Loading...
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
41static struct kvm_s390_gib *gib;
42
43/* handle external calls via sigp interpretation facility */
44static 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
76static 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
119static 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
150int psw_extint_disabled(struct kvm_vcpu *vcpu)
151{
152 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
153}
154
155static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
156{
157 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
158}
159
160static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
161{
162 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
163}
164
165static 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
172static 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
183static 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
197static 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
203static 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
210static uint64_t isc_to_isc_bits(int isc)
211{
212 return (0x80 >> isc) << 24;
213}
214
215static inline u32 isc_to_int_word(u8 isc)
216{
217 return ((u32)isc << 27) | 0x80000000;
218}
219
220static 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 */
246static 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 */
269static 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 */
289static 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
306static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
307{
308 return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa;
309}
310
311static 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
316static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
317{
318 return READ_ONCE(gisa->ipm);
319}
320
321static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
322{
323 return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
324}
325
326static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
327{
328 unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs |
329 vcpu->arch.local_int.pending_irqs;
330
331 pending &= ~vcpu->kvm->arch.float_int.masked_irqs;
332 return pending;
333}
334
335static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
336{
337 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
338 unsigned long pending_mask;
339
340 pending_mask = pending_irqs_no_gisa(vcpu);
341 if (gi->origin)
342 pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
343 return pending_mask;
344}
345
346static inline int isc_to_irq_type(unsigned long isc)
347{
348 return IRQ_PEND_IO_ISC_0 - isc;
349}
350
351static inline int irq_type_to_isc(unsigned long irq_type)
352{
353 return IRQ_PEND_IO_ISC_0 - irq_type;
354}
355
356static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
357 unsigned long active_mask)
358{
359 int i;
360
361 for (i = 0; i <= MAX_ISC; i++)
362 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
363 active_mask &= ~(1UL << (isc_to_irq_type(i)));
364
365 return active_mask;
366}
367
368static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
369{
370 unsigned long active_mask;
371
372 active_mask = pending_irqs(vcpu);
373 if (!active_mask)
374 return 0;
375
376 if (psw_extint_disabled(vcpu))
377 active_mask &= ~IRQ_PEND_EXT_MASK;
378 if (psw_ioint_disabled(vcpu))
379 active_mask &= ~IRQ_PEND_IO_MASK;
380 else
381 active_mask = disable_iscs(vcpu, active_mask);
382 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
383 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
384 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
385 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
386 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
387 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
388 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
389 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
390 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) {
391 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
392 __clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask);
393 }
394 if (psw_mchk_disabled(vcpu))
395 active_mask &= ~IRQ_PEND_MCHK_MASK;
396 /* PV guest cpus can have a single interruption injected at a time. */
397 if (kvm_s390_pv_cpu_get_handle(vcpu) &&
398 vcpu->arch.sie_block->iictl != IICTL_CODE_NONE)
399 active_mask &= ~(IRQ_PEND_EXT_II_MASK |
400 IRQ_PEND_IO_MASK |
401 IRQ_PEND_MCHK_MASK);
402 /*
403 * Check both floating and local interrupt's cr14 because
404 * bit IRQ_PEND_MCHK_REP could be set in both cases.
405 */
406 if (!(vcpu->arch.sie_block->gcr[14] &
407 (vcpu->kvm->arch.float_int.mchk.cr14 |
408 vcpu->arch.local_int.irq.mchk.cr14)))
409 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
410
411 /*
412 * STOP irqs will never be actively delivered. They are triggered via
413 * intercept requests and cleared when the stop intercept is performed.
414 */
415 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
416
417 return active_mask;
418}
419
420static void __set_cpu_idle(struct kvm_vcpu *vcpu)
421{
422 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
423 set_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
424}
425
426static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
427{
428 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
429 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
430}
431
432static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
433{
434 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
435 CPUSTAT_STOP_INT);
436 vcpu->arch.sie_block->lctl = 0x0000;
437 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
438
439 if (guestdbg_enabled(vcpu)) {
440 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
441 LCTL_CR10 | LCTL_CR11);
442 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
443 }
444}
445
446static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
447{
448 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
449 return;
450 if (psw_ioint_disabled(vcpu))
451 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
452 else
453 vcpu->arch.sie_block->lctl |= LCTL_CR6;
454}
455
456static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
457{
458 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
459 return;
460 if (psw_extint_disabled(vcpu))
461 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
462 else
463 vcpu->arch.sie_block->lctl |= LCTL_CR0;
464}
465
466static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
467{
468 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
469 return;
470 if (psw_mchk_disabled(vcpu))
471 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
472 else
473 vcpu->arch.sie_block->lctl |= LCTL_CR14;
474}
475
476static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
477{
478 if (kvm_s390_is_stop_irq_pending(vcpu))
479 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
480}
481
482/* Set interception request for non-deliverable interrupts */
483static void set_intercept_indicators(struct kvm_vcpu *vcpu)
484{
485 set_intercept_indicators_io(vcpu);
486 set_intercept_indicators_ext(vcpu);
487 set_intercept_indicators_mchk(vcpu);
488 set_intercept_indicators_stop(vcpu);
489}
490
491static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
492{
493 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
494 int rc = 0;
495
496 vcpu->stat.deliver_cputm++;
497 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
498 0, 0);
499 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
500 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
501 vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER;
502 } else {
503 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
504 (u16 *)__LC_EXT_INT_CODE);
505 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
506 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
507 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
508 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
509 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
510 }
511 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
512 return rc ? -EFAULT : 0;
513}
514
515static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
516{
517 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
518 int rc = 0;
519
520 vcpu->stat.deliver_ckc++;
521 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
522 0, 0);
523 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
524 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
525 vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP;
526 } else {
527 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
528 (u16 __user *)__LC_EXT_INT_CODE);
529 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
530 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
531 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
532 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
533 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
534 }
535 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
536 return rc ? -EFAULT : 0;
537}
538
539static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
540{
541 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
542 struct kvm_s390_ext_info ext;
543 int rc;
544
545 spin_lock(&li->lock);
546 ext = li->irq.ext;
547 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
548 li->irq.ext.ext_params2 = 0;
549 spin_unlock(&li->lock);
550
551 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
552 ext.ext_params2);
553 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
554 KVM_S390_INT_PFAULT_INIT,
555 0, ext.ext_params2);
556
557 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
558 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
559 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
560 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
561 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
562 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
563 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
564 return rc ? -EFAULT : 0;
565}
566
567static int __write_machine_check(struct kvm_vcpu *vcpu,
568 struct kvm_s390_mchk_info *mchk)
569{
570 unsigned long ext_sa_addr;
571 unsigned long lc;
572 freg_t fprs[NUM_FPRS];
573 union mci mci;
574 int rc;
575
576 /*
577 * All other possible payload for a machine check (e.g. the register
578 * contents in the save area) will be handled by the ultravisor, as
579 * the hypervisor does not not have the needed information for
580 * protected guests.
581 */
582 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
583 vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK;
584 vcpu->arch.sie_block->mcic = mchk->mcic;
585 vcpu->arch.sie_block->faddr = mchk->failing_storage_address;
586 vcpu->arch.sie_block->edc = mchk->ext_damage_code;
587 return 0;
588 }
589
590 mci.val = mchk->mcic;
591 /* take care of lazy register loading */
592 save_fpu_regs();
593 save_access_regs(vcpu->run->s.regs.acrs);
594 if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
595 save_gs_cb(current->thread.gs_cb);
596
597 /* Extended save area */
598 rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
599 sizeof(unsigned long));
600 /* Only bits 0 through 63-LC are used for address formation */
601 lc = ext_sa_addr & MCESA_LC_MASK;
602 if (test_kvm_facility(vcpu->kvm, 133)) {
603 switch (lc) {
604 case 0:
605 case 10:
606 ext_sa_addr &= ~0x3ffUL;
607 break;
608 case 11:
609 ext_sa_addr &= ~0x7ffUL;
610 break;
611 case 12:
612 ext_sa_addr &= ~0xfffUL;
613 break;
614 default:
615 ext_sa_addr = 0;
616 break;
617 }
618 } else {
619 ext_sa_addr &= ~0x3ffUL;
620 }
621
622 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
623 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
624 512))
625 mci.vr = 0;
626 } else {
627 mci.vr = 0;
628 }
629 if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
630 && (lc == 11 || lc == 12)) {
631 if (write_guest_abs(vcpu, ext_sa_addr + 1024,
632 &vcpu->run->s.regs.gscb, 32))
633 mci.gs = 0;
634 } else {
635 mci.gs = 0;
636 }
637
638 /* General interruption information */
639 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
640 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
641 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
642 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
643 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
644 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
645
646 /* Register-save areas */
647 if (MACHINE_HAS_VX) {
648 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
649 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
650 } else {
651 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
652 vcpu->run->s.regs.fprs, 128);
653 }
654 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
655 vcpu->run->s.regs.gprs, 128);
656 rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
657 (u32 __user *) __LC_FP_CREG_SAVE_AREA);
658 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
659 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
660 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
661 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
662 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
663 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
664 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
665 &vcpu->run->s.regs.acrs, 64);
666 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
667 &vcpu->arch.sie_block->gcr, 128);
668
669 /* Extended interruption information */
670 rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
671 (u32 __user *) __LC_EXT_DAMAGE_CODE);
672 rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
673 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
674 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
675 sizeof(mchk->fixed_logout));
676 return rc ? -EFAULT : 0;
677}
678
679static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
680{
681 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
682 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
683 struct kvm_s390_mchk_info mchk = {};
684 int deliver = 0;
685 int rc = 0;
686
687 spin_lock(&fi->lock);
688 spin_lock(&li->lock);
689 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
690 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
691 /*
692 * If there was an exigent machine check pending, then any
693 * repressible machine checks that might have been pending
694 * are indicated along with it, so always clear bits for
695 * repressible and exigent interrupts
696 */
697 mchk = li->irq.mchk;
698 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
699 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
700 memset(&li->irq.mchk, 0, sizeof(mchk));
701 deliver = 1;
702 }
703 /*
704 * We indicate floating repressible conditions along with
705 * other pending conditions. Channel Report Pending and Channel
706 * Subsystem damage are the only two and are indicated by
707 * bits in mcic and masked in cr14.
708 */
709 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
710 mchk.mcic |= fi->mchk.mcic;
711 mchk.cr14 |= fi->mchk.cr14;
712 memset(&fi->mchk, 0, sizeof(mchk));
713 deliver = 1;
714 }
715 spin_unlock(&li->lock);
716 spin_unlock(&fi->lock);
717
718 if (deliver) {
719 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
720 mchk.mcic);
721 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
722 KVM_S390_MCHK,
723 mchk.cr14, mchk.mcic);
724 vcpu->stat.deliver_machine_check++;
725 rc = __write_machine_check(vcpu, &mchk);
726 }
727 return rc;
728}
729
730static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
731{
732 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
733 int rc = 0;
734
735 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
736 vcpu->stat.deliver_restart_signal++;
737 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
738
739 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
740 vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART;
741 } else {
742 rc = write_guest_lc(vcpu,
743 offsetof(struct lowcore, restart_old_psw),
744 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
745 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
746 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
747 }
748 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
749 return rc ? -EFAULT : 0;
750}
751
752static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
753{
754 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
755 struct kvm_s390_prefix_info prefix;
756
757 spin_lock(&li->lock);
758 prefix = li->irq.prefix;
759 li->irq.prefix.address = 0;
760 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
761 spin_unlock(&li->lock);
762
763 vcpu->stat.deliver_prefix_signal++;
764 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
765 KVM_S390_SIGP_SET_PREFIX,
766 prefix.address, 0);
767
768 kvm_s390_set_prefix(vcpu, prefix.address);
769 return 0;
770}
771
772static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
773{
774 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
775 int rc;
776 int cpu_addr;
777
778 spin_lock(&li->lock);
779 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
780 clear_bit(cpu_addr, li->sigp_emerg_pending);
781 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
782 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
783 spin_unlock(&li->lock);
784
785 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
786 vcpu->stat.deliver_emergency_signal++;
787 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
788 cpu_addr, 0);
789 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
790 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
791 vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG;
792 vcpu->arch.sie_block->extcpuaddr = cpu_addr;
793 return 0;
794 }
795
796 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
797 (u16 *)__LC_EXT_INT_CODE);
798 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
799 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
800 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
801 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
802 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
803 return rc ? -EFAULT : 0;
804}
805
806static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
807{
808 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
809 struct kvm_s390_extcall_info extcall;
810 int rc;
811
812 spin_lock(&li->lock);
813 extcall = li->irq.extcall;
814 li->irq.extcall.code = 0;
815 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
816 spin_unlock(&li->lock);
817
818 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
819 vcpu->stat.deliver_external_call++;
820 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
821 KVM_S390_INT_EXTERNAL_CALL,
822 extcall.code, 0);
823 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
824 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
825 vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL;
826 vcpu->arch.sie_block->extcpuaddr = extcall.code;
827 return 0;
828 }
829
830 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
831 (u16 *)__LC_EXT_INT_CODE);
832 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
833 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
834 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
835 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
836 sizeof(psw_t));
837 return rc ? -EFAULT : 0;
838}
839
840static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code)
841{
842 switch (code) {
843 case PGM_SPECIFICATION:
844 vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION;
845 break;
846 case PGM_OPERAND:
847 vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND;
848 break;
849 default:
850 return -EINVAL;
851 }
852 return 0;
853}
854
855static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
856{
857 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
858 struct kvm_s390_pgm_info pgm_info;
859 int rc = 0, nullifying = false;
860 u16 ilen;
861
862 spin_lock(&li->lock);
863 pgm_info = li->irq.pgm;
864 clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
865 memset(&li->irq.pgm, 0, sizeof(pgm_info));
866 spin_unlock(&li->lock);
867
868 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
869 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
870 pgm_info.code, ilen);
871 vcpu->stat.deliver_program++;
872 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
873 pgm_info.code, 0);
874
875 /* PER is handled by the ultravisor */
876 if (kvm_s390_pv_cpu_is_protected(vcpu))
877 return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER);
878
879 switch (pgm_info.code & ~PGM_PER) {
880 case PGM_AFX_TRANSLATION:
881 case PGM_ASX_TRANSLATION:
882 case PGM_EX_TRANSLATION:
883 case PGM_LFX_TRANSLATION:
884 case PGM_LSTE_SEQUENCE:
885 case PGM_LSX_TRANSLATION:
886 case PGM_LX_TRANSLATION:
887 case PGM_PRIMARY_AUTHORITY:
888 case PGM_SECONDARY_AUTHORITY:
889 nullifying = true;
890 fallthrough;
891 case PGM_SPACE_SWITCH:
892 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
893 (u64 *)__LC_TRANS_EXC_CODE);
894 break;
895 case PGM_ALEN_TRANSLATION:
896 case PGM_ALE_SEQUENCE:
897 case PGM_ASTE_INSTANCE:
898 case PGM_ASTE_SEQUENCE:
899 case PGM_ASTE_VALIDITY:
900 case PGM_EXTENDED_AUTHORITY:
901 rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
902 (u8 *)__LC_EXC_ACCESS_ID);
903 nullifying = true;
904 break;
905 case PGM_ASCE_TYPE:
906 case PGM_PAGE_TRANSLATION:
907 case PGM_REGION_FIRST_TRANS:
908 case PGM_REGION_SECOND_TRANS:
909 case PGM_REGION_THIRD_TRANS:
910 case PGM_SEGMENT_TRANSLATION:
911 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
912 (u64 *)__LC_TRANS_EXC_CODE);
913 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
914 (u8 *)__LC_EXC_ACCESS_ID);
915 rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
916 (u8 *)__LC_OP_ACCESS_ID);
917 nullifying = true;
918 break;
919 case PGM_MONITOR:
920 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
921 (u16 *)__LC_MON_CLASS_NR);
922 rc |= put_guest_lc(vcpu, pgm_info.mon_code,
923 (u64 *)__LC_MON_CODE);
924 break;
925 case PGM_VECTOR_PROCESSING:
926 case PGM_DATA:
927 rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
928 (u32 *)__LC_DATA_EXC_CODE);
929 break;
930 case PGM_PROTECTION:
931 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
932 (u64 *)__LC_TRANS_EXC_CODE);
933 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
934 (u8 *)__LC_EXC_ACCESS_ID);
935 break;
936 case PGM_STACK_FULL:
937 case PGM_STACK_EMPTY:
938 case PGM_STACK_SPECIFICATION:
939 case PGM_STACK_TYPE:
940 case PGM_STACK_OPERATION:
941 case PGM_TRACE_TABEL:
942 case PGM_CRYPTO_OPERATION:
943 nullifying = true;
944 break;
