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/access-regs.h>
23#include <asm/asm-offsets.h>
24#include <asm/dis.h>
25#include <linux/uaccess.h>
26#include <asm/sclp.h>
27#include <asm/isc.h>
28#include <asm/gmap.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 if (!kvm_s390_use_sca_entries())
122 return;
123 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_ECALL_PEND);
124 read_lock(&vcpu->kvm->arch.sca_lock);
125 if (vcpu->kvm->arch.use_esca) {
126 struct esca_block *sca = vcpu->kvm->arch.sca;
127 union esca_sigp_ctrl *sigp_ctrl =
128 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
129
130 WRITE_ONCE(sigp_ctrl->value, 0);
131 } else {
132 struct bsca_block *sca = vcpu->kvm->arch.sca;
133 union bsca_sigp_ctrl *sigp_ctrl =
134 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
135
136 WRITE_ONCE(sigp_ctrl->value, 0);
137 }
138 read_unlock(&vcpu->kvm->arch.sca_lock);
139}
140
141int psw_extint_disabled(struct kvm_vcpu *vcpu)
142{
143 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
144}
145
146static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
147{
148 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
149}
150
151static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
152{
153 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
154}
155
156static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
157{
158 return psw_extint_disabled(vcpu) &&
159 psw_ioint_disabled(vcpu) &&
160 psw_mchk_disabled(vcpu);
161}
162
163static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
164{
165 if (psw_extint_disabled(vcpu) ||
166 !(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
167 return 0;
168 if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
169 /* No timer interrupts when single stepping */
170 return 0;
171 return 1;
172}
173
174static int ckc_irq_pending(struct kvm_vcpu *vcpu)
175{
176 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
177 const u64 ckc = vcpu->arch.sie_block->ckc;
178
179 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
180 if ((s64)ckc >= (s64)now)
181 return 0;
182 } else if (ckc >= now) {
183 return 0;
184 }
185 return ckc_interrupts_enabled(vcpu);
186}
187
188static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
189{
190 return !psw_extint_disabled(vcpu) &&
191 (vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK);
192}
193
194static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
195{
196 if (!cpu_timer_interrupts_enabled(vcpu))
197 return 0;
198 return kvm_s390_get_cpu_timer(vcpu) >> 63;
199}
200
201static uint64_t isc_to_isc_bits(int isc)
202{
203 return (0x80 >> isc) << 24;
204}
205
206static inline u32 isc_to_int_word(u8 isc)
207{
208 return ((u32)isc << 27) | 0x80000000;
209}
210
211static inline u8 int_word_to_isc(u32 int_word)
212{
213 return (int_word & 0x38000000) >> 27;
214}
215
216/*
217 * To use atomic bitmap functions, we have to provide a bitmap address
218 * that is u64 aligned. However, the ipm might be u32 aligned.
219 * Therefore, we logically start the bitmap at the very beginning of the
220 * struct and fixup the bit number.
221 */
222#define IPM_BIT_OFFSET (offsetof(struct kvm_s390_gisa, ipm) * BITS_PER_BYTE)
223
224/**
225 * gisa_set_iam - change the GISA interruption alert mask
226 *
227 * @gisa: gisa to operate on
228 * @iam: new IAM value to use
229 *
230 * Change the IAM atomically with the next alert address and the IPM
231 * of the GISA if the GISA is not part of the GIB alert list. All three
232 * fields are located in the first long word of the GISA.
233 *
234 * Returns: 0 on success
235 * -EBUSY in case the gisa is part of the alert list
236 */
237static inline int gisa_set_iam(struct kvm_s390_gisa *gisa, u8 iam)
238{
239 u64 word, _word;
240
241 word = READ_ONCE(gisa->u64.word[0]);
242 do {
243 if ((u64)gisa != word >> 32)
244 return -EBUSY;
245 _word = (word & ~0xffUL) | iam;
246 } while (!try_cmpxchg(&gisa->u64.word[0], &word, _word));
247
248 return 0;
249}
250
251/**
252 * gisa_clear_ipm - clear the GISA interruption pending mask
253 *
254 * @gisa: gisa to operate on
255 *
256 * Clear the IPM atomically with the next alert address and the IAM
257 * of the GISA unconditionally. All three fields are located in the
258 * first long word of the GISA.
259 */
260static inline void gisa_clear_ipm(struct kvm_s390_gisa *gisa)
261{
262 u64 word, _word;
263
264 word = READ_ONCE(gisa->u64.word[0]);
265 do {
266 _word = word & ~(0xffUL << 24);
267 } while (!try_cmpxchg(&gisa->u64.word[0], &word, _word));
268}
269
270/**
271 * gisa_get_ipm_or_restore_iam - return IPM or restore GISA IAM
272 *
273 * @gi: gisa interrupt struct to work on
274 *
275 * Atomically restores the interruption alert mask if none of the
276 * relevant ISCs are pending and return the IPM.
277 *
278 * Returns: the relevant pending ISCs
279 */
280static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
281{
282 u8 pending_mask, alert_mask;
283 u64 word, _word;
284
285 word = READ_ONCE(gi->origin->u64.word[0]);
286 do {
287 alert_mask = READ_ONCE(gi->alert.mask);
288 pending_mask = (u8)(word >> 24) & alert_mask;
289 if (pending_mask)
290 return pending_mask;
291 _word = (word & ~0xffUL) | alert_mask;
292 } while (!try_cmpxchg(&gi->origin->u64.word[0], &word, _word));
293
294 return 0;
295}
296
297static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
298{
299 set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
300}
301
302static inline u8 gisa_get_ipm(struct kvm_s390_gisa *gisa)
303{
304 return READ_ONCE(gisa->ipm);
305}
306
307static inline int gisa_tac_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
308{
309 return test_and_clear_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
310}
311
312static inline unsigned long pending_irqs_no_gisa(struct kvm_vcpu *vcpu)
313{
314 unsigned long pending = vcpu->kvm->arch.float_int.pending_irqs |
315 vcpu->arch.local_int.pending_irqs;
316
317 pending &= ~vcpu->kvm->arch.float_int.masked_irqs;
318 return pending;
319}
320
321static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
322{
323 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
324 unsigned long pending_mask;
325
326 pending_mask = pending_irqs_no_gisa(vcpu);
327 if (gi->origin)
328 pending_mask |= gisa_get_ipm(gi->origin) << IRQ_PEND_IO_ISC_7;
329 return pending_mask;
330}
331
332static inline int isc_to_irq_type(unsigned long isc)
333{
334 return IRQ_PEND_IO_ISC_0 - isc;
335}
336
337static inline int irq_type_to_isc(unsigned long irq_type)
338{
339 return IRQ_PEND_IO_ISC_0 - irq_type;
340}
341
342static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
343 unsigned long active_mask)
344{
345 int i;
346
347 for (i = 0; i <= MAX_ISC; i++)
348 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
349 active_mask &= ~(1UL << (isc_to_irq_type(i)));
350
351 return active_mask;
352}
353
354static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
355{
356 unsigned long active_mask;
357
358 active_mask = pending_irqs(vcpu);
359 if (!active_mask)
360 return 0;
361
362 if (psw_extint_disabled(vcpu))
363 active_mask &= ~IRQ_PEND_EXT_MASK;
364 if (psw_ioint_disabled(vcpu))
365 active_mask &= ~IRQ_PEND_IO_MASK;
366 else
367 active_mask = disable_iscs(vcpu, active_mask);
368 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
369 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
370 if (!(vcpu->arch.sie_block->gcr[0] & CR0_EMERGENCY_SIGNAL_SUBMASK))
371 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
372 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SUBMASK))
373 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
374 if (!(vcpu->arch.sie_block->gcr[0] & CR0_CPU_TIMER_SUBMASK))
375 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
376 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK)) {
377 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
378 __clear_bit(IRQ_PEND_EXT_SERVICE_EV, &active_mask);
379 }
380 if (psw_mchk_disabled(vcpu))
381 active_mask &= ~IRQ_PEND_MCHK_MASK;
382 /* PV guest cpus can have a single interruption injected at a time. */
383 if (kvm_s390_pv_cpu_get_handle(vcpu) &&
384 vcpu->arch.sie_block->iictl != IICTL_CODE_NONE)
385 active_mask &= ~(IRQ_PEND_EXT_II_MASK |
386 IRQ_PEND_IO_MASK |
387 IRQ_PEND_MCHK_MASK);
388 /*
389 * Check both floating and local interrupt's cr14 because
390 * bit IRQ_PEND_MCHK_REP could be set in both cases.
391 */
392 if (!(vcpu->arch.sie_block->gcr[14] &
393 (vcpu->kvm->arch.float_int.mchk.cr14 |
394 vcpu->arch.local_int.irq.mchk.cr14)))
395 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
396
397 /*
398 * STOP irqs will never be actively delivered. They are triggered via
399 * intercept requests and cleared when the stop intercept is performed.
