = word & ~(0xffUL << 24);
276 	} while (cmpxchg(&gisa->u64.word[0], word, _word) != word);
277 }
278 
279 /**
280  * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM
281  *
282  * @gi: gisa interrupt struct to work on
283  *
284  * Atomically restores the interruption alert mask if none of the
285  * relevant ISCs are pending and return the IPM.
286  *
287  * Returns: the relevant pending ISCs
288  */
gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt * gi)289 static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
290 {
291 	u8 pending_mask, alert_mask;
292 	u64 word, _word;
293 
294 	do {
295 		word = READ_ONCE(gi->origin->u64.word[0]);
296 		alert_mask = READ_ONCE(gi->alert.mask);
297 		pending_mask = (u8)(word >> 24) & alert_mask;
298 		if (pending_mask)
299 			return pending_mask;
300 		_word = (word & ~0xffUL) | alert_mask;
301 	} while (cmpxchg(&gi->origin->u64.word[0], word, _word) != word);
302 
303 	return 0;
304 }
305 
gisa_in_alert_list(struct kvm_s390_gisa * gisa)306 static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
307 {
308 	return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa;
309 }
310 
gisa_set_ipm_gisc(struct kvm_s390_gisa * gisa,u32 gisc)311 static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
312 {
313 	set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
314 }
315 
gisa_get_ipm(struct kvm_s390_gisa * gisa)316 static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
317 {
318 	return READ_ONCE(gisa->ipm);
319 }
320 
gisa_clear_ipm_gisc(struct kvm_s390_gisa * gisa,u32 gisc)321 static inline void gisa_clear_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
322 {
323 	clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
324 }
325 
gisa_tac_ipm_gisc(struct kvm_s390_gisa * gisa,u32 gisc)326 static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
327 {
328 	return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
329 }
330 
pending_irqs_no_gisa(struct kvm_vcpu * vcpu)331 static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
332 {
333 	unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs |
334 				vcpu->arch.local_int.pending_irqs;
335 
336 	pending &= ~vcpu->kvm->arch.float_int.masked_irqs;
337 	return pending;
338 }
339 
pending_irqs(struct kvm_vcpu * vcpu)340 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
341 {
342 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
343 	unsigned long pending_mask;
344 
345 	pending_mask = pending_irqs_no_gisa(vcpu);
346 	if (gi->origin)
347 		pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
348 	return pending_mask;
349 }
350 
isc_to_irq_type(unsigned long isc)351 static inline int isc_to_irq_type(unsigned long isc)
352 {
353 	return IRQ_PEND_IO_ISC_0 - isc;
354 }
355 
irq_type_to_isc(unsigned long irq_type)356 static inline int irq_type_to_isc(unsigned long irq_type)
357 {
358 	return IRQ_PEND_IO_ISC_0 - irq_type;
359 }
360 
disable_iscs(struct kvm_vcpu * vcpu,unsigned long active_mask)361 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
362 				   unsigned long active_mask)
363 {
364 	int i;
365 
366 	for (i = 0; i <= MAX_ISC; i++)
367 		if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
368 			active_mask &= ~(1UL << (isc_to_irq_type(i)));
369 
370 	return active_mask;
371 }
372 
deliverable_irqs(struct kvm_vcpu * vcpu)373 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
374 {
375 	unsigned long active_mask;
376 
377 	active_mask = pending_irqs(vcpu);
378 	if (!active_mask)
379 		return 0;
380 
381 	if (psw_extint_disabled(vcpu))
382 		active_mask &= ~IRQ_PEND_EXT_MASK;
383 	if (psw_ioint_disabled(vcpu))
384 		active_mask &= ~IRQ_PEND_IO_MASK;
385 	else
386 		active_mask = disable_iscs(vcpu, active_mask);
387 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
388 		__clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
389 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
390 		__clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
391 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
392 		__clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
393 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
394 		__clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
395 	if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) {
396 		__clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
397 		__clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask);
398 	}
399 	if (psw_mchk_disabled(vcpu))
400 		active_mask &= ~IRQ_PEND_MCHK_MASK;
401 	/* PV guest cpus can have a single interruption injected at a time. */
402 	if (kvm_s390_pv_cpu_get_handle(vcpu) &&
403 	    vcpu->arch.sie_block->iictl != IICTL_CODE_NONE)
404 		active_mask &= ~(IRQ_PEND_EXT_II_MASK |
405 				 IRQ_PEND_IO_MASK |
406 				 IRQ_PEND_MCHK_MASK);
407 	/*
408 	 * Check both floating and local interrupt's cr14 because
409 	 * bit IRQ_PEND_MCHK_REP could be set in both cases.
410 	 */
411 	if (!(vcpu->arch.sie_block->gcr[14] &
412 	   (vcpu->kvm->arch.float_int.mchk.cr14 |
413 	   vcpu->arch.local_int.irq.mchk.cr14)))
414 		__clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
415 
416 	/*
417 	 * STOP irqs will never be actively delivered. They are triggered via
418 	 * intercept requests and cleared when the stop intercept is performed.
419 	 */
420 	__clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
421 
422 	return active_mask;
423 }
424 
__set_cpu_idle(struct kvm_vcpu * vcpu)425 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
426 {
427 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
428 	set_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
429 }
430 
__unset_cpu_idle(struct kvm_vcpu * vcpu)431 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
432 {
433 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
434 	clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
435 }
436 
__reset_intercept_indicators(struct kvm_vcpu * vcpu)437 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
438 {
439 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
440 				      CPUSTAT_STOP_INT);
441 	vcpu->arch.sie_block->lctl = 0x0000;
442 	vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
443 
444 	if (guestdbg_enabled(vcpu)) {
445 		vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
446 					       LCTL_CR10 | LCTL_CR11);
447 		vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
448 	}
449 }
450 
set_intercept_indicators_io(struct kvm_vcpu * vcpu)451 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
452 {
453 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
454 		return;
455 	if (psw_ioint_disabled(vcpu))
456 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
457 	else
458 		vcpu->arch.sie_block->lctl |= LCTL_CR6;
459 }
460 
set_intercept_indicators_ext(struct kvm_vcpu * vcpu)461 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
462 {
463 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
464 		return;
465 	if (psw_extint_disabled(vcpu))
466 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
467 	else
468 		vcpu->arch.sie_block->lctl |= LCTL_CR0;
469 }
470 
set_intercept_indicators_mchk(struct kvm_vcpu * vcpu)471 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
472 {
473 	if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
474 		return;
475 	if (psw_mchk_disabled(vcpu))
476 		vcpu->arch.sie_block->ictl |= ICTL_LPSW;
477 	else
478 		vcpu->arch.sie_block->lctl |= LCTL_CR14;
479 }
480 
set_intercept_indicators_stop(struct kvm_vcpu * vcpu)481 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
482 {
483 	if (kvm_s390_is_stop_irq_pending(vcpu))
484 		kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
485 }
486 
487 /* Set interception request for non-deliverable interrupts */
set_intercept_indicators(struct kvm_vcpu * vcpu)488 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
489 {
490 	set_intercept_indicators_io(vcpu);
491 	set_intercept_indicators_ext(vcpu);
492 	set_intercept_indicators_mchk(vcpu);
493 	set_intercept_indicators_stop(vcpu);
494 }
495 
__deliver_cpu_timer(struct kvm_vcpu * vcpu)496 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
497 {
498 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
499 	int rc = 0;
500 
501 	vcpu->stat.deliver_cputm++;
502 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
503 					 0, 0);
504 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
505 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
506 		vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER;
507 	} else {
508 		rc  = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
509 				   (u16 *)__LC_EXT_INT_CODE);
510 		rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
511 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
512 				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
513 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
514 				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
515 	}
516 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
517 	return rc ? -EFAULT : 0;
518 }
519 
__deliver_ckc(struct kvm_vcpu * vcpu)520 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
521 {
522 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
523 	int rc = 0;
524 
525 	vcpu->stat.deliver_ckc++;
526 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
527 					 0, 0);
528 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
529 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
530 		vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP;
531 	} else {
532 		rc  = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
533 				   (u16 __user *)__LC_EXT_INT_CODE);
534 		rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
535 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
536 				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
537 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
538 				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
539 	}
540 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
541 	return rc ? -EFAULT : 0;
542 }
543 
__deliver_pfault_init(struct kvm_vcpu * vcpu)544 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
545 {
546 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
547 	struct kvm_s390_ext_info ext;
548 	int rc;
549 
550 	spin_lock(&li->lock);
551 	ext = li->irq.ext;
552 	clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
553 	li->irq.ext.ext_params2 = 0;
554 	spin_unlock(&li->lock);
555 
556 	VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
557 		   ext.ext_params2);
558 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
559 					 KVM_S390_INT_PFAULT_INIT,
560 					 0, ext.ext_params2);
561 
562 	rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
563 	rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
564 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
565 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
566 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
567 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
568 	rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
569 	return rc ? -EFAULT : 0;
570 }
571 
__write_machine_check(struct kvm_vcpu * vcpu,struct kvm_s390_mchk_info * mchk)572 static int __write_machine_check(struct kvm_vcpu *vcpu,
573 				 struct kvm_s390_mchk_info *mchk)
574 {
575 	unsigned long ext_sa_addr;
576 	unsigned long lc;
577 	freg_t fprs[NUM_FPRS];
578 	union mci mci;
579 	int rc;
580 
581 	/*
582 	 * All other possible payload for a machine check (e.g. the register
583 	 * contents in the save area) will be handled by the ultravisor, as
584 	 * the hypervisor does not not have the needed information for
585 	 * protected guests.
586 	 */
587 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
588 		vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK;
589 		vcpu->arch.sie_block->mcic = mchk->mcic;
590 		vcpu->arch.sie_block->faddr = mchk->failing_storage_address;
591 		vcpu->arch.sie_block->edc = mchk->ext_damage_code;
592 		return 0;
593 	}
594 
595 	mci.val = mchk->mcic;
596 	/* take care of lazy register loading */
597 	save_fpu_regs();
598 	save_access_regs(vcpu->run->s.regs.acrs);
599 	if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
600 		save_gs_cb(current->thread.gs_cb);
601 
602 	/* Extended save area */
603 	rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
604 			   sizeof(unsigned long));
605 	/* Only bits 0 through 63-LC are used for address formation */
606 	lc = ext_sa_addr & MCESA_LC_MASK;
607 	if (test_kvm_facility(vcpu->kvm, 133)) {
608 		switch (lc) {
609 		case 0:
610 		case 10:
611 			ext_sa_addr &= ~0x3ffUL;
612 			break;
613 		case 11:
614 			ext_sa_addr &= ~0x7ffUL;
615 			break;
616 		case 12:
617 			ext_sa_addr &= ~0xfffUL;
618 			break;
619 		default:
620 			ext_sa_addr = 0;
621 			break;
622 		}
623 	} else {
624 		ext_sa_addr &= ~0x3ffUL;
625 	}
626 
627 	if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
628 		if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
629 				    512))
630 			mci.vr = 0;
631 	} else {
632 		mci.vr = 0;
633 	}
634 	if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
635 	    && (lc == 11 || lc == 12)) {
636 		if (write_guest_abs(vcpu, ext_sa_addr + 1024,
637 				    &vcpu->run->s.regs.gscb, 32))
638 			mci.gs = 0;
639 	} else {
640 		mci.gs = 0;
641 	}
642 
643 	/* General interruption information */
644 	rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
645 	rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
646 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
647 	rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
648 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
649 	rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
650 
651 	/* Register-save areas */
652 	if (MACHINE_HAS_VX) {
653 		convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
654 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
655 	} else {
656 		rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
657 				     vcpu->run->s.regs.fprs, 128);
658 	}
659 	rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
660 			     vcpu->run->s.regs.gprs, 128);
661 	rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
662 			   (u32 __user *) __LC_FP_CREG_SAVE_AREA);
663 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
664 			   (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
665 	rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
666 			   (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
667 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
668 			   (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
669 	rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
670 			     &vcpu->run->s.regs.acrs, 64);
671 	rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
672 			     &vcpu->arch.sie_block->gcr, 128);
673 
674 	/* Extended interruption information */
675 	rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
676 			   (u32 __user *) __LC_EXT_DAMAGE_CODE);
677 	rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
678 			   (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
679 	rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
680 			     sizeof(mchk->fixed_logout));
681 	return rc ? -EFAULT : 0;
682 }
683 
__deliver_machine_check(struct kvm_vcpu * vcpu)684 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
685 {
686 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
687 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
688 	struct kvm_s390_mchk_info mchk = {};
689 	int deliver = 0;
690 	int rc = 0;
691 
692 	spin_lock(&fi->lock);
693 	spin_lock(&li->lock);
694 	if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
695 	    test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
696 		/*
697 		 * If there was an exigent machine check pending, then any
698 		 * repressible machine checks that might have been pending
699 		 * are indicated along with it, so always clear bits for
700 		 * repressible and exigent interrupts
701 		 */
702 		mchk = li->irq.mchk;
703 		clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
704 		clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
705 		memset(&li->irq.mchk, 0, sizeof(mchk));
706 		deliver = 1;
707 	}
708 	/*
709 	 * We indicate floating repressible conditions along with
710 	 * other pending conditions. Channel Report Pending and Channel
711 	 * Subsystem damage are the only two and are indicated by
712 	 * bits in mcic and masked in cr14.