945 }
946
947 if (pgm_info.code & PGM_PER) {
948 rc |= put_guest_lc(vcpu, pgm_info.per_code,
949 (u8 *) __LC_PER_CODE);
950 rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
951 (u8 *)__LC_PER_ATMID);
952 rc |= put_guest_lc(vcpu, pgm_info.per_address,
953 (u64 *) __LC_PER_ADDRESS);
954 rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
955 (u8 *) __LC_PER_ACCESS_ID);
956 }
957
958 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
959 kvm_s390_rewind_psw(vcpu, ilen);
960
961 /* bit 1+2 of the target are the ilc, so we can directly use ilen */
962 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
963 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
964 (u64 *) __LC_PGM_LAST_BREAK);
965 rc |= put_guest_lc(vcpu, pgm_info.code,
966 (u16 *)__LC_PGM_INT_CODE);
967 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
968 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
969 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
970 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
971 return rc ? -EFAULT : 0;
972}
973
974#define SCCB_MASK 0xFFFFFFF8
975#define SCCB_EVENT_PENDING 0x3
976
977static int write_sclp(struct kvm_vcpu *vcpu, u32 parm)
978{
979 int rc;
980
981 if (kvm_s390_pv_cpu_get_handle(vcpu)) {
982 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
983 vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG;
984 vcpu->arch.sie_block->eiparams = parm;
985 return 0;
986 }
987
988 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
989 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
990 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
991 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
992 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
993 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
994 rc |= put_guest_lc(vcpu, parm,
995 (u32 *)__LC_EXT_PARAMS);
996
997 return rc ? -EFAULT : 0;
998}
999
1000static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
1001{
1002 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1003 struct kvm_s390_ext_info ext;
1004
1005 spin_lock(&fi->lock);
1006 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) ||
1007 !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
1008 spin_unlock(&fi->lock);
1009 return 0;
1010 }
1011 ext = fi->srv_signal;
1012 memset(&fi->srv_signal, 0, sizeof(ext));
1013 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1014 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1015 if (kvm_s390_pv_cpu_is_protected(vcpu))
1016 set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs);
1017 spin_unlock(&fi->lock);
1018
1019 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
1020 ext.ext_params);
1021 vcpu->stat.deliver_service_signal++;
1022 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1023 ext.ext_params, 0);
1024
1025 return write_sclp(vcpu, ext.ext_params);
1026}
1027
1028static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)
1029{
1030 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1031 struct kvm_s390_ext_info ext;
1032
1033 spin_lock(&fi->lock);
1034 if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) {
1035 spin_unlock(&fi->lock);
1036 return 0;
1037 }
1038 ext = fi->srv_signal;
1039 /* only clear the event bit */
1040 fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING;
1041 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1042 spin_unlock(&fi->lock);
1043
1044 VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event");
1045 vcpu->stat.deliver_service_signal++;
1046 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1047 ext.ext_params, 0);
1048
1049 return write_sclp(vcpu, SCCB_EVENT_PENDING);
1050}
1051
1052static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
1053{
1054 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1055 struct kvm_s390_interrupt_info *inti;
1056 int rc = 0;
1057
1058 spin_lock(&fi->lock);
1059 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
1060 struct kvm_s390_interrupt_info,
1061 list);
1062 if (inti) {
1063 list_del(&inti->list);
1064 fi->counters[FIRQ_CNTR_PFAULT] -= 1;
1065 }
1066 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
1067 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1068 spin_unlock(&fi->lock);
1069
1070 if (inti) {
1071 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1072 KVM_S390_INT_PFAULT_DONE, 0,
1073 inti->ext.ext_params2);
1074 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
1075 inti->ext.ext_params2);
1076
1077 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1078 (u16 *)__LC_EXT_INT_CODE);
1079 rc |= put_guest_lc(vcpu, PFAULT_DONE,
1080 (u16 *)__LC_EXT_CPU_ADDR);
1081 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1082 &vcpu->arch.sie_block->gpsw,
1083 sizeof(psw_t));
1084 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1085 &vcpu->arch.sie_block->gpsw,
1086 sizeof(psw_t));
1087 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1088 (u64 *)__LC_EXT_PARAMS2);
1089 kfree(inti);
1090 }
1091 return rc ? -EFAULT : 0;
1092}
1093
1094static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
1095{
1096 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1097 struct kvm_s390_interrupt_info *inti;
1098 int rc = 0;
1099
1100 spin_lock(&fi->lock);
1101 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
1102 struct kvm_s390_interrupt_info,
1103 list);
1104 if (inti) {
1105 VCPU_EVENT(vcpu, 4,
1106 "deliver: virtio parm: 0x%x,parm64: 0x%llx",
1107 inti->ext.ext_params, inti->ext.ext_params2);
1108 vcpu->stat.deliver_virtio++;
1109 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1110 inti->type,
1111 inti->ext.ext_params,
1112 inti->ext.ext_params2);
1113 list_del(&inti->list);
1114 fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1115 }
1116 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1117 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1118 spin_unlock(&fi->lock);
1119
1120 if (inti) {
1121 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1122 (u16 *)__LC_EXT_INT_CODE);
1123 rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1124 (u16 *)__LC_EXT_CPU_ADDR);
1125 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1126 &vcpu->arch.sie_block->gpsw,
1127 sizeof(psw_t));
1128 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1129 &vcpu->arch.sie_block->gpsw,
1130 sizeof(psw_t));
1131 rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1132 (u32 *)__LC_EXT_PARAMS);
1133 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1134 (u64 *)__LC_EXT_PARAMS2);
1135 kfree(inti);
1136 }
1137 return rc ? -EFAULT : 0;
1138}
1139
1140static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1141{
1142 int rc;
1143
1144 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
1145 vcpu->arch.sie_block->iictl = IICTL_CODE_IO;
1146 vcpu->arch.sie_block->subchannel_id = io->subchannel_id;
1147 vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr;
1148 vcpu->arch.sie_block->io_int_parm = io->io_int_parm;
1149 vcpu->arch.sie_block->io_int_word = io->io_int_word;
1150 return 0;
1151 }
1152
1153 rc = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1154 rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1155 rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1156 rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1157 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1158 &vcpu->arch.sie_block->gpsw,
1159 sizeof(psw_t));
1160 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1161 &vcpu->arch.sie_block->gpsw,
1162 sizeof(psw_t));
1163 return rc ? -EFAULT : 0;
1164}
1165
1166static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1167 unsigned long irq_type)
1168{
1169 struct list_head *isc_list;
1170 struct kvm_s390_float_interrupt *fi;
1171 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1172 struct kvm_s390_interrupt_info *inti = NULL;
1173 struct kvm_s390_io_info io;
1174 u32 isc;
1175 int rc = 0;
1176
1177 fi = &vcpu->kvm->arch.float_int;
1178
1179 spin_lock(&fi->lock);
1180 isc = irq_type_to_isc(irq_type);
1181 isc_list = &fi->lists[isc];
1182 inti = list_first_entry_or_null(isc_list,
1183 struct kvm_s390_interrupt_info,
1184 list);
1185 if (inti) {
1186 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1187 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1188 else
1189 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1190 inti->io.subchannel_id >> 8,
1191 inti->io.subchannel_id >> 1 & 0x3,
1192 inti->io.subchannel_nr);
1193
1194 vcpu->stat.deliver_io++;
1195 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1196 inti->type,
1197 ((__u32)inti->io.subchannel_id << 16) |
1198 inti->io.subchannel_nr,
1199 ((__u64)inti->io.io_int_parm << 32) |
1200 inti->io.io_int_word);
1201 list_del(&inti->list);
1202 fi->counters[FIRQ_CNTR_IO] -= 1;
1203 }
1204 if (list_empty(isc_list))
1205 clear_bit(irq_type, &fi->pending_irqs);
1206 spin_unlock(&fi->lock);
1207
1208 if (inti) {
1209 rc = __do_deliver_io(vcpu, &(inti->io));
1210 kfree(inti);
1211 goto out;
1212 }
1213
1214 if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1215 /*
1216 * in case an adapter interrupt was not delivered
1217 * in SIE context KVM will handle the delivery
1218 */
1219 VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1220 memset(&io, 0, sizeof(io));
1221 io.io_int_word = isc_to_int_word(isc);
1222 vcpu->stat.deliver_io++;
1223 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1224 KVM_S390_INT_IO(1, 0, 0, 0),
1225 ((__u32)io.subchannel_id << 16) |
1226 io.subchannel_nr,
1227 ((__u64)io.io_int_parm << 32) |
1228 io.io_int_word);
1229 rc = __do_deliver_io(vcpu, &io);
1230 }
1231out:
1232 return rc;
1233}
1234
1235/* Check whether an external call is pending (deliverable or not) */
1236int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1237{
1238 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1239
1240 if (!sclp.has_sigpif)
1241 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1242
1243 return sca_ext_call_pending(vcpu, NULL);
1244}
1245
1246int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1247{
1248 if (deliverable_irqs(vcpu))
1249 return 1;
1250
1251 if (kvm_cpu_has_pending_timer(vcpu))
1252 return 1;
1253
1254 /* external call pending and deliverable */
1255 if (kvm_s390_ext_call_pending(vcpu) &&
1256 !psw_extint_disabled(vcpu) &&
1257 (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1258 return 1;
1259
1260 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1261 return 1;
1262 return 0;
1263}
1264
1265int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1266{
1267 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1268}
1269
1270static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1271{
1272 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1273 const u64 ckc = vcpu->arch.sie_block->ckc;
1274 u64 cputm, sltime = 0;
1275
1276 if (ckc_interrupts_enabled(vcpu)) {
1277 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1278 if ((s64)now < (s64)ckc)
1279 sltime = tod_to_ns((s64)ckc - (s64)now);
1280 } else if (now < ckc) {
1281 sltime = tod_to_ns(ckc - now);
1282 }
1283 /* already expired */
1284 if (!sltime)
1285 return 0;
1286 if (cpu_timer_interrupts_enabled(vcpu)) {
1287 cputm = kvm_s390_get_cpu_timer(vcpu);
1288 /* already expired? */
1289 if (cputm >> 63)
1290 return 0;
1291 return min_t(u64, sltime, tod_to_ns(cputm));
1292 }
1293 } else if (cpu_timer_interrupts_enabled(vcpu)) {
1294 sltime = kvm_s390_get_cpu_timer(vcpu);
1295 /* already expired? */
1296 if (sltime >> 63)
1297 return 0;
1298 }
1299 return sltime;
1300}
1301
1302int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1303{
1304 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1305 u64 sltime;
1306
1307 vcpu->stat.exit_wait_state++;
1308
1309 /* fast path */
1310 if (kvm_arch_vcpu_runnable(vcpu))
1311 return 0;
1312
1313 if (psw_interrupts_disabled(vcpu)) {
1314 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1315 return -EOPNOTSUPP; /* disabled wait */
1316 }
1317
1318 if (gi->origin &&
1319 (gisa_get_ipm_or_restore_iam(gi) &
1320 vcpu->arch.sie_block->gcr[6] >> 24))
1321 return 0;
1322
1323 if (!ckc_interrupts_enabled(vcpu) &&
1324 !cpu_timer_interrupts_enabled(vcpu)) {
1325 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1326 __set_cpu_idle(vcpu);
1327 goto no_timer;
1328 }
1329
1330 sltime = __calculate_sltime(vcpu);
1331 if (!sltime)
1332 return 0;
1333
1334 __set_cpu_idle(vcpu);
1335 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1336 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1337no_timer:
1338 kvm_vcpu_srcu_read_unlock(vcpu);
1339 kvm_vcpu_halt(vcpu);
1340 vcpu->valid_wakeup = false;
1341 __unset_cpu_idle(vcpu);
1342 kvm_vcpu_srcu_read_lock(vcpu);
1343
1344 hrtimer_cancel(&vcpu->arch.ckc_timer);
1345 return 0;
1346}
1347
1348void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1349{
1350 vcpu->valid_wakeup = true;
1351 kvm_vcpu_wake_up(vcpu);
1352
1353 /*
1354 * The VCPU might not be sleeping but rather executing VSIE. Let's
1355 * kick it, so it leaves the SIE to process the request.
1356 */
1357 kvm_s390_vsie_kick(vcpu);
1358}
1359
1360enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1361{
1362 struct kvm_vcpu *vcpu;
1363 u64 sltime;
1364
1365 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1366 sltime = __calculate_sltime(vcpu);
1367
1368 /*
1369 * If the monotonic clock runs faster than the tod clock we might be
1370 * woken up too early and have to go back to sleep to avoid deadlocks.
1371 */
1372 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1373 return HRTIMER_RESTART;
1374 kvm_s390_vcpu_wakeup(vcpu);
1375 return HRTIMER_NORESTART;
1376}
1377
1378void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1379{
1380 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1381
1382 spin_lock(&li->lock);
1383 li->pending_irqs = 0;
1384 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1385 memset(&li->irq, 0, sizeof(li->irq));
1386 spin_unlock(&li->lock);
1387
1388 sca_clear_ext_call(vcpu);
1389}
1390
1391int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1392{
1393 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1394 int rc = 0;
1395 unsigned long irq_type;
1396 unsigned long irqs;
1397
1398 __reset_intercept_indicators(vcpu);
1399
1400 /* pending ckc conditions might have been invalidated */
1401 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1402 if (ckc_irq_pending(vcpu))
1403 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1404
1405 /* pending cpu timer conditions might have been invalidated */
1406 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1407 if (cpu_timer_irq_pending(vcpu))
1408 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1409
1410 while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1411 /* bits are in the reverse order of interrupt priority */
1412 irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1413 switch (irq_type) {
1414 case IRQ_PEND_IO_ISC_0:
1415 case IRQ_PEND_IO_ISC_1:
1416 case IRQ_PEND_IO_ISC_2:
1417 case IRQ_PEND_IO_ISC_3:
1418 case IRQ_PEND_IO_ISC_4:
1419 case IRQ_PEND_IO_ISC_5:
1420 case IRQ_PEND_IO_ISC_6:
1421 case IRQ_PEND_IO_ISC_7:
1422 rc = __deliver_io(vcpu, irq_type);
1423 break;
1424 case IRQ_PEND_MCHK_EX:
1425 case IRQ_PEND_MCHK_REP:
1426 rc = __deliver_machine_check(vcpu);
1427 break;
1428 case IRQ_PEND_PROG:
1429 rc = __deliver_prog(vcpu);
1430 break;
1431 case IRQ_PEND_EXT_EMERGENCY:
1432 rc = __deliver_emergency_signal(vcpu);
1433 break;
1434 case IRQ_PEND_EXT_EXTERNAL:
1435 rc = __deliver_external_call(vcpu);
1436 break;
1437 case IRQ_PEND_EXT_CLOCK_COMP:
1438 rc = __deliver_ckc(vcpu);
1439 break;
1440 case IRQ_PEND_EXT_CPU_TIMER:
1441 rc = __deliver_cpu_timer(vcpu);
1442 break;
1443 case IRQ_PEND_RESTART:
1444 rc = __deliver_restart(vcpu);
1445 break;
1446 case IRQ_PEND_SET_PREFIX:
1447 rc = __deliver_set_prefix(vcpu);
1448 break;
1449 case IRQ_PEND_PFAULT_INIT:
1450 rc = __deliver_pfault_init(vcpu);
1451 break;
1452 case IRQ_PEND_EXT_SERVICE:
1453 rc = __deliver_service(vcpu);
1454 break;
1455 case IRQ_PEND_EXT_SERVICE_EV:
1456 rc = __deliver_service_ev(vcpu);
1457 break;
1458 case IRQ_PEND_PFAULT_DONE:
1459 rc = __deliver_pfault_done(vcpu);
1460 break;
1461 case IRQ_PEND_VIRTIO:
1462 rc = __deliver_virtio(vcpu);
1463 break;
1464 default:
1465 WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1466 clear_bit(irq_type, &li->pending_irqs);
1467 }
1468 }
1469
1470 set_intercept_indicators(vcpu);
1471
1472 return rc;
1473}
1474
1475static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1476{
1477 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1478
1479 vcpu->stat.inject_program++;
1480 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1481 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1482 irq->u.pgm.code, 0);
1483
1484 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1485 /* auto detection if no valid ILC was given */
1486 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1487 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1488 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1489 }
1490
1491 if (irq->u.pgm.code == PGM_PER) {
1492 li->irq.pgm.code |= PGM_PER;
1493 li->irq.pgm.flags = irq->u.pgm.flags;
1494 /* only modify PER related information */
1495 li->irq.pgm.per_address = irq->u.pgm.per_address;
1496 li->irq.pgm.per_code = irq->u.pgm.per_code;
1497 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1498 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1499 } else if (!(irq->u.pgm.code & PGM_PER)) {
1500 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1501 irq->u.pgm.code;
1502 li->irq.pgm.flags = irq->u.pgm.flags;
1503 /* only modify non-PER information */
1504 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1505 li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1506 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1507 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1508 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1509 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1510 } else {
1511 li->irq.pgm = irq->u.pgm;
1512 }
1513 set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1514 return 0;
1515}
1516
1517static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1518{
1519 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1520
1521 vcpu->stat.inject_pfault_init++;
1522 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1523 irq->u.ext.ext_params2);
1524 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1525 irq->u.ext.ext_params,
1526 irq->u.ext.ext_params2);
1527
1528 li->irq.ext = irq->u.ext;
1529 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1530 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1531 return 0;
1532}
1533
1534static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1535{
1536 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1537 struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1538 uint16_t src_id = irq->u.extcall.code;
1539
1540 vcpu->stat.inject_external_call++;
1541 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1542 src_id);
1543 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1544 src_id, 0);
1545
1546 /* sending vcpu invalid */
1547 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1548 return -EINVAL;
1549
1550 if (sclp.has_sigpif && !kvm_s390_pv_cpu_get_handle(vcpu))
1551 return sca_inject_ext_call(vcpu, src_id);
1552
1553 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1554 return -EBUSY;
1555 *extcall = irq->u.extcall;
1556 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1557 return 0;
1558}
1559
1560static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1561{
1562 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1563 struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1564
1565 vcpu->stat.inject_set_prefix++;
1566 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1567 irq->u.prefix.address);
1568 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1569 irq->u.prefix.address, 0);
1570
1571 if (!is_vcpu_stopped(vcpu))
1572 return -EBUSY;
1573
1574 *prefix = irq->u.prefix;
1575 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1576 return 0;
1577}
1578
1579#define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1580static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1581{
1582 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1583 struct kvm_s390_stop_info *stop = &li->irq.stop;
1584 int rc = 0;
1585
1586 vcpu->stat.inject_stop_signal++;
1587 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1588
1589 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1590 return -EINVAL;
1591
1592 if (is_vcpu_stopped(vcpu)) {
1593 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1594 rc = kvm_s390_store_status_unloaded(vcpu,
1595 KVM_S390_STORE_STATUS_NOADDR);
1596 return rc;
1597 }
1598
1599 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1600 return -EBUSY;
1601 stop->flags = irq->u.stop.flags;
1602 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
1603 return 0;
1604}
1605
1606static int __inject_sigp_restart(struct kvm_vcpu *vcpu)
1607{
1608 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1609
1610 vcpu->stat.inject_restart++;
1611 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1612 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1613
1614 set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1615 return 0;
1616}
1617
1618static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1619 struct kvm_s390_irq *irq)
1620{
1621 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1622
1623 vcpu->stat.inject_emergency_signal++;
1624 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1625 irq->u.emerg.code);
1626 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1627 irq->u.emerg.code, 0);
1628
1629 /* sending vcpu invalid */
1630 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1631 return -EINVAL;
1632
1633 set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1634 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1635 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1636 return 0;
1637}
1638
1639static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1640{
1641 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1642 struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1643
1644 vcpu->stat.inject_mchk++;
1645 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1646 irq->u.mchk.mcic);
1647 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1648 irq->u.mchk.mcic);
1649
1650 /*
1651 * Because repressible machine checks can be indicated along with
1652 * exigent machine checks (PoP, Chapter 11, Interruption action)
1653 * we need to combine cr14, mcic and external damage code.