400 */
401 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
402
403 return active_mask;
404}
405
406static void __set_cpu_idle(struct kvm_vcpu *vcpu)
407{
408 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
409 set_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
410}
411
412static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
413{
414 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
415 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
416}
417
418static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
419{
420 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IO_INT | CPUSTAT_EXT_INT |
421 CPUSTAT_STOP_INT);
422 vcpu->arch.sie_block->lctl = 0x0000;
423 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
424
425 if (guestdbg_enabled(vcpu)) {
426 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
427 LCTL_CR10 | LCTL_CR11);
428 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
429 }
430}
431
432static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
433{
434 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_IO_MASK))
435 return;
436 if (psw_ioint_disabled(vcpu))
437 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IO_INT);
438 else
439 vcpu->arch.sie_block->lctl |= LCTL_CR6;
440}
441
442static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
443{
444 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_EXT_MASK))
445 return;
446 if (psw_extint_disabled(vcpu))
447 kvm_s390_set_cpuflags(vcpu, CPUSTAT_EXT_INT);
448 else
449 vcpu->arch.sie_block->lctl |= LCTL_CR0;
450}
451
452static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
453{
454 if (!(pending_irqs_no_gisa(vcpu) & IRQ_PEND_MCHK_MASK))
455 return;
456 if (psw_mchk_disabled(vcpu))
457 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
458 else
459 vcpu->arch.sie_block->lctl |= LCTL_CR14;
460}
461
462static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
463{
464 if (kvm_s390_is_stop_irq_pending(vcpu))
465 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
466}
467
468/* Set interception request for non-deliverable interrupts */
469static void set_intercept_indicators(struct kvm_vcpu *vcpu)
470{
471 set_intercept_indicators_io(vcpu);
472 set_intercept_indicators_ext(vcpu);
473 set_intercept_indicators_mchk(vcpu);
474 set_intercept_indicators_stop(vcpu);
475}
476
477static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
478{
479 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
480 int rc = 0;
481
482 vcpu->stat.deliver_cputm++;
483 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
484 0, 0);
485 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
486 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
487 vcpu->arch.sie_block->eic = EXT_IRQ_CPU_TIMER;
488 } else {
489 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
490 (u16 *)__LC_EXT_INT_CODE);
491 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
492 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
493 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
494 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
495 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
496 }
497 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
498 return rc ? -EFAULT : 0;
499}
500
501static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
502{
503 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
504 int rc = 0;
505
506 vcpu->stat.deliver_ckc++;
507 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
508 0, 0);
509 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
510 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
511 vcpu->arch.sie_block->eic = EXT_IRQ_CLK_COMP;
512 } else {
513 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
514 (u16 __user *)__LC_EXT_INT_CODE);
515 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
516 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
517 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
518 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
519 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
520 }
521 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
522 return rc ? -EFAULT : 0;
523}
524
525static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
526{
527 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
528 struct kvm_s390_ext_info ext;
529 int rc;
530
531 spin_lock(&li->lock);
532 ext = li->irq.ext;
533 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
534 li->irq.ext.ext_params2 = 0;
535 spin_unlock(&li->lock);
536
537 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
538 ext.ext_params2);
539 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
540 KVM_S390_INT_PFAULT_INIT,
541 0, ext.ext_params2);
542
543 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
544 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
545 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
546 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
547 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
548 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
549 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
550 return rc ? -EFAULT : 0;
551}
552
553static int __write_machine_check(struct kvm_vcpu *vcpu,
554 struct kvm_s390_mchk_info *mchk)
555{
556 unsigned long ext_sa_addr;
557 unsigned long lc;
558 freg_t fprs[NUM_FPRS];
559 union mci mci;
560 int rc;
561
562 /*
563 * All other possible payload for a machine check (e.g. the register
564 * contents in the save area) will be handled by the ultravisor, as
565 * the hypervisor does not not have the needed information for
566 * protected guests.
567 */
568 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
569 vcpu->arch.sie_block->iictl = IICTL_CODE_MCHK;
570 vcpu->arch.sie_block->mcic = mchk->mcic;
571 vcpu->arch.sie_block->faddr = mchk->failing_storage_address;
572 vcpu->arch.sie_block->edc = mchk->ext_damage_code;
573 return 0;
574 }
575
576 mci.val = mchk->mcic;
577 /* take care of lazy register loading */
578 kvm_s390_fpu_store(vcpu->run);
579 save_access_regs(vcpu->run->s.regs.acrs);
580 if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
581 save_gs_cb(current->thread.gs_cb);
582
583 /* Extended save area */
584 rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
585 sizeof(unsigned long));
586 /* Only bits 0 through 63-LC are used for address formation */
587 lc = ext_sa_addr & MCESA_LC_MASK;
588 if (test_kvm_facility(vcpu->kvm, 133)) {
589 switch (lc) {
590 case 0:
591 case 10:
592 ext_sa_addr &= ~0x3ffUL;
593 break;
594 case 11:
595 ext_sa_addr &= ~0x7ffUL;
596 break;
597 case 12:
598 ext_sa_addr &= ~0xfffUL;
599 break;
600 default:
601 ext_sa_addr = 0;
602 break;
603 }
604 } else {
605 ext_sa_addr &= ~0x3ffUL;
606 }
607
608 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
609 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
610 512))
611 mci.vr = 0;
612 } else {
613 mci.vr = 0;
614 }
615 if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
616 && (lc == 11 || lc == 12)) {
617 if (write_guest_abs(vcpu, ext_sa_addr + 1024,
618 &vcpu->run->s.regs.gscb, 32))
619 mci.gs = 0;
620 } else {
621 mci.gs = 0;
622 }
623
624 /* General interruption information */
625 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
626 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
627 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
628 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
629 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
630 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
631
632 /* Register-save areas */
633 if (cpu_has_vx()) {
634 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
635 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
636 } else {
637 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
638 vcpu->run->s.regs.fprs, 128);
639 }
640 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
641 vcpu->run->s.regs.gprs, 128);
642 rc |= put_guest_lc(vcpu, vcpu->run->s.regs.fpc,
643 (u32 __user *) __LC_FP_CREG_SAVE_AREA);
644 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
645 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
646 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
647 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
648 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
649 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
650 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
651 &vcpu->run->s.regs.acrs, 64);
652 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
653 &vcpu->arch.sie_block->gcr, 128);
654
655 /* Extended interruption information */
656 rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
657 (u32 __user *) __LC_EXT_DAMAGE_CODE);
658 rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
659 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
660 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
661 sizeof(mchk->fixed_logout));
662 return rc ? -EFAULT : 0;
663}
664
665static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
666{
667 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
668 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
669 struct kvm_s390_mchk_info mchk = {};
670 int deliver = 0;
671 int rc = 0;
672
673 spin_lock(&fi->lock);
674 spin_lock(&li->lock);
675 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
676 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
677 /*
678 * If there was an exigent machine check pending, then any
679 * repressible machine checks that might have been pending
680 * are indicated along with it, so always clear bits for
681 * repressible and exigent interrupts
682 */
683 mchk = li->irq.mchk;
684 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
685 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
686 memset(&li->irq.mchk, 0, sizeof(mchk));
687 deliver = 1;
688 }
689 /*
690 * We indicate floating repressible conditions along with
691 * other pending conditions. Channel Report Pending and Channel
692 * Subsystem damage are the only two and are indicated by
693 * bits in mcic and masked in cr14.