713 	 */
714 	if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
715 		mchk.mcic |= fi->mchk.mcic;
716 		mchk.cr14 |= fi->mchk.cr14;
717 		memset(&fi->mchk, 0, sizeof(mchk));
718 		deliver = 1;
719 	}
720 	spin_unlock(&li->lock);
721 	spin_unlock(&fi->lock);
722 
723 	if (deliver) {
724 		VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
725 			   mchk.mcic);
726 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
727 						 KVM_S390_MCHK,
728 						 mchk.cr14, mchk.mcic);
729 		vcpu->stat.deliver_machine_check++;
730 		rc = __write_machine_check(vcpu, &mchk);
731 	}
732 	return rc;
733 }
734 
__deliver_restart(struct kvm_vcpu * vcpu)735 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
736 {
737 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
738 	int rc = 0;
739 
740 	VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
741 	vcpu->stat.deliver_restart_signal++;
742 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
743 
744 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
745 		vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART;
746 	} else {
747 		rc  = write_guest_lc(vcpu,
748 				     offsetof(struct lowcore, restart_old_psw),
749 				     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
750 		rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
751 				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
752 	}
753 	clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
754 	return rc ? -EFAULT : 0;
755 }
756 
__deliver_set_prefix(struct kvm_vcpu * vcpu)757 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
758 {
759 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
760 	struct kvm_s390_prefix_info prefix;
761 
762 	spin_lock(&li->lock);
763 	prefix = li->irq.prefix;
764 	li->irq.prefix.address = 0;
765 	clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
766 	spin_unlock(&li->lock);
767 
768 	vcpu->stat.deliver_prefix_signal++;
769 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
770 					 KVM_S390_SIGP_SET_PREFIX,
771 					 prefix.address, 0);
772 
773 	kvm_s390_set_prefix(vcpu, prefix.address);
774 	return 0;
775 }
776 
__deliver_emergency_signal(struct kvm_vcpu * vcpu)777 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
778 {
779 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
780 	int rc;
781 	int cpu_addr;
782 
783 	spin_lock(&li->lock);
784 	cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
785 	clear_bit(cpu_addr, li->sigp_emerg_pending);
786 	if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
787 		clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
788 	spin_unlock(&li->lock);
789 
790 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
791 	vcpu->stat.deliver_emergency_signal++;
792 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
793 					 cpu_addr, 0);
794 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
795 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
796 		vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG;
797 		vcpu->arch.sie_block->extcpuaddr = cpu_addr;
798 		return 0;
799 	}
800 
801 	rc  = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
802 			   (u16 *)__LC_EXT_INT_CODE);
803 	rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
804 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
805 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
806 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
807 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
808 	return rc ? -EFAULT : 0;
809 }
810 
__deliver_external_call(struct kvm_vcpu * vcpu)811 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
812 {
813 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
814 	struct kvm_s390_extcall_info extcall;
815 	int rc;
816 
817 	spin_lock(&li->lock);
818 	extcall = li->irq.extcall;
819 	li->irq.extcall.code = 0;
820 	clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
821 	spin_unlock(&li->lock);
822 
823 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
824 	vcpu->stat.deliver_external_call++;
825 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
826 					 KVM_S390_INT_EXTERNAL_CALL,
827 					 extcall.code, 0);
828 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
829 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
830 		vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL;
831 		vcpu->arch.sie_block->extcpuaddr = extcall.code;
832 		return 0;
833 	}
834 
835 	rc  = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
836 			   (u16 *)__LC_EXT_INT_CODE);
837 	rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
838 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
839 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
840 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
841 			    sizeof(psw_t));
842 	return rc ? -EFAULT : 0;
843 }
844 
__deliver_prog_pv(struct kvm_vcpu * vcpu,u16 code)845 static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code)
846 {
847 	switch (code) {
848 	case PGM_SPECIFICATION:
849 		vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION;
850 		break;
851 	case PGM_OPERAND:
852 		vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND;
853 		break;
854 	default:
855 		return -EINVAL;
856 	}
857 	return 0;
858 }
859 
__deliver_prog(struct kvm_vcpu * vcpu)860 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
861 {
862 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
863 	struct kvm_s390_pgm_info pgm_info;
864 	int rc = 0, nullifying = false;
865 	u16 ilen;
866 
867 	spin_lock(&li->lock);
868 	pgm_info = li->irq.pgm;
869 	clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
870 	memset(&li->irq.pgm, 0, sizeof(pgm_info));
871 	spin_unlock(&li->lock);
872 
873 	ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
874 	VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
875 		   pgm_info.code, ilen);
876 	vcpu->stat.deliver_program++;
877 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
878 					 pgm_info.code, 0);
879 
880 	/* PER is handled by the ultravisor */
881 	if (kvm_s390_pv_cpu_is_protected(vcpu))
882 		return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER);
883 
884 	switch (pgm_info.code & ~PGM_PER) {
885 	case PGM_AFX_TRANSLATION:
886 	case PGM_ASX_TRANSLATION:
887 	case PGM_EX_TRANSLATION:
888 	case PGM_LFX_TRANSLATION:
889 	case PGM_LSTE_SEQUENCE:
890 	case PGM_LSX_TRANSLATION:
891 	case PGM_LX_TRANSLATION:
892 	case PGM_PRIMARY_AUTHORITY:
893 	case PGM_SECONDARY_AUTHORITY:
894 		nullifying = true;
895 		fallthrough;
896 	case PGM_SPACE_SWITCH:
897 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
898 				  (u64 *)__LC_TRANS_EXC_CODE);
899 		break;
900 	case PGM_ALEN_TRANSLATION:
901 	case PGM_ALE_SEQUENCE:
902 	case PGM_ASTE_INSTANCE:
903 	case PGM_ASTE_SEQUENCE:
904 	case PGM_ASTE_VALIDITY:
905 	case PGM_EXTENDED_AUTHORITY:
906 		rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
907 				  (u8 *)__LC_EXC_ACCESS_ID);
908 		nullifying = true;
909 		break;
910 	case PGM_ASCE_TYPE:
911 	case PGM_PAGE_TRANSLATION:
912 	case PGM_REGION_FIRST_TRANS:
913 	case PGM_REGION_SECOND_TRANS:
914 	case PGM_REGION_THIRD_TRANS:
915 	case PGM_SEGMENT_TRANSLATION:
916 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
917 				  (u64 *)__LC_TRANS_EXC_CODE);
918 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
919 				   (u8 *)__LC_EXC_ACCESS_ID);
920 		rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
921 				   (u8 *)__LC_OP_ACCESS_ID);
922 		nullifying = true;
923 		break;
924 	case PGM_MONITOR:
925 		rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
926 				  (u16 *)__LC_MON_CLASS_NR);
927 		rc |= put_guest_lc(vcpu, pgm_info.mon_code,
928 				   (u64 *)__LC_MON_CODE);
929 		break;
930 	case PGM_VECTOR_PROCESSING:
931 	case PGM_DATA:
932 		rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
933 				  (u32 *)__LC_DATA_EXC_CODE);
934 		break;
935 	case PGM_PROTECTION:
936 		rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
937 				  (u64 *)__LC_TRANS_EXC_CODE);
938 		rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
939 				   (u8 *)__LC_EXC_ACCESS_ID);
940 		break;
941 	case PGM_STACK_FULL:
942 	case PGM_STACK_EMPTY:
943 	case PGM_STACK_SPECIFICATION:
944 	case PGM_STACK_TYPE:
945 	case PGM_STACK_OPERATION:
946 	case PGM_TRACE_TABEL:
947 	case PGM_CRYPTO_OPERATION:
948 		nullifying = true;
949 		break;
950 	}
951 
952 	if (pgm_info.code & PGM_PER) {
953 		rc |= put_guest_lc(vcpu, pgm_info.per_code,
954 				   (u8 *) __LC_PER_CODE);
955 		rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
956 				   (u8 *)__LC_PER_ATMID);
957 		rc |= put_guest_lc(vcpu, pgm_info.per_address,
958 				   (u64 *) __LC_PER_ADDRESS);
959 		rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
960 				   (u8 *) __LC_PER_ACCESS_ID);
961 	}
962 
963 	if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
964 		kvm_s390_rewind_psw(vcpu, ilen);
965 
966 	/* bit 1+2 of the target are the ilc, so we can directly use ilen */
967 	rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
968 	rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
969 				 (u64 *) __LC_PGM_LAST_BREAK);
970 	rc |= put_guest_lc(vcpu, pgm_info.code,
971 			   (u16 *)__LC_PGM_INT_CODE);
972 	rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
973 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
974 	rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
975 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
976 	return rc ? -EFAULT : 0;
977 }
978 
979 #define SCCB_MASK 0xFFFFFFF8
980 #define SCCB_EVENT_PENDING 0x3
981 
write_sclp(struct kvm_vcpu * vcpu,u32 parm)982 static int write_sclp(struct kvm_vcpu *vcpu, u32 parm)
983 {
984 	int rc;
985 
986 	if (kvm_s390_pv_cpu_get_handle(vcpu)) {
987 		vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
988 		vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG;
989 		vcpu->arch.sie_block->eiparams = parm;
990 		return 0;
991 	}
992 
993 	rc  = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
994 	rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
995 	rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
996 			     &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
997 	rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
998 			    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
999 	rc |= put_guest_lc(vcpu, parm,
1000 			   (u32 *)__LC_EXT_PARAMS);
1001 
1002 	return rc ? -EFAULT : 0;
1003 }
1004 
__deliver_service(struct kvm_vcpu * vcpu)1005 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
1006 {
1007 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1008 	struct kvm_s390_ext_info ext;
1009 
1010 	spin_lock(&fi->lock);
1011 	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) ||
1012 	    !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
1013 		spin_unlock(&fi->lock);
1014 		return 0;
1015 	}
1016 	ext = fi->srv_signal;
1017 	memset(&fi->srv_signal, 0, sizeof(ext));
1018 	clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1019 	clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1020 	if (kvm_s390_pv_cpu_is_protected(vcpu))
1021 		set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs);
1022 	spin_unlock(&fi->lock);
1023 
1024 	VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
1025 		   ext.ext_params);
1026 	vcpu->stat.deliver_service_signal++;
1027 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1028 					 ext.ext_params, 0);
1029 
1030 	return write_sclp(vcpu, ext.ext_params);
1031 }
1032 
__deliver_service_ev(struct kvm_vcpu * vcpu)1033 static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)
1034 {