1654 * Failing storage address and the logout area should not be or'ed
1655 * together, we just indicate the last occurrence of the corresponding
1656 * machine check
1657 */
1658 mchk->cr14 |= irq->u.mchk.cr14;
1659 mchk->mcic |= irq->u.mchk.mcic;
1660 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1661 mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1662 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1663 sizeof(mchk->fixed_logout));
1664 if (mchk->mcic & MCHK_EX_MASK)
1665 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1666 else if (mchk->mcic & MCHK_REP_MASK)
1667 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
1668 return 0;
1669}
1670
1671static int __inject_ckc(struct kvm_vcpu *vcpu)
1672{
1673 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1674
1675 vcpu->stat.inject_ckc++;
1676 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1677 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1678 0, 0);
1679
1680 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1681 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1682 return 0;
1683}
1684
1685static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1686{
1687 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1688
1689 vcpu->stat.inject_cputm++;
1690 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1691 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1692 0, 0);
1693
1694 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1695 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
1696 return 0;
1697}
1698
1699static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1700 int isc, u32 schid)
1701{
1702 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1703 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1704 struct kvm_s390_interrupt_info *iter;
1705 u16 id = (schid & 0xffff0000U) >> 16;
1706 u16 nr = schid & 0x0000ffffU;
1707
1708 spin_lock(&fi->lock);
1709 list_for_each_entry(iter, isc_list, list) {
1710 if (schid && (id != iter->io.subchannel_id ||
1711 nr != iter->io.subchannel_nr))
1712 continue;
1713 /* found an appropriate entry */
1714 list_del_init(&iter->list);
1715 fi->counters[FIRQ_CNTR_IO] -= 1;
1716 if (list_empty(isc_list))
1717 clear_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1718 spin_unlock(&fi->lock);
1719 return iter;
1720 }
1721 spin_unlock(&fi->lock);
1722 return NULL;
1723}
1724
1725static struct kvm_s390_interrupt_info *get_top_io_int(struct kvm *kvm,
1726 u64 isc_mask, u32 schid)
1727{
1728 struct kvm_s390_interrupt_info *inti = NULL;
1729 int isc;
1730
1731 for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1732 if (isc_mask & isc_to_isc_bits(isc))
1733 inti = get_io_int(kvm, isc, schid);
1734 }
1735 return inti;
1736}
1737
1738static int get_top_gisa_isc(struct kvm *kvm, u64 isc_mask, u32 schid)
1739{
1740 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1741 unsigned long active_mask;
1742 int isc;
1743
1744 if (schid)
1745 goto out;
1746 if (!gi->origin)
1747 goto out;
1748
1749 active_mask = (isc_mask & gisa_get_ipm(gi->origin) << 24) << 32;
1750 while (active_mask) {
1751 isc = __fls(active_mask) ^ (BITS_PER_LONG - 1);
1752 if (gisa_tac_ipm_gisc(gi->origin, isc))
1753 return isc;
1754 clear_bit_inv(isc, &active_mask);
1755 }
1756out:
1757 return -EINVAL;
1758}
1759
1760/*
1761 * Dequeue and return an I/O interrupt matching any of the interruption
1762 * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1763 * Take into account the interrupts pending in the interrupt list and in GISA.
1764 *
1765 * Note that for a guest that does not enable I/O interrupts
1766 * but relies on TPI, a flood of classic interrupts may starve
1767 * out adapter interrupts on the same isc. Linux does not do
1768 * that, and it is possible to work around the issue by configuring
1769 * different iscs for classic and adapter interrupts in the guest,
1770 * but we may want to revisit this in the future.
1771 */
1772struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1773 u64 isc_mask, u32 schid)
1774{
1775 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1776 struct kvm_s390_interrupt_info *inti, *tmp_inti;
1777 int isc;
1778
1779 inti = get_top_io_int(kvm, isc_mask, schid);
1780
1781 isc = get_top_gisa_isc(kvm, isc_mask, schid);
1782 if (isc < 0)
1783 /* no AI in GISA */
1784 goto out;
1785
1786 if (!inti)
1787 /* AI in GISA but no classical IO int */
1788 goto gisa_out;
1789
1790 /* both types of interrupts present */
1791 if (int_word_to_isc(inti->io.io_int_word) <= isc) {
1792 /* classical IO int with higher priority */
1793 gisa_set_ipm_gisc(gi->origin, isc);
1794 goto out;
1795 }
1796gisa_out:
1797 tmp_inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
1798 if (tmp_inti) {
1799 tmp_inti->type = KVM_S390_INT_IO(1, 0, 0, 0);
1800 tmp_inti->io.io_int_word = isc_to_int_word(isc);
1801 if (inti)
1802 kvm_s390_reinject_io_int(kvm, inti);
1803 inti = tmp_inti;
1804 } else
1805 gisa_set_ipm_gisc(gi->origin, isc);
1806out:
1807 return inti;
1808}
1809
1810static int __inject_service(struct kvm *kvm,
1811 struct kvm_s390_interrupt_info *inti)
1812{
1813 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1814
1815 kvm->stat.inject_service_signal++;
1816 spin_lock(&fi->lock);
1817 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1818
1819 /* We always allow events, track them separately from the sccb ints */
1820 if (fi->srv_signal.ext_params & SCCB_EVENT_PENDING)
1821 set_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1822
1823 /*
1824 * Early versions of the QEMU s390 bios will inject several
1825 * service interrupts after another without handling a
1826 * condition code indicating busy.
1827 * We will silently ignore those superfluous sccb values.
1828 * A future version of QEMU will take care of serialization
1829 * of servc requests
1830 */
1831 if (fi->srv_signal.ext_params & SCCB_MASK)
1832 goto out;
1833 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1834 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1835out:
1836 spin_unlock(&fi->lock);
1837 kfree(inti);
1838 return 0;
1839}
1840
1841static int __inject_virtio(struct kvm *kvm,
1842 struct kvm_s390_interrupt_info *inti)
1843{
1844 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1845
1846 kvm->stat.inject_virtio++;
1847 spin_lock(&fi->lock);
1848 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1849 spin_unlock(&fi->lock);
1850 return -EBUSY;
1851 }
1852 fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1853 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1854 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1855 spin_unlock(&fi->lock);
1856 return 0;
1857}
1858
1859static int __inject_pfault_done(struct kvm *kvm,
1860 struct kvm_s390_interrupt_info *inti)
1861{
1862 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1863
1864 kvm->stat.inject_pfault_done++;
1865 spin_lock(&fi->lock);
1866 if (fi->counters[FIRQ_CNTR_PFAULT] >=
1867 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1868 spin_unlock(&fi->lock);
1869 return -EBUSY;
1870 }
1871 fi->counters[FIRQ_CNTR_PFAULT] += 1;
1872 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1873 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1874 spin_unlock(&fi->lock);
1875 return 0;
1876}
1877
1878#define CR_PENDING_SUBCLASS 28
1879static int __inject_float_mchk(struct kvm *kvm,
1880 struct kvm_s390_interrupt_info *inti)
1881{
1882 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1883
1884 kvm->stat.inject_float_mchk++;
1885 spin_lock(&fi->lock);
1886 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1887 fi->mchk.mcic |= inti->mchk.mcic;
1888 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1889 spin_unlock(&fi->lock);
1890 kfree(inti);
1891 return 0;
1892}
1893
1894static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1895{
1896 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
1897 struct kvm_s390_float_interrupt *fi;
1898 struct list_head *list;
1899 int isc;
1900
1901 kvm->stat.inject_io++;
1902 isc = int_word_to_isc(inti->io.io_int_word);
1903
1904 /*
1905 * We do not use the lock checking variant as this is just a
1906 * performance optimization and we do not hold the lock here.
1907 * This is ok as the code will pick interrupts from both "lists"
1908 * for delivery.
1909 */
1910 if (gi->origin && inti->type & KVM_S390_INT_IO_AI_MASK) {
1911 VM_EVENT(kvm, 4, "%s isc %1u", "inject: I/O (AI/gisa)", isc);
1912 gisa_set_ipm_gisc(gi->origin, isc);
1913 kfree(inti);
1914 return 0;
1915 }
1916
1917 fi = &kvm->arch.float_int;
1918 spin_lock(&fi->lock);
1919 if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1920 spin_unlock(&fi->lock);
1921 return -EBUSY;
1922 }
1923 fi->counters[FIRQ_CNTR_IO] += 1;
1924
1925 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1926 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1927 else
1928 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1929 inti->io.subchannel_id >> 8,
1930 inti->io.subchannel_id >> 1 & 0x3,
1931 inti->io.subchannel_nr);
1932 list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1933 list_add_tail(&inti->list, list);
1934 set_bit(isc_to_irq_type(isc), &fi->pending_irqs);
1935 spin_unlock(&fi->lock);
1936 return 0;
1937}
1938
1939/*
1940 * Find a destination VCPU for a floating irq and kick it.
1941 */
1942static void __floating_irq_kick(struct kvm *kvm, u64 type)
1943{
1944 struct kvm_vcpu *dst_vcpu;
1945 int sigcpu, online_vcpus, nr_tries = 0;
1946
1947 online_vcpus = atomic_read(&kvm->online_vcpus);
1948 if (!online_vcpus)
1949 return;
1950
1951 /* find idle VCPUs first, then round robin */
1952 sigcpu = find_first_bit(kvm->arch.idle_mask, online_vcpus);
1953 if (sigcpu == online_vcpus) {
1954 do {
1955 sigcpu = kvm->arch.float_int.next_rr_cpu++;
1956 kvm->arch.float_int.next_rr_cpu %= online_vcpus;
1957 /* avoid endless loops if all vcpus are stopped */
1958 if (nr_tries++ >= online_vcpus)
1959 return;
1960 } while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1961 }
1962 dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1963
1964 /* make the VCPU drop out of the SIE, or wake it up if sleeping */
1965 switch (type) {
1966 case KVM_S390_MCHK:
1967 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_STOP_INT);
1968 break;
1969 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1970 if (!(type & KVM_S390_INT_IO_AI_MASK &&
1971 kvm->arch.gisa_int.origin) ||
1972 kvm_s390_pv_cpu_get_handle(dst_vcpu))
1973 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_IO_INT);
1974 break;
1975 default:
1976 kvm_s390_set_cpuflags(dst_vcpu, CPUSTAT_EXT_INT);
1977 break;
1978 }
1979 kvm_s390_vcpu_wakeup(dst_vcpu);
1980}
1981
1982static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1983{
1984 u64 type = READ_ONCE(inti->type);
1985 int rc;
1986
1987 switch (type) {
1988 case KVM_S390_MCHK:
1989 rc = __inject_float_mchk(kvm, inti);
1990 break;
1991 case KVM_S390_INT_VIRTIO:
1992 rc = __inject_virtio(kvm, inti);
1993 break;
1994 case KVM_S390_INT_SERVICE:
1995 rc = __inject_service(kvm, inti);
1996 break;
1997 case KVM_S390_INT_PFAULT_DONE:
1998 rc = __inject_pfault_done(kvm, inti);
1999 break;
2000 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2001 rc = __inject_io(kvm, inti);
2002 break;
2003 default:
2004 rc = -EINVAL;
2005 }
2006 if (rc)
2007 return rc;
2008
2009 __floating_irq_kick(kvm, type);
2010 return 0;
2011}
2012
2013int kvm_s390_inject_vm(struct kvm *kvm,
2014 struct kvm_s390_interrupt *s390int)
2015{
2016 struct kvm_s390_interrupt_info *inti;
2017 int rc;
2018
2019 inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2020 if (!inti)
2021 return -ENOMEM;
2022
2023 inti->type = s390int->type;
2024 switch (inti->type) {
2025 case KVM_S390_INT_VIRTIO:
2026 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
2027 s390int->parm, s390int->parm64);
2028 inti->ext.ext_params = s390int->parm;
2029 inti->ext.ext_params2 = s390int->parm64;
2030 break;
2031 case KVM_S390_INT_SERVICE:
2032 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
2033 inti->ext.ext_params = s390int->parm;
2034 break;
2035 case KVM_S390_INT_PFAULT_DONE:
2036 inti->ext.ext_params2 = s390int->parm64;
2037 break;
2038 case KVM_S390_MCHK:
2039 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
2040 s390int->parm64);
2041 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
2042 inti->mchk.mcic = s390int->parm64;
2043 break;
2044 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2045 inti->io.subchannel_id = s390int->parm >> 16;
2046 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
2047 inti->io.io_int_parm = s390int->parm64 >> 32;
2048 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
2049 break;
2050 default:
2051 kfree(inti);
2052 return -EINVAL;
2053 }
2054 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
2055 2);
2056
2057 rc = __inject_vm(kvm, inti);
2058 if (rc)
2059 kfree(inti);
2060 return rc;
2061}
2062
2063int kvm_s390_reinject_io_int(struct kvm *kvm,
2064 struct kvm_s390_interrupt_info *inti)
2065{
2066 return __inject_vm(kvm, inti);
2067}
2068
2069int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
2070 struct kvm_s390_irq *irq)
2071{
2072 irq->type = s390int->type;
2073 switch (irq->type) {
2074 case KVM_S390_PROGRAM_INT:
2075 if (s390int->parm & 0xffff0000)
2076 return -EINVAL;
2077 irq->u.pgm.code = s390int->parm;
2078 break;
2079 case KVM_S390_SIGP_SET_PREFIX:
2080 irq->u.prefix.address = s390int->parm;
2081 break;
2082 case KVM_S390_SIGP_STOP:
2083 irq->u.stop.flags = s390int->parm;
2084 break;
2085 case KVM_S390_INT_EXTERNAL_CALL:
2086 if (s390int->parm & 0xffff0000)
2087 return -EINVAL;
2088 irq->u.extcall.code = s390int->parm;
2089 break;
2090 case KVM_S390_INT_EMERGENCY:
2091 if (s390int->parm & 0xffff0000)
2092 return -EINVAL;
2093 irq->u.emerg.code = s390int->parm;
2094 break;
2095 case KVM_S390_MCHK:
2096 irq->u.mchk.mcic = s390int->parm64;
2097 break;
2098 case KVM_S390_INT_PFAULT_INIT:
2099 irq->u.ext.ext_params = s390int->parm;
2100 irq->u.ext.ext_params2 = s390int->parm64;
2101 break;
2102 case KVM_S390_RESTART:
2103 case KVM_S390_INT_CLOCK_COMP:
2104 case KVM_S390_INT_CPU_TIMER:
2105 break;
2106 default:
2107 return -EINVAL;
2108 }
2109 return 0;
2110}
2111
2112int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
2113{
2114 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2115
2116 return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2117}
2118
2119int kvm_s390_is_restart_irq_pending(struct kvm_vcpu *vcpu)
2120{
2121 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2122
2123 return test_bit(IRQ_PEND_RESTART, &li->pending_irqs);
2124}
2125
2126void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
2127{
2128 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2129
2130 spin_lock(&li->lock);
2131 li->irq.stop.flags = 0;
2132 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
2133 spin_unlock(&li->lock);
2134}
2135
2136static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2137{
2138 int rc;
2139
2140 switch (irq->type) {
2141 case KVM_S390_PROGRAM_INT:
2142 rc = __inject_prog(vcpu, irq);
2143 break;
2144 case KVM_S390_SIGP_SET_PREFIX:
2145 rc = __inject_set_prefix(vcpu, irq);
2146 break;
2147 case KVM_S390_SIGP_STOP:
2148 rc = __inject_sigp_stop(vcpu, irq);
2149 break;
2150 case KVM_S390_RESTART:
2151 rc = __inject_sigp_restart(vcpu);
2152 break;
2153 case KVM_S390_INT_CLOCK_COMP:
2154 rc = __inject_ckc(vcpu);
2155 break;
2156 case KVM_S390_INT_CPU_TIMER:
2157 rc = __inject_cpu_timer(vcpu);
2158 break;
2159 case KVM_S390_INT_EXTERNAL_CALL:
2160 rc = __inject_extcall(vcpu, irq);
2161 break;
2162 case KVM_S390_INT_EMERGENCY:
2163 rc = __inject_sigp_emergency(vcpu, irq);
2164 break;
2165 case KVM_S390_MCHK:
2166 rc = __inject_mchk(vcpu, irq);
2167 break;
2168 case KVM_S390_INT_PFAULT_INIT:
2169 rc = __inject_pfault_init(vcpu, irq);
2170 break;
2171 case KVM_S390_INT_VIRTIO:
2172 case KVM_S390_INT_SERVICE:
2173 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2174 default:
2175 rc = -EINVAL;
2176 }
2177
2178 return rc;
2179}
2180
2181int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
2182{
2183 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2184 int rc;
2185
2186 spin_lock(&li->lock);
2187 rc = do_inject_vcpu(vcpu, irq);
2188 spin_unlock(&li->lock);
2189 if (!rc)
2190 kvm_s390_vcpu_wakeup(vcpu);
2191 return rc;
2192}
2193
2194static inline void clear_irq_list(struct list_head *_list)
2195{
2196 struct kvm_s390_interrupt_info *inti, *n;
2197
2198 list_for_each_entry_safe(inti, n, _list, list) {
2199 list_del(&inti->list);
2200 kfree(inti);
2201 }
2202}
2203
2204static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
2205 struct kvm_s390_irq *irq)
2206{
2207 irq->type = inti->type;
2208 switch (inti->type) {
2209 case KVM_S390_INT_PFAULT_INIT:
2210 case KVM_S390_INT_PFAULT_DONE:
2211 case KVM_S390_INT_VIRTIO:
2212 irq->u.ext = inti->ext;
2213 break;
2214 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2215 irq->u.io = inti->io;
2216 break;
2217 }
2218}
2219
2220void kvm_s390_clear_float_irqs(struct kvm *kvm)
2221{