694 */
695 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
696 mchk.mcic |= fi->mchk.mcic;
697 mchk.cr14 |= fi->mchk.cr14;
698 memset(&fi->mchk, 0, sizeof(mchk));
699 deliver = 1;
700 }
701 spin_unlock(&li->lock);
702 spin_unlock(&fi->lock);
703
704 if (deliver) {
705 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
706 mchk.mcic);
707 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
708 KVM_S390_MCHK,
709 mchk.cr14, mchk.mcic);
710 vcpu->stat.deliver_machine_check++;
711 rc = __write_machine_check(vcpu, &mchk);
712 }
713 return rc;
714}
715
716static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
717{
718 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
719 int rc = 0;
720
721 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
722 vcpu->stat.deliver_restart_signal++;
723 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
724
725 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
726 vcpu->arch.sie_block->iictl = IICTL_CODE_RESTART;
727 } else {
728 rc = write_guest_lc(vcpu,
729 offsetof(struct lowcore, restart_old_psw),
730 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
731 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
732 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
733 }
734 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
735 return rc ? -EFAULT : 0;
736}
737
738static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
739{
740 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
741 struct kvm_s390_prefix_info prefix;
742
743 spin_lock(&li->lock);
744 prefix = li->irq.prefix;
745 li->irq.prefix.address = 0;
746 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
747 spin_unlock(&li->lock);
748
749 vcpu->stat.deliver_prefix_signal++;
750 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
751 KVM_S390_SIGP_SET_PREFIX,
752 prefix.address, 0);
753
754 kvm_s390_set_prefix(vcpu, prefix.address);
755 return 0;
756}
757
758static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
759{
760 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
761 int rc;
762 int cpu_addr;
763
764 spin_lock(&li->lock);
765 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
766 clear_bit(cpu_addr, li->sigp_emerg_pending);
767 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
768 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
769 spin_unlock(&li->lock);
770
771 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
772 vcpu->stat.deliver_emergency_signal++;
773 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
774 cpu_addr, 0);
775 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
776 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
777 vcpu->arch.sie_block->eic = EXT_IRQ_EMERGENCY_SIG;
778 vcpu->arch.sie_block->extcpuaddr = cpu_addr;
779 return 0;
780 }
781
782 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
783 (u16 *)__LC_EXT_INT_CODE);
784 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
785 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
786 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
787 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
788 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
789 return rc ? -EFAULT : 0;
790}
791
792static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
793{
794 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
795 struct kvm_s390_extcall_info extcall;
796 int rc;
797
798 spin_lock(&li->lock);
799 extcall = li->irq.extcall;
800 li->irq.extcall.code = 0;
801 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
802 spin_unlock(&li->lock);
803
804 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
805 vcpu->stat.deliver_external_call++;
806 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
807 KVM_S390_INT_EXTERNAL_CALL,
808 extcall.code, 0);
809 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
810 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
811 vcpu->arch.sie_block->eic = EXT_IRQ_EXTERNAL_CALL;
812 vcpu->arch.sie_block->extcpuaddr = extcall.code;
813 return 0;
814 }
815
816 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
817 (u16 *)__LC_EXT_INT_CODE);
818 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
819 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
820 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
821 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
822 sizeof(psw_t));
823 return rc ? -EFAULT : 0;
824}
825
826static int __deliver_prog_pv(struct kvm_vcpu *vcpu, u16 code)
827{
828 switch (code) {
829 case PGM_SPECIFICATION:
830 vcpu->arch.sie_block->iictl = IICTL_CODE_SPECIFICATION;
831 break;
832 case PGM_OPERAND:
833 vcpu->arch.sie_block->iictl = IICTL_CODE_OPERAND;
834 break;
835 default:
836 return -EINVAL;
837 }
838 return 0;
839}
840
841static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
842{
843 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
844 struct kvm_s390_pgm_info pgm_info;
845 int rc = 0, nullifying = false;
846 u16 ilen;
847
848 spin_lock(&li->lock);
849 pgm_info = li->irq.pgm;
850 clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
851 memset(&li->irq.pgm, 0, sizeof(pgm_info));
852 spin_unlock(&li->lock);
853
854 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
855 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
856 pgm_info.code, ilen);
857 vcpu->stat.deliver_program++;
858 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
859 pgm_info.code, 0);
860
861 /* PER is handled by the ultravisor */
862 if (kvm_s390_pv_cpu_is_protected(vcpu))
863 return __deliver_prog_pv(vcpu, pgm_info.code & ~PGM_PER);
864
865 switch (pgm_info.code & ~PGM_PER) {
866 case PGM_AFX_TRANSLATION:
867 case PGM_ASX_TRANSLATION:
868 case PGM_EX_TRANSLATION:
869 case PGM_LFX_TRANSLATION:
870 case PGM_LSTE_SEQUENCE:
871 case PGM_LSX_TRANSLATION:
872 case PGM_LX_TRANSLATION:
873 case PGM_PRIMARY_AUTHORITY:
874 case PGM_SECONDARY_AUTHORITY:
875 nullifying = true;
876 fallthrough;
877 case PGM_SPACE_SWITCH:
878 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
879 (u64 *)__LC_TRANS_EXC_CODE);
880 break;
881 case PGM_ALEN_TRANSLATION:
882 case PGM_ALE_SEQUENCE:
883 case PGM_ASTE_INSTANCE:
884 case PGM_ASTE_SEQUENCE:
885 case PGM_ASTE_VALIDITY:
886 case PGM_EXTENDED_AUTHORITY:
887 rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
888 (u8 *)__LC_EXC_ACCESS_ID);
889 nullifying = true;
890 break;
891 case PGM_ASCE_TYPE:
892 case PGM_PAGE_TRANSLATION:
893 case PGM_REGION_FIRST_TRANS:
894 case PGM_REGION_SECOND_TRANS:
895 case PGM_REGION_THIRD_TRANS:
896 case PGM_SEGMENT_TRANSLATION:
897 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
898 (u64 *)__LC_TRANS_EXC_CODE);
899 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
900 (u8 *)__LC_EXC_ACCESS_ID);
901 rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
902 (u8 *)__LC_OP_ACCESS_ID);
903 nullifying = true;
904 break;
905 case PGM_MONITOR:
906 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
907 (u16 *)__LC_MON_CLASS_NR);
908 rc |= put_guest_lc(vcpu, pgm_info.mon_code,
909 (u64 *)__LC_MON_CODE);
910 break;
911 case PGM_VECTOR_PROCESSING:
912 case PGM_DATA:
913 rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
914 (u32 *)__LC_DATA_EXC_CODE);
915 break;
916 case PGM_PROTECTION:
917 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
918 (u64 *)__LC_TRANS_EXC_CODE);
919 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
920 (u8 *)__LC_EXC_ACCESS_ID);
921 break;
922 case PGM_STACK_FULL:
923 case PGM_STACK_EMPTY:
924 case PGM_STACK_SPECIFICATION:
925 case PGM_STACK_TYPE:
926 case PGM_STACK_OPERATION:
927 case PGM_TRACE_TABEL:
928 case PGM_CRYPTO_OPERATION:
929 nullifying = true;
930 break;
931 }
932
933 if (pgm_info.code & PGM_PER) {
934 rc |= put_guest_lc(vcpu, pgm_info.per_code,
935 (u8 *) __LC_PER_CODE);
936 rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
937 (u8 *)__LC_PER_ATMID);
938 rc |= put_guest_lc(vcpu, pgm_info.per_address,
939 (u64 *) __LC_PER_ADDRESS);
940 rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
941 (u8 *) __LC_PER_ACCESS_ID);
942 }
943
944 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
945 kvm_s390_rewind_psw(vcpu, ilen);
946
947 /* bit 1+2 of the target are the ilc, so we can directly use ilen */
948 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
949 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
950 (u64 *) __LC_PGM_LAST_BREAK);
951 rc |= put_guest_lc(vcpu, pgm_info.code,
952 (u16 *)__LC_PGM_INT_CODE);
953 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
954 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
955 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
956 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
957 return rc ? -EFAULT : 0;
958}
959
960#define SCCB_MASK 0xFFFFFFF8
961#define SCCB_EVENT_PENDING 0x3
962
963static int write_sclp(struct kvm_vcpu *vcpu, u32 parm)
964{
965 int rc;
966
967 if (kvm_s390_pv_cpu_get_handle(vcpu)) {
968 vcpu->arch.sie_block->iictl = IICTL_CODE_EXT;
969 vcpu->arch.sie_block->eic = EXT_IRQ_SERVICE_SIG;
970 vcpu->arch.sie_block->eiparams = parm;
971 return 0;
972 }
973
974 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
975 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
976 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
977 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
978 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
979 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
980 rc |= put_guest_lc(vcpu, parm,
981 (u32 *)__LC_EXT_PARAMS);
982
983 return rc ? -EFAULT : 0;
984}
985
986static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
987{
988 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
989 struct kvm_s390_ext_info ext;
990
991 spin_lock(&fi->lock);
992 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs) ||
993 !(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
994 spin_unlock(&fi->lock);
995 return 0;
996 }
997 ext = fi->srv_signal;
998 memset(&fi->srv_signal, 0, sizeof(ext));
999 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1000 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1001 if (kvm_s390_pv_cpu_is_protected(vcpu))
1002 set_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs);
1003 spin_unlock(&fi->lock);
1004
1005 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
1006 ext.ext_params);
1007 vcpu->stat.deliver_service_signal++;
1008 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1009 ext.ext_params, 0);