1035 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1036 	struct kvm_s390_ext_info ext;
1037 
1038 	spin_lock(&fi->lock);
1039 	if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) {
1040 		spin_unlock(&fi->lock);
1041 		return 0;
1042 	}
1043 	ext = fi->srv_signal;
1044 	/* only clear the event bit */
1045 	fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING;
1046 	clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1047 	spin_unlock(&fi->lock);
1048 
1049 	VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event");
1050 	vcpu->stat.deliver_service_signal++;
1051 	trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1052 					 ext.ext_params, 0);
1053 
1054 	return write_sclp(vcpu, SCCB_EVENT_PENDING);
1055 }
1056 
__deliver_pfault_done(struct kvm_vcpu * vcpu)1057 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
1058 {
1059 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1060 	struct kvm_s390_interrupt_info *inti;
1061 	int rc = 0;
1062 
1063 	spin_lock(&fi->lock);
1064 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
1065 					struct kvm_s390_interrupt_info,
1066 					list);
1067 	if (inti) {
1068 		list_del(&inti->list);
1069 		fi->counters[FIRQ_CNTR_PFAULT] -= 1;
1070 	}
1071 	if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
1072 		clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1073 	spin_unlock(&fi->lock);
1074 
1075 	if (inti) {
1076 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1077 						 KVM_S390_INT_PFAULT_DONE, 0,
1078 						 inti->ext.ext_params2);
1079 		VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
1080 			   inti->ext.ext_params2);
1081 
1082 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1083 				(u16 *)__LC_EXT_INT_CODE);
1084 		rc |= put_guest_lc(vcpu, PFAULT_DONE,
1085 				(u16 *)__LC_EXT_CPU_ADDR);
1086 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1087 				&vcpu->arch.sie_block->gpsw,
1088 				sizeof(psw_t));
1089 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1090 				&vcpu->arch.sie_block->gpsw,
1091 				sizeof(psw_t));
1092 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1093 				(u64 *)__LC_EXT_PARAMS2);
1094 		kfree(inti);
1095 	}
1096 	return rc ? -EFAULT : 0;
1097 }
1098 
__deliver_virtio(struct kvm_vcpu * vcpu)1099 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
1100 {
1101 	struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1102 	struct kvm_s390_interrupt_info *inti;
1103 	int rc = 0;
1104 
1105 	spin_lock(&fi->lock);
1106 	inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
1107 					struct kvm_s390_interrupt_info,
1108 					list);
1109 	if (inti) {
1110 		VCPU_EVENT(vcpu, 4,
1111 			   "deliver: virtio parm: 0x%x,parm64: 0x%llx",
1112 			   inti->ext.ext_params, inti->ext.ext_params2);
1113 		vcpu->stat.deliver_virtio++;
1114 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1115 				inti->type,
1116 				inti->ext.ext_params,
1117 				inti->ext.ext_params2);
1118 		list_del(&inti->list);
1119 		fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1120 	}
1121 	if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1122 		clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1123 	spin_unlock(&fi->lock);
1124 
1125 	if (inti) {
1126 		rc  = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1127 				(u16 *)__LC_EXT_INT_CODE);
1128 		rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1129 				(u16 *)__LC_EXT_CPU_ADDR);
1130 		rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1131 				&vcpu->arch.sie_block->gpsw,
1132 				sizeof(psw_t));
1133 		rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1134 				&vcpu->arch.sie_block->gpsw,
1135 				sizeof(psw_t));
1136 		rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1137 				(u32 *)__LC_EXT_PARAMS);
1138 		rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1139 				(u64 *)__LC_EXT_PARAMS2);
1140 		kfree(inti);
1141 	}
1142 	return rc ? -EFAULT : 0;
1143 }
1144 
__do_deliver_io(struct kvm_vcpu * vcpu,struct kvm_s390_io_info * io)1145 static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1146 {
1147 	int rc;
1148 
1149 	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
1150 		vcpu->arch.sie_block->iictl = IICTL_CODE_IO;
1151 		vcpu->arch.sie_block->subchannel_id = io->subchannel_id;
1152 		vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr;
1153 		vcpu->arch.sie_block->io_int_parm = io->io_int_parm;
1154 		vcpu->arch.sie_block->io_int_word = io->io_int_word;
1155 		return 0;
1156 	}
1157 
1158 	rc  = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1159 	rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1160 	rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1161 	rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1162 	rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1163 			     &vcpu->arch.sie_block->gpsw,
1164 			     sizeof(psw_t));
1165 	rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1166 			    &vcpu->arch.sie_block->gpsw,
1167 			    sizeof(psw_t));
1168 	return rc ? -EFAULT : 0;
1169 }
1170 
__deliver_io(struct kvm_vcpu * vcpu,unsigned long irq_type)1171 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1172 				     unsigned long irq_type)
1173 {
1174 	struct list_head *isc_list;
1175 	struct kvm_s390_float_interrupt *fi;
1176 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1177 	struct kvm_s390_interrupt_info *inti = NULL;
1178 	struct kvm_s390_io_info io;
1179 	u32 isc;
1180 	int rc = 0;
1181 
1182 	fi = &vcpu->kvm->arch.float_int;
1183 
1184 	spin_lock(&fi->lock);
1185 	isc = irq_type_to_isc(irq_type);
1186 	isc_list = &fi->lists[isc];
1187 	inti = list_first_entry_or_null(isc_list,
1188 					struct kvm_s390_interrupt_info,
1189 					list);
1190 	if (inti) {
1191 		if (inti->type & KVM_S390_INT_IO_AI_MASK)
1192 			VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1193 		else
1194 			VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1195 			inti->io.subchannel_id >> 8,
1196 			inti->io.subchannel_id >> 1 & 0x3,
1197 			inti->io.subchannel_nr);
1198 
1199 		vcpu->stat.deliver_io++;
1200 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1201 				inti->type,
1202 				((__u32)inti->io.subchannel_id << 16) |
1203 				inti->io.subchannel_nr,
1204 				((__u64)inti->io.io_int_parm << 32) |
1205 				inti->io.io_int_word);
1206 		list_del(&inti->list);
1207 		fi->counters[FIRQ_CNTR_IO] -= 1;
1208 	}
1209 	if (list_empty(isc_list))
1210 		clear_bit(irq_type, &fi->pending_irqs);
1211 	spin_unlock(&fi->lock);
1212 
1213 	if (inti) {
1214 		rc = __do_deliver_io(vcpu, &(inti->io));
1215 		kfree(inti);
1216 		goto out;
1217 	}
1218 
1219 	if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1220 		/*
1221 		 * in case an adapter interrupt was not delivered
1222 		 * in SIE context KVM will handle the delivery
1223 		 */
1224 		VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1225 		memset(&io, 0, sizeof(io));
1226 		io.io_int_word = isc_to_int_word(isc);
1227 		vcpu->stat.deliver_io++;
1228 		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1229 			KVM_S390_INT_IO(1, 0, 0, 0),
1230 			((__u32)io.subchannel_id << 16) |
1231 			io.subchannel_nr,
1232 			((__u64)io.io_int_parm << 32) |
1233 			io.io_int_word);
1234 		rc = __do_deliver_io(vcpu, &io);
1235 	}
1236 out:
1237 	return rc;
1238 }
1239 
1240 /* Check whether an external call is pending (deliverable or not) */
kvm_s390_ext_call_pending(struct kvm_vcpu * vcpu)1241 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1242 {
1243 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1244 
1245 	if (!sclp.has_sigpif)
1246 		return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1247 
1248 	return sca_ext_call_pending(vcpu, NULL);
1249 }
1250 
kvm_s390_vcpu_has_irq(struct kvm_vcpu * vcpu,int exclude_stop)1251 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1252 {
1253 	if (deliverable_irqs(vcpu))
1254 		return 1;
1255 
1256 	if (kvm_cpu_has_pending_timer(vcpu))
1257 		return 1;
1258 
1259 	/* external call pending and deliverable */
1260 	if (kvm_s390_ext_call_pending(vcpu) &&
1261 	    !psw_extint_disabled(vcpu) &&
1262 	    (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1263 		return 1;
1264 
1265 	if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1266 		return 1;
1267 	return 0;
1268 }
1269 
kvm_cpu_has_pending_timer(struct kvm_vcpu * vcpu)1270 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1271 {
1272 	return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1273 }
1274 
__calculate_sltime(struct kvm_vcpu * vcpu)1275 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1276 {
1277 	const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1278 	const u64 ckc = vcpu->arch.sie_block->ckc;
1279 	u64 cputm, sltime = 0;
1280 
1281 	if (ckc_interrupts_enabled(vcpu)) {
1282 		if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1283 			if ((s64)now < (s64)ckc)
1284 				sltime = tod_to_ns((s64)ckc - (s64)now);
1285 		} else if (now < ckc) {
1286 			sltime = tod_to_ns(ckc - now);
1287 		}
1288 		/* already expired */
1289 		if (!sltime)
1290 			return 0;
1291 		if (cpu_timer_interrupts_enabled(vcpu)) {
1292 			cputm = kvm_s390_get_cpu_timer(vcpu);
1293 			/* already expired? */
1294 			if (cputm >> 63)
1295 				return 0;
1296 			return min_t(u64, sltime, tod_to_ns(cputm));
1297 		}
1298 	} else if (cpu_timer_interrupts_enabled(vcpu)) {
1299 		sltime = kvm_s390_get_cpu_timer(vcpu);
1300 		/* already expired? */
1301 		if (sltime >> 63)
1302 			return 0;
1303 	}
1304 	return sltime;
1305 }
1306 
kvm_s390_handle_wait(struct kvm_vcpu * vcpu)1307 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1308 {
1309 	struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1310 	u64 sltime;
1311 
1312 	vcpu->stat.exit_wait_state++;
1313 
1314 	/* fast path */
1315 	if (kvm_arch_vcpu_runnable(vcpu))
1316 		return 0;
1317 
1318 	if (psw_interrupts_disabled(vcpu)) {
1319 		VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1320 		return -EOPNOTSUPP; /* disabled wait */
1321 	}
1322 
1323 	if (gi->origin &&
1324 	    (gisa_get_ipm_or_restore_iam(gi) &
1325 	     vcpu->arch.sie_block->gcr[6] >> 24))
1326 		return 0;
1327 
1328 	if (!ckc_interrupts_enabled(vcpu) &&
1329 	    !cpu_timer_interrupts_enabled(vcpu)) {
1330 		VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1331 		__set_cpu_idle(vcpu);
1332 		goto no_timer;
1333 	}
1334 
1335 	sltime = __calculate_sltime(vcpu);
1336 	if (!sltime)
1337 		return 0;
1338 
1339 	__set_cpu_idle(vcpu);
1340 	hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1341 	VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1342 no_timer:
1343 	kvm_vcpu_srcu_read_unlock(vcpu);
1344 	kvm_vcpu_halt(vcpu);
1345 	vcpu->valid_wakeup = false;
1346 	__unset_cpu_idle(vcpu);
1347 	kvm_vcpu_srcu_read_lock(vcpu);
1348 
1349 	hrtimer_cancel(&vcpu->arch.ckc_timer);
1350 	return 0;
1351 }
1352 
kvm_s390_vcpu_wakeup(struct kvm_vcpu * vcpu)1353 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1354 {
1355 	vcpu->valid_wakeup = true;
1356 	kvm_vcpu_wake_up(vcpu);
1357 
1358 	/*
1359 	 * The VCPU might not be sleeping but rather executing VSIE. Let's
1360 	 * kick it, so it leaves the SIE to process the request.
1361 	 */
1362 	kvm_s390_vsie_kick(vcpu);
1363 }
1364 
kvm_s390_idle_wakeup(struct hrtimer * timer)1365 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1366 {
1367 	struct kvm_vcpu *vcpu;
1368 	u64 sltime;
1369 
1370 	vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1371 	sltime = __calculate_sltime(vcpu);
1372 
1373 	/*
1374 	 * If the monotonic clock runs faster than the tod clock we might be
1375 	 * woken up too early and have to go back to sleep to avoid deadlocks.