2222 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2223 int i;
2224
2225 mutex_lock(&kvm->lock);
2226 if (!kvm_s390_pv_is_protected(kvm))
2227 fi->masked_irqs = 0;
2228 mutex_unlock(&kvm->lock);
2229 spin_lock(&fi->lock);
2230 fi->pending_irqs = 0;
2231 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
2232 memset(&fi->mchk, 0, sizeof(fi->mchk));
2233 for (i = 0; i < FIRQ_LIST_COUNT; i++)
2234 clear_irq_list(&fi->lists[i]);
2235 for (i = 0; i < FIRQ_MAX_COUNT; i++)
2236 fi->counters[i] = 0;
2237 spin_unlock(&fi->lock);
2238 kvm_s390_gisa_clear(kvm);
2239};
2240
2241static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
2242{
2243 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
2244 struct kvm_s390_interrupt_info *inti;
2245 struct kvm_s390_float_interrupt *fi;
2246 struct kvm_s390_irq *buf;
2247 struct kvm_s390_irq *irq;
2248 int max_irqs;
2249 int ret = 0;
2250 int n = 0;
2251 int i;
2252
2253 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
2254 return -EINVAL;
2255
2256 /*
2257 * We are already using -ENOMEM to signal
2258 * userspace it may retry with a bigger buffer,
2259 * so we need to use something else for this case
2260 */
2261 buf = vzalloc(len);
2262 if (!buf)
2263 return -ENOBUFS;
2264
2265 max_irqs = len / sizeof(struct kvm_s390_irq);
2266
2267 if (gi->origin && gisa_get_ipm(gi->origin)) {
2268 for (i = 0; i <= MAX_ISC; i++) {
2269 if (n == max_irqs) {
2270 /* signal userspace to try again */
2271 ret = -ENOMEM;
2272 goto out_nolock;
2273 }
2274 if (gisa_tac_ipm_gisc(gi->origin, i)) {
2275 irq = (struct kvm_s390_irq *) &buf[n];
2276 irq->type = KVM_S390_INT_IO(1, 0, 0, 0);
2277 irq->u.io.io_int_word = isc_to_int_word(i);
2278 n++;
2279 }
2280 }
2281 }
2282 fi = &kvm->arch.float_int;
2283 spin_lock(&fi->lock);
2284 for (i = 0; i < FIRQ_LIST_COUNT; i++) {
2285 list_for_each_entry(inti, &fi->lists[i], list) {
2286 if (n == max_irqs) {
2287 /* signal userspace to try again */
2288 ret = -ENOMEM;
2289 goto out;
2290 }
2291 inti_to_irq(inti, &buf[n]);
2292 n++;
2293 }
2294 }
2295 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs) ||
2296 test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs)) {
2297 if (n == max_irqs) {
2298 /* signal userspace to try again */
2299 ret = -ENOMEM;
2300 goto out;
2301 }
2302 irq = (struct kvm_s390_irq *) &buf[n];
2303 irq->type = KVM_S390_INT_SERVICE;
2304 irq->u.ext = fi->srv_signal;
2305 n++;
2306 }
2307 if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
2308 if (n == max_irqs) {
2309 /* signal userspace to try again */
2310 ret = -ENOMEM;
2311 goto out;
2312 }
2313 irq = (struct kvm_s390_irq *) &buf[n];
2314 irq->type = KVM_S390_MCHK;
2315 irq->u.mchk = fi->mchk;
2316 n++;
2317}
2318
2319out:
2320 spin_unlock(&fi->lock);
2321out_nolock:
2322 if (!ret && n > 0) {
2323 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
2324 ret = -EFAULT;
2325 }
2326 vfree(buf);
2327
2328 return ret < 0 ? ret : n;
2329}
2330
2331static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
2332{
2333 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2334 struct kvm_s390_ais_all ais;
2335
2336 if (attr->attr < sizeof(ais))
2337 return -EINVAL;
2338
2339 if (!test_kvm_facility(kvm, 72))
2340 return -EOPNOTSUPP;
2341
2342 mutex_lock(&fi->ais_lock);
2343 ais.simm = fi->simm;
2344 ais.nimm = fi->nimm;
2345 mutex_unlock(&fi->ais_lock);
2346
2347 if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
2348 return -EFAULT;
2349
2350 return 0;
2351}
2352
2353static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2354{
2355 int r;
2356
2357 switch (attr->group) {
2358 case KVM_DEV_FLIC_GET_ALL_IRQS:
2359 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
2360 attr->attr);
2361 break;
2362 case KVM_DEV_FLIC_AISM_ALL:
2363 r = flic_ais_mode_get_all(dev->kvm, attr);
2364 break;
2365 default:
2366 r = -EINVAL;
2367 }
2368
2369 return r;
2370}
2371
2372static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
2373 u64 addr)
2374{
2375 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
2376 void *target = NULL;
2377 void __user *source;
2378 u64 size;
2379
2380 if (get_user(inti->type, (u64 __user *)addr))
2381 return -EFAULT;
2382
2383 switch (inti->type) {
2384 case KVM_S390_INT_PFAULT_INIT:
2385 case KVM_S390_INT_PFAULT_DONE:
2386 case KVM_S390_INT_VIRTIO:
2387 case KVM_S390_INT_SERVICE:
2388 target = (void *) &inti->ext;
2389 source = &uptr->u.ext;
2390 size = sizeof(inti->ext);
2391 break;
2392 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
2393 target = (void *) &inti->io;
2394 source = &uptr->u.io;
2395 size = sizeof(inti->io);
2396 break;
2397 case KVM_S390_MCHK:
2398 target = (void *) &inti->mchk;
2399 source = &uptr->u.mchk;
2400 size = sizeof(inti->mchk);
2401 break;
2402 default:
2403 return -EINVAL;
2404 }
2405
2406 if (copy_from_user(target, source, size))
2407 return -EFAULT;
2408
2409 return 0;
2410}
2411
2412static int enqueue_floating_irq(struct kvm_device *dev,
2413 struct kvm_device_attr *attr)
2414{
2415 struct kvm_s390_interrupt_info *inti = NULL;
2416 int r = 0;
2417 int len = attr->attr;
2418
2419 if (len % sizeof(struct kvm_s390_irq) != 0)
2420 return -EINVAL;
2421 else if (len > KVM_S390_FLIC_MAX_BUFFER)
2422 return -EINVAL;
2423
2424 while (len >= sizeof(struct kvm_s390_irq)) {
2425 inti = kzalloc(sizeof(*inti), GFP_KERNEL_ACCOUNT);
2426 if (!inti)
2427 return -ENOMEM;
2428
2429 r = copy_irq_from_user(inti, attr->addr);
2430 if (r) {
2431 kfree(inti);
2432 return r;
2433 }
2434 r = __inject_vm(dev->kvm, inti);
2435 if (r) {
2436 kfree(inti);
2437 return r;
2438 }
2439 len -= sizeof(struct kvm_s390_irq);
2440 attr->addr += sizeof(struct kvm_s390_irq);
2441 }
2442
2443 return r;
2444}
2445
2446static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2447{
2448 if (id >= MAX_S390_IO_ADAPTERS)
2449 return NULL;
2450 id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
2451 return kvm->arch.adapters[id];
2452}
2453
2454static int register_io_adapter(struct kvm_device *dev,
2455 struct kvm_device_attr *attr)
2456{
2457 struct s390_io_adapter *adapter;
2458 struct kvm_s390_io_adapter adapter_info;
2459
2460 if (copy_from_user(&adapter_info,
2461 (void __user *)attr->addr, sizeof(adapter_info)))
2462 return -EFAULT;
2463
2464 if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
2465 return -EINVAL;
2466
2467 adapter_info.id = array_index_nospec(adapter_info.id,
2468 MAX_S390_IO_ADAPTERS);
2469
2470 if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
2471 return -EINVAL;
2472
2473 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL_ACCOUNT);
2474 if (!adapter)
2475 return -ENOMEM;
2476
2477 adapter->id = adapter_info.id;
2478 adapter->isc = adapter_info.isc;
2479 adapter->maskable = adapter_info.maskable;
2480 adapter->masked = false;
2481 adapter->swap = adapter_info.swap;
2482 adapter->suppressible = (adapter_info.flags) &
2483 KVM_S390_ADAPTER_SUPPRESSIBLE;
2484 dev->kvm->arch.adapters[adapter->id] = adapter;
2485
2486 return 0;
2487}
2488
2489int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2490{
2491 int ret;
2492 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2493
2494 if (!adapter || !adapter->maskable)
2495 return -EINVAL;
2496 ret = adapter->masked;
2497 adapter->masked = masked;
2498 return ret;
2499}
2500
2501void kvm_s390_destroy_adapters(struct kvm *kvm)
2502{
2503 int i;
2504
2505 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++)
2506 kfree(kvm->arch.adapters[i]);
2507}
2508
2509static int modify_io_adapter(struct kvm_device *dev,
2510 struct kvm_device_attr *attr)
2511{
2512 struct kvm_s390_io_adapter_req req;
2513 struct s390_io_adapter *adapter;
2514 int ret;
2515
2516 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2517 return -EFAULT;
2518
2519 adapter = get_io_adapter(dev->kvm, req.id);
2520 if (!adapter)
2521 return -EINVAL;
2522 switch (req.type) {
2523 case KVM_S390_IO_ADAPTER_MASK:
2524 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2525 if (ret > 0)
2526 ret = 0;
2527 break;
2528 /*
2529 * The following operations are no longer needed and therefore no-ops.
2530 * The gpa to hva translation is done when an IRQ route is set up. The
2531 * set_irq code uses get_user_pages_remote() to do the actual write.
2532 */
2533 case KVM_S390_IO_ADAPTER_MAP:
2534 case KVM_S390_IO_ADAPTER_UNMAP:
2535 ret = 0;
2536 break;
2537 default:
2538 ret = -EINVAL;
2539 }
2540
2541 return ret;
2542}
2543
2544static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2545
2546{
2547 const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2548 u32 schid;
2549
2550 if (attr->flags)
2551 return -EINVAL;
2552 if (attr->attr != sizeof(schid))
2553 return -EINVAL;
2554 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2555 return -EFAULT;
2556 if (!schid)
2557 return -EINVAL;
2558 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2559 /*
2560 * If userspace is conforming to the architecture, we can have at most
2561 * one pending I/O interrupt per subchannel, so this is effectively a
2562 * clear all.
2563 */
2564 return 0;
2565}
2566
2567static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2568{
2569 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2570 struct kvm_s390_ais_req req;
2571 int ret = 0;
2572
2573 if (!test_kvm_facility(kvm, 72))
2574 return -EOPNOTSUPP;
2575
2576 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2577 return -EFAULT;
2578
2579 if (req.isc > MAX_ISC)
2580 return -EINVAL;
2581
2582 trace_kvm_s390_modify_ais_mode(req.isc,
2583 (fi->simm & AIS_MODE_MASK(req.isc)) ?
2584 (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2585 2 : KVM_S390_AIS_MODE_SINGLE :
2586 KVM_S390_AIS_MODE_ALL, req.mode);
2587
2588 mutex_lock(&fi->ais_lock);
2589 switch (req.mode) {
2590 case KVM_S390_AIS_MODE_ALL:
2591 fi->simm &= ~AIS_MODE_MASK(req.isc);
2592 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2593 break;
2594 case KVM_S390_AIS_MODE_SINGLE:
2595 fi->simm |= AIS_MODE_MASK(req.isc);
2596 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2597 break;
2598 default:
2599 ret = -EINVAL;
2600 }
2601 mutex_unlock(&fi->ais_lock);
2602
2603 return ret;
2604}
2605
2606static int kvm_s390_inject_airq(struct kvm *kvm,
2607 struct s390_io_adapter *adapter)
2608{
2609 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2610 struct kvm_s390_interrupt s390int = {
2611 .type = KVM_S390_INT_IO(1, 0, 0, 0),
2612 .parm = 0,
2613 .parm64 = isc_to_int_word(adapter->isc),
2614 };
2615 int ret = 0;
2616
2617 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2618 return kvm_s390_inject_vm(kvm, &s390int);
2619
2620 mutex_lock(&fi->ais_lock);
2621 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2622 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2623 goto out;
2624 }
2625
2626 ret = kvm_s390_inject_vm(kvm, &s390int);
2627 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2628 fi->nimm |= AIS_MODE_MASK(adapter->isc);
2629 trace_kvm_s390_modify_ais_mode(adapter->isc,
2630 KVM_S390_AIS_MODE_SINGLE, 2);
2631 }
2632out:
2633 mutex_unlock(&fi->ais_lock);
2634 return ret;
2635}
2636
2637static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2638{
2639 unsigned int id = attr->attr;
2640 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2641
2642 if (!adapter)
2643 return -EINVAL;
2644
2645 return kvm_s390_inject_airq(kvm, adapter);
2646}
2647
2648static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2649{
2650 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2651 struct kvm_s390_ais_all ais;
2652
2653 if (!test_kvm_facility(kvm, 72))
2654 return -EOPNOTSUPP;
2655
2656 if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2657 return -EFAULT;
2658
2659 mutex_lock(&fi->ais_lock);
2660 fi->simm = ais.simm;
2661 fi->nimm = ais.nimm;
2662 mutex_unlock(&fi->ais_lock);
2663
2664 return 0;
2665}
2666
2667static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2668{
2669 int r = 0;
2670 unsigned long i;
2671 struct kvm_vcpu *vcpu;
2672
2673 switch (attr->group) {
2674 case KVM_DEV_FLIC_ENQUEUE:
2675 r = enqueue_floating_irq(dev, attr);
2676 break;
2677 case KVM_DEV_FLIC_CLEAR_IRQS:
2678 kvm_s390_clear_float_irqs(dev->kvm);
2679 break;
2680 case KVM_DEV_FLIC_APF_ENABLE:
2681 dev->kvm->arch.gmap->pfault_enabled = 1;
2682 break;
2683 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2684 dev->kvm->arch.gmap->pfault_enabled = 0;
2685 /*
2686 * Make sure no async faults are in transition when
2687 * clearing the queues. So we don't need to worry
2688 * about late coming workers.
2689 */
2690 synchronize_srcu(&dev->kvm->srcu);
2691 kvm_for_each_vcpu(i, vcpu, dev->kvm)
2692 kvm_clear_async_pf_completion_queue(vcpu);
2693 break;
2694 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2695 r = register_io_adapter(dev, attr);
2696 break;
2697 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2698 r = modify_io_adapter(dev, attr);
2699 break;
2700 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2701 r = clear_io_irq(dev->kvm, attr);
2702 break;
2703 case KVM_DEV_FLIC_AISM:
2704 r = modify_ais_mode(dev->kvm, attr);
2705 break;
2706 case KVM_DEV_FLIC_AIRQ_INJECT:
2707 r = flic_inject_airq(dev->kvm, attr);
2708 break;
2709 case KVM_DEV_FLIC_AISM_ALL:
2710 r = flic_ais_mode_set_all(dev->kvm, attr);
2711 break;
2712 default:
2713 r = -EINVAL;
2714 }
2715
2716 return r;
2717}
2718
2719static int flic_has_attr(struct kvm_device *dev,
2720 struct kvm_device_attr *attr)
2721{
2722 switch (attr->group) {
2723 case KVM_DEV_FLIC_GET_ALL_IRQS:
2724 case KVM_DEV_FLIC_ENQUEUE:
2725 case KVM_DEV_FLIC_CLEAR_IRQS:
2726 case KVM_DEV_FLIC_APF_ENABLE:
2727 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2728 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2729 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2730 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2731 case KVM_DEV_FLIC_AISM:
2732 case KVM_DEV_FLIC_AIRQ_INJECT:
2733 case KVM_DEV_FLIC_AISM_ALL:
2734 return 0;
2735 }
2736 return -ENXIO;
2737}
2738
2739static int flic_create(struct kvm_device *dev, u32 type)
2740{
2741 if (!dev)
2742 return -EINVAL;
2743 if (dev->kvm->arch.flic)
2744 return -EINVAL;
2745 dev->kvm->arch.flic = dev;
2746 return 0;
2747}
2748
2749static void flic_destroy(struct kvm_device *dev)
2750{
2751 dev->kvm->arch.flic = NULL;
2752 kfree(dev);
2753}
2754
2755/* s390 floating irq controller (flic) */
2756struct kvm_device_ops kvm_flic_ops = {
2757 .name = "kvm-flic",
2758 .get_attr = flic_get_attr,
2759 .set_attr = flic_set_attr,
2760 .has_attr = flic_has_attr,
2761 .create = flic_create,
2762 .destroy = flic_destroy,
2763};
2764
2765static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2766{
2767 unsigned long bit;
2768
2769 bit = bit_nr + (addr % PAGE_SIZE) * 8;
2770
2771 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2772}
2773
2774static struct page *get_map_page(struct kvm *kvm, u64 uaddr)
2775{
2776 struct page *page = NULL;
2777
2778 mmap_read_lock(kvm->mm);
2779 get_user_pages_remote(kvm->mm, uaddr, 1, FOLL_WRITE,
2780 &page, NULL, NULL);
2781 mmap_read_unlock(kvm->mm);
2782 return page;
2783}
2784
2785static int adapter_indicators_set(struct kvm *kvm,
2786 struct s390_io_adapter *adapter,
2787 struct kvm_s390_adapter_int *adapter_int)
2788{
2789 unsigned long bit;
2790 int summary_set, idx;
2791 struct page *ind_page, *summary_page;
2792 void *map;
2793
2794 ind_page = get_map_page(kvm, adapter_int->ind_addr);
2795 if (!ind_page)
2796 return -1;
2797 summary_page = get_map_page(kvm, adapter_int->summary_addr);
2798 if (!summary_page) {
2799 put_page(ind_page);
2800 return -1;
2801 }
2802
2803 idx = srcu_read_lock(&kvm->srcu);
2804 map = page_address(ind_page);
2805 bit = get_ind_bit(adapter_int->ind_addr,
2806 adapter_int->ind_offset, adapter->swap);
2807 set_bit(bit, map);
2808 mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT);
2809 set_page_dirty_lock(ind_page);
2810 map = page_address(summary_page);
2811 bit = get_ind_bit(adapter_int->summary_addr,
2812 adapter_int->summary_offset, adapter->swap);
2813 summary_set = test_and_set_bit(bit, map);
2814 mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT);
2815 set_page_dirty_lock(summary_page);
2816 srcu_read_unlock(&kvm->srcu, idx);
2817
2818 put_page(ind_page);
2819 put_page(summary_page);
2820 return summary_set ? 0 : 1;
2821}
2822
2823/*
2824 * < 0 - not injected due to error
2825 * = 0 - coalesced, summary indicator already active
2826 * > 0 - injected interrupt
2827 */
2828static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2829 struct kvm *kvm, int irq_source_id, int level,
2830 bool line_status)
2831{
2832 int ret;
2833 struct s390_io_adapter *adapter;
2834
2835 /* We're only interested in the 0->1 transition. */
2836 if (!level)
2837 return 0;
2838 adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2839 if (!adapter)
2840 return -1;
2841 ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2842 if ((ret > 0) && !adapter->masked) {
2843 ret = kvm_s390_inject_airq(kvm, adapter);
2844 if (ret == 0)
2845 ret = 1;
2846 }
2847 return ret;
2848}
2849
2850/*
2851 * Inject the machine check to the guest.