1010
1011 return write_sclp(vcpu, ext.ext_params);
1012}
1013
1014static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)
1015{
1016 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1017 struct kvm_s390_ext_info ext;
1018
1019 spin_lock(&fi->lock);
1020 if (!(test_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs))) {
1021 spin_unlock(&fi->lock);
1022 return 0;
1023 }
1024 ext = fi->srv_signal;
1025 /* only clear the event bits */
1026 fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING;
1027 clear_bit(IRQ_PEND_EXT_SERVICE_EV, &fi->pending_irqs);
1028 spin_unlock(&fi->lock);
1029
1030 VCPU_EVENT(vcpu, 4, "%s", "deliver: sclp parameter event");
1031 vcpu->stat.deliver_service_signal++;
1032 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
1033 ext.ext_params, 0);
1034
1035 return write_sclp(vcpu, ext.ext_params & SCCB_EVENT_PENDING);
1036}
1037
1038static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
1039{
1040 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1041 struct kvm_s390_interrupt_info *inti;
1042 int rc = 0;
1043
1044 spin_lock(&fi->lock);
1045 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
1046 struct kvm_s390_interrupt_info,
1047 list);
1048 if (inti) {
1049 list_del(&inti->list);
1050 fi->counters[FIRQ_CNTR_PFAULT] -= 1;
1051 }
1052 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
1053 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1054 spin_unlock(&fi->lock);
1055
1056 if (inti) {
1057 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1058 KVM_S390_INT_PFAULT_DONE, 0,
1059 inti->ext.ext_params2);
1060 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
1061 inti->ext.ext_params2);
1062
1063 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1064 (u16 *)__LC_EXT_INT_CODE);
1065 rc |= put_guest_lc(vcpu, PFAULT_DONE,
1066 (u16 *)__LC_EXT_CPU_ADDR);
1067 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1068 &vcpu->arch.sie_block->gpsw,
1069 sizeof(psw_t));
1070 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1071 &vcpu->arch.sie_block->gpsw,
1072 sizeof(psw_t));
1073 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1074 (u64 *)__LC_EXT_PARAMS2);
1075 kfree(inti);
1076 }
1077 return rc ? -EFAULT : 0;
1078}
1079
1080static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
1081{
1082 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
1083 struct kvm_s390_interrupt_info *inti;
1084 int rc = 0;
1085
1086 spin_lock(&fi->lock);
1087 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
1088 struct kvm_s390_interrupt_info,
1089 list);
1090 if (inti) {
1091 VCPU_EVENT(vcpu, 4,
1092 "deliver: virtio parm: 0x%x,parm64: 0x%llx",
1093 inti->ext.ext_params, inti->ext.ext_params2);
1094 vcpu->stat.deliver_virtio++;
1095 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1096 inti->type,
1097 inti->ext.ext_params,
1098 inti->ext.ext_params2);
1099 list_del(&inti->list);
1100 fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
1101 }
1102 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
1103 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1104 spin_unlock(&fi->lock);
1105
1106 if (inti) {
1107 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
1108 (u16 *)__LC_EXT_INT_CODE);
1109 rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
1110 (u16 *)__LC_EXT_CPU_ADDR);
1111 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
1112 &vcpu->arch.sie_block->gpsw,
1113 sizeof(psw_t));
1114 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
1115 &vcpu->arch.sie_block->gpsw,
1116 sizeof(psw_t));
1117 rc |= put_guest_lc(vcpu, inti->ext.ext_params,
1118 (u32 *)__LC_EXT_PARAMS);
1119 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
1120 (u64 *)__LC_EXT_PARAMS2);
1121 kfree(inti);
1122 }
1123 return rc ? -EFAULT : 0;
1124}
1125
1126static int __do_deliver_io(struct kvm_vcpu *vcpu, struct kvm_s390_io_info *io)
1127{
1128 int rc;
1129
1130 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
1131 vcpu->arch.sie_block->iictl = IICTL_CODE_IO;
1132 vcpu->arch.sie_block->subchannel_id = io->subchannel_id;
1133 vcpu->arch.sie_block->subchannel_nr = io->subchannel_nr;
1134 vcpu->arch.sie_block->io_int_parm = io->io_int_parm;
1135 vcpu->arch.sie_block->io_int_word = io->io_int_word;
1136 return 0;
1137 }
1138
1139 rc = put_guest_lc(vcpu, io->subchannel_id, (u16 *)__LC_SUBCHANNEL_ID);
1140 rc |= put_guest_lc(vcpu, io->subchannel_nr, (u16 *)__LC_SUBCHANNEL_NR);
1141 rc |= put_guest_lc(vcpu, io->io_int_parm, (u32 *)__LC_IO_INT_PARM);
1142 rc |= put_guest_lc(vcpu, io->io_int_word, (u32 *)__LC_IO_INT_WORD);
1143 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
1144 &vcpu->arch.sie_block->gpsw,
1145 sizeof(psw_t));
1146 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
1147 &vcpu->arch.sie_block->gpsw,
1148 sizeof(psw_t));
1149 return rc ? -EFAULT : 0;
1150}
1151
1152static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
1153 unsigned long irq_type)
1154{
1155 struct list_head *isc_list;
1156 struct kvm_s390_float_interrupt *fi;
1157 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1158 struct kvm_s390_interrupt_info *inti = NULL;
1159 struct kvm_s390_io_info io;
1160 u32 isc;
1161 int rc = 0;
1162
1163 fi = &vcpu->kvm->arch.float_int;
1164
1165 spin_lock(&fi->lock);
1166 isc = irq_type_to_isc(irq_type);
1167 isc_list = &fi->lists[isc];
1168 inti = list_first_entry_or_null(isc_list,
1169 struct kvm_s390_interrupt_info,
1170 list);
1171 if (inti) {
1172 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1173 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
1174 else
1175 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
1176 inti->io.subchannel_id >> 8,
1177 inti->io.subchannel_id >> 1 & 0x3,
1178 inti->io.subchannel_nr);
1179
1180 vcpu->stat.deliver_io++;
1181 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1182 inti->type,
1183 ((__u32)inti->io.subchannel_id << 16) |
1184 inti->io.subchannel_nr,
1185 ((__u64)inti->io.io_int_parm << 32) |
1186 inti->io.io_int_word);
1187 list_del(&inti->list);
1188 fi->counters[FIRQ_CNTR_IO] -= 1;
1189 }
1190 if (list_empty(isc_list))
1191 clear_bit(irq_type, &fi->pending_irqs);
1192 spin_unlock(&fi->lock);
1193
1194 if (inti) {
1195 rc = __do_deliver_io(vcpu, &(inti->io));
1196 kfree(inti);
1197 goto out;
1198 }
1199
1200 if (gi->origin && gisa_tac_ipm_gisc(gi->origin, isc)) {
1201 /*
1202 * in case an adapter interrupt was not delivered
1203 * in SIE context KVM will handle the delivery
1204 */
1205 VCPU_EVENT(vcpu, 4, "%s isc %u", "deliver: I/O (AI/gisa)", isc);
1206 memset(&io, 0, sizeof(io));
1207 io.io_int_word = isc_to_int_word(isc);
1208 vcpu->stat.deliver_io++;
1209 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
1210 KVM_S390_INT_IO(1, 0, 0, 0),
1211 ((__u32)io.subchannel_id << 16) |
1212 io.subchannel_nr,
1213 ((__u64)io.io_int_parm << 32) |
1214 io.io_int_word);
1215 rc = __do_deliver_io(vcpu, &io);
1216 }
1217out:
1218 return rc;
1219}
1220
1221/* Check whether an external call is pending (deliverable or not) */
1222int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
1223{
1224 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1225
1226 if (!sclp.has_sigpif)
1227 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
1228
1229 return sca_ext_call_pending(vcpu, NULL);
1230}
1231
1232int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
1233{
1234 if (deliverable_irqs(vcpu))
1235 return 1;
1236
1237 if (kvm_cpu_has_pending_timer(vcpu))
1238 return 1;
1239
1240 /* external call pending and deliverable */
1241 if (kvm_s390_ext_call_pending(vcpu) &&
1242 !psw_extint_disabled(vcpu) &&
1243 (vcpu->arch.sie_block->gcr[0] & CR0_EXTERNAL_CALL_SUBMASK))
1244 return 1;
1245
1246 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1247 return 1;
1248 return 0;
1249}
1250
1251int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1252{
1253 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1254}
1255
1256static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1257{
1258 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1259 const u64 ckc = vcpu->arch.sie_block->ckc;
1260 u64 cputm, sltime = 0;
1261
1262 if (ckc_interrupts_enabled(vcpu)) {
1263 if (vcpu->arch.sie_block->gcr[0] & CR0_CLOCK_COMPARATOR_SIGN) {
1264 if ((s64)now < (s64)ckc)
1265 sltime = tod_to_ns((s64)ckc - (s64)now);
1266 } else if (now < ckc) {
1267 sltime = tod_to_ns(ckc - now);
1268 }
1269 /* already expired */
1270 if (!sltime)
1271 return 0;
1272 if (cpu_timer_interrupts_enabled(vcpu)) {
1273 cputm = kvm_s390_get_cpu_timer(vcpu);
1274 /* already expired? */
1275 if (cputm >> 63)
1276 return 0;
1277 return min_t(u64, sltime, tod_to_ns(cputm));
1278 }
1279 } else if (cpu_timer_interrupts_enabled(vcpu)) {
1280 sltime = kvm_s390_get_cpu_timer(vcpu);
1281 /* already expired? */
1282 if (sltime >> 63)
1283 return 0;
1284 }
1285 return sltime;
1286}
1287
1288int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1289{
1290 struct kvm_s390_gisa_interrupt *gi = &vcpu->kvm->arch.gisa_int;
1291 u64 sltime;
1292
1293 vcpu->stat.exit_wait_state++;
1294
1295 /* fast path */
1296 if (kvm_arch_vcpu_runnable(vcpu))
1297 return 0;
1298
1299 if (psw_interrupts_disabled(vcpu)) {
1300 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1301 return -EOPNOTSUPP; /* disabled wait */
1302 }
1303
1304 if (gi->origin &&
1305 (gisa_get_ipm_or_restore_iam(gi) &
1306 vcpu->arch.sie_block->gcr[6] >> 24))
1307 return 0;
1308
1309 if (!ckc_interrupts_enabled(vcpu) &&
1310 !cpu_timer_interrupts_enabled(vcpu)) {
1311 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1312 __set_cpu_idle(vcpu);
1313 goto no_timer;
1314 }
1315
1316 sltime = __calculate_sltime(vcpu);
1317 if (!sltime)
1318 return 0;
1319
1320 __set_cpu_idle(vcpu);
1321 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1322 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1323no_timer:
1324 kvm_vcpu_srcu_read_unlock(vcpu);
1325 kvm_vcpu_halt(vcpu);
1326 vcpu->valid_wakeup = false;
1327 __unset_cpu_idle(vcpu);
1328 kvm_vcpu_srcu_read_lock(vcpu);
1329
1330 hrtimer_cancel(&vcpu->arch.ckc_timer);
1331 return 0;
1332}
1333
1334void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1335{
1336 vcpu->valid_wakeup = true;
1337 kvm_vcpu_wake_up(vcpu);
1338
1339 /*
1340 * The VCPU might not be sleeping but rather executing VSIE. Let's
1341 * kick it, so it leaves the SIE to process the request.