1376 	 */
1377 	if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1378 		return HRTIMER_RESTART;
1379 	kvm_s390_vcpu_wakeup(vcpu);
1380 	return HRTIMER_NORESTART;
1381 }
1382 
kvm_s390_clear_local_irqs(struct kvm_vcpu * vcpu)1383 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1384 {
1385 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1386 
1387 	spin_lock(&li->lock);
1388 	li->pending_irqs = 0;
1389 	bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1390 	memset(&li->irq, 0, sizeof(li->irq));
1391 	spin_unlock(&li->lock);
1392 
1393 	sca_clear_ext_call(vcpu);
1394 }
1395 
kvm_s390_deliver_pending_interrupts(struct kvm_vcpu * vcpu)1396 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1397 {
1398 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1399 	int rc = 0;
1400 	unsigned long irq_type;
1401 	unsigned long irqs;
1402 
1403 	__reset_intercept_indicators(vcpu);
1404 
1405 	/* pending ckc conditions might have been invalidated */
1406 	clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1407 	if (ckc_irq_pending(vcpu))
1408 		set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1409 
1410 	/* pending cpu timer conditions might have been invalidated */
1411 	clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1412 	if (cpu_timer_irq_pending(vcpu))
1413 		set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1414 
1415 	while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1416 		/* bits are in the reverse order of interrupt priority */
1417 		irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1418 		switch (irq_type) {
1419 		case IRQ_PEND_IO_ISC_0:
1420 		case IRQ_PEND_IO_ISC_1:
1421 		case IRQ_PEND_IO_ISC_2:
1422 		case IRQ_PEND_IO_ISC_3:
1423 		case IRQ_PEND_IO_ISC_4:
1424 		case IRQ_PEND_IO_ISC_5:
1425 		case IRQ_PEND_IO_ISC_6:
1426 		case IRQ_PEND_IO_ISC_7:
1427 			rc = __deliver_io(vcpu, irq_type);
1428 			break;
1429 		case IRQ_PEND_MCHK_EX:
1430 		case IRQ_PEND_MCHK_REP:
1431 			rc = __deliver_machine_check(vcpu);
1432 			break;
1433 		case IRQ_PEND_PROG:
1434 			rc = __deliver_prog(vcpu);
1435 			break;
1436 		case IRQ_PEND_EXT_EMERGENCY:
1437 			rc = __deliver_emergency_signal(vcpu);
1438 			break;
1439 		case IRQ_PEND_EXT_EXTERNAL:
1440 			rc = __deliver_external_call(vcpu);
1441 			break;
1442 		case IRQ_PEND_EXT_CLOCK_COMP:
1443 			rc = __deliver_ckc(vcpu);
1444 			break;
1445 		case IRQ_PEND_EXT_CPU_TIMER:
1446 			rc = __deliver_cpu_timer(vcpu);
1447 			break;
1448 		case IRQ_PEND_RESTART:
1449 			rc = __deliver_restart(vcpu);
1450 			break;
1451 		case IRQ_PEND_SET_PREFIX:
1452 			rc = __deliver_set_prefix(vcpu);
1453 			break;
1454 		case IRQ_PEND_PFAULT_INIT:
1455 			rc = __deliver_pfault_init(vcpu);
1456 			break;
1457 		case IRQ_PEND_EXT_SERVICE:
1458 			rc = __deliver_service(vcpu);
1459 			break;
1460 		case IRQ_PEND_EXT_SERVICE_EV:
1461 			rc = __deliver_service_ev(vcpu);
1462 			break;
1463 		case IRQ_PEND_PFAULT_DONE:
1464 			rc = __deliver_pfault_done(vcpu);
1465 			break;
1466 		case IRQ_PEND_VIRTIO:
1467 			rc = __deliver_virtio(vcpu);
1468 			break;
1469 		default:
1470 			WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1471 			clear_bit(irq_type, &li->pending_irqs);
1472 		}
1473 	}
1474 
1475 	set_intercept_indicators(vcpu);
1476 
1477 	return rc;
1478 }
1479 
__inject_prog(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1480 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1481 {
1482 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1483 
1484 	vcpu->stat.inject_program++;
1485 	VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1486 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1487 				   irq->u.pgm.code, 0);
1488 
1489 	if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1490 		/* auto detection if no valid ILC was given */
1491 		irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1492 		irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1493 		irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1494 	}
1495 
1496 	if (irq->u.pgm.code == PGM_PER) {
1497 		li->irq.pgm.code |= PGM_PER;
1498 		li->irq.pgm.flags = irq->u.pgm.flags;
1499 		/* only modify PER related information */
1500 		li->irq.pgm.per_address = irq->u.pgm.per_address;
1501 		li->irq.pgm.per_code = irq->u.pgm.per_code;
1502 		li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1503 		li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1504 	} else if (!(irq->u.pgm.code & PGM_PER)) {
1505 		li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1506 				   irq->u.pgm.code;
1507 		li->irq.pgm.flags = irq->u.pgm.flags;
1508 		/* only modify non-PER information */
1509 		li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1510 		li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1511 		li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1512 		li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1513 		li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1514 		li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1515 	} else {
1516 		li->irq.pgm = irq->u.pgm;
1517 	}
1518 	set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1519 	return 0;
1520 }
1521 
__inject_pfault_init(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1522 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1523 {
1524 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1525 
1526 	vcpu->stat.inject_pfault_init++;
1527 	VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1528 		   irq->u.ext.ext_params2);
1529 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1530 				   irq->u.ext.ext_params,
1531 				   irq->u.ext.ext_params2);
1532 
1533 	li->irq.ext = irq->u.ext;
1534 	set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1535 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1536 	return 0;
1537 }
1538 
__inject_extcall(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1539 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1540 {
1541 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1542 	struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1543 	uint16_t src_id = irq->u.extcall.code;
1544 
1545 	vcpu->stat.inject_external_call++;
1546 	VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1547 		   src_id);
1548 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1549 				   src_id, 0);
1550 
1551 	/* sending vcpu invalid */
1552 	if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1553 		return -EINVAL;
1554 
1555 	if (sclp.has_sigpif && !kvm_s390_pv_cpu_get_handle(vcpu))
1556 		return sca_inject_ext_call(vcpu, src_id);
1557 
1558 	if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1559 		return -EBUSY;
1560 	*extcall = irq->u.extcall;
1561 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1562 	return 0;
1563 }
1564 
__inject_set_prefix(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1565 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1566 {
1567 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1568 	struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1569 
1570 	vcpu->stat.inject_set_prefix++;
1571 	VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1572 		   irq->u.prefix.address);
1573 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1574 				   irq->u.prefix.address, 0);
1575 
1576 	if (!is_vcpu_stopped(vcpu))
1577 		return -EBUSY;
1578 
1579 	*prefix = irq->u.prefix;
1580 	set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1581 	return 0;
1582 }
1583 
1584 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
__inject_sigp_stop(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1585 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1586 {
1587 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1588 	struct kvm_s390_stop_info *stop = &li->irq.stop;
1589 	int rc = 0;
1590 
1591 	vcpu->stat.inject_stop_signal++;
1592 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1593 
1594 	if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1595 		return -EINVAL;
1596 
1597 	if (is_vcpu_stopped(vcpu)) {
1598 		if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1599 			rc = kvm_s390_store_status_unloaded(vcpu,
1600 						KVM_S390_STORE_STATUS_NOADDR);
1601 		return rc;
1602 	}
1603 
1604 	if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1605 		return -EBUSY;
1606 	stop->flags = irq->u.stop.flags;
1607 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1608 	return 0;
1609 }
1610 
__inject_sigp_restart(struct kvm_vcpu * vcpu)1611 static int __inject_sigp_restart(struct kvm_vcpu *vcpu)
1612 {
1613 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1614 
1615 	vcpu->stat.inject_restart++;
1616 	VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1617 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1618 
1619 	set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1620 	return 0;
1621 }
1622 
__inject_sigp_emergency(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1623 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1624 				   struct kvm_s390_irq *irq)
1625 {
1626 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1627 
1628 	vcpu->stat.inject_emergency_signal++;
1629 	VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1630 		   irq->u.emerg.code);
1631 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1632 				   irq->u.emerg.code, 0);
1633 
1634 	/* sending vcpu invalid */
1635 	if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1636 		return -EINVAL;
1637 
1638 	set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1639 	set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1640 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1641 	return 0;
1642 }
1643 
__inject_mchk(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)1644 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1645 {
1646 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1647 	struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1648 
1649 	vcpu->stat.inject_mchk++;
1650 	VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1651 		   irq->u.mchk.mcic);
1652 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1653 				   irq->u.mchk.mcic);
1654 
1655 	/*
1656 	 * Because repressible machine checks can be indicated along with
1657 	 * exigent machine checks (PoP, Chapter 11, Interruption action)
1658 	 * we need to combine cr14, mcic and external damage code.
1659 	 * Failing storage address and the logout area should not be or'ed
1660 	 * together, we just indicate the last occurrence of the corresponding
1661 	 * machine check
1662 	 */
1663 	mchk->cr14 |= irq->u.mchk.cr14;
1664 	mchk->mcic |= irq->u.mchk.mcic;
1665 	mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1666 	mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1667 	memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1668 	       sizeof(mchk->fixed_logout));
1669 	if (mchk->mcic & MCHK_EX_MASK)
1670 		set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1671 	else if (mchk->mcic & MCHK_REP_MASK)
1672 		set_bit(IRQ_PEND_MCHK_REP,  &li->pending_irqs);
1673 	return 0;
1674 }
1675 
__inject_ckc(struct kvm_vcpu * vcpu)1676 static int __inject_ckc(struct kvm_vcpu *vcpu)
1677 {
1678 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1679 
1680 	vcpu->stat.inject_ckc++;
1681 	VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1682 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1683 				   0, 0);
1684 
1685 	set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1686 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1687 	return 0;
1688 }
1689 
__inject_cpu_timer(struct kvm_vcpu * vcpu)1690 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1691 {
1692 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1693 
1694 	vcpu->stat.inject_cputm++;
1695 	VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1696 	trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1697 				   0, 0);
1698 
1699 	set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1700 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1701 	return 0;
1702 }
1703 
get_io_int(struct kvm * kvm,int isc,u32 schid)1704 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1705 						  int isc, u32 schid)
1706 {
1707 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1708 	struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1709 	struct kvm_s390_interrupt_info *iter;
1710 	u16 id = (schid & 0xffff0000U) >> 16;
1711 	u16 nr = schid & 0x0000ffffU;
1712 
1713 	spin_lock(&fi->lock);
1714 	list_for_each_entry(iter, isc_list, list) {
1715 		if (schid && (id != iter->io.subchannel_id ||
1716 			      nr != iter->io.subchannel_nr))
1717 			continue;
1718 		/* found an appropriate entry */
1719 		list_del_init(&iter->list);
1720 		fi->counters[FIRQ_CNTR_IO] -= 1;
1721 		if (list_empty(isc_list))
1722 			clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1723 		spin_unlock(&fi->lock);
1724 		return iter;
1725 	}
1726 	spin_unlock(&fi->lock);
1727 	return NULL;
1728 }
1729 
get_top_io_int(struct kvm * kvm,u64 isc_mask,u32 schid)1730 static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1731 						      u64 isc_mask, u32 schid)
1732 {
1733 	struct kvm_s390_interrupt_info *inti = NULL;
1734 	int isc;
1735 
1736 	for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1737 		if (isc_mask & isc_to_isc_bits(isc))
1738 			inti = get_io_int(kvm, isc, schid);
1739 	}
1740 	return inti;
1741 }
1742 
get_top_gisa_isc(struct kvm * kvm,u64 isc_mask,u32 schid)1743 static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1744 {
1745 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1746 	unsigned long active_mask;
1747 	int isc;
1748 
1749 	if (schid)
1750 		goto out;
1751 	if (!gi->origin)
1752 		goto out;
1753 
1754 	active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1755 	while (active_mask) {
1756 		isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1757 		if (gisa_tac_ipm_gisc(gi->origin, isc))
1758 			return isc;
1759 		clear_bit_inv(isc, &active_mask);
1760 	}
1761 out:
1762 	return -EINVAL;
1763 }
1764 
1765 /*
1766  * Dequeue and return an I/O interrupt matching any of the interruption
1767  * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1768  * Take into account the interrupts pending in the interrupt list and in GISA.
1769  *
1770  * Note that for a guest that does not enable I/O interrupts
1771  * but relies on TPI, a flood of classic interrupts may starve
1772  * out adapter interrupts on the same isc. Linux does not do
1773  * that, and it is possible to work around the issue by configuring
1774  * different iscs for classic and adapter interrupts in the guest,
1775  * but we may want to revisit this in the future.
1776  */
kvm_s390_get_io_int(struct kvm * kvm,u64 isc_mask,u32 schid)1777 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1778 						    u64 isc_mask, u32 schid)
1779 {
1780 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1781 	struct kvm_s390_interrupt_info *inti, *tmp_inti;
1782 	int isc;
1783 
1784 	inti = get_top_io_int(kvm, isc_mask, schid);
1785 
1786 	isc = get_top_gisa_isc(kvm, isc_mask, schid);
1787 	if (isc < 0)
1788 		/* no AI in GISA */
1789 		goto out;
1790 
1791 	if (!inti)
1792 		/* AI in GISA but no classical IO int */
1793 		goto gisa_out;
1794 
1795 	/* both types of interrupts present */
1796 	if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1797 		/* classical IO int with higher priority */
1798 		gisa_set_ipm_gisc(gi->origin, isc);
1799 		goto out;
1800 	}
1801 gisa_out:
1802 	tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
1803 	if (tmp_inti) {
1804 		tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1805 		tmp_inti->io.io_int_word = isc_to_int_word(isc);
1806 		if (inti)
1807 			kvm_s390_reinject_io_int(kvm, inti);
1808 		inti = tmp_inti;
1809 	} else
1810 		gisa_set_ipm_gisc(gi->origin, isc);
1811 out:
1812 	return inti;
1813 }
1814 
__inject_service(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1815 static int __inject_service(struct kvm *kvm,
1816 			     struct kvm_s390_interrupt_info *inti)
1817 {
1818 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1819 
1820 	kvm->stat.inject_service_signal++;
1821 	spin_lock(&fi->lock);
1822 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1823 
1824 	/* We always allow events, track them separately from the sccb ints */
1825 	if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING)
1826 		set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1827 
1828 	/*
1829 	 * Early versions of the QEMU s390 bios will inject several
1830 	 * service interrupts after another without handling a
1831 	 * condition code indicating busy.
1832 	 * We will silently ignore those superfluous sccb values.