2852 */
2853void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2854 struct mcck_volatile_info *mcck_info)
2855{
2856 struct kvm_s390_interrupt_info inti;
2857 struct kvm_s390_irq irq;
2858 struct kvm_s390_mchk_info *mchk;
2859 union mci mci;
2860 __u64 cr14 = 0; /* upper bits are not used */
2861 int rc;
2862
2863 mci.val = mcck_info->mcic;
2864 if (mci.sr)
2865 cr14 |= CR14_RECOVERY_SUBMASK;
2866 if (mci.dg)
2867 cr14 |= CR14_DEGRADATION_SUBMASK;
2868 if (mci.w)
2869 cr14 |= CR14_WARNING_SUBMASK;
2870
2871 mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2872 mchk->cr14 = cr14;
2873 mchk->mcic = mcck_info->mcic;
2874 mchk->ext_damage_code = mcck_info->ext_damage_code;
2875 mchk->failing_storage_address = mcck_info->failing_storage_address;
2876 if (mci.ck) {
2877 /* Inject the floating machine check */
2878 inti.type = KVM_S390_MCHK;
2879 rc = __inject_vm(vcpu->kvm, &inti);
2880 } else {
2881 /* Inject the machine check to specified vcpu */
2882 irq.type = KVM_S390_MCHK;
2883 rc = kvm_s390_inject_vcpu(vcpu, &irq);
2884 }
2885 WARN_ON_ONCE(rc);
2886}
2887
2888int kvm_set_routing_entry(struct kvm *kvm,
2889 struct kvm_kernel_irq_routing_entry *e,
2890 const struct kvm_irq_routing_entry *ue)
2891{
2892 u64 uaddr;
2893
2894 switch (ue->type) {
2895 /* we store the userspace addresses instead of the guest addresses */
2896 case KVM_IRQ_ROUTING_S390_ADAPTER:
2897 e->set = set_adapter_int;
2898 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.summary_addr);
2899 if (uaddr == -EFAULT)
2900 return -EFAULT;
2901 e->adapter.summary_addr = uaddr;
2902 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.ind_addr);
2903 if (uaddr == -EFAULT)
2904 return -EFAULT;
2905 e->adapter.ind_addr = uaddr;
2906 e->adapter.summary_offset = ue->u.adapter.summary_offset;
2907 e->adapter.ind_offset = ue->u.adapter.ind_offset;
2908 e->adapter.adapter_id = ue->u.adapter.adapter_id;
2909 return 0;
2910 default:
2911 return -EINVAL;
2912 }
2913}
2914
2915int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2916 int irq_source_id, int level, bool line_status)
2917{
2918 return -EINVAL;
2919}
2920
2921int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2922{
2923 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2924 struct kvm_s390_irq *buf;
2925 int r = 0;
2926 int n;
2927
2928 buf = vmalloc(len);
2929 if (!buf)
2930 return -ENOMEM;
2931
2932 if (copy_from_user((void *) buf, irqstate, len)) {
2933 r = -EFAULT;
2934 goto out_free;
2935 }
2936
2937 /*
2938 * Don't allow setting the interrupt state
2939 * when there are already interrupts pending
2940 */
2941 spin_lock(&li->lock);
2942 if (li->pending_irqs) {
2943 r = -EBUSY;
2944 goto out_unlock;
2945 }
2946
2947 for (n = 0; n < len / sizeof(*buf); n++) {
2948 r = do_inject_vcpu(vcpu, &buf[n]);
2949 if (r)
2950 break;
2951 }
2952
2953out_unlock:
2954 spin_unlock(&li->lock);
2955out_free:
2956 vfree(buf);
2957
2958 return r;
2959}
2960
2961static void store_local_irq(struct kvm_s390_local_interrupt *li,
2962 struct kvm_s390_irq *irq,
2963 unsigned long irq_type)
2964{
2965 switch (irq_type) {
2966 case IRQ_PEND_MCHK_EX:
2967 case IRQ_PEND_MCHK_REP:
2968 irq->type = KVM_S390_MCHK;
2969 irq->u.mchk = li->irq.mchk;
2970 break;
2971 case IRQ_PEND_PROG:
2972 irq->type = KVM_S390_PROGRAM_INT;
2973 irq->u.pgm = li->irq.pgm;
2974 break;
2975 case IRQ_PEND_PFAULT_INIT:
2976 irq->type = KVM_S390_INT_PFAULT_INIT;
2977 irq->u.ext = li->irq.ext;
2978 break;
2979 case IRQ_PEND_EXT_EXTERNAL:
2980 irq->type = KVM_S390_INT_EXTERNAL_CALL;
2981 irq->u.extcall = li->irq.extcall;
2982 break;
2983 case IRQ_PEND_EXT_CLOCK_COMP:
2984 irq->type = KVM_S390_INT_CLOCK_COMP;
2985 break;
2986 case IRQ_PEND_EXT_CPU_TIMER:
2987 irq->type = KVM_S390_INT_CPU_TIMER;
2988 break;
2989 case IRQ_PEND_SIGP_STOP:
2990 irq->type = KVM_S390_SIGP_STOP;
2991 irq->u.stop = li->irq.stop;
2992 break;
2993 case IRQ_PEND_RESTART:
2994 irq->type = KVM_S390_RESTART;
2995 break;
2996 case IRQ_PEND_SET_PREFIX:
2997 irq->type = KVM_S390_SIGP_SET_PREFIX;
2998 irq->u.prefix = li->irq.prefix;
2999 break;
3000 }
3001}
3002
3003int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
3004{
3005 int scn;
3006 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
3007 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
3008 unsigned long pending_irqs;
3009 struct kvm_s390_irq irq;
3010 unsigned long irq_type;
3011 int cpuaddr;
3012 int n = 0;
3013
3014 spin_lock(&li->lock);
3015 pending_irqs = li->pending_irqs;
3016 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
3017 sizeof(sigp_emerg_pending));
3018 spin_unlock(&li->lock);
3019
3020 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
3021 memset(&irq, 0, sizeof(irq));
3022 if (irq_type == IRQ_PEND_EXT_EMERGENCY)
3023 continue;
3024 if (n + sizeof(irq) > len)
3025 return -ENOBUFS;
3026 store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
3027 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3028 return -EFAULT;
3029 n += sizeof(irq);
3030 }
3031
3032 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
3033 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
3034 memset(&irq, 0, sizeof(irq));
3035 if (n + sizeof(irq) > len)
3036 return -ENOBUFS;
3037 irq.type = KVM_S390_INT_EMERGENCY;
3038 irq.u.emerg.code = cpuaddr;
3039 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3040 return -EFAULT;
3041 n += sizeof(irq);
3042 }
3043 }
3044
3045 if (sca_ext_call_pending(vcpu, &scn)) {
3046 if (n + sizeof(irq) > len)
3047 return -ENOBUFS;
3048 memset(&irq, 0, sizeof(irq));
3049 irq.type = KVM_S390_INT_EXTERNAL_CALL;
3050 irq.u.extcall.code = scn;
3051 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3052 return -EFAULT;
3053 n += sizeof(irq);
3054 }
3055
3056 return n;
3057}
3058
3059static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
3060{
3061 int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus);
3062 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3063 struct kvm_vcpu *vcpu;
3064 u8 vcpu_isc_mask;
3065
3066 for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) {
3067 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
3068 if (psw_ioint_disabled(vcpu))
3069 continue;
3070 vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
3071 if (deliverable_mask & vcpu_isc_mask) {
3072 /* lately kicked but not yet running */
3073 if (test_and_set_bit(vcpu_idx, gi->kicked_mask))
3074 return;
3075 kvm_s390_vcpu_wakeup(vcpu);
3076 return;
3077 }
3078 }
3079}
3080
3081static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3082{
3083 struct kvm_s390_gisa_interrupt *gi =
3084 container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3085 struct kvm *kvm =
3086 container_of(gi->origin, struct sie_page2, gisa)->kvm;
3087 u8 pending_mask;
3088
3089 pending_mask = gisa_get_ipm_or_restore_iam(gi);
3090 if (pending_mask) {
3091 __airqs_kick_single_vcpu(kvm, pending_mask);
3092 hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3093 return HRTIMER_RESTART;
3094 }
3095
3096 return HRTIMER_NORESTART;
3097}
3098
3099#define NULL_GISA_ADDR 0x00000000UL
3100#define NONE_GISA_ADDR 0x00000001UL
3101#define GISA_ADDR_MASK 0xfffff000UL
3102
3103static void process_gib_alert_list(void)
3104{
3105 struct kvm_s390_gisa_interrupt *gi;
3106 struct kvm_s390_gisa *gisa;
3107 struct kvm *kvm;
3108 u32 final, origin = 0UL;
3109
3110 do {
3111 /*
3112 * If the NONE_GISA_ADDR is still stored in the alert list
3113 * origin, we will leave the outer loop. No further GISA has
3114 * been added to the alert list by millicode while processing
3115 * the current alert list.
3116 */
3117 final = (origin & NONE_GISA_ADDR);
3118 /*
3119 * Cut off the alert list and store the NONE_GISA_ADDR in the
3120 * alert list origin to avoid further GAL interruptions.
3121 * A new alert list can be build up by millicode in parallel
3122 * for guests not in the yet cut-off alert list. When in the
3123 * final loop, store the NULL_GISA_ADDR instead. This will re-
3124 * enable GAL interruptions on the host again.
3125 */
3126 origin = xchg(&gib->alert_list_origin,
3127 (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3128 /*
3129 * Loop through the just cut-off alert list and start the
3130 * gisa timers to kick idle vcpus to consume the pending
3131 * interruptions asap.
3132 */
3133 while (origin & GISA_ADDR_MASK) {
3134 gisa = (struct kvm_s390_gisa *)(u64)origin;
3135 origin = gisa->next_alert;
3136 gisa->next_alert = (u32)(u64)gisa;
3137 kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3138 gi = &kvm->arch.gisa_int;
3139 if (hrtimer_active(&gi->timer))
3140 hrtimer_cancel(&gi->timer);
3141 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3142 }
3143 } while (!final);
3144
3145}
3146
3147void kvm_s390_gisa_clear(struct kvm *kvm)
3148{
3149 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3150
3151 if (!gi->origin)
3152 return;
3153 gisa_clear_ipm(gi->origin);
3154 VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3155}
3156
3157void kvm_s390_gisa_init(struct kvm *kvm)
3158{
3159 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3160
3161 if (!css_general_characteristics.aiv)
3162 return;
3163 gi->origin = &kvm->arch.sie_page2->gisa;
3164 gi->alert.mask = 0;
3165 spin_lock_init(&gi->alert.ref_lock);
3166 gi->expires = 50 * 1000; /* 50 usec */
3167 hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3168 gi->timer.function = gisa_vcpu_kicker;
3169 memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3170 gi->origin->next_alert = (u32)(u64)gi->origin;
3171 VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3172}
3173
3174void kvm_s390_gisa_enable(struct kvm *kvm)
3175{
3176 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3177 struct kvm_vcpu *vcpu;
3178 unsigned long i;
3179 u32 gisa_desc;
3180
3181 if (gi->origin)
3182 return;
3183 kvm_s390_gisa_init(kvm);
3184 gisa_desc = kvm_s390_get_gisa_desc(kvm);
3185 if (!gisa_desc)
3186 return;
3187 kvm_for_each_vcpu(i, vcpu, kvm) {
3188 mutex_lock(&vcpu->mutex);
3189 vcpu->arch.sie_block->gd = gisa_desc;
3190 vcpu->arch.sie_block->eca |= ECA_AIV;
3191 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3192 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3193 mutex_unlock(&vcpu->mutex);
3194 }
3195}
3196
3197void kvm_s390_gisa_destroy(struct kvm *kvm)
3198{
3199 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3200 struct kvm_s390_gisa *gisa = gi->origin;
3201
3202 if (!gi->origin)
3203 return;
3204 if (gi->alert.mask)
3205 KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
3206 kvm, gi->alert.mask);
3207 while (gisa_in_alert_list(gi->origin))
3208 cpu_relax();
3209 hrtimer_cancel(&gi->timer);
3210 gi->origin = NULL;
3211 VM_EVENT(kvm, 3, "gisa 0x%pK destroyed", gisa);
3212}
3213
3214void kvm_s390_gisa_disable(struct kvm *kvm)
3215{
3216 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3217 struct kvm_vcpu *vcpu;
3218 unsigned long i;
3219
3220 if (!gi->origin)
3221 return;
3222 kvm_for_each_vcpu(i, vcpu, kvm) {
3223 mutex_lock(&vcpu->mutex);
3224 vcpu->arch.sie_block->eca &= ~ECA_AIV;
3225 vcpu->arch.sie_block->gd = 0U;
3226 mutex_unlock(&vcpu->mutex);
3227 VCPU_EVENT(vcpu, 3, "AIV disabled for cpu %03u", vcpu->vcpu_id);
3228 }
3229 kvm_s390_gisa_destroy(kvm);
3230}
3231
3232/**
3233 * kvm_s390_gisc_register - register a guest ISC
3234 *
3235 * @kvm: the kernel vm to work with
3236 * @gisc: the guest interruption sub class to register
3237 *
3238 * The function extends the vm specific alert mask to use.
3239 * The effective IAM mask in the GISA is updated as well
3240 * in case the GISA is not part of the GIB alert list.
3241 * It will be updated latest when the IAM gets restored
3242 * by gisa_get_ipm_or_restore_iam().
3243 *
3244 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3245 * has registered with the channel subsystem.
3246 * -ENODEV in case the vm uses no GISA
3247 * -ERANGE in case the guest ISC is invalid
3248 */
3249int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3250{
3251 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3252
3253 if (!gi->origin)
3254 return -ENODEV;
3255 if (gisc > MAX_ISC)
3256 return -ERANGE;
3257
3258 spin_lock(&gi->alert.ref_lock);
3259 gi->alert.ref_count[gisc]++;
3260 if (gi->alert.ref_count[gisc] == 1) {
3261 gi->alert.mask |= 0x80 >> gisc;
3262 gisa_set_iam(gi->origin, gi->alert.mask);
3263 }
3264 spin_unlock(&gi->alert.ref_lock);
3265
3266 return gib->nisc;
3267}
3268EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3269
3270/**
3271 * kvm_s390_gisc_unregister - unregister a guest ISC
3272 *
3273 * @kvm: the kernel vm to work with
3274 * @gisc: the guest interruption sub class to register
3275 *
3276 * The function reduces the vm specific alert mask to use.
3277 * The effective IAM mask in the GISA is updated as well
3278 * in case the GISA is not part of the GIB alert list.
3279 * It will be updated latest when the IAM gets restored
3280 * by gisa_get_ipm_or_restore_iam().
3281 *
3282 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3283 * has registered with the channel subsystem.