1342 */
1343 kvm_s390_vsie_kick(vcpu);
1344}
1345
1346enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1347{
1348 struct kvm_vcpu *vcpu;
1349 u64 sltime;
1350
1351 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1352 sltime = __calculate_sltime(vcpu);
1353
1354 /*
1355 * If the monotonic clock runs faster than the tod clock we might be
1356 * woken up too early and have to go back to sleep to avoid deadlocks.
1357 */
1358 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1359 return HRTIMER_RESTART;
1360 kvm_s390_vcpu_wakeup(vcpu);
1361 return HRTIMER_NORESTART;
1362}
1363
1364void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1365{
1366 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1367
1368 spin_lock(&li->lock);
1369 li->pending_irqs = 0;
1370 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1371 memset(&li->irq, 0, sizeof(li->irq));
1372 spin_unlock(&li->lock);
1373
1374 sca_clear_ext_call(vcpu);
1375}
1376
1377int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1378{
1379 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1380 int rc = 0;
1381 bool delivered = false;
1382 unsigned long irq_type;
1383 unsigned long irqs;
1384
1385 __reset_intercept_indicators(vcpu);
1386
1387 /* pending ckc conditions might have been invalidated */
1388 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1389 if (ckc_irq_pending(vcpu))
1390 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1391
1392 /* pending cpu timer conditions might have been invalidated */
1393 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1394 if (cpu_timer_irq_pending(vcpu))
1395 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1396
1397 while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1398 /* bits are in the reverse order of interrupt priority */
1399 irq_type = find_last_bit(&irqs, IRQ_PEND_COUNT);
1400 switch (irq_type) {
1401 case IRQ_PEND_IO_ISC_0:
1402 case IRQ_PEND_IO_ISC_1:
1403 case IRQ_PEND_IO_ISC_2:
1404 case IRQ_PEND_IO_ISC_3:
1405 case IRQ_PEND_IO_ISC_4:
1406 case IRQ_PEND_IO_ISC_5:
1407 case IRQ_PEND_IO_ISC_6:
1408 case IRQ_PEND_IO_ISC_7:
1409 rc = __deliver_io(vcpu, irq_type);
1410 break;
1411 case IRQ_PEND_MCHK_EX:
1412 case IRQ_PEND_MCHK_REP:
1413 rc = __deliver_machine_check(vcpu);
1414 break;
1415 case IRQ_PEND_PROG:
1416 rc = __deliver_prog(vcpu);
1417 break;
1418 case IRQ_PEND_EXT_EMERGENCY:
1419 rc = __deliver_emergency_signal(vcpu);
1420 break;
1421 case IRQ_PEND_EXT_EXTERNAL:
1422 rc = __deliver_external_call(vcpu);
1423 break;
1424 case IRQ_PEND_EXT_CLOCK_COMP:
1425 rc = __deliver_ckc(vcpu);
1426 break;
1427 case IRQ_PEND_EXT_CPU_TIMER:
1428 rc = __deliver_cpu_timer(vcpu);
1429 break;
1430 case IRQ_PEND_RESTART:
1431 rc = __deliver_restart(vcpu);
1432 break;
1433 case IRQ_PEND_SET_PREFIX:
1434 rc = __deliver_set_prefix(vcpu);
1435 break;
1436 case IRQ_PEND_PFAULT_INIT:
1437 rc = __deliver_pfault_init(vcpu);
1438 break;
1439 case IRQ_PEND_EXT_SERVICE:
1440 rc = __deliver_service(vcpu);
1441 break;
1442 case IRQ_PEND_EXT_SERVICE_EV:
1443 rc = __deliver_service_ev(vcpu);
1444 break;
1445 case IRQ_PEND_PFAULT_DONE:
1446 rc = __deliver_pfault_done(vcpu);
1447 break;
1448 case IRQ_PEND_VIRTIO:
1449 rc = __deliver_virtio(vcpu);
1450 break;
1451 default:
1452 WARN_ONCE(1, "Unknown pending irq type %ld", irq_type);
1453 clear_bit(irq_type, &li->pending_irqs);
1454 }
1455 delivered |= !rc;
1456 }
1457
1458 /*
1459 * We delivered at least one interrupt and modified the PC. Force a
1460 * singlestep event now.
1461 */
1462 if (delivered && guestdbg_sstep_enabled(vcpu)) {
1463 struct kvm_debug_exit_arch *debug_exit = &vcpu->run->debug.arch;
1464
1465 debug_exit->addr = vcpu->arch.sie_block->gpsw.addr;
1466 debug_exit->type = KVM_SINGLESTEP;
1467 vcpu->guest_debug |= KVM_GUESTDBG_EXIT_PENDING;
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 if (kvm_is_ucontrol(dev->kvm))
2682 return -EINVAL;
2683 dev->kvm->arch.gmap->pfault_enabled = 1;
2684 break;
2685 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2686 if (kvm_is_ucontrol(dev->kvm))
2687 return -EINVAL;
2688 dev->kvm->arch.gmap->pfault_enabled = 0;
2689 /*
2690 * Make sure no async faults are in transition when
2691 * clearing the queues. So we don't need to worry
2692 * about late coming workers.
2693 */
2694 synchronize_srcu(&dev->kvm->srcu);
2695 kvm_for_each_vcpu(i, vcpu, dev->kvm)
2696 kvm_clear_async_pf_completion_queue(vcpu);
2697 break;
2698 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2699 r = register_io_adapter(dev, attr);
2700 break;
2701 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2702 r = modify_io_adapter(dev, attr);
2703 break;
2704 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2705 r = clear_io_irq(dev->kvm, attr);
2706 break;
2707 case KVM_DEV_FLIC_AISM:
2708 r = modify_ais_mode(dev->kvm, attr);
2709 break;
2710 case KVM_DEV_FLIC_AIRQ_INJECT:
2711 r = flic_inject_airq(dev->kvm, attr);
2712 break;
2713 case KVM_DEV_FLIC_AISM_ALL:
2714 r = flic_ais_mode_set_all(dev->kvm, attr);
2715 break;
2716 default:
2717 r = -EINVAL;
2718 }
2719
2720 return r;
2721}
2722
2723static int flic_has_attr(struct kvm_device *dev,
2724 struct kvm_device_attr *attr)
2725{
2726 switch (attr->group) {
2727 case KVM_DEV_FLIC_GET_ALL_IRQS:
2728 case KVM_DEV_FLIC_ENQUEUE:
2729 case KVM_DEV_FLIC_CLEAR_IRQS:
2730 case KVM_DEV_FLIC_APF_ENABLE:
2731 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2732 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2733 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2734 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2735 case KVM_DEV_FLIC_AISM:
2736 case KVM_DEV_FLIC_AIRQ_INJECT:
2737 case KVM_DEV_FLIC_AISM_ALL:
2738 return 0;
2739 }
2740 return -ENXIO;
2741}
2742
2743static int flic_create(struct kvm_device *dev, u32 type)
2744{
2745 if (!dev)
2746 return -EINVAL;
2747 if (dev->kvm->arch.flic)
2748 return -EINVAL;
2749 dev->kvm->arch.flic = dev;
2750 return 0;
2751}
2752
2753static void flic_destroy(struct kvm_device *dev)
2754{
2755 dev->kvm->arch.flic = NULL;
2756 kfree(dev);
2757}
2758
2759/* s390 floating irq controller (flic) */
2760struct kvm_device_ops kvm_flic_ops = {
2761 .name = "kvm-flic",
2762 .get_attr = flic_get_attr,
2763 .set_attr = flic_set_attr,
2764 .has_attr = flic_has_attr,
2765 .create = flic_create,
2766 .destroy = flic_destroy,
2767};
2768
2769static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2770{
2771 unsigned long bit;
2772
2773 bit = bit_nr + (addr % PAGE_SIZE) * 8;
2774
2775 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2776}
2777
2778static struct page *get_map_page(struct kvm *kvm, u64 uaddr)
2779{
2780 struct page *page = NULL;
2781
2782 mmap_read_lock(kvm->mm);
2783 get_user_pages_remote(kvm->mm, uaddr, 1, FOLL_WRITE,
2784 &page, NULL);
2785 mmap_read_unlock(kvm->mm);
2786 return page;
2787}
2788
2789static int adapter_indicators_set(struct kvm *kvm,
2790 struct s390_io_adapter *adapter,
2791 struct kvm_s390_adapter_int *adapter_int)
2792{
2793 unsigned long bit;
2794 int summary_set, idx;
2795 struct page *ind_page, *summary_page;
2796 void *map;
2797
2798 ind_page = get_map_page(kvm, adapter_int->ind_addr);
2799 if (!ind_page)
2800 return -1;
2801 summary_page = get_map_page(kvm, adapter_int->summary_addr);
2802 if (!summary_page) {
2803 put_page(ind_page);
2804 return -1;
2805 }
2806
2807 idx = srcu_read_lock(&kvm->srcu);
2808 map = page_address(ind_page);
2809 bit = get_ind_bit(adapter_int->ind_addr,
2810 adapter_int->ind_offset, adapter->swap);
2811 set_bit(bit, map);
2812 mark_page_dirty(kvm, adapter_int->ind_addr >> PAGE_SHIFT);
2813 set_page_dirty_lock(ind_page);
2814 map = page_address(summary_page);
2815 bit = get_ind_bit(adapter_int->summary_addr,
2816 adapter_int->summary_offset, adapter->swap);
2817 summary_set = test_and_set_bit(bit, map);
2818 mark_page_dirty(kvm, adapter_int->summary_addr >> PAGE_SHIFT);
2819 set_page_dirty_lock(summary_page);
2820 srcu_read_unlock(&kvm->srcu, idx);
2821
2822 put_page(ind_page);
2823 put_page(summary_page);
2824 return summary_set ? 0 : 1;
2825}
2826
2827/*
2828 * < 0 - not injected due to error
2829 * = 0 - coalesced, summary indicator already active
2830 * > 0 - injected interrupt
2831 */
2832static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2833 struct kvm *kvm, int irq_source_id, int level,
2834 bool line_status)
2835{
2836 int ret;
2837 struct s390_io_adapter *adapter;
2838
2839 /* We're only interested in the 0->1 transition. */
2840 if (!level)
2841 return 0;
2842 adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2843 if (!adapter)
2844 return -1;
2845 ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2846 if ((ret > 0) && !adapter->masked) {
2847 ret = kvm_s390_inject_airq(kvm, adapter);
2848 if (ret == 0)
2849 ret = 1;
2850 }
2851 return ret;
2852}
2853
2854/*
2855 * Inject the machine check to the guest.