1833 	 * A future version of QEMU will take care of serialization
1834 	 * of servc requests
1835 	 */
1836 	if (fi->srv_signal.ext_params & SCCB_MASK)
1837 		goto out;
1838 	fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1839 	set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1840 out:
1841 	spin_unlock(&fi->lock);
1842 	kfree(inti);
1843 	return 0;
1844 }
1845 
__inject_virtio(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1846 static int __inject_virtio(struct kvm *kvm,
1847 			    struct kvm_s390_interrupt_info *inti)
1848 {
1849 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1850 
1851 	kvm->stat.inject_virtio++;
1852 	spin_lock(&fi->lock);
1853 	if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1854 		spin_unlock(&fi->lock);
1855 		return -EBUSY;
1856 	}
1857 	fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1858 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1859 	set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1860 	spin_unlock(&fi->lock);
1861 	return 0;
1862 }
1863 
__inject_pfault_done(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1864 static int __inject_pfault_done(struct kvm *kvm,
1865 				 struct kvm_s390_interrupt_info *inti)
1866 {
1867 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1868 
1869 	kvm->stat.inject_pfault_done++;
1870 	spin_lock(&fi->lock);
1871 	if (fi->counters[FIRQ_CNTR_PFAULT] >=
1872 		(ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1873 		spin_unlock(&fi->lock);
1874 		return -EBUSY;
1875 	}
1876 	fi->counters[FIRQ_CNTR_PFAULT] += 1;
1877 	list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1878 	set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1879 	spin_unlock(&fi->lock);
1880 	return 0;
1881 }
1882 
1883 #define CR_PENDING_SUBCLASS 28
__inject_float_mchk(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1884 static int __inject_float_mchk(struct kvm *kvm,
1885 				struct kvm_s390_interrupt_info *inti)
1886 {
1887 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1888 
1889 	kvm->stat.inject_float_mchk++;
1890 	spin_lock(&fi->lock);
1891 	fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1892 	fi->mchk.mcic |= inti->mchk.mcic;
1893 	set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1894 	spin_unlock(&fi->lock);
1895 	kfree(inti);
1896 	return 0;
1897 }
1898 
__inject_io(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1899 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1900 {
1901 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1902 	struct kvm_s390_float_interrupt *fi;
1903 	struct list_head *list;
1904 	int isc;
1905 
1906 	kvm->stat.inject_io++;
1907 	isc = int_word_to_isc(inti->io.io_int_word);
1908 
1909 	/*
1910 	 * We do not use the lock checking variant as this is just a
1911 	 * performance optimization and we do not hold the lock here.
1912 	 * This is ok as the code will pick interrupts from both "lists"
1913 	 * for delivery.
1914 	 */
1915 	if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1916 		VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1917 		gisa_set_ipm_gisc(gi->origin, isc);
1918 		kfree(inti);
1919 		return 0;
1920 	}
1921 
1922 	fi = &kvm->arch.float_int;
1923 	spin_lock(&fi->lock);
1924 	if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1925 		spin_unlock(&fi->lock);
1926 		return -EBUSY;
1927 	}
1928 	fi->counters[FIRQ_CNTR_IO] += 1;
1929 
1930 	if (inti->type & KVM_S390_INT_IO_AI_MASK)
1931 		VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1932 	else
1933 		VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1934 			inti->io.subchannel_id >> 8,
1935 			inti->io.subchannel_id >> 1 & 0x3,
1936 			inti->io.subchannel_nr);
1937 	list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1938 	list_add_tail(&inti->list, list);
1939 	set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1940 	spin_unlock(&fi->lock);
1941 	return 0;
1942 }
1943 
1944 /*
1945  * Find a destination VCPU for a floating irq and kick it.
1946  */
__floating_irq_kick(struct kvm * kvm,u64 type)1947 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1948 {
1949 	struct kvm_vcpu *dst_vcpu;
1950 	int sigcpu, online_vcpus, nr_tries = 0;
1951 
1952 	online_vcpus = atomic_read(&kvm->online_vcpus);
1953 	if (!online_vcpus)
1954 		return;
1955 
1956 	/* find idle VCPUs first, then round robin */
1957 	sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1958 	if (sigcpu == online_vcpus) {
1959 		do {
1960 			sigcpu = kvm->arch.float_int.next_rr_cpu++;
1961 			kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1962 			/* avoid endless loops if all vcpus are stopped */
1963 			if (nr_tries++ >= online_vcpus)
1964 				return;
1965 		} while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1966 	}
1967 	dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1968 
1969 	/* make the VCPU drop out of the SIE, or wake it up if sleeping */
1970 	switch (type) {
1971 	case KVM_S390_MCHK:
1972 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1973 		break;
1974 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1975 		if (!(type & KVM_S390_INT_IO_AI_MASK &&
1976 		      kvm->arch.gisa_int.origin) ||
1977 		      kvm_s390_pv_cpu_get_handle(dst_vcpu))
1978 			kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1979 		break;
1980 	default:
1981 		kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1982 		break;
1983 	}
1984 	kvm_s390_vcpu_wakeup(dst_vcpu);
1985 }
1986 
__inject_vm(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)1987 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1988 {
1989 	u64 type = READ_ONCE(inti->type);
1990 	int rc;
1991 
1992 	switch (type) {
1993 	case KVM_S390_MCHK:
1994 		rc = __inject_float_mchk(kvm, inti);
1995 		break;
1996 	case KVM_S390_INT_VIRTIO:
1997 		rc = __inject_virtio(kvm, inti);
1998 		break;
1999 	case KVM_S390_INT_SERVICE:
2000 		rc = __inject_service(kvm, inti);
2001 		break;
2002 	case KVM_S390_INT_PFAULT_DONE:
2003 		rc = __inject_pfault_done(kvm, inti);
2004 		break;
2005 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2006 		rc = __inject_io(kvm, inti);
2007 		break;
2008 	default:
2009 		rc = -EINVAL;
2010 	}
2011 	if (rc)
2012 		return rc;
2013 
2014 	__floating_irq_kick(kvm, type);
2015 	return 0;
2016 }
2017 
kvm_s390_inject_vm(struct kvm * kvm,struct kvm_s390_interrupt * s390int)2018 int kvm_s390_inject_vm(struct kvm *kvm,
2019 		       struct kvm_s390_interrupt *s390int)
2020 {
2021 	struct kvm_s390_interrupt_info *inti;
2022 	int rc;
2023 
2024 	inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2025 	if (!inti)
2026 		return -ENOMEM;
2027 
2028 	inti->type = s390int->type;
2029 	switch (inti->type) {
2030 	case KVM_S390_INT_VIRTIO:
2031 		VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
2032 			 s390int->parm, s390int->parm64);
2033 		inti->ext.ext_params = s390int->parm;
2034 		inti->ext.ext_params2 = s390int->parm64;
2035 		break;
2036 	case KVM_S390_INT_SERVICE:
2037 		VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
2038 		inti->ext.ext_params = s390int->parm;
2039 		break;
2040 	case KVM_S390_INT_PFAULT_DONE:
2041 		inti->ext.ext_params2 = s390int->parm64;
2042 		break;
2043 	case KVM_S390_MCHK:
2044 		VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
2045 			 s390int->parm64);
2046 		inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
2047 		inti->mchk.mcic = s390int->parm64;
2048 		break;
2049 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2050 		inti->io.subchannel_id = s390int->parm >> 16;
2051 		inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
2052 		inti->io.io_int_parm = s390int->parm64 >> 32;
2053 		inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
2054 		break;
2055 	default:
2056 		kfree(inti);
2057 		return -EINVAL;
2058 	}
2059 	trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
2060 				 2);
2061 
2062 	rc = __inject_vm(kvm, inti);
2063 	if (rc)
2064 		kfree(inti);
2065 	return rc;
2066 }
2067 
kvm_s390_reinject_io_int(struct kvm * kvm,struct kvm_s390_interrupt_info * inti)2068 int kvm_s390_reinject_io_int(struct kvm *kvm,
2069 			      struct kvm_s390_interrupt_info *inti)
2070 {
2071 	return __inject_vm(kvm, inti);
2072 }
2073 
s390int_to_s390irq(struct kvm_s390_interrupt * s390int,struct kvm_s390_irq * irq)2074 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
2075 		       struct kvm_s390_irq *irq)
2076 {
2077 	irq->type = s390int->type;
2078 	switch (irq->type) {
2079 	case KVM_S390_PROGRAM_INT:
2080 		if (s390int->parm & 0xffff0000)
2081 			return -EINVAL;
2082 		irq->u.pgm.code = s390int->parm;
2083 		break;
2084 	case KVM_S390_SIGP_SET_PREFIX:
2085 		irq->u.prefix.address = s390int->parm;
2086 		break;
2087 	case KVM_S390_SIGP_STOP:
2088 		irq->u.stop.flags = s390int->parm;
2089 		break;
2090 	case KVM_S390_INT_EXTERNAL_CALL:
2091 		if (s390int->parm & 0xffff0000)
2092 			return -EINVAL;
2093 		irq->u.extcall.code = s390int->parm;
2094 		break;
2095 	case KVM_S390_INT_EMERGENCY:
2096 		if (s390int->parm & 0xffff0000)
2097 			return -EINVAL;
2098 		irq->u.emerg.code = s390int->parm;
2099 		break;
2100 	case KVM_S390_MCHK:
2101 		irq->u.mchk.mcic = s390int->parm64;
2102 		break;
2103 	case KVM_S390_INT_PFAULT_INIT:
2104 		irq->u.ext.ext_params = s390int->parm;
2105 		irq->u.ext.ext_params2 = s390int->parm64;
2106 		break;
2107 	case KVM_S390_RESTART:
2108 	case KVM_S390_INT_CLOCK_COMP:
2109 	case KVM_S390_INT_CPU_TIMER:
2110 		break;
2111 	default:
2112 		return -EINVAL;
2113 	}
2114 	return 0;
2115 }
2116 
kvm_s390_is_stop_irq_pending(struct kvm_vcpu * vcpu)2117 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
2118 {
2119 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2120 
2121 	return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2122 }
2123 
kvm_s390_is_restart_irq_pending(struct kvm_vcpu * vcpu)2124 int kvm_s390_is_restart_irq_pending(struct kvm_vcpu *vcpu)
2125 {
2126 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2127 
2128 	return test_bit(IRQ_PEND_RESTART, &li->pending_irqs);
2129 }
2130 
kvm_s390_clear_stop_irq(struct kvm_vcpu * vcpu)2131 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
2132 {
2133 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2134 
2135 	spin_lock(&li->lock);
2136 	li->irq.stop.flags = 0;
2137 	clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2138 	spin_unlock(&li->lock);
2139 }
2140 
do_inject_vcpu(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)2141 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2142 {
2143 	int rc;
2144 
2145 	switch (irq->type) {
2146 	case KVM_S390_PROGRAM_INT:
2147 		rc = __inject_prog(vcpu, irq);
2148 		break;
2149 	case KVM_S390_SIGP_SET_PREFIX:
2150 		rc = __inject_set_prefix(vcpu, irq);
2151 		break;
2152 	case KVM_S390_SIGP_STOP:
2153 		rc = __inject_sigp_stop(vcpu, irq);
2154 		break;
2155 	case KVM_S390_RESTART:
2156 		rc = __inject_sigp_restart(vcpu);
2157 		break;
2158 	case KVM_S390_INT_CLOCK_COMP:
2159 		rc = __inject_ckc(vcpu);
2160 		break;
2161 	case KVM_S390_INT_CPU_TIMER:
2162 		rc = __inject_cpu_timer(vcpu);
2163 		break;
2164 	case KVM_S390_INT_EXTERNAL_CALL:
2165 		rc = __inject_extcall(vcpu, irq);
2166 		break;
2167 	case KVM_S390_INT_EMERGENCY:
2168 		rc = __inject_sigp_emergency(vcpu, irq);
2169 		break;
2170 	case KVM_S390_MCHK:
2171 		rc = __inject_mchk(vcpu, irq);
2172 		break;
2173 	case KVM_S390_INT_PFAULT_INIT:
2174 		rc = __inject_pfault_init(vcpu, irq);
2175 		break;
2176 	case KVM_S390_INT_VIRTIO:
2177 	case KVM_S390_INT_SERVICE:
2178 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2179 	default:
2180 		rc = -EINVAL;
2181 	}
2182 
2183 	return rc;
2184 }
2185 
kvm_s390_inject_vcpu(struct kvm_vcpu * vcpu,struct kvm_s390_irq * irq)2186 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2187 {
2188 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2189 	int rc;
2190 
2191 	spin_lock(&li->lock);