3284 * -ENODEV in case the vm uses no GISA
3285 * -ERANGE in case the guest ISC is invalid
3286 * -EINVAL in case the guest ISC is not registered
3287 */
3288int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3289{
3290 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3291 int rc = 0;
3292
3293 if (!gi->origin)
3294 return -ENODEV;
3295 if (gisc > MAX_ISC)
3296 return -ERANGE;
3297
3298 spin_lock(&gi->alert.ref_lock);
3299 if (gi->alert.ref_count[gisc] == 0) {
3300 rc = -EINVAL;
3301 goto out;
3302 }
3303 gi->alert.ref_count[gisc]--;
3304 if (gi->alert.ref_count[gisc] == 0) {
3305 gi->alert.mask &= ~(0x80 >> gisc);
3306 gisa_set_iam(gi->origin, gi->alert.mask);
3307 }
3308out:
3309 spin_unlock(&gi->alert.ref_lock);
3310
3311 return rc;
3312}
3313EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3314
3315static void aen_host_forward(unsigned long si)
3316{
3317 struct kvm_s390_gisa_interrupt *gi;
3318 struct zpci_gaite *gaite;
3319 struct kvm *kvm;
3320
3321 gaite = (struct zpci_gaite *)aift->gait +
3322 (si * sizeof(struct zpci_gaite));
3323 if (gaite->count == 0)
3324 return;
3325 if (gaite->aisb != 0)
3326 set_bit_inv(gaite->aisbo, phys_to_virt(gaite->aisb));
3327
3328 kvm = kvm_s390_pci_si_to_kvm(aift, si);
3329 if (!kvm)
3330 return;
3331 gi = &kvm->arch.gisa_int;
3332
3333 if (!(gi->origin->g1.simm & AIS_MODE_MASK(gaite->gisc)) ||
3334 !(gi->origin->g1.nimm & AIS_MODE_MASK(gaite->gisc))) {
3335 gisa_set_ipm_gisc(gi->origin, gaite->gisc);
3336 if (hrtimer_active(&gi->timer))
3337 hrtimer_cancel(&gi->timer);
3338 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3339 kvm->stat.aen_forward++;
3340 }
3341}
3342
3343static void aen_process_gait(u8 isc)
3344{
3345 bool found = false, first = true;
3346 union zpci_sic_iib iib = {{0}};
3347 unsigned long si, flags;
3348
3349 spin_lock_irqsave(&aift->gait_lock, flags);
3350
3351 if (!aift->gait) {
3352 spin_unlock_irqrestore(&aift->gait_lock, flags);
3353 return;
3354 }
3355
3356 for (si = 0;;) {
3357 /* Scan adapter summary indicator bit vector */
3358 si = airq_iv_scan(aift->sbv, si, airq_iv_end(aift->sbv));
3359 if (si == -1UL) {
3360 if (first || found) {
3361 /* Re-enable interrupts. */
3362 zpci_set_irq_ctrl(SIC_IRQ_MODE_SINGLE, isc,
3363 &iib);
3364 first = found = false;
3365 } else {
3366 /* Interrupts on and all bits processed */
3367 break;
3368 }
3369 found = false;
3370 si = 0;
3371 /* Scan again after re-enabling interrupts */
3372 continue;
3373 }
3374 found = true;
3375 aen_host_forward(si);
3376 }
3377
3378 spin_unlock_irqrestore(&aift->gait_lock, flags);
3379}
3380
3381static void gib_alert_irq_handler(struct airq_struct *airq,
3382 struct tpi_info *tpi_info)
3383{
3384 struct tpi_adapter_info *info = (struct tpi_adapter_info *)tpi_info;
3385
3386 inc_irq_stat(IRQIO_GAL);
3387
3388 if ((info->forward || info->error) &&
3389 IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3390 aen_process_gait(info->isc);
3391 if (info->aism != 0)
3392 process_gib_alert_list();
3393 } else {
3394 process_gib_alert_list();
3395 }
3396}
3397
3398static struct airq_struct gib_alert_irq = {
3399 .handler = gib_alert_irq_handler,
3400 .lsi_ptr = &gib_alert_irq.lsi_mask,
3401};
3402
3403void kvm_s390_gib_destroy(void)
3404{
3405 if (!gib)
3406 return;
3407 if (kvm_s390_pci_interp_allowed() && aift) {
3408 mutex_lock(&aift->aift_lock);
3409 kvm_s390_pci_aen_exit();
3410 mutex_unlock(&aift->aift_lock);
3411 }
3412 chsc_sgib(0);
3413 unregister_adapter_interrupt(&gib_alert_irq);
3414 free_page((unsigned long)gib);
3415 gib = NULL;
3416}
3417
3418int kvm_s390_gib_init(u8 nisc)
3419{
3420 int rc = 0;
3421
3422 if (!css_general_characteristics.aiv) {
3423 KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3424 goto out;
3425 }
3426
3427 gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3428 if (!gib) {
3429 rc = -ENOMEM;
3430 goto out;
3431 }
3432
3433 gib_alert_irq.isc = nisc;
3434 if (register_adapter_interrupt(&gib_alert_irq)) {
3435 pr_err("Registering the GIB alert interruption handler failed\n");
3436 rc = -EIO;
3437 goto out_free_gib;
3438 }
3439
3440 gib->nisc = nisc;
3441 if (chsc_sgib((u32)(u64)gib)) {
3442 pr_err("Associating the GIB with the AIV facility failed\n");
3443 free_page((unsigned long)gib);
3444 gib = NULL;
3445 rc = -EIO;
3446 goto out_unreg_gal;
3447 }
3448
3449 if (kvm_s390_pci_interp_allowed()) {
3450 if (kvm_s390_pci_aen_init(nisc)) {
3451 pr_err("Initializing AEN for PCI failed\n");
3452 rc = -EIO;
3453 goto out_unreg_gal;
3454 }
3455 }
3456
3457 KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3458 goto out;
3459
3460out_unreg_gal:
3461 unregister_adapter_interrupt(&gib_alert_irq);
3462out_free_gib:
3463 free_page((unsigned long)gib);
3464 gib = NULL;
3465out:
3466 return rc;
3467}
1/*
2 * handling kvm guest interrupts
3 *
4 * Copyright IBM Corp. 2008,2014
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License (version 2 only)
8 * as published by the Free Software Foundation.
9 *
10 * Author(s): Carsten Otte <cotte@de.ibm.com>
11 */
12
13#include <linux/interrupt.h>
14#include <linux/kvm_host.h>
15#include <linux/hrtimer.h>
16#include <linux/mmu_context.h>
17#include <linux/signal.h>
18#include <linux/slab.h>
19#include <asm/asm-offsets.h>
20#include <asm/uaccess.h>
21#include "kvm-s390.h"
22#include "gaccess.h"
23#include "trace-s390.h"
24
25#define IOINT_SCHID_MASK 0x0000ffff
26#define IOINT_SSID_MASK 0x00030000
27#define IOINT_CSSID_MASK 0x03fc0000
28#define IOINT_AI_MASK 0x04000000
29
30static int is_ioint(u64 type)
31{
32 return ((type & 0xfffe0000u) != 0xfffe0000u);
33}
34
35int psw_extint_disabled(struct kvm_vcpu *vcpu)
36{
37 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
38}
39
40static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
41{
42 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
43}
44
45static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
46{
47 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
48}
49
50static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
51{
52 if ((vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PER) ||
53 (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO) ||
54 (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT))
55 return 0;
56 return 1;
57}
58
59static u64 int_word_to_isc_bits(u32 int_word)
60{
61 u8 isc = (int_word & 0x38000000) >> 27;
62
63 return (0x80 >> isc) << 24;
64}
65
66static int __interrupt_is_deliverable(struct kvm_vcpu *vcpu,
67 struct kvm_s390_interrupt_info *inti)
68{
69 switch (inti->type) {
70 case KVM_S390_INT_EXTERNAL_CALL:
71 if (psw_extint_disabled(vcpu))
72 return 0;
73 if (vcpu->arch.sie_block->gcr[0] & 0x2000ul)
74 return 1;
75 case KVM_S390_INT_EMERGENCY:
76 if (psw_extint_disabled(vcpu))
77 return 0;
78 if (vcpu->arch.sie_block->gcr[0] & 0x4000ul)
79 return 1;
80 return 0;
81 case KVM_S390_INT_SERVICE:
82 case KVM_S390_INT_PFAULT_INIT:
83 case KVM_S390_INT_PFAULT_DONE:
84 case KVM_S390_INT_VIRTIO:
85 if (psw_extint_disabled(vcpu))
86 return 0;
87 if (vcpu->arch.sie_block->gcr[0] & 0x200ul)
88 return 1;
89 return 0;
90 case KVM_S390_PROGRAM_INT:
91 case KVM_S390_SIGP_STOP:
92 case KVM_S390_SIGP_SET_PREFIX:
93 case KVM_S390_RESTART:
94 return 1;
95 case KVM_S390_MCHK:
96 if (psw_mchk_disabled(vcpu))
97 return 0;
98 if (vcpu->arch.sie_block->gcr[14] & inti->mchk.cr14)
99 return 1;
100 return 0;
101 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
102 if (psw_ioint_disabled(vcpu))
103 return 0;
104 if (vcpu->arch.sie_block->gcr[6] &
105 int_word_to_isc_bits(inti->io.io_int_word))
106 return 1;
107 return 0;
108 default:
109 printk(KERN_WARNING "illegal interrupt type %llx\n",
110 inti->type);
111 BUG();
112 }
113 return 0;
114}
115
116static void __set_cpu_idle(struct kvm_vcpu *vcpu)
117{
118 atomic_set_mask(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
119 set_bit(vcpu->vcpu_id, vcpu->arch.local_int.float_int->idle_mask);
120}
121
122static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
123{
124 atomic_clear_mask(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
125 clear_bit(vcpu->vcpu_id, vcpu->arch.local_int.float_int->idle_mask);
126}
127
128static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
129{
130 atomic_clear_mask(CPUSTAT_ECALL_PEND |
131 CPUSTAT_IO_INT | CPUSTAT_EXT_INT | CPUSTAT_STOP_INT,
132 &vcpu->arch.sie_block->cpuflags);
133 vcpu->arch.sie_block->lctl = 0x0000;
134 vcpu->arch.sie_block->ictl &= ~ICTL_LPSW;
135}
136
137static void __set_cpuflag(struct kvm_vcpu *vcpu, u32 flag)
138{
139 atomic_set_mask(flag, &vcpu->arch.sie_block->cpuflags);
140}
141
142static void __set_intercept_indicator(struct kvm_vcpu *vcpu,
143 struct kvm_s390_interrupt_info *inti)
144{
145 switch (inti->type) {
146 case KVM_S390_INT_EXTERNAL_CALL:
147 case KVM_S390_INT_EMERGENCY:
148 case KVM_S390_INT_SERVICE:
149 case KVM_S390_INT_PFAULT_INIT:
150 case KVM_S390_INT_PFAULT_DONE:
151 case KVM_S390_INT_VIRTIO:
152 if (psw_extint_disabled(vcpu))
153 __set_cpuflag(vcpu, CPUSTAT_EXT_INT);
154 else
155 vcpu->arch.sie_block->lctl |= LCTL_CR0;
156 break;
157 case KVM_S390_SIGP_STOP:
158 __set_cpuflag(vcpu, CPUSTAT_STOP_INT);
159 break;
160 case KVM_S390_MCHK:
161 if (psw_mchk_disabled(vcpu))
162 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
163 else
164 vcpu->arch.sie_block->lctl |= LCTL_CR14;
165 break;
166 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
167 if (psw_ioint_disabled(vcpu))
168 __set_cpuflag(vcpu, CPUSTAT_IO_INT);
169 else
170 vcpu->arch.sie_block->lctl |= LCTL_CR6;
171 break;
172 default:
173 BUG();
174 }
175}
176
177static void __do_deliver_interrupt(struct kvm_vcpu *vcpu,
178 struct kvm_s390_interrupt_info *inti)
179{
180 const unsigned short table[] = { 2, 4, 4, 6 };
181 int rc = 0;
182
183 switch (inti->type) {
184 case KVM_S390_INT_EMERGENCY:
185 VCPU_EVENT(vcpu, 4, "%s", "interrupt: sigp emerg");
186 vcpu->stat.deliver_emergency_signal++;
187 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type,
188 inti->emerg.code, 0);
189 rc = put_guest(vcpu, 0x1201, (u16 __user *)__LC_EXT_INT_CODE);
190 rc |= put_guest(vcpu, inti->emerg.code,
191 (u16 __user *)__LC_EXT_CPU_ADDR);
192 rc |= copy_to_guest(vcpu, __LC_EXT_OLD_PSW,
193 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
194 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
195 __LC_EXT_NEW_PSW, sizeof(psw_t));
196 break;
197 case KVM_S390_INT_EXTERNAL_CALL:
198 VCPU_EVENT(vcpu, 4, "%s", "interrupt: sigp ext call");
199 vcpu->stat.deliver_external_call++;
200 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type,
201 inti->extcall.code, 0);
202 rc = put_guest(vcpu, 0x1202, (u16 __user *)__LC_EXT_INT_CODE);
203 rc |= put_guest(vcpu, inti->extcall.code,
204 (u16 __user *)__LC_EXT_CPU_ADDR);
205 rc |= copy_to_guest(vcpu, __LC_EXT_OLD_PSW,
206 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
207 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
208 __LC_EXT_NEW_PSW, sizeof(psw_t));
209 break;
210 case KVM_S390_INT_SERVICE:
211 VCPU_EVENT(vcpu, 4, "interrupt: sclp parm:%x",
212 inti->ext.ext_params);
213 vcpu->stat.deliver_service_signal++;
214 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type,
215 inti->ext.ext_params, 0);
216 rc = put_guest(vcpu, 0x2401, (u16 __user *)__LC_EXT_INT_CODE);
217 rc |= copy_to_guest(vcpu, __LC_EXT_OLD_PSW,
218 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
219 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
220 __LC_EXT_NEW_PSW, sizeof(psw_t));
221 rc |= put_guest(vcpu, inti->ext.ext_params,
222 (u32 __user *)__LC_EXT_PARAMS);
223 break;
224 case KVM_S390_INT_PFAULT_INIT:
225 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type, 0,
226 inti->ext.ext_params2);
227 rc = put_guest(vcpu, 0x2603, (u16 __user *) __LC_EXT_INT_CODE);
228 rc |= put_guest(vcpu, 0x0600, (u16 __user *) __LC_EXT_CPU_ADDR);
229 rc |= copy_to_guest(vcpu, __LC_EXT_OLD_PSW,
230 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
231 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
232 __LC_EXT_NEW_PSW, sizeof(psw_t));
233 rc |= put_guest(vcpu, inti->ext.ext_params2,
234 (u64 __user *) __LC_EXT_PARAMS2);
235 break;
236 case KVM_S390_INT_PFAULT_DONE:
237 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type, 0,
238 inti->ext.ext_params2);
239 rc = put_guest(vcpu, 0x2603, (u16 __user *) __LC_EXT_INT_CODE);
240 rc |= put_guest(vcpu, 0x0680, (u16 __user *) __LC_EXT_CPU_ADDR);
241 rc |= copy_to_guest(vcpu, __LC_EXT_OLD_PSW,
242 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
243 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
244 __LC_EXT_NEW_PSW, sizeof(psw_t));
245 rc |= put_guest(vcpu, inti->ext.ext_params2,
246 (u64 __user *) __LC_EXT_PARAMS2);
247 break;
248 case KVM_S390_INT_VIRTIO:
249 VCPU_EVENT(vcpu, 4, "interrupt: virtio parm:%x,parm64:%llx",
250 inti->ext.ext_params, inti->ext.ext_params2);
251 vcpu->stat.deliver_virtio_interrupt++;
252 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type,
253 inti->ext.ext_params,
254 inti->ext.ext_params2);
255 rc = put_guest(vcpu, 0x2603, (u16 __user *)__LC_EXT_INT_CODE);
256 rc |= put_guest(vcpu, 0x0d00, (u16 __user *)__LC_EXT_CPU_ADDR);
257 rc |= copy_to_guest(vcpu, __LC_EXT_OLD_PSW,
258 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
259 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
260 __LC_EXT_NEW_PSW, sizeof(psw_t));
261 rc |= put_guest(vcpu, inti->ext.ext_params,
262 (u32 __user *)__LC_EXT_PARAMS);
263 rc |= put_guest(vcpu, inti->ext.ext_params2,
264 (u64 __user *)__LC_EXT_PARAMS2);
265 break;
266 case KVM_S390_SIGP_STOP:
267 VCPU_EVENT(vcpu, 4, "%s", "interrupt: cpu stop");
268 vcpu->stat.deliver_stop_signal++;
269 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type,
270 0, 0);
271 __set_intercept_indicator(vcpu, inti);
272 break;
273
274 case KVM_S390_SIGP_SET_PREFIX:
275 VCPU_EVENT(vcpu, 4, "interrupt: set prefix to %x",
276 inti->prefix.address);
277 vcpu->stat.deliver_prefix_signal++;
278 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type,
279 inti->prefix.address, 0);
280 kvm_s390_set_prefix(vcpu, inti->prefix.address);
281 break;
282
283 case KVM_S390_RESTART:
284 VCPU_EVENT(vcpu, 4, "%s", "interrupt: cpu restart");
285 vcpu->stat.deliver_restart_signal++;
286 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type,
287 0, 0);
288 rc = copy_to_guest(vcpu,
289 offsetof(struct _lowcore, restart_old_psw),
290 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
291 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
292 offsetof(struct _lowcore, restart_psw),
293 sizeof(psw_t));
294 atomic_clear_mask(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
295 break;
296 case KVM_S390_PROGRAM_INT:
297 VCPU_EVENT(vcpu, 4, "interrupt: pgm check code:%x, ilc:%x",
298 inti->pgm.code,
299 table[vcpu->arch.sie_block->ipa >> 14]);
300 vcpu->stat.deliver_program_int++;
301 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type,
302 inti->pgm.code, 0);
303 rc = put_guest(vcpu, inti->pgm.code, (u16 __user *)__LC_PGM_INT_CODE);
304 rc |= put_guest(vcpu, table[vcpu->arch.sie_block->ipa >> 14],
305 (u16 __user *)__LC_PGM_ILC);
306 rc |= copy_to_guest(vcpu, __LC_PGM_OLD_PSW,
307 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
308 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
309 __LC_PGM_NEW_PSW, sizeof(psw_t));
310 break;
311
312 case KVM_S390_MCHK:
313 VCPU_EVENT(vcpu, 4, "interrupt: machine check mcic=%llx",