2856 */
2857void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2858 struct mcck_volatile_info *mcck_info)
2859{
2860 struct kvm_s390_interrupt_info inti;
2861 struct kvm_s390_irq irq;
2862 struct kvm_s390_mchk_info *mchk;
2863 union mci mci;
2864 __u64 cr14 = 0; /* upper bits are not used */
2865 int rc;
2866
2867 mci.val = mcck_info->mcic;
2868 if (mci.sr)
2869 cr14 |= CR14_RECOVERY_SUBMASK;
2870 if (mci.dg)
2871 cr14 |= CR14_DEGRADATION_SUBMASK;
2872 if (mci.w)
2873 cr14 |= CR14_WARNING_SUBMASK;
2874
2875 mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2876 mchk->cr14 = cr14;
2877 mchk->mcic = mcck_info->mcic;
2878 mchk->ext_damage_code = mcck_info->ext_damage_code;
2879 mchk->failing_storage_address = mcck_info->failing_storage_address;
2880 if (mci.ck) {
2881 /* Inject the floating machine check */
2882 inti.type = KVM_S390_MCHK;
2883 rc = __inject_vm(vcpu->kvm, &inti);
2884 } else {
2885 /* Inject the machine check to specified vcpu */
2886 irq.type = KVM_S390_MCHK;
2887 rc = kvm_s390_inject_vcpu(vcpu, &irq);
2888 }
2889 WARN_ON_ONCE(rc);
2890}
2891
2892int kvm_set_routing_entry(struct kvm *kvm,
2893 struct kvm_kernel_irq_routing_entry *e,
2894 const struct kvm_irq_routing_entry *ue)
2895{
2896 u64 uaddr;
2897
2898 switch (ue->type) {
2899 /* we store the userspace addresses instead of the guest addresses */
2900 case KVM_IRQ_ROUTING_S390_ADAPTER:
2901 if (kvm_is_ucontrol(kvm))
2902 return -EINVAL;
2903 e->set = set_adapter_int;
2904 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.summary_addr);
2905 if (uaddr == -EFAULT)
2906 return -EFAULT;
2907 e->adapter.summary_addr = uaddr;
2908 uaddr = gmap_translate(kvm->arch.gmap, ue->u.adapter.ind_addr);
2909 if (uaddr == -EFAULT)
2910 return -EFAULT;
2911 e->adapter.ind_addr = uaddr;
2912 e->adapter.summary_offset = ue->u.adapter.summary_offset;
2913 e->adapter.ind_offset = ue->u.adapter.ind_offset;
2914 e->adapter.adapter_id = ue->u.adapter.adapter_id;
2915 return 0;
2916 default:
2917 return -EINVAL;
2918 }
2919}
2920
2921int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2922 int irq_source_id, int level, bool line_status)
2923{
2924 return -EINVAL;
2925}
2926
2927int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2928{
2929 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2930 struct kvm_s390_irq *buf;
2931 int r = 0;
2932 int n;
2933
2934 buf = vmalloc(len);
2935 if (!buf)
2936 return -ENOMEM;
2937
2938 if (copy_from_user((void *) buf, irqstate, len)) {
2939 r = -EFAULT;
2940 goto out_free;
2941 }
2942
2943 /*
2944 * Don't allow setting the interrupt state
2945 * when there are already interrupts pending
2946 */
2947 spin_lock(&li->lock);
2948 if (li->pending_irqs) {
2949 r = -EBUSY;
2950 goto out_unlock;
2951 }
2952
2953 for (n = 0; n < len / sizeof(*buf); n++) {
2954 r = do_inject_vcpu(vcpu, &buf[n]);
2955 if (r)
2956 break;
2957 }
2958
2959out_unlock:
2960 spin_unlock(&li->lock);
2961out_free:
2962 vfree(buf);
2963
2964 return r;
2965}
2966
2967static void store_local_irq(struct kvm_s390_local_interrupt *li,
2968 struct kvm_s390_irq *irq,
2969 unsigned long irq_type)
2970{
2971 switch (irq_type) {
2972 case IRQ_PEND_MCHK_EX:
2973 case IRQ_PEND_MCHK_REP:
2974 irq->type = KVM_S390_MCHK;
2975 irq->u.mchk = li->irq.mchk;
2976 break;
2977 case IRQ_PEND_PROG:
2978 irq->type = KVM_S390_PROGRAM_INT;
2979 irq->u.pgm = li->irq.pgm;
2980 break;
2981 case IRQ_PEND_PFAULT_INIT:
2982 irq->type = KVM_S390_INT_PFAULT_INIT;
2983 irq->u.ext = li->irq.ext;
2984 break;
2985 case IRQ_PEND_EXT_EXTERNAL:
2986 irq->type = KVM_S390_INT_EXTERNAL_CALL;
2987 irq->u.extcall = li->irq.extcall;
2988 break;
2989 case IRQ_PEND_EXT_CLOCK_COMP:
2990 irq->type = KVM_S390_INT_CLOCK_COMP;
2991 break;
2992 case IRQ_PEND_EXT_CPU_TIMER:
2993 irq->type = KVM_S390_INT_CPU_TIMER;
2994 break;
2995 case IRQ_PEND_SIGP_STOP:
2996 irq->type = KVM_S390_SIGP_STOP;
2997 irq->u.stop = li->irq.stop;
2998 break;
2999 case IRQ_PEND_RESTART:
3000 irq->type = KVM_S390_RESTART;
3001 break;
3002 case IRQ_PEND_SET_PREFIX:
3003 irq->type = KVM_S390_SIGP_SET_PREFIX;
3004 irq->u.prefix = li->irq.prefix;
3005 break;
3006 }
3007}
3008
3009int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
3010{
3011 int scn;
3012 DECLARE_BITMAP(sigp_emerg_pending, KVM_MAX_VCPUS);
3013 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
3014 unsigned long pending_irqs;
3015 struct kvm_s390_irq irq;
3016 unsigned long irq_type;
3017 int cpuaddr;
3018 int n = 0;
3019
3020 spin_lock(&li->lock);
3021 pending_irqs = li->pending_irqs;
3022 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
3023 sizeof(sigp_emerg_pending));
3024 spin_unlock(&li->lock);
3025
3026 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
3027 memset(&irq, 0, sizeof(irq));
3028 if (irq_type == IRQ_PEND_EXT_EMERGENCY)
3029 continue;
3030 if (n + sizeof(irq) > len)
3031 return -ENOBUFS;
3032 store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
3033 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3034 return -EFAULT;
3035 n += sizeof(irq);
3036 }
3037
3038 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
3039 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
3040 memset(&irq, 0, sizeof(irq));
3041 if (n + sizeof(irq) > len)
3042 return -ENOBUFS;
3043 irq.type = KVM_S390_INT_EMERGENCY;
3044 irq.u.emerg.code = cpuaddr;
3045 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3046 return -EFAULT;
3047 n += sizeof(irq);
3048 }
3049 }
3050
3051 if (sca_ext_call_pending(vcpu, &scn)) {
3052 if (n + sizeof(irq) > len)
3053 return -ENOBUFS;
3054 memset(&irq, 0, sizeof(irq));
3055 irq.type = KVM_S390_INT_EXTERNAL_CALL;
3056 irq.u.extcall.code = scn;
3057 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
3058 return -EFAULT;
3059 n += sizeof(irq);
3060 }
3061
3062 return n;
3063}
3064
3065static void __airqs_kick_single_vcpu(struct kvm *kvm, u8 deliverable_mask)
3066{
3067 int vcpu_idx, online_vcpus = atomic_read(&kvm->online_vcpus);
3068 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3069 struct kvm_vcpu *vcpu;
3070 u8 vcpu_isc_mask;
3071
3072 for_each_set_bit(vcpu_idx, kvm->arch.idle_mask, online_vcpus) {
3073 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
3074 if (psw_ioint_disabled(vcpu))
3075 continue;
3076 vcpu_isc_mask = (u8)(vcpu->arch.sie_block->gcr[6] >> 24);
3077 if (deliverable_mask & vcpu_isc_mask) {
3078 /* lately kicked but not yet running */
3079 if (test_and_set_bit(vcpu_idx, gi->kicked_mask))
3080 return;
3081 kvm_s390_vcpu_wakeup(vcpu);
3082 return;
3083 }
3084 }
3085}
3086
3087static enum hrtimer_restart gisa_vcpu_kicker(struct hrtimer *timer)
3088{
3089 struct kvm_s390_gisa_interrupt *gi =
3090 container_of(timer, struct kvm_s390_gisa_interrupt, timer);
3091 struct kvm *kvm =
3092 container_of(gi->origin, struct sie_page2, gisa)->kvm;
3093 u8 pending_mask;
3094
3095 pending_mask = gisa_get_ipm_or_restore_iam(gi);
3096 if (pending_mask) {
3097 __airqs_kick_single_vcpu(kvm, pending_mask);
3098 hrtimer_forward_now(timer, ns_to_ktime(gi->expires));
3099 return HRTIMER_RESTART;
3100 }
3101
3102 return HRTIMER_NORESTART;
3103}
3104
3105#define NULL_GISA_ADDR 0x00000000UL
3106#define NONE_GISA_ADDR 0x00000001UL
3107#define GISA_ADDR_MASK 0xfffff000UL
3108
3109static void process_gib_alert_list(void)
3110{
3111 struct kvm_s390_gisa_interrupt *gi;
3112 u32 final, gisa_phys, origin = 0UL;
3113 struct kvm_s390_gisa *gisa;
3114 struct kvm *kvm;
3115
3116 do {
3117 /*
3118 * If the NONE_GISA_ADDR is still stored in the alert list
3119 * origin, we will leave the outer loop. No further GISA has
3120 * been added to the alert list by millicode while processing
3121 * the current alert list.