2192 	rc = do_inject_vcpu(vcpu, irq);
2193 	spin_unlock(&li->lock);
2194 	if (!rc)
2195 		kvm_s390_vcpu_wakeup(vcpu);
2196 	return rc;
2197 }
2198 
clear_irq_list(struct list_head * _list)2199 static inline void clear_irq_list(struct list_head *_list)
2200 {
2201 	struct kvm_s390_interrupt_info *inti, *n;
2202 
2203 	list_for_each_entry_safe(inti, n, _list, list) {
2204 		list_del(&inti->list);
2205 		kfree(inti);
2206 	}
2207 }
2208 
inti_to_irq(struct kvm_s390_interrupt_info * inti,struct kvm_s390_irq * irq)2209 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2210 		       struct kvm_s390_irq *irq)
2211 {
2212 	irq->type = inti->type;
2213 	switch (inti->type) {
2214 	case KVM_S390_INT_PFAULT_INIT:
2215 	case KVM_S390_INT_PFAULT_DONE:
2216 	case KVM_S390_INT_VIRTIO:
2217 		irq->u.ext = inti->ext;
2218 		break;
2219 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2220 		irq->u.io = inti->io;
2221 		break;
2222 	}
2223 }
2224 
kvm_s390_clear_float_irqs(struct kvm * kvm)2225 void kvm_s390_clear_float_irqs(struct kvm *kvm)
2226 {
2227 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2228 	int i;
2229 
2230 	mutex_lock(&kvm->lock);
2231 	if (!kvm_s390_pv_is_protected(kvm))
2232 		fi->masked_irqs = 0;
2233 	mutex_unlock(&kvm->lock);
2234 	spin_lock(&fi->lock);
2235 	fi->pending_irqs = 0;
2236 	memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2237 	memset(&fi->mchk, 0, sizeof(fi->mchk));
2238 	for (i = 0; i < FIRQ_LIST_COUNT; i++)
2239 		clear_irq_list(&fi->lists[i]);
2240 	for (i = 0; i < FIRQ_MAX_COUNT; i++)
2241 		fi->counters[i] = 0;
2242 	spin_unlock(&fi->lock);
2243 	kvm_s390_gisa_clear(kvm);
2244 };
2245 
get_all_floating_irqs(struct kvm * kvm,u8 __user * usrbuf,u64 len)2246 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2247 {
2248 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2249 	struct kvm_s390_interrupt_info *inti;
2250 	struct kvm_s390_float_interrupt *fi;
2251 	struct kvm_s390_irq *buf;
2252 	struct kvm_s390_irq *irq;
2253 	int max_irqs;
2254 	int ret = 0;
2255 	int n = 0;
2256 	int i;
2257 
2258 	if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2259 		return -EINVAL;
2260 
2261 	/*
2262 	 * We are already using -ENOMEM to signal
2263 	 * userspace it may retry with a bigger buffer,
2264 	 * so we need to use something else for this case
2265 	 */
2266 	buf = vzalloc(len);
2267 	if (!buf)
2268 		return -ENOBUFS;
2269 
2270 	max_irqs = len / sizeof(struct kvm_s390_irq);
2271 
2272 	if (gi->origin && gisa_get_ipm(gi->origin)) {
2273 		for (i = 0; i <= MAX_ISC; i++) {
2274 			if (n == max_irqs) {
2275 				/* signal userspace to try again */
2276 				ret = -ENOMEM;
2277 				goto out_nolock;
2278 			}
2279 			if (gisa_tac_ipm_gisc(gi->origin, i)) {
2280 				irq = (struct kvm_s390_irq *) &buf[n];
2281 				irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2282 				irq->u.io.io_int_word = isc_to_int_word(i);
2283 				n++;
2284 			}
2285 		}
2286 	}
2287 	fi = &kvm->arch.float_int;
2288 	spin_lock(&fi->lock);
2289 	for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2290 		list_for_each_entry(inti, &fi->lists[i], list) {
2291 			if (n == max_irqs) {
2292 				/* signal userspace to try again */
2293 				ret = -ENOMEM;
2294 				goto out;
2295 			}
2296 			inti_to_irq(inti, &buf[n]);
2297 			n++;
2298 		}
2299 	}
2300 	if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) ||
2301 	    test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) {
2302 		if (n == max_irqs) {
2303 			/* signal userspace to try again */
2304 			ret = -ENOMEM;
2305 			goto out;
2306 		}
2307 		irq = (struct kvm_s390_irq *) &buf[n];
2308 		irq->type = KVM_S390_INT_SERVICE;
2309 		irq->u.ext = fi->srv_signal;
2310 		n++;
2311 	}
2312 	if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2313 		if (n == max_irqs) {
2314 				/* signal userspace to try again */
2315 				ret = -ENOMEM;
2316 				goto out;
2317 		}
2318 		irq = (struct kvm_s390_irq *) &buf[n];
2319 		irq->type = KVM_S390_MCHK;
2320 		irq->u.mchk = fi->mchk;
2321 		n++;
2322 }
2323 
2324 out:
2325 	spin_unlock(&fi->lock);
2326 out_nolock:
2327 	if (!ret && n > 0) {
2328 		if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2329 			ret = -EFAULT;
2330 	}
2331 	vfree(buf);
2332 
2333 	return ret < 0 ? ret : n;
2334 }
2335 
flic_ais_mode_get_all(struct kvm * kvm,struct kvm_device_attr * attr)2336 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2337 {
2338 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2339 	struct kvm_s390_ais_all ais;
2340 
2341 	if (attr->attr < sizeof(ais))
2342 		return -EINVAL;
2343 
2344 	if (!test_kvm_facility(kvm, 72))
2345 		return -EOPNOTSUPP;
2346 
2347 	mutex_lock(&fi->ais_lock);
2348 	ais.simm = fi->simm;
2349 	ais.nimm = fi->nimm;
2350 	mutex_unlock(&fi->ais_lock);
2351 
2352 	if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2353 		return -EFAULT;
2354 
2355 	return 0;
2356 }
2357 
flic_get_attr(struct kvm_device * dev,struct kvm_device_attr * attr)2358 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2359 {
2360 	int r;
2361 
2362 	switch (attr->group) {
2363 	case KVM_DEV_FLIC_GET_ALL_IRQS:
2364 		r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2365 					  attr->attr);
2366 		break;
2367 	case KVM_DEV_FLIC_AISM_ALL:
2368 		r = flic_ais_mode_get_all(dev->kvm, attr);
2369 		break;
2370 	default:
2371 		r = -EINVAL;
2372 	}
2373 
2374 	return r;
2375 }
2376 
copy_irq_from_user(struct kvm_s390_interrupt_info * inti,u64 addr)2377 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2378 				     u64 addr)
2379 {
2380 	struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2381 	void *target = NULL;
2382 	void __user *source;
2383 	u64 size;
2384 
2385 	if (get_user(inti->type, (u64 __user *)addr))
2386 		return -EFAULT;
2387 
2388 	switch (inti->type) {
2389 	case KVM_S390_INT_PFAULT_INIT:
2390 	case KVM_S390_INT_PFAULT_DONE:
2391 	case KVM_S390_INT_VIRTIO:
2392 	case KVM_S390_INT_SERVICE:
2393 		target = (void *) &inti->ext;
2394 		source = &uptr->u.ext;
2395 		size = sizeof(inti->ext);
2396 		break;
2397 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2398 		target = (void *) &inti->io;
2399 		source = &uptr->u.io;
2400 		size = sizeof(inti->io);
2401 		break;
2402 	case KVM_S390_MCHK:
2403 		target = (void *) &inti->mchk;
2404 		source = &uptr->u.mchk;
2405 		size = sizeof(inti->mchk);
2406 		break;
2407 	default:
2408 		return -EINVAL;
2409 	}
2410 
2411 	if (copy_from_user(target, source, size))
2412 		return -EFAULT;
2413 
2414 	return 0;
2415 }
2416 
enqueue_floating_irq(struct kvm_device * dev,struct kvm_device_attr * attr)2417 static int enqueue_floating_irq(struct kvm_device *dev,
2418 				struct kvm_device_attr *attr)
2419 {
2420 	struct kvm_s390_interrupt_info *inti = NULL;
2421 	int r = 0;
2422 	int len = attr->attr;
2423 
2424 	if (len % sizeof(struct kvm_s390_irq) != 0)
2425 		return -EINVAL;
2426 	else if (len > KVM_S390_FLIC_MAX_BUFFER)
2427 		return -EINVAL;
2428 
2429 	while (len >= sizeof(struct kvm_s390_irq)) {
2430 		inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2431 		if (!inti)
2432 			return -ENOMEM;
2433 
2434 		r = copy_irq_from_user(inti, attr->addr);
2435 		if (r) {
2436 			kfree(inti);
2437 			return r;
2438 		}
2439 		r = __inject_vm(dev->kvm, inti);
2440 		if (r) {
2441 			kfree(inti);
2442 			return r;
2443 		}
2444 		len -= sizeof(struct kvm_s390_irq);
2445 		attr->addr += sizeof(struct kvm_s390_irq);
2446 	}
2447 
2448 	return r;
2449 }
2450 
get_io_adapter(struct kvm * kvm,unsigned int id)2451 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2452 {
2453 	if (id >= MAX_S390_IO_ADAPTERS)
2454 		return NULL;
2455 	id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2456 	return kvm->arch.adapters[id];
2457 }
2458 
register_io_adapter(struct kvm_device * dev,struct kvm_device_attr * attr)2459 static int register_io_adapter(struct kvm_device *dev,
2460 			       struct kvm_device_attr *attr)
2461 {
2462 	struct s390_io_adapter *adapter;
2463 	struct kvm_s390_io_adapter adapter_info;
2464 
2465 	if (copy_from_user(&adapter_info,
2466 			   (void __user *)attr->addr, sizeof(adapter_info)))
2467 		return -EFAULT;
2468 
2469 	if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2470 		return -EINVAL;
2471 
2472 	adapter_info.id = array_index_nospec(adapter_info.id,
2473 					     MAX_S390_IO_ADAPTERS);
2474 
2475 	if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2476 		return -EINVAL;
2477 
2478 	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL_ACCOUNT);
2479 	if (!adapter)
2480 		return -ENOMEM;
2481 
2482 	adapter->id = adapter_info.id;
2483 	adapter->isc = adapter_info.isc;
2484 	adapter->maskable = adapter_info.maskable;
2485 	adapter->masked = false;
2486 	adapter->swap = adapter_info.swap;
2487 	adapter->suppressible = (adapter_info.flags) &
2488 				KVM_S390_ADAPTER_SUPPRESSIBLE;
2489 	dev->kvm->arch.adapters[adapter->id] = adapter;
2490 
2491 	return 0;
2492 }
2493 
kvm_s390_mask_adapter(struct kvm * kvm,unsigned int id,bool masked)2494 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2495 {
2496 	int ret;
2497 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2498 
2499 	if (!adapter || !adapter->maskable)
2500 		return -EINVAL;
2501 	ret = adapter->masked;
2502 	adapter->masked = masked;
2503 	return ret;
2504 }
2505 
kvm_s390_destroy_adapters(struct kvm * kvm)2506 void kvm_s390_destroy_adapters(struct kvm *kvm)
2507 {
2508 	int i;
2509 
2510 	for (i = 0; i < MAX_S390_IO_ADAPTERS; i++)
2511 		kfree(kvm->arch.adapters[i]);
2512 }
2513 
modify_io_adapter(struct kvm_device * dev,struct kvm_device_attr * attr)2514 static int modify_io_adapter(struct kvm_device *dev,
2515 			     struct kvm_device_attr *attr)
2516 {
2517 	struct kvm_s390_io_adapter_req req;
2518 	struct s390_io_adapter *adapter;
2519 	int ret;
2520 
2521 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2522 		return -EFAULT;
2523 
2524 	adapter = get_io_adapter(dev->kvm, req.id);
2525 	if (!adapter)
2526 		return -EINVAL;
2527 	switch (req.type) {
2528 	case KVM_S390_IO_ADAPTER_MASK:
2529 		ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2530 		if (ret > 0)
2531 			ret = 0;
2532 		break;
2533 	/*
2534 	 * The following operations are no longer needed and therefore no-ops.
2535 	 * The gpa to hva translation is done when an IRQ route is set up. The
2536 	 * set_irq code uses get_user_pages_remote() to do the actual write.
2537 	 */
2538 	case KVM_S390_IO_ADAPTER_MAP:
2539 	case KVM_S390_IO_ADAPTER_UNMAP:
2540 		ret = 0;
2541 		break;
2542 	default:
2543 		ret = -EINVAL;
2544 	}
2545 
2546 	return ret;
2547 }
2548 
clear_io_irq(struct kvm * kvm,struct kvm_device_attr * attr)2549 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2550 
2551 {
2552 	const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2553 	u32 schid;
2554 
2555 	if (attr->flags)
2556 		return -EINVAL;
2557 	if (attr->attr != sizeof(schid))
2558 		return -EINVAL;
2559 	if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2560 		return -EFAULT;
2561 	if (!schid)
2562 		return -EINVAL;
2563 	kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2564 	/*
2565 	 * If userspace is conforming to the architecture, we can have at most
2566 	 * one pending I/O interrupt per subchannel, so this is effectively a
2567 	 * clear all.
2568 	 */
2569 	return 0;
2570 }
2571 
modify_ais_mode(struct kvm * kvm,struct kvm_device_attr * attr)2572 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2573 {
2574 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2575 	struct kvm_s390_ais_req req;
2576 	int ret = 0;
2577 
2578 	if (!test_kvm_facility(kvm, 72))
2579 		return -EOPNOTSUPP;
2580 
2581 	if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2582 		return -EFAULT;
2583 
2584 	if (req.isc > MAX_ISC)
2585 		return -EINVAL;
2586 
2587 	trace_kvm_s390_modify_ais_mode(req.isc,
2588 				       (fi->simm & AIS_MODE_MASK(req.isc)) ?