314 inti->mchk.mcic);
315 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type,
316 inti->mchk.cr14,
317 inti->mchk.mcic);
318 rc = kvm_s390_vcpu_store_status(vcpu,
319 KVM_S390_STORE_STATUS_PREFIXED);
320 rc |= put_guest(vcpu, inti->mchk.mcic, (u64 __user *) __LC_MCCK_CODE);
321 rc |= copy_to_guest(vcpu, __LC_MCK_OLD_PSW,
322 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
323 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
324 __LC_MCK_NEW_PSW, sizeof(psw_t));
325 break;
326
327 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
328 {
329 __u32 param0 = ((__u32)inti->io.subchannel_id << 16) |
330 inti->io.subchannel_nr;
331 __u64 param1 = ((__u64)inti->io.io_int_parm << 32) |
332 inti->io.io_int_word;
333 VCPU_EVENT(vcpu, 4, "interrupt: I/O %llx", inti->type);
334 vcpu->stat.deliver_io_int++;
335 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type,
336 param0, param1);
337 rc = put_guest(vcpu, inti->io.subchannel_id,
338 (u16 __user *) __LC_SUBCHANNEL_ID);
339 rc |= put_guest(vcpu, inti->io.subchannel_nr,
340 (u16 __user *) __LC_SUBCHANNEL_NR);
341 rc |= put_guest(vcpu, inti->io.io_int_parm,
342 (u32 __user *) __LC_IO_INT_PARM);
343 rc |= put_guest(vcpu, inti->io.io_int_word,
344 (u32 __user *) __LC_IO_INT_WORD);
345 rc |= copy_to_guest(vcpu, __LC_IO_OLD_PSW,
346 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
347 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
348 __LC_IO_NEW_PSW, sizeof(psw_t));
349 break;
350 }
351 default:
352 BUG();
353 }
354 if (rc) {
355 printk("kvm: The guest lowcore is not mapped during interrupt "
356 "delivery, killing userspace\n");
357 do_exit(SIGKILL);
358 }
359}
360
361static int __try_deliver_ckc_interrupt(struct kvm_vcpu *vcpu)
362{
363 int rc;
364
365 if (psw_extint_disabled(vcpu))
366 return 0;
367 if (!(vcpu->arch.sie_block->gcr[0] & 0x800ul))
368 return 0;
369 rc = put_guest(vcpu, 0x1004, (u16 __user *)__LC_EXT_INT_CODE);
370 rc |= copy_to_guest(vcpu, __LC_EXT_OLD_PSW,
371 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
372 rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
373 __LC_EXT_NEW_PSW, sizeof(psw_t));
374 if (rc) {
375 printk("kvm: The guest lowcore is not mapped during interrupt "
376 "delivery, killing userspace\n");
377 do_exit(SIGKILL);
378 }
379 return 1;
380}
381
382int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu)
383{
384 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
385 struct kvm_s390_float_interrupt *fi = vcpu->arch.local_int.float_int;
386 struct kvm_s390_interrupt_info *inti;
387 int rc = 0;
388
389 if (atomic_read(&li->active)) {
390 spin_lock_bh(&li->lock);
391 list_for_each_entry(inti, &li->list, list)
392 if (__interrupt_is_deliverable(vcpu, inti)) {
393 rc = 1;
394 break;
395 }
396 spin_unlock_bh(&li->lock);
397 }
398
399 if ((!rc) && atomic_read(&fi->active)) {
400 spin_lock(&fi->lock);
401 list_for_each_entry(inti, &fi->list, list)
402 if (__interrupt_is_deliverable(vcpu, inti)) {
403 rc = 1;
404 break;
405 }
406 spin_unlock(&fi->lock);
407 }
408
409 if ((!rc) && (vcpu->arch.sie_block->ckc <
410 get_tod_clock_fast() + vcpu->arch.sie_block->epoch)) {
411 if ((!psw_extint_disabled(vcpu)) &&
412 (vcpu->arch.sie_block->gcr[0] & 0x800ul))
413 rc = 1;
414 }
415
416 return rc;
417}
418
419int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
420{
421 return 0;
422}
423
424int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
425{
426 u64 now, sltime;
427 DECLARE_WAITQUEUE(wait, current);
428
429 vcpu->stat.exit_wait_state++;
430 if (kvm_cpu_has_interrupt(vcpu))
431 return 0;
432
433 __set_cpu_idle(vcpu);
434 spin_lock_bh(&vcpu->arch.local_int.lock);
435 vcpu->arch.local_int.timer_due = 0;
436 spin_unlock_bh(&vcpu->arch.local_int.lock);
437
438 if (psw_interrupts_disabled(vcpu)) {
439 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
440 __unset_cpu_idle(vcpu);
441 return -EOPNOTSUPP; /* disabled wait */
442 }
443
444 if (psw_extint_disabled(vcpu) ||
445 (!(vcpu->arch.sie_block->gcr[0] & 0x800ul))) {
446 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
447 goto no_timer;
448 }
449
450 now = get_tod_clock_fast() + vcpu->arch.sie_block->epoch;
451 if (vcpu->arch.sie_block->ckc < now) {
452 __unset_cpu_idle(vcpu);
453 return 0;
454 }
455
456 sltime = tod_to_ns(vcpu->arch.sie_block->ckc - now);
457
458 hrtimer_start(&vcpu->arch.ckc_timer, ktime_set (0, sltime) , HRTIMER_MODE_REL);
459 VCPU_EVENT(vcpu, 5, "enabled wait via clock comparator: %llx ns", sltime);
460no_timer:
461 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
462 spin_lock(&vcpu->arch.local_int.float_int->lock);
463 spin_lock_bh(&vcpu->arch.local_int.lock);
464 add_wait_queue(&vcpu->wq, &wait);
465 while (list_empty(&vcpu->arch.local_int.list) &&
466 list_empty(&vcpu->arch.local_int.float_int->list) &&
467 (!vcpu->arch.local_int.timer_due) &&
468 !signal_pending(current)) {
469 set_current_state(TASK_INTERRUPTIBLE);
470 spin_unlock_bh(&vcpu->arch.local_int.lock);
471 spin_unlock(&vcpu->arch.local_int.float_int->lock);
472 schedule();
473 spin_lock(&vcpu->arch.local_int.float_int->lock);
474 spin_lock_bh(&vcpu->arch.local_int.lock);
475 }
476 __unset_cpu_idle(vcpu);
477 __set_current_state(TASK_RUNNING);
478 remove_wait_queue(&vcpu->wq, &wait);
479 spin_unlock_bh(&vcpu->arch.local_int.lock);
480 spin_unlock(&vcpu->arch.local_int.float_int->lock);
481 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
482
483 hrtimer_try_to_cancel(&vcpu->arch.ckc_timer);
484 return 0;
485}
486
487void kvm_s390_tasklet(unsigned long parm)
488{
489 struct kvm_vcpu *vcpu = (struct kvm_vcpu *) parm;
490
491 spin_lock(&vcpu->arch.local_int.lock);
492 vcpu->arch.local_int.timer_due = 1;
493 if (waitqueue_active(&vcpu->wq))
494 wake_up_interruptible(&vcpu->wq);
495 spin_unlock(&vcpu->arch.local_int.lock);
496}
497
498/*
499 * low level hrtimer wake routine. Because this runs in hardirq context
500 * we schedule a tasklet to do the real work.
501 */
502enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
503{
504 struct kvm_vcpu *vcpu;
505
506 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
507 vcpu->preempted = true;
508 tasklet_schedule(&vcpu->arch.tasklet);
509
510 return HRTIMER_NORESTART;
511}
512
513void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
514{
515 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
516 struct kvm_s390_interrupt_info *n, *inti = NULL;
517
518 spin_lock_bh(&li->lock);
519 list_for_each_entry_safe(inti, n, &li->list, list) {
520 list_del(&inti->list);
521 kfree(inti);
522 }
523 atomic_set(&li->active, 0);
524 spin_unlock_bh(&li->lock);
525}
526
527void kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
528{
529 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
530 struct kvm_s390_float_interrupt *fi = vcpu->arch.local_int.float_int;
531 struct kvm_s390_interrupt_info *n, *inti = NULL;
532 int deliver;
533
534 __reset_intercept_indicators(vcpu);
535 if (atomic_read(&li->active)) {
536 do {
537 deliver = 0;
538 spin_lock_bh(&li->lock);
539 list_for_each_entry_safe(inti, n, &li->list, list) {
540 if (__interrupt_is_deliverable(vcpu, inti)) {
541 list_del(&inti->list);
542 deliver = 1;
543 break;
544 }
545 __set_intercept_indicator(vcpu, inti);
546 }
547 if (list_empty(&li->list))
548 atomic_set(&li->active, 0);
549 spin_unlock_bh(&li->lock);
550 if (deliver) {
551 __do_deliver_interrupt(vcpu, inti);
552 kfree(inti);
553 }
554 } while (deliver);
555 }
556
557 if ((vcpu->arch.sie_block->ckc <
558 get_tod_clock_fast() + vcpu->arch.sie_block->epoch))
559 __try_deliver_ckc_interrupt(vcpu);
560
561 if (atomic_read(&fi->active)) {
562 do {
563 deliver = 0;
564 spin_lock(&fi->lock);
565 list_for_each_entry_safe(inti, n, &fi->list, list) {
566 if (__interrupt_is_deliverable(vcpu, inti)) {
567 list_del(&inti->list);
568 fi->irq_count--;
569 deliver = 1;
570 break;
571 }
572 __set_intercept_indicator(vcpu, inti);
573 }
574 if (list_empty(&fi->list))
575 atomic_set(&fi->active, 0);
576 spin_unlock(&fi->lock);
577 if (deliver) {
578 __do_deliver_interrupt(vcpu, inti);
579 kfree(inti);
580 }
581 } while (deliver);
582 }
583}
584
585void kvm_s390_deliver_pending_machine_checks(struct kvm_vcpu *vcpu)
586{
587 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
588 struct kvm_s390_float_interrupt *fi = vcpu->arch.local_int.float_int;
589 struct kvm_s390_interrupt_info *n, *inti = NULL;
590 int deliver;
591
592 __reset_intercept_indicators(vcpu);
593 if (atomic_read(&li->active)) {
594 do {
595 deliver = 0;
596 spin_lock_bh(&li->lock);
597 list_for_each_entry_safe(inti, n, &li->list, list) {
598 if ((inti->type == KVM_S390_MCHK) &&
599 __interrupt_is_deliverable(vcpu, inti)) {
600 list_del(&inti->list);
601 deliver = 1;
602 break;
603 }
604 __set_intercept_indicator(vcpu, inti);
605 }
606 if (list_empty(&li->list))
607 atomic_set(&li->active, 0);
608 spin_unlock_bh(&li->lock);
609 if (deliver) {
610 __do_deliver_interrupt(vcpu, inti);
611 kfree(inti);
612 }
613 } while (deliver);
614 }
615
616 if (atomic_read(&fi->active)) {
617 do {
618 deliver = 0;
619 spin_lock(&fi->lock);
620 list_for_each_entry_safe(inti, n, &fi->list, list) {
621 if ((inti->type == KVM_S390_MCHK) &&
622 __interrupt_is_deliverable(vcpu, inti)) {
623 list_del(&inti->list);
624 fi->irq_count--;
625 deliver = 1;
626 break;
627 }
628 __set_intercept_indicator(vcpu, inti);
629 }
630 if (list_empty(&fi->list))
631 atomic_set(&fi->active, 0);
632 spin_unlock(&fi->lock);
633 if (deliver) {
634 __do_deliver_interrupt(vcpu, inti);
635 kfree(inti);
636 }
637 } while (deliver);
638 }
639}
640
641int kvm_s390_inject_program_int(struct kvm_vcpu *vcpu, u16 code)
642{
643 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
644 struct kvm_s390_interrupt_info *inti;
645
646 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
647 if (!inti)
648 return -ENOMEM;
649
650 inti->type = KVM_S390_PROGRAM_INT;
651 inti->pgm.code = code;
652
653 VCPU_EVENT(vcpu, 3, "inject: program check %d (from kernel)", code);
654 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, inti->type, code, 0, 1);
655 spin_lock_bh(&li->lock);
656 list_add(&inti->list, &li->list);
657 atomic_set(&li->active, 1);
658 BUG_ON(waitqueue_active(li->wq));
659 spin_unlock_bh(&li->lock);
660 return 0;
661}
662
663struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
664 u64 cr6, u64 schid)
665{
666 struct kvm_s390_float_interrupt *fi;
667 struct kvm_s390_interrupt_info *inti, *iter;
668
669 if ((!schid && !cr6) || (schid && cr6))
670 return NULL;
671 mutex_lock(&kvm->lock);
672 fi = &kvm->arch.float_int;
673 spin_lock(&fi->lock);
674 inti = NULL;
675 list_for_each_entry(iter, &fi->list, list) {
676 if (!is_ioint(iter->type))
677 continue;
678 if (cr6 &&
679 ((cr6 & int_word_to_isc_bits(iter->io.io_int_word)) == 0))
680 continue;
681 if (schid) {
682 if (((schid & 0x00000000ffff0000) >> 16) !=
683 iter->io.subchannel_id)
684 continue;
685 if ((schid & 0x000000000000ffff) !=
686 iter->io.subchannel_nr)
687 continue;
688 }
689 inti = iter;
690 break;
691 }
692 if (inti) {
693 list_del_init(&inti->list);
694 fi->irq_count--;
695 }
696 if (list_empty(&fi->list))
697 atomic_set(&fi->active, 0);
698 spin_unlock(&fi->lock);
699 mutex_unlock(&kvm->lock);
700 return inti;
701}
702
703static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
704{
705 struct kvm_s390_local_interrupt *li;
706 struct kvm_s390_float_interrupt *fi;
707 struct kvm_s390_interrupt_info *iter;
708 struct kvm_vcpu *dst_vcpu = NULL;
709 int sigcpu;
710 int rc = 0;
711
712 mutex_lock(&kvm->lock);
713 fi = &kvm->arch.float_int;
714 spin_lock(&fi->lock);
715 if (fi->irq_count >= KVM_S390_MAX_FLOAT_IRQS) {
716 rc = -EINVAL;
717 goto unlock_fi;
718 }
719 fi->irq_count++;
720 if (!is_ioint(inti->type)) {
721 list_add_tail(&inti->list, &fi->list);
722 } else {
723 u64 isc_bits = int_word_to_isc_bits(inti->io.io_int_word);
724
725 /* Keep I/O interrupts sorted in isc order. */
726 list_for_each_entry(iter, &fi->list, list) {
727 if (!is_ioint(iter->type))
728 continue;
729 if (int_word_to_isc_bits(iter->io.io_int_word)
730 <= isc_bits)
731 continue;
732 break;
733 }
734 list_add_tail(&inti->list, &iter->list);
735 }
736 atomic_set(&fi->active, 1);
737 sigcpu = find_first_bit(fi->idle_mask, KVM_MAX_VCPUS);
738 if (sigcpu == KVM_MAX_VCPUS) {
739 do {
740 sigcpu = fi->next_rr_cpu++;
741 if (sigcpu == KVM_MAX_VCPUS)
742 sigcpu = fi->next_rr_cpu = 0;
743 } while (kvm_get_vcpu(kvm, sigcpu) == NULL);
744 }
745 dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
746 li = &dst_vcpu->arch.local_int;
747 spin_lock_bh(&li->lock);
748 atomic_set_mask(CPUSTAT_EXT_INT, li->cpuflags);
749 if (waitqueue_active(li->wq))
750 wake_up_interruptible(li->wq);
751 kvm_get_vcpu(kvm, sigcpu)->preempted = true;
752 spin_unlock_bh(&li->lock);
753unlock_fi:
754 spin_unlock(&fi->lock);
755 mutex_unlock(&kvm->lock);
756 return rc;
757}
758
759int kvm_s390_inject_vm(struct kvm *kvm,
760 struct kvm_s390_interrupt *s390int)
761{
762 struct kvm_s390_interrupt_info *inti;
763
764 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
765 if (!inti)
766 return -ENOMEM;
767
768 inti->type = s390int->type;
769 switch (inti->type) {
770 case KVM_S390_INT_VIRTIO:
771 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
772 s390int->parm, s390int->parm64);
773 inti->ext.ext_params = s390int->parm;
774 inti->ext.ext_params2 = s390int->parm64;
775 break;
776 case KVM_S390_INT_SERVICE:
777 VM_EVENT(kvm, 5, "inject: sclp parm:%x", s390int->parm);
778 inti->ext.ext_params = s390int->parm;
779 break;
780 case KVM_S390_INT_PFAULT_DONE:
781 inti->type = s390int->type;
782 inti->ext.ext_params2 = s390int->parm64;
783 break;
784 case KVM_S390_MCHK:
785 VM_EVENT(kvm, 5, "inject: machine check parm64:%llx",
786 s390int->parm64);
787 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
788 inti->mchk.mcic = s390int->parm64;
789 break;
790 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
791 if (inti->type & IOINT_AI_MASK)
792 VM_EVENT(kvm, 5, "%s", "inject: I/O (AI)");
793 else
794 VM_EVENT(kvm, 5, "inject: I/O css %x ss %x schid %04x",
795 s390int->type & IOINT_CSSID_MASK,
796 s390int->type & IOINT_SSID_MASK,
797 s390int->type & IOINT_SCHID_MASK);
798 inti->io.subchannel_id = s390int->parm >> 16;
799 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
800 inti->io.io_int_parm = s390int->parm64 >> 32;
801 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
802 break;
803 default:
804 kfree(inti);
805 return -EINVAL;
806 }
807 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
808 2);
809
810 return __inject_vm(kvm, inti);
811}
812
813int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu,
814 struct kvm_s390_interrupt *s390int)
815{
816 struct kvm_s390_local_interrupt *li;
817 struct kvm_s390_interrupt_info *inti;
818
819 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
820 if (!inti)
821 return -ENOMEM;
822
823 switch (s390int->type) {
824 case KVM_S390_PROGRAM_INT:
825 if (s390int->parm & 0xffff0000) {
826 kfree(inti);
827 return -EINVAL;
828 }
829 inti->type = s390int->type;
830 inti->pgm.code = s390int->parm;
831 VCPU_EVENT(vcpu, 3, "inject: program check %d (from user)",
832 s390int->parm);
833 break;
834 case KVM_S390_SIGP_SET_PREFIX:
835 inti->prefix.address = s390int->parm;
836 inti->type = s390int->type;
837 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x (from user)",