3122 */
3123 final = (origin & NONE_GISA_ADDR);
3124 /*
3125 * Cut off the alert list and store the NONE_GISA_ADDR in the
3126 * alert list origin to avoid further GAL interruptions.
3127 * A new alert list can be build up by millicode in parallel
3128 * for guests not in the yet cut-off alert list. When in the
3129 * final loop, store the NULL_GISA_ADDR instead. This will re-
3130 * enable GAL interruptions on the host again.
3131 */
3132 origin = xchg(&gib->alert_list_origin,
3133 (!final) ? NONE_GISA_ADDR : NULL_GISA_ADDR);
3134 /*
3135 * Loop through the just cut-off alert list and start the
3136 * gisa timers to kick idle vcpus to consume the pending
3137 * interruptions asap.
3138 */
3139 while (origin & GISA_ADDR_MASK) {
3140 gisa_phys = origin;
3141 gisa = phys_to_virt(gisa_phys);
3142 origin = gisa->next_alert;
3143 gisa->next_alert = gisa_phys;
3144 kvm = container_of(gisa, struct sie_page2, gisa)->kvm;
3145 gi = &kvm->arch.gisa_int;
3146 if (hrtimer_active(&gi->timer))
3147 hrtimer_cancel(&gi->timer);
3148 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3149 }
3150 } while (!final);
3151
3152}
3153
3154void kvm_s390_gisa_clear(struct kvm *kvm)
3155{
3156 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3157
3158 if (!gi->origin)
3159 return;
3160 gisa_clear_ipm(gi->origin);
3161 VM_EVENT(kvm, 3, "gisa 0x%pK cleared", gi->origin);
3162}
3163
3164void kvm_s390_gisa_init(struct kvm *kvm)
3165{
3166 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3167
3168 if (!css_general_characteristics.aiv)
3169 return;
3170 gi->origin = &kvm->arch.sie_page2->gisa;
3171 gi->alert.mask = 0;
3172 spin_lock_init(&gi->alert.ref_lock);
3173 gi->expires = 50 * 1000; /* 50 usec */
3174 hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3175 gi->timer.function = gisa_vcpu_kicker;
3176 memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
3177 gi->origin->next_alert = (u32)virt_to_phys(gi->origin);
3178 VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
3179}
3180
3181void kvm_s390_gisa_enable(struct kvm *kvm)
3182{
3183 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3184 struct kvm_vcpu *vcpu;
3185 unsigned long i;
3186 u32 gisa_desc;
3187
3188 if (gi->origin)
3189 return;
3190 kvm_s390_gisa_init(kvm);
3191 gisa_desc = kvm_s390_get_gisa_desc(kvm);
3192 if (!gisa_desc)
3193 return;
3194 kvm_for_each_vcpu(i, vcpu, kvm) {
3195 mutex_lock(&vcpu->mutex);
3196 vcpu->arch.sie_block->gd = gisa_desc;
3197 vcpu->arch.sie_block->eca |= ECA_AIV;
3198 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3199 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3200 mutex_unlock(&vcpu->mutex);
3201 }
3202}
3203
3204void kvm_s390_gisa_destroy(struct kvm *kvm)
3205{
3206 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3207 struct kvm_s390_gisa *gisa = gi->origin;
3208
3209 if (!gi->origin)
3210 return;
3211 WARN(gi->alert.mask != 0x00,
3212 "unexpected non zero alert.mask 0x%02x",
3213 gi->alert.mask);
3214 gi->alert.mask = 0x00;
3215 if (gisa_set_iam(gi->origin, gi->alert.mask))
3216 process_gib_alert_list();
3217 hrtimer_cancel(&gi->timer);
3218 gi->origin = NULL;
3219 VM_EVENT(kvm, 3, "gisa 0x%pK destroyed", gisa);
3220}
3221
3222void kvm_s390_gisa_disable(struct kvm *kvm)
3223{
3224 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3225 struct kvm_vcpu *vcpu;
3226 unsigned long i;
3227
3228 if (!gi->origin)
3229 return;
3230 kvm_for_each_vcpu(i, vcpu, kvm) {
3231 mutex_lock(&vcpu->mutex);
3232 vcpu->arch.sie_block->eca &= ~ECA_AIV;
3233 vcpu->arch.sie_block->gd = 0U;
3234 mutex_unlock(&vcpu->mutex);
3235 VCPU_EVENT(vcpu, 3, "AIV disabled for cpu %03u", vcpu->vcpu_id);
3236 }
3237 kvm_s390_gisa_destroy(kvm);
3238}
3239
3240/**
3241 * kvm_s390_gisc_register - register a guest ISC
3242 *
3243 * @kvm: the kernel vm to work with
3244 * @gisc: the guest interruption sub class to register
3245 *
3246 * The function extends the vm specific alert mask to use.
3247 * The effective IAM mask in the GISA is updated as well
3248 * in case the GISA is not part of the GIB alert list.
3249 * It will be updated latest when the IAM gets restored
3250 * by gisa_get_ipm_or_restore_iam().
3251 *
3252 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3253 * has registered with the channel subsystem.
3254 * -ENODEV in case the vm uses no GISA
3255 * -ERANGE in case the guest ISC is invalid
3256 */
3257int kvm_s390_gisc_register(struct kvm *kvm, u32 gisc)
3258{
3259 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3260
3261 if (!gi->origin)
3262 return -ENODEV;
3263 if (gisc > MAX_ISC)
3264 return -ERANGE;
3265
3266 spin_lock(&gi->alert.ref_lock);
3267 gi->alert.ref_count[gisc]++;
3268 if (gi->alert.ref_count[gisc] == 1) {
3269 gi->alert.mask |= 0x80 >> gisc;
3270 gisa_set_iam(gi->origin, gi->alert.mask);
3271 }
3272 spin_unlock(&gi->alert.ref_lock);
3273
3274 return gib->nisc;
3275}
3276EXPORT_SYMBOL_GPL(kvm_s390_gisc_register);
3277
3278/**
3279 * kvm_s390_gisc_unregister - unregister a guest ISC
3280 *
3281 * @kvm: the kernel vm to work with
3282 * @gisc: the guest interruption sub class to register
3283 *
3284 * The function reduces the vm specific alert mask to use.
3285 * The effective IAM mask in the GISA is updated as well
3286 * in case the GISA is not part of the GIB alert list.
3287 * It will be updated latest when the IAM gets restored
3288 * by gisa_get_ipm_or_restore_iam().
3289 *
3290 * Returns: the nonspecific ISC (NISC) the gib alert mechanism
3291 * has registered with the channel subsystem.