2589 				       (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2590 				       2 : KVM_S390_AIS_MODE_SINGLE :
2591 				       KVM_S390_AIS_MODE_ALL, req.mode);
2592 
2593 	mutex_lock(&fi->ais_lock);
2594 	switch (req.mode) {
2595 	case KVM_S390_AIS_MODE_ALL:
2596 		fi->simm &= ~AIS_MODE_MASK(req.isc);
2597 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2598 		break;
2599 	case KVM_S390_AIS_MODE_SINGLE:
2600 		fi->simm |= AIS_MODE_MASK(req.isc);
2601 		fi->nimm &= ~AIS_MODE_MASK(req.isc);
2602 		break;
2603 	default:
2604 		ret = -EINVAL;
2605 	}
2606 	mutex_unlock(&fi->ais_lock);
2607 
2608 	return ret;
2609 }
2610 
kvm_s390_inject_airq(struct kvm * kvm,struct s390_io_adapter * adapter)2611 static int kvm_s390_inject_airq(struct kvm *kvm,
2612 				struct s390_io_adapter *adapter)
2613 {
2614 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2615 	struct kvm_s390_interrupt s390int = {
2616 		.type = KVM_S390_INT_IO(1, 0, 0, 0),
2617 		.parm = 0,
2618 		.parm64 = isc_to_int_word(adapter->isc),
2619 	};
2620 	int ret = 0;
2621 
2622 	if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2623 		return kvm_s390_inject_vm(kvm, &s390int);
2624 
2625 	mutex_lock(&fi->ais_lock);
2626 	if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2627 		trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2628 		goto out;
2629 	}
2630 
2631 	ret = kvm_s390_inject_vm(kvm, &s390int);
2632 	if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2633 		fi->nimm |= AIS_MODE_MASK(adapter->isc);
2634 		trace_kvm_s390_modify_ais_mode(adapter->isc,
2635 					       KVM_S390_AIS_MODE_SINGLE, 2);
2636 	}
2637 out:
2638 	mutex_unlock(&fi->ais_lock);
2639 	return ret;
2640 }
2641 
flic_inject_airq(struct kvm * kvm,struct kvm_device_attr * attr)2642 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2643 {
2644 	unsigned int id = attr->attr;
2645 	struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2646 
2647 	if (!adapter)
2648 		return -EINVAL;
2649 
2650 	return kvm_s390_inject_airq(kvm, adapter);
2651 }
2652 
flic_ais_mode_set_all(struct kvm * kvm,struct kvm_device_attr * attr)2653 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2654 {
2655 	struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2656 	struct kvm_s390_ais_all ais;
2657 
2658 	if (!test_kvm_facility(kvm, 72))
2659 		return -EOPNOTSUPP;
2660 
2661 	if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2662 		return -EFAULT;
2663 
2664 	mutex_lock(&fi->ais_lock);
2665 	fi->simm = ais.simm;
2666 	fi->nimm = ais.nimm;
2667 	mutex_unlock(&fi->ais_lock);
2668 
2669 	return 0;
2670 }
2671 
flic_set_attr(struct kvm_device * dev,struct kvm_device_attr * attr)2672 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2673 {
2674 	int r = 0;
2675 	unsigned long i;
2676 	struct kvm_vcpu *vcpu;
2677 
2678 	switch (attr->group) {
2679 	case KVM_DEV_FLIC_ENQUEUE:
2680 		r = enqueue_floating_irq(dev, attr);
2681 		break;
2682 	case KVM_DEV_FLIC_CLEAR_IRQS:
2683 		kvm_s390_clear_float_irqs(dev->kvm);
2684 		break;
2685 	case KVM_DEV_FLIC_APF_ENABLE:
2686 		dev->kvm->arch.gmap->pfault_enabled = 1;
2687 		break;
2688 	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
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, 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;
2898 
2899 	switch (ue->type) {
2900 	/* we store the userspace addresses instead of the guest addresses */
2901 	case KVM_IRQ_ROUTING_S390_ADAPTER:
2902 		e->set = set_adapter_int;
2903 		uaddr =  gmap_translate(kvm->arch.gmap, ue->u.adapter.summary_addr);
2904 		if (uaddr == -EFAULT)
2905 			return -EFAULT;
2906 		e->adapter.summary_addr = uaddr;
2907 		uaddr =  gmap_translate(kvm->arch.gmap, ue->u.adapter.ind_addr);
2908 		if (uaddr == -EFAULT)
2909 			return -EFAULT;
2910 		e->adapter.ind_addr = uaddr;
2911 		e->adapter.summary_offset = ue->u.adapter.summary_offset;
2912 		e->adapter.ind_offset = ue->u.adapter.ind_offset;
2913 		e->adapter.adapter_id = ue->u.adapter.adapter_id;
2914 		return 0;
2915 	default:
2916 		return -EINVAL;
2917 	}
2918 }
2919 
kvm_set_msi(struct kvm_kernel_irq_routing_entry * e,struct kvm * kvm,int irq_source_id,int level,bool line_status)2920 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2921 		int irq_source_id, int level, bool line_status)
2922 {
2923 	return -EINVAL;
2924 }
2925 
kvm_s390_set_irq_state(struct kvm_vcpu * vcpu,void __user * irqstate,int len)2926 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2927 {
2928 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2929 	struct kvm_s390_irq *buf;
2930 	int r = 0;
2931 	int n;
2932 
2933 	buf = vmalloc(len);
2934 	if (!buf)
2935 		return -ENOMEM;
2936 
2937 	if (copy_from_user((void *) buf, irqstate, len)) {
2938 		r = -EFAULT;
2939 		goto out_free;
2940 	}
2941 
2942 	/*
2943 	 * Don't allow setting the interrupt state
2944 	 * when there are already interrupts pending
2945 	 */
2946 	spin_lock(&li->lock);
2947 	if (li->pending_irqs) {
2948 		r = -EBUSY;
2949 		goto out_unlock;
2950 	}
2951 
2952 	for (n = 0; n < len / sizeof(*buf); n++) {
2953 		r = do_inject_vcpu(vcpu, &buf[n]);
2954 		if (r)
2955 			break;
2956 	}
2957 
2958 out_unlock:
2959 	spin_unlock(&li->lock);
2960 out_free:
2961 	vfree(buf);
2962 
2963 	return r;
2964 }
2965 
store_local_irq(struct kvm_s390_local_interrupt * li,struct kvm_s390_irq * irq,unsigned long irq_type)2966 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2967 			    struct kvm_s390_irq *irq,
2968 			    unsigned long irq_type)
2969 {
2970 	switch (irq_type) {
2971 	case IRQ_PEND_MCHK_EX:
2972 	case IRQ_PEND_MCHK_REP:
2973 		irq->type = KVM_S390_MCHK;
2974 		irq->u.mchk = li->irq.mchk;
2975 		break;
2976 	case IRQ_PEND_PROG:
2977 		irq->type = KVM_S390_PROGRAM_INT;
2978 		irq->u.pgm = li->irq.pgm;
2979 		break;
2980 	case IRQ_PEND_PFAULT_INIT:
2981 		irq->type = KVM_S390_INT_PFAULT_INIT;
2982 		irq->u.ext = li->irq.ext;
2983 		break;
2984 	case IRQ_PEND_EXT_EXTERNAL:
2985 		irq->type = KVM_S390_INT_EXTERNAL_CALL;
2986 		irq->u.extcall = li->irq.extcall;
2987 		break;
2988 	case IRQ_PEND_EXT_CLOCK_COMP:
2989 		irq->type = KVM_S390_INT_CLOCK_COMP;
2990 		break;
2991 	case IRQ_PEND_EXT_CPU_TIMER:
2992 		irq->type = KVM_S390_INT_CPU_TIMER;
2993 		break;
2994 	case IRQ_PEND_SIGP_STOP:
2995 		irq->type = KVM_S390_SIGP_STOP;
2996 		irq->u.stop = li->irq.stop;
2997 		break;
2998 	case IRQ_PEND_RESTART:
2999 		irq->type = KVM_S390_RESTART;
3000 		break;
3001 	case IRQ_PEND_SET_PREFIX:
3002 		irq->type = KVM_S390_SIGP_SET_PREFIX;
3003 		irq->u.prefix = li->irq.prefix;
3004 		break;
3005 	}
3006 }
3007 
kvm_s390_get_irq_state(struct kvm_vcpu * vcpu,__u8 __user * buf,int len)3008 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
3009 {
3010 	int scn;
3011 	DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
3012 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
3013 	unsigned long pending_irqs;
3014 	struct kvm_s390_irq irq;
3015 	unsigned long irq_type;
3016 	int cpuaddr;
3017 	int n = 0;
3018 
3019 	spin_lock(&li->lock);
3020 	pending_irqs = li->pending_irqs;
3021 	memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
3022 	       sizeof(sigp_emerg_pending));
3023 	spin_unlock(&li->lock);
3024 
3025 	for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
3026 		memset(&irq, 0, sizeof(irq));
3027 		if (irq_type == IRQ_PEND_EXT_EMERGENCY)
3028 			continue;
3029 		if (n + sizeof(irq) > len)
3030 			return -ENOBUFS;
3031 		store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
3032 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3033 			return -EFAULT;
3034 		n += sizeof(irq);
3035 	}
3036 
3037 	if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
3038 		for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
3039 			memset(&irq, 0, sizeof(irq));
3040 			if (n + sizeof(irq) > len)
3041 				return -ENOBUFS;
3042 			irq.type = KVM_S390_INT_EMERGENCY;
3043 			irq.u.emerg.code = cpuaddr;
3044 			if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3045 				return -EFAULT;
3046 			n += sizeof(irq);
3047 		}
3048 	}
3049 
3050 	if (sca_ext_call_pending(vcpu, &scn)) {
3051 		if (n + sizeof(irq) > len)
3052 			return -ENOBUFS;
3053 		memset(&irq, 0, sizeof(irq));
3054 		irq.type = KVM_S390_INT_EXTERNAL_CALL;
3055 		irq.u.extcall.code = scn;
3056 		if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3057 			return -EFAULT;
3058 		n += sizeof(irq);
3059 	}
3060 
3061 	return n;
3062 }
3063 
__airqs_kick_single_vcpu(struct kvm * kvm,u8 deliverable_mask)3064 static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
3065 {
3066 	int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus);
3067 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3068 	struct kvm_vcpu *vcpu;
3069 	u8 vcpu_isc_mask;
3070 
3071 	for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) {
3072 		vcpu = kvm_get_vcpu(kvm, vcpu_idx);
3073 		if (psw_ioint_disabled(vcpu))
3074 			continue;
3075 		vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
3076 		if (deliverable_mask & vcpu_isc_mask) {
3077 			/* lately kicked but not yet running */
3078 			if (test_and_set_bit(vcpu_idx, gi->kicked_mask))
3079 				return;
3080 			kvm_s390_vcpu_wakeup(vcpu);
3081 			return;
3082 		}
3083 	}
3084 }
3085 
gisa_vcpu_kicker(struct hrtimer * timer)3086 static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3087 {
3088 	struct kvm_s390_gisa_interrupt *gi =
3089 		container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3090 	struct kvm *kvm =
3091 		container_of(gi->origin, struct sie_page2, gisa)->kvm;
3092 	u8 pending_mask;
3093 
3094 	pending_mask = gisa_get_ipm_or_restore_iam(gi);
3095 	if (pending_mask) {
3096 		__airqs_kick_single_vcpu(kvm, pending_mask);
3097 		hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3098 		return HRTIMER_RESTART;
3099 	}
3100 
3101 	return HRTIMER_NORESTART;
3102 }
3103 
3104 #define NULL_GISA_ADDR 0x00000000UL
3105 #define NONE_GISA_ADDR 0x00000001UL
3106 #define GISA_ADDR_MASK 0xfffff000UL
3107 
process_gib_alert_list(void)3108 static void process_gib_alert_list(void)
3109 {
3110 	struct kvm_s390_gisa_interrupt *gi;
3111 	struct kvm_s390_gisa *gisa;
3112 	struct kvm *kvm;
3113 	u32 final, origin = 0UL;
3114 
3115 	do {
3116 		/*
3117 		 * If the NONE_GISA_ADDR is still stored in the alert list
3118 		 * origin, we will leave the outer loop. No further GISA has
3119 		 * been added to the alert list by millicode while processing
3120 		 * the current alert list.
3121 		 */
3122 		final = (origin & NONE_GISA_ADDR);
3123 		/*
3124 		 * Cut off the alert list and store the NONE_GISA_ADDR in the
3125 		 * alert list origin to avoid further GAL interruptions.
3126 		 * A new alert list can be build up by millicode in parallel
3127 		 * for guests not in the yet cut-off alert list. When in the
3128 		 * final loop, store the NULL_GISA_ADDR instead. This will re-
3129 		 * enable GAL interruptions on the host again.
3130 		 */
3131 		origin = xchg(&gib->alert_list_origin,
3132 			      (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3133 		/*
3134 		 * Loop through the just cut-off alert list and start the
3135 		 * gisa timers to kick idle vcpus to consume the pending
3136 		 * interruptions asap.