838 s390int->parm);
839 break;
840 case KVM_S390_SIGP_STOP:
841 case KVM_S390_RESTART:
842 VCPU_EVENT(vcpu, 3, "inject: type %x", s390int->type);
843 inti->type = s390int->type;
844 break;
845 case KVM_S390_INT_EXTERNAL_CALL:
846 if (s390int->parm & 0xffff0000) {
847 kfree(inti);
848 return -EINVAL;
849 }
850 VCPU_EVENT(vcpu, 3, "inject: external call source-cpu:%u",
851 s390int->parm);
852 inti->type = s390int->type;
853 inti->extcall.code = s390int->parm;
854 break;
855 case KVM_S390_INT_EMERGENCY:
856 if (s390int->parm & 0xffff0000) {
857 kfree(inti);
858 return -EINVAL;
859 }
860 VCPU_EVENT(vcpu, 3, "inject: emergency %u\n", s390int->parm);
861 inti->type = s390int->type;
862 inti->emerg.code = s390int->parm;
863 break;
864 case KVM_S390_MCHK:
865 VCPU_EVENT(vcpu, 5, "inject: machine check parm64:%llx",
866 s390int->parm64);
867 inti->type = s390int->type;
868 inti->mchk.mcic = s390int->parm64;
869 break;
870 case KVM_S390_INT_PFAULT_INIT:
871 inti->type = s390int->type;
872 inti->ext.ext_params2 = s390int->parm64;
873 break;
874 case KVM_S390_INT_VIRTIO:
875 case KVM_S390_INT_SERVICE:
876 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
877 default:
878 kfree(inti);
879 return -EINVAL;
880 }
881 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, s390int->type, s390int->parm,
882 s390int->parm64, 2);
883
884 mutex_lock(&vcpu->kvm->lock);
885 li = &vcpu->arch.local_int;
886 spin_lock_bh(&li->lock);
887 if (inti->type == KVM_S390_PROGRAM_INT)
888 list_add(&inti->list, &li->list);
889 else
890 list_add_tail(&inti->list, &li->list);
891 atomic_set(&li->active, 1);
892 if (inti->type == KVM_S390_SIGP_STOP)
893 li->action_bits |= ACTION_STOP_ON_STOP;
894 atomic_set_mask(CPUSTAT_EXT_INT, li->cpuflags);
895 if (waitqueue_active(&vcpu->wq))
896 wake_up_interruptible(&vcpu->wq);
897 vcpu->preempted = true;
898 spin_unlock_bh(&li->lock);
899 mutex_unlock(&vcpu->kvm->lock);
900 return 0;
901}
902
903static void clear_floating_interrupts(struct kvm *kvm)
904{
905 struct kvm_s390_float_interrupt *fi;
906 struct kvm_s390_interrupt_info *n, *inti = NULL;
907
908 mutex_lock(&kvm->lock);
909 fi = &kvm->arch.float_int;
910 spin_lock(&fi->lock);
911 list_for_each_entry_safe(inti, n, &fi->list, list) {
912 list_del(&inti->list);
913 kfree(inti);
914 }
915 fi->irq_count = 0;
916 atomic_set(&fi->active, 0);
917 spin_unlock(&fi->lock);
918 mutex_unlock(&kvm->lock);
919}
920
921static inline int copy_irq_to_user(struct kvm_s390_interrupt_info *inti,
922 u8 *addr)
923{
924 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
925 struct kvm_s390_irq irq = {0};
926
927 irq.type = inti->type;
928 switch (inti->type) {
929 case KVM_S390_INT_PFAULT_INIT:
930 case KVM_S390_INT_PFAULT_DONE:
931 case KVM_S390_INT_VIRTIO:
932 case KVM_S390_INT_SERVICE:
933 irq.u.ext = inti->ext;
934 break;
935 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
936 irq.u.io = inti->io;
937 break;
938 case KVM_S390_MCHK:
939 irq.u.mchk = inti->mchk;
940 break;
941 default:
942 return -EINVAL;
943 }
944
945 if (copy_to_user(uptr, &irq, sizeof(irq)))
946 return -EFAULT;
947
948 return 0;
949}
950
951static int get_all_floating_irqs(struct kvm *kvm, __u8 *buf, __u64 len)
952{
953 struct kvm_s390_interrupt_info *inti;
954 struct kvm_s390_float_interrupt *fi;
955 int ret = 0;
956 int n = 0;
957
958 mutex_lock(&kvm->lock);
959 fi = &kvm->arch.float_int;
960 spin_lock(&fi->lock);
961
962 list_for_each_entry(inti, &fi->list, list) {
963 if (len < sizeof(struct kvm_s390_irq)) {
964 /* signal userspace to try again */
965 ret = -ENOMEM;
966 break;
967 }
968 ret = copy_irq_to_user(inti, buf);
969 if (ret)
970 break;
971 buf += sizeof(struct kvm_s390_irq);
972 len -= sizeof(struct kvm_s390_irq);
973 n++;
974 }
975
976 spin_unlock(&fi->lock);
977 mutex_unlock(&kvm->lock);
978
979 return ret < 0 ? ret : n;
980}
981
982static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
983{
984 int r;
985
986 switch (attr->group) {
987 case KVM_DEV_FLIC_GET_ALL_IRQS:
988 r = get_all_floating_irqs(dev->kvm, (u8 *) attr->addr,
989 attr->attr);
990 break;
991 default:
992 r = -EINVAL;
993 }
994
995 return r;
996}
997
998static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
999 u64 addr)
1000{
1001 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
1002 void *target = NULL;
1003 void __user *source;
1004 u64 size;
1005
1006 if (get_user(inti->type, (u64 __user *)addr))
1007 return -EFAULT;
1008
1009 switch (inti->type) {
1010 case KVM_S390_INT_PFAULT_INIT:
1011 case KVM_S390_INT_PFAULT_DONE:
1012 case KVM_S390_INT_VIRTIO:
1013 case KVM_S390_INT_SERVICE:
1014 target = (void *) &inti->ext;
1015 source = &uptr->u.ext;
1016 size = sizeof(inti->ext);
1017 break;
1018 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1019 target = (void *) &inti->io;
1020 source = &uptr->u.io;
1021 size = sizeof(inti->io);
1022 break;
1023 case KVM_S390_MCHK:
1024 target = (void *) &inti->mchk;
1025 source = &uptr->u.mchk;
1026 size = sizeof(inti->mchk);
1027 break;
1028 default:
1029 return -EINVAL;
1030 }
1031
1032 if (copy_from_user(target, source, size))
1033 return -EFAULT;
1034
1035 return 0;
1036}
1037
1038static int enqueue_floating_irq(struct kvm_device *dev,
1039 struct kvm_device_attr *attr)
1040{
1041 struct kvm_s390_interrupt_info *inti = NULL;
1042 int r = 0;
1043 int len = attr->attr;
1044
1045 if (len % sizeof(struct kvm_s390_irq) != 0)
1046 return -EINVAL;
1047 else if (len > KVM_S390_FLIC_MAX_BUFFER)
1048 return -EINVAL;
1049
1050 while (len >= sizeof(struct kvm_s390_irq)) {
1051 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1052 if (!inti)
1053 return -ENOMEM;
1054
1055 r = copy_irq_from_user(inti, attr->addr);
1056 if (r) {
1057 kfree(inti);
1058 return r;
1059 }
1060 r = __inject_vm(dev->kvm, inti);
1061 if (r) {
1062 kfree(inti);
1063 return r;
1064 }
1065 len -= sizeof(struct kvm_s390_irq);
1066 attr->addr += sizeof(struct kvm_s390_irq);
1067 }
1068
1069 return r;
1070}
1071
1072static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
1073{
1074 if (id >= MAX_S390_IO_ADAPTERS)
1075 return NULL;
1076 return kvm->arch.adapters[id];
1077}
1078
1079static int register_io_adapter(struct kvm_device *dev,
1080 struct kvm_device_attr *attr)
1081{
1082 struct s390_io_adapter *adapter;
1083 struct kvm_s390_io_adapter adapter_info;
1084
1085 if (copy_from_user(&adapter_info,
1086 (void __user *)attr->addr, sizeof(adapter_info)))
1087 return -EFAULT;
1088
1089 if ((adapter_info.id >= MAX_S390_IO_ADAPTERS) ||
1090 (dev->kvm->arch.adapters[adapter_info.id] != NULL))
1091 return -EINVAL;
1092
1093 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
1094 if (!adapter)
1095 return -ENOMEM;
1096
1097 INIT_LIST_HEAD(&adapter->maps);
1098 init_rwsem(&adapter->maps_lock);
1099 atomic_set(&adapter->nr_maps, 0);
1100 adapter->id = adapter_info.id;
1101 adapter->isc = adapter_info.isc;
1102 adapter->maskable = adapter_info.maskable;
1103 adapter->masked = false;
1104 adapter->swap = adapter_info.swap;
1105 dev->kvm->arch.adapters[adapter->id] = adapter;
1106
1107 return 0;
1108}
1109
1110int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
1111{
1112 int ret;
1113 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
1114
1115 if (!adapter || !adapter->maskable)
1116 return -EINVAL;
1117 ret = adapter->masked;
1118 adapter->masked = masked;
1119 return ret;
1120}
1121
1122static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr)
1123{
1124 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
1125 struct s390_map_info *map;
1126 int ret;
1127
1128 if (!adapter || !addr)
1129 return -EINVAL;
1130
1131 map = kzalloc(sizeof(*map), GFP_KERNEL);
1132 if (!map) {
1133 ret = -ENOMEM;
1134 goto out;
1135 }
1136 INIT_LIST_HEAD(&map->list);
1137 map->guest_addr = addr;
1138 map->addr = gmap_translate(addr, kvm->arch.gmap);
1139 if (map->addr == -EFAULT) {
1140 ret = -EFAULT;
1141 goto out;
1142 }
1143 ret = get_user_pages_fast(map->addr, 1, 1, &map->page);
1144 if (ret < 0)
1145 goto out;
1146 BUG_ON(ret != 1);
1147 down_write(&adapter->maps_lock);
1148 if (atomic_inc_return(&adapter->nr_maps) < MAX_S390_ADAPTER_MAPS) {
1149 list_add_tail(&map->list, &adapter->maps);
1150 ret = 0;
1151 } else {
1152 put_page(map->page);
1153 ret = -EINVAL;
1154 }
1155 up_write(&adapter->maps_lock);
1156out:
1157 if (ret)
1158 kfree(map);
1159 return ret;
1160}
1161
1162static int kvm_s390_adapter_unmap(struct kvm *kvm, unsigned int id, __u64 addr)
1163{
1164 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
1165 struct s390_map_info *map, *tmp;
1166 int found = 0;
1167
1168 if (!adapter || !addr)
1169 return -EINVAL;
1170
1171 down_write(&adapter->maps_lock);
1172 list_for_each_entry_safe(map, tmp, &adapter->maps, list) {
1173 if (map->guest_addr == addr) {
1174 found = 1;
1175 atomic_dec(&adapter->nr_maps);
1176 list_del(&map->list);
1177 put_page(map->page);
1178 kfree(map);
1179 break;
1180 }
1181 }
1182 up_write(&adapter->maps_lock);
1183
1184 return found ? 0 : -EINVAL;
1185}
1186
1187void kvm_s390_destroy_adapters(struct kvm *kvm)
1188{
1189 int i;
1190 struct s390_map_info *map, *tmp;
1191
1192 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) {
1193 if (!kvm->arch.adapters[i])
1194 continue;
1195 list_for_each_entry_safe(map, tmp,
1196 &kvm->arch.adapters[i]->maps, list) {
1197 list_del(&map->list);
1198 put_page(map->page);
1199 kfree(map);
1200 }
1201 kfree(kvm->arch.adapters[i]);
1202 }
1203}
1204
1205static int modify_io_adapter(struct kvm_device *dev,
1206 struct kvm_device_attr *attr)
1207{
1208 struct kvm_s390_io_adapter_req req;
1209 struct s390_io_adapter *adapter;
1210 int ret;
1211
1212 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
1213 return -EFAULT;
1214
1215 adapter = get_io_adapter(dev->kvm, req.id);
1216 if (!adapter)
1217 return -EINVAL;
1218 switch (req.type) {
1219 case KVM_S390_IO_ADAPTER_MASK:
1220 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
1221 if (ret > 0)
1222 ret = 0;
1223 break;
1224 case KVM_S390_IO_ADAPTER_MAP:
1225 ret = kvm_s390_adapter_map(dev->kvm, req.id, req.addr);
1226 break;
1227 case KVM_S390_IO_ADAPTER_UNMAP:
1228 ret = kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr);
1229 break;
1230 default:
1231 ret = -EINVAL;
1232 }
1233
1234 return ret;
1235}
1236
1237static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1238{
1239 int r = 0;
1240 unsigned int i;
1241 struct kvm_vcpu *vcpu;
1242
1243 switch (attr->group) {
1244 case KVM_DEV_FLIC_ENQUEUE:
1245 r = enqueue_floating_irq(dev, attr);
1246 break;
1247 case KVM_DEV_FLIC_CLEAR_IRQS:
1248 r = 0;
1249 clear_floating_interrupts(dev->kvm);
1250 break;
1251 case KVM_DEV_FLIC_APF_ENABLE:
1252 dev->kvm->arch.gmap->pfault_enabled = 1;
1253 break;
1254 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
1255 dev->kvm->arch.gmap->pfault_enabled = 0;
1256 /*
1257 * Make sure no async faults are in transition when
1258 * clearing the queues. So we don't need to worry
1259 * about late coming workers.
1260 */
1261 synchronize_srcu(&dev->kvm->srcu);
1262 kvm_for_each_vcpu(i, vcpu, dev->kvm)
1263 kvm_clear_async_pf_completion_queue(vcpu);
1264 break;
1265 case KVM_DEV_FLIC_ADAPTER_REGISTER:
1266 r = register_io_adapter(dev, attr);
1267 break;
1268 case KVM_DEV_FLIC_ADAPTER_MODIFY:
1269 r = modify_io_adapter(dev, attr);
1270 break;
1271 default:
1272 r = -EINVAL;
1273 }
1274
1275 return r;
1276}
1277
1278static int flic_create(struct kvm_device *dev, u32 type)
1279{
1280 if (!dev)
1281 return -EINVAL;
1282 if (dev->kvm->arch.flic)
1283 return -EINVAL;
1284 dev->kvm->arch.flic = dev;
1285 return 0;
1286}
1287
1288static void flic_destroy(struct kvm_device *dev)
1289{
1290 dev->kvm->arch.flic = NULL;
1291 kfree(dev);
1292}
1293
1294/* s390 floating irq controller (flic) */
1295struct kvm_device_ops kvm_flic_ops = {
1296 .name = "kvm-flic",
1297 .get_attr = flic_get_attr,
1298 .set_attr = flic_set_attr,
1299 .create = flic_create,
1300 .destroy = flic_destroy,
1301};
1302
1303static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
1304{
1305 unsigned long bit;
1306
1307 bit = bit_nr + (addr % PAGE_SIZE) * 8;
1308
1309 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
1310}
1311
1312static struct s390_map_info *get_map_info(struct s390_io_adapter *adapter,
1313 u64 addr)
1314{
1315 struct s390_map_info *map;
1316
1317 if (!adapter)
1318 return NULL;
1319
1320 list_for_each_entry(map, &adapter->maps, list) {
1321 if (map->guest_addr == addr)
1322 return map;
1323 }
1324 return NULL;
1325}
1326
1327static int adapter_indicators_set(struct kvm *kvm,
1328 struct s390_io_adapter *adapter,
1329 struct kvm_s390_adapter_int *adapter_int)
1330{
1331 unsigned long bit;
1332 int summary_set, idx;
1333 struct s390_map_info *info;
1334 void *map;
1335
1336 info = get_map_info(adapter, adapter_int->ind_addr);
1337 if (!info)
1338 return -1;
1339 map = page_address(info->page);
1340 bit = get_ind_bit(info->addr, adapter_int->ind_offset, adapter->swap);
1341 set_bit(bit, map);
1342 idx = srcu_read_lock(&kvm->srcu);
1343 mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
1344 set_page_dirty_lock(info->page);
1345 info = get_map_info(adapter, adapter_int->summary_addr);
1346 if (!info) {
1347 srcu_read_unlock(&kvm->srcu, idx);
1348 return -1;
1349 }
1350 map = page_address(info->page);
1351 bit = get_ind_bit(info->addr, adapter_int->summary_offset,
1352 adapter->swap);
1353 summary_set = test_and_set_bit(bit, map);
1354 mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
1355 set_page_dirty_lock(info->page);
1356 srcu_read_unlock(&kvm->srcu, idx);
1357 return summary_set ? 0 : 1;
1358}
1359
1360/*
1361 * < 0 - not injected due to error
1362 * = 0 - coalesced, summary indicator already active
1363 * > 0 - injected interrupt
1364 */
1365static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
1366 struct kvm *kvm, int irq_source_id, int level,
1367 bool line_status)
1368{
1369 int ret;
1370 struct s390_io_adapter *adapter;
1371
1372 /* We're only interested in the 0->1 transition. */
1373 if (!level)
1374 return 0;
1375 adapter = get_io_adapter(kvm, e->adapter.adapter_id);
1376 if (!adapter)
1377 return -1;
1378 down_read(&adapter->maps_lock);
1379 ret = adapter_indicators_set(kvm, adapter, &e->adapter);
1380 up_read(&adapter->maps_lock);
1381 if ((ret > 0) && !adapter->masked) {
1382 struct kvm_s390_interrupt s390int = {
1383 .type = KVM_S390_INT_IO(1, 0, 0, 0),
1384 .parm = 0,
1385 .parm64 = (adapter->isc << 27) | 0x80000000,
1386 };
1387 ret = kvm_s390_inject_vm(kvm, &s390int);
1388 if (ret == 0)
1389 ret = 1;
1390 }
1391 return ret;
1392}
1393
1394int kvm_set_routing_entry(struct kvm_irq_routing_table *rt,
1395 struct kvm_kernel_irq_routing_entry *e,
1396 const struct kvm_irq_routing_entry *ue)
1397{
1398 int ret;
1399
1400 switch (ue->type) {
1401 case KVM_IRQ_ROUTING_S390_ADAPTER:
1402 e->set = set_adapter_int;
1403 e->adapter.summary_addr = ue->u.adapter.summary_addr;
1404 e->adapter.ind_addr = ue->u.adapter.ind_addr;
1405 e->adapter.summary_offset = ue->u.adapter.summary_offset;
1406 e->adapter.ind_offset = ue->u.adapter.ind_offset;
1407 e->adapter.adapter_id = ue->u.adapter.adapter_id;
1408 ret = 0;
1409 break;
1410 default:
1411 ret = -EINVAL;
1412 }
1413
1414 return ret;
1415}
1416
1417int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
1418 int irq_source_id, int level, bool line_status)
1419{
1420 return -EINVAL;
1421}