3292 * -ENODEV in case the vm uses no GISA
3293 * -ERANGE in case the guest ISC is invalid
3294 * -EINVAL in case the guest ISC is not registered
3295 */
3296int kvm_s390_gisc_unregister(struct kvm *kvm, u32 gisc)
3297{
3298 struct kvm_s390_gisa_interrupt *gi = &kvm->arch.gisa_int;
3299 int rc = 0;
3300
3301 if (!gi->origin)
3302 return -ENODEV;
3303 if (gisc > MAX_ISC)
3304 return -ERANGE;
3305
3306 spin_lock(&gi->alert.ref_lock);
3307 if (gi->alert.ref_count[gisc] == 0) {
3308 rc = -EINVAL;
3309 goto out;
3310 }
3311 gi->alert.ref_count[gisc]--;
3312 if (gi->alert.ref_count[gisc] == 0) {
3313 gi->alert.mask &= ~(0x80 >> gisc);
3314 gisa_set_iam(gi->origin, gi->alert.mask);
3315 }
3316out:
3317 spin_unlock(&gi->alert.ref_lock);
3318
3319 return rc;
3320}
3321EXPORT_SYMBOL_GPL(kvm_s390_gisc_unregister);
3322
3323static void aen_host_forward(unsigned long si)
3324{
3325 struct kvm_s390_gisa_interrupt *gi;
3326 struct zpci_gaite *gaite;
3327 struct kvm *kvm;
3328
3329 gaite = (struct zpci_gaite *)aift->gait +
3330 (si * sizeof(struct zpci_gaite));
3331 if (gaite->count == 0)
3332 return;
3333 if (gaite->aisb != 0)
3334 set_bit_inv(gaite->aisbo, phys_to_virt(gaite->aisb));
3335
3336 kvm = kvm_s390_pci_si_to_kvm(aift, si);
3337 if (!kvm)
3338 return;
3339 gi = &kvm->arch.gisa_int;
3340
3341 if (!(gi->origin->g1.simm & AIS_MODE_MASK(gaite->gisc)) ||
3342 !(gi->origin->g1.nimm & AIS_MODE_MASK(gaite->gisc))) {
3343 gisa_set_ipm_gisc(gi->origin, gaite->gisc);
3344 if (hrtimer_active(&gi->timer))
3345 hrtimer_cancel(&gi->timer);
3346 hrtimer_start(&gi->timer, 0, HRTIMER_MODE_REL);
3347 kvm->stat.aen_forward++;
3348 }
3349}
3350
3351static void aen_process_gait(u8 isc)
3352{
3353 bool found = false, first = true;
3354 union zpci_sic_iib iib = {{0}};
3355 unsigned long si, flags;
3356
3357 spin_lock_irqsave(&aift->gait_lock, flags);
3358
3359 if (!aift->gait) {
3360 spin_unlock_irqrestore(&aift->gait_lock, flags);
3361 return;
3362 }
3363
3364 for (si = 0;;) {
3365 /* Scan adapter summary indicator bit vector */
3366 si = airq_iv_scan(aift->sbv, si, airq_iv_end(aift->sbv));
3367 if (si == -1UL) {
3368 if (first || found) {
3369 /* Re-enable interrupts. */
3370 zpci_set_irq_ctrl(SIC_IRQ_MODE_SINGLE, isc,
3371 &iib);
3372 first = found = false;
3373 } else {
3374 /* Interrupts on and all bits processed */
3375 break;
3376 }
3377 found = false;
3378 si = 0;
3379 /* Scan again after re-enabling interrupts */
3380 continue;
3381 }
3382 found = true;
3383 aen_host_forward(si);
3384 }
3385
3386 spin_unlock_irqrestore(&aift->gait_lock, flags);
3387}
3388
3389static void gib_alert_irq_handler(struct airq_struct *airq,
3390 struct tpi_info *tpi_info)
3391{
3392 struct tpi_adapter_info *info = (struct tpi_adapter_info *)tpi_info;
3393
3394 inc_irq_stat(IRQIO_GAL);
3395
3396 if ((info->forward || info->error) &&
3397 IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3398 aen_process_gait(info->isc);
3399 if (info->aism != 0)
3400 process_gib_alert_list();
3401 } else {
3402 process_gib_alert_list();
3403 }
3404}
3405
3406static struct airq_struct gib_alert_irq = {
3407 .handler = gib_alert_irq_handler,
3408};
3409
3410void kvm_s390_gib_destroy(void)
3411{
3412 if (!gib)
3413 return;
3414 if (kvm_s390_pci_interp_allowed() && aift) {
3415 mutex_lock(&aift->aift_lock);
3416 kvm_s390_pci_aen_exit();
3417 mutex_unlock(&aift->aift_lock);
3418 }
3419 chsc_sgib(0);
3420 unregister_adapter_interrupt(&gib_alert_irq);
3421 free_page((unsigned long)gib);
3422 gib = NULL;
3423}
3424
3425int __init kvm_s390_gib_init(u8 nisc)
3426{
3427 u32 gib_origin;
3428 int rc = 0;
3429
3430 if (!css_general_characteristics.aiv) {
3431 KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
3432 goto out;
3433 }
3434
3435 gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3436 if (!gib) {
3437 rc = -ENOMEM;
3438 goto out;
3439 }
3440
3441 gib_alert_irq.isc = nisc;
3442 if (register_adapter_interrupt(&gib_alert_irq)) {
3443 pr_err("Registering the GIB alert interruption handler failed\n");
3444 rc = -EIO;
3445 goto out_free_gib;
3446 }
3447 /* adapter interrupts used for AP (applicable here) don't use the LSI */
3448 *gib_alert_irq.lsi_ptr = 0xff;
3449
3450 gib->nisc = nisc;
3451 gib_origin = virt_to_phys(gib);
3452 if (chsc_sgib(gib_origin)) {
3453 pr_err("Associating the GIB with the AIV facility failed\n");
3454 free_page((unsigned long)gib);
3455 gib = NULL;
3456 rc = -EIO;
3457 goto out_unreg_gal;
3458 }
3459
3460 if (kvm_s390_pci_interp_allowed()) {
3461 if (kvm_s390_pci_aen_init(nisc)) {
3462 pr_err("Initializing AEN for PCI failed\n");
3463 rc = -EIO;
3464 goto out_unreg_gal;
3465 }
3466 }
3467
3468 KVM_EVENT(3, "gib 0x%pK (nisc=%d) initialized", gib, gib->nisc);
3469 goto out;
3470
3471out_unreg_gal:
3472 unregister_adapter_interrupt(&gib_alert_irq);
3473out_free_gib:
3474 free_page((unsigned long)gib);
3475 gib = NULL;
3476out:
3477 return rc;
3478}
1/*
2 * handling kvm guest interrupts
3 *
4 * Copyright IBM Corp. 2008, 2015
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License (version 2 only)
8 * as published by the Free Software Foundation.
9 *
10 * Author(s): Carsten Otte <cotte@de.ibm.com>
11 */
12
13#include <linux/interrupt.h>
14#include <linux/kvm_host.h>
15#include <linux/hrtimer.h>
16#include <linux/mmu_context.h>
17#include <linux/signal.h>
18#include <linux/slab.h>
19#include <linux/bitmap.h>
20#include <linux/vmalloc.h>
21#include <asm/asm-offsets.h>
22#include <asm/dis.h>
23#include <linux/uaccess.h>
24#include <asm/sclp.h>
25#include <asm/isc.h>
26#include <asm/gmap.h>
27#include <asm/switch_to.h>
28#include <asm/nmi.h>
29#include "kvm-s390.h"
30#include "gaccess.h"
31#include "trace-s390.h"
32
33#define PFAULT_INIT 0x0600
34#define PFAULT_DONE 0x0680
35#define VIRTIO_PARAM 0x0d00
36
37/* handle external calls via sigp interpretation facility */
38static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
39{
40 int c, scn;
41
42 if (!(atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_ECALL_PEND))
43 return 0;
44
45 BUG_ON(!kvm_s390_use_sca_entries());
46 read_lock(&vcpu->kvm->arch.sca_lock);
47 if (vcpu->kvm->arch.use_esca) {
48 struct esca_block *sca = vcpu->kvm->arch.sca;
49 union esca_sigp_ctrl sigp_ctrl =
50 sca->cpu[vcpu->vcpu_id].sigp_ctrl;
51
52 c = sigp_ctrl.c;
53 scn = sigp_ctrl.scn;
54 } else {
55 struct bsca_block *sca = vcpu->kvm->arch.sca;
56 union bsca_sigp_ctrl sigp_ctrl =
57 sca->cpu[vcpu->vcpu_id].sigp_ctrl;
58
59 c = sigp_ctrl.c;
60 scn = sigp_ctrl.scn;
61 }
62 read_unlock(&vcpu->kvm->arch.sca_lock);
63
64 if (src_id)
65 *src_id = scn;
66
67 return c;
68}
69
70static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
71{
72 int expect, rc;
73
74 BUG_ON(!kvm_s390_use_sca_entries());
75 read_lock(&vcpu->kvm->arch.sca_lock);
76 if (vcpu->kvm->arch.use_esca) {
77 struct esca_block *sca = vcpu->kvm->arch.sca;
78 union esca_sigp_ctrl *sigp_ctrl =
79 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
80 union esca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
81
82 new_val.scn = src_id;
83 new_val.c = 1;
84 old_val.c = 0;
85
86 expect = old_val.value;
87 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
88 } else {
89 struct bsca_block *sca = vcpu->kvm->arch.sca;
90 union bsca_sigp_ctrl *sigp_ctrl =
91 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
92 union bsca_sigp_ctrl new_val = {0}, old_val = *sigp_ctrl;
93
94 new_val.scn = src_id;
95 new_val.c = 1;
96 old_val.c = 0;
97
98 expect = old_val.value;
99 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
100 }
101 read_unlock(&vcpu->kvm->arch.sca_lock);
102
103 if (rc != expect) {
104 /* another external call is pending */
105 return -EBUSY;
106 }
107 atomic_or(CPUSTAT_ECALL_PEND, &vcpu->arch.sie_block->cpuflags);
108 return 0;
109}
110
111static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
112{
113 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
114 int rc, expect;
115
116 if (!kvm_s390_use_sca_entries())
117 return;
118 atomic_andnot(CPUSTAT_ECALL_PEND, li->cpuflags);
119 read_lock(&vcpu->kvm->arch.sca_lock);
120 if (vcpu->kvm->arch.use_esca) {
121 struct esca_block *sca = vcpu->kvm->arch.sca;
122 union esca_sigp_ctrl *sigp_ctrl