3137 		 */
3138 		while (origin & GISA_ADDR_MASK) {
3139 			gisa = (struct kvm_s390_gisa *)(u64)origin;
3140 			origin = gisa->next_alert;
3141 			gisa->next_alert = (u32)(u64)gisa;
3142 			kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3143 			gi = &kvm->arch.gisa_int;
3144 			if (hrtimer_active(&gi->timer))
3145 				hrtimer_cancel(&gi->timer);
3146 			hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3147 		}
3148 	} while (!final);
3149 
3150 }
3151 
kvm_s390_gisa_clear(struct kvm * kvm)3152 void kvm_s390_gisa_clear(struct kvm *kvm)
3153 {
3154 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3155 
3156 	if (!gi->origin)
3157 		return;
3158 	gisa_clear_ipm(gi->origin);
3159 	VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3160 }
3161 
kvm_s390_gisa_init(struct kvm * kvm)3162 void kvm_s390_gisa_init(struct kvm *kvm)
3163 {
3164 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3165 
3166 	if (!css_general_characteristics.aiv)
3167 		return;
3168 	gi->origin = &kvm->arch.sie_page2->gisa;
3169 	gi->alert.mask = 0;
3170 	spin_lock_init(&gi->alert.ref_lock);
3171 	gi->expires = 50 * 1000; /* 50 usec */
3172 	hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3173 	gi->timer.function = gisa_vcpu_kicker;
3174 	memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3175 	gi->origin->next_alert = (u32)(u64)gi->origin;
3176 	VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3177 }
3178 
kvm_s390_gisa_enable(struct kvm * kvm)3179 void kvm_s390_gisa_enable(struct kvm *kvm)
3180 {
3181 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3182 	struct kvm_vcpu *vcpu;
3183 	unsigned long i;
3184 	u32 gisa_desc;
3185 
3186 	if (gi->origin)
3187 		return;
3188 	kvm_s390_gisa_init(kvm);
3189 	gisa_desc = kvm_s390_get_gisa_desc(kvm);
3190 	if (!gisa_desc)
3191 		return;
3192 	kvm_for_each_vcpu(i, vcpu, kvm) {
3193 		mutex_lock(&vcpu->mutex);
3194 		vcpu->arch.sie_block->gd = gisa_desc;
3195 		vcpu->arch.sie_block->eca |= ECA_AIV;
3196 		VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3197 			   vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3198 		mutex_unlock(&vcpu->mutex);
3199 	}
3200 }
3201 
kvm_s390_gisa_destroy(struct kvm * kvm)3202 void kvm_s390_gisa_destroy(struct kvm *kvm)
3203 {
3204 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3205 	struct kvm_s390_gisa *gisa = gi->origin;
3206 
3207 	if (!gi->origin)
3208 		return;
3209 	if (gi->alert.mask)
3210 		KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
3211 			  kvm, gi->alert.mask);
3212 	while (gisa_in_alert_list(gi->origin))
3213 		cpu_relax();
3214 	hrtimer_cancel(&gi->timer);
3215 	gi->origin = NULL;
3216 	VM_EVENT(kvm, 3, "gisa 0x%pK destroyed", gisa);
3217 }
3218 
kvm_s390_gisa_disable(struct kvm * kvm)3219 void kvm_s390_gisa_disable(struct kvm *kvm)
3220 {
3221 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3222 	struct kvm_vcpu *vcpu;
3223 	unsigned long i;
3224 
3225 	if (!gi->origin)
3226 		return;
3227 	kvm_for_each_vcpu(i, vcpu, kvm) {
3228 		mutex_lock(&vcpu->mutex);
3229 		vcpu->arch.sie_block->eca &= ~ECA_AIV;
3230 		vcpu->arch.sie_block->gd = 0U;
3231 		mutex_unlock(&vcpu->mutex);
3232 		VCPU_EVENT(vcpu, 3, "AIV disabled for cpu %03u", vcpu->vcpu_id);
3233 	}
3234 	kvm_s390_gisa_destroy(kvm);
3235 }
3236 
3237 /**
3238  * kvm_s390_gisc_register - register a guest ISC
3239  *
3240  * @kvm:  the kernel vm to work with
3241  * @gisc: the guest interruption sub class to register
3242  *
3243  * The function extends the vm specific alert mask to use.
3244  * The effective IAM mask in the GISA is updated as well
3245  * in case the GISA is not part of the GIB alert list.
3246  * It will be updated latest when the IAM gets restored
3247  * by gisa_get_ipm_or_restore_iam().
3248  *
3249  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3250  *          has registered with the channel subsystem.
3251  *          -ENODEV in case the vm uses no GISA
3252  *          -ERANGE in case the guest ISC is invalid
3253  */
kvm_s390_gisc_register(struct kvm * kvm,u32 gisc)3254 int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3255 {
3256 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3257 
3258 	if (!gi->origin)
3259 		return -ENODEV;
3260 	if (gisc > MAX_ISC)
3261 		return -ERANGE;
3262 
3263 	spin_lock(&gi->alert.ref_lock);
3264 	gi->alert.ref_count[gisc]++;
3265 	if (gi->alert.ref_count[gisc] == 1) {
3266 		gi->alert.mask |= 0x80 >> gisc;
3267 		gisa_set_iam(gi->origin, gi->alert.mask);
3268 	}
3269 	spin_unlock(&gi->alert.ref_lock);
3270 
3271 	return gib->nisc;
3272 }
3273 EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3274 
3275 /**
3276  * kvm_s390_gisc_unregister - unregister a guest ISC
3277  *
3278  * @kvm:  the kernel vm to work with
3279  * @gisc: the guest interruption sub class to register
3280  *
3281  * The function reduces the vm specific alert mask to use.
3282  * The effective IAM mask in the GISA is updated as well
3283  * in case the GISA is not part of the GIB alert list.
3284  * It will be updated latest when the IAM gets restored
3285  * by gisa_get_ipm_or_restore_iam().
3286  *
3287  * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3288  *          has registered with the channel subsystem.
3289  *          -ENODEV in case the vm uses no GISA
3290  *          -ERANGE in case the guest ISC is invalid
3291  *          -EINVAL in case the guest ISC is not registered
3292  */
kvm_s390_gisc_unregister(struct kvm * kvm,u32 gisc)3293 int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3294 {
3295 	struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3296 	int rc = 0;
3297 
3298 	if (!gi->origin)
3299 		return -ENODEV;
3300 	if (gisc > MAX_ISC)
3301 		return -ERANGE;
3302 
3303 	spin_lock(&gi->alert.ref_lock);
3304 	if (gi->alert.ref_count[gisc] == 0) {
3305 		rc = -EINVAL;
3306 		goto out;
3307 	}
3308 	gi->alert.ref_count[gisc]--;
3309 	if (gi->alert.ref_count[gisc] == 0) {
3310 		gi->alert.mask &= ~(0x80 >> gisc);
3311 		gisa_set_iam(gi->origin, gi->alert.mask);
3312 	}
3313 out:
3314 	spin_unlock(&gi->alert.ref_lock);
3315 
3316 	return rc;
3317 }
3318 EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3319 
aen_host_forward(unsigned long si)3320 static void aen_host_forward(unsigned long si)
3321 {
3322 	struct kvm_s390_gisa_interrupt *gi;
3323 	struct zpci_gaite *gaite;
3324 	struct kvm *kvm;
3325 
3326 	gaite = (struct zpci_gaite *)aift->gait +
3327 		(si * sizeof(struct zpci_gaite));
3328 	if (gaite->count == 0)
3329 		return;
3330 	if (gaite->aisb != 0)
3331 		set_bit_inv(gaite->aisbo, phys_to_virt(gaite->aisb));
3332 
3333 	kvm = kvm_s390_pci_si_to_kvm(aift, si);
3334 	if (!kvm)
3335 		return;
3336 	gi = &kvm->arch.gisa_int;
3337 
3338 	if (!(gi->origin->g1.simm & AIS_MODE_MASK(gaite->gisc)) ||
3339 	    !(gi->origin->g1.nimm & AIS_MODE_MASK(gaite->gisc))) {
3340 		gisa_set_ipm_gisc(gi->origin, gaite->gisc);
3341 		if (hrtimer_active(&gi->timer))
3342 			hrtimer_cancel(&gi->timer);
3343 		hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3344 		kvm->stat.aen_forward++;
3345 	}
3346 }
3347 
aen_process_gait(u8 isc)3348 static void aen_process_gait(u8 isc)
3349 {
3350 	bool found = false, first = true;
3351 	union zpci_sic_iib iib = {{0}};
3352 	unsigned long si, flags;
3353 
3354 	spin_lock_irqsave(&aift->gait_lock, flags);
3355 
3356 	if (!aift->gait) {
3357 		spin_unlock_irqrestore(&aift->gait_lock, flags);
3358 		return;
3359 	}
3360 
3361 	for (si = 0;;) {
3362 		/* Scan adapter summary indicator bit vector */
3363 		si = airq_iv_scan(aift->sbv, si, airq_iv_end(aift->sbv));
3364 		if (si == -1UL) {
3365 			if (first || found) {
3366 				/* Re-enable interrupts. */
3367 				zpci_set_irq_ctrl(SIC_IRQ_MODE_SINGLE, isc,
3368 						  &iib);
3369 				first = found = false;
3370 			} else {
3371 				/* Interrupts on and all bits processed */
3372 				break;
3373 			}
3374 			found = false;
3375 			si = 0;
3376 			/* Scan again after re-enabling interrupts */
3377 			continue;
3378 		}
3379 		found = true;
3380 		aen_host_forward(si);
3381 	}
3382 
3383 	spin_unlock_irqrestore(&aift->gait_lock, flags);
3384 }
3385 
gib_alert_irq_handler(struct airq_struct * airq,struct tpi_info * tpi_info)3386 static void gib_alert_irq_handler(struct airq_struct *airq,
3387 				  struct tpi_info *tpi_info)
3388 {
3389 	struct tpi_adapter_info *info = (struct tpi_adapter_info *)tpi_info;
3390 
3391 	inc_irq_stat(IRQIO_GAL);
3392 
3393 	if ((info->forward || info->error) &&
3394 	    IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3395 		aen_process_gait(info->isc);
3396 		if (info->aism != 0)
3397 			process_gib_alert_list();
3398 	} else {
3399 		process_gib_alert_list();
3400 	}
3401 }
3402 
3403 static struct airq_struct gib_alert_irq = {
3404 	.handler = gib_alert_irq_handler,
3405 	.lsi_ptr = &gib_alert_irq.lsi_mask,
3406 };
3407 
kvm_s390_gib_destroy(void)3408 void kvm_s390_gib_destroy(void)
3409 {
3410 	if (!gib)
3411 		return;
3412 	if (kvm_s390_pci_interp_allowed() && aift) {
3413 		mutex_lock(&aift->aift_lock);
3414 		kvm_s390_pci_aen_exit();
3415 		mutex_unlock(&aift->aift_lock);
3416 	}
3417 	chsc_sgib(0);
3418 	unregister_adapter_interrupt(&gib_alert_irq);
3419 	free_page((unsigned long)gib);
3420 	gib = NULL;
3421 }
3422 
kvm_s390_gib_init(u8 nisc)3423 int kvm_s390_gib_init(u8 nisc)
3424 {
3425 	int rc = 0;
3426 
3427 	if (!css_general_characteristics.aiv) {
3428 		KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3429 		goto out;
3430 	}
3431 
3432 	gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3433 	if (!gib) {
3434 		rc = -ENOMEM;
3435 		goto out;
3436 	}
3437 
3438 	gib_alert_irq.isc = nisc;
3439 	if (register_adapter_interrupt(&gib_alert_irq)) {
3440 		pr_err("Registering the GIB alert interruption handler failed\n");
3441 		rc = -EIO;
3442 		goto out_free_gib;
3443 	}
3444 
3445 	gib->nisc = nisc;
3446 	if (chsc_sgib((u32)(u64)gib)) {
3447 		pr_err("Associating the GIB with the AIV facility failed\n");
3448 		free_page((unsigned long)gib);
3449 		gib = NULL;
3450 		rc = -EIO;
3451 		goto out_unreg_gal;
3452 	}
3453 
3454 	if (kvm_s390_pci_interp_allowed()) {
3455 		if (kvm_s390_pci_aen_init(nisc)) {
3456 			pr_err("Initializing AEN for PCI failed\n");
3457 			rc = -EIO;
3458 			goto out_unreg_gal;
3459 		}
3460 	}
3461 
3462 	KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3463 	goto out;
3464 
3465 out_unreg_gal:
3466 	unregister_adapter_interrupt(&gib_alert_irq);
3467 out_free_gib:
3468 	free_page((unsigned long)gib);
3469 	gib = NULL;
3470 out:
3471 	return rc;
3472 }
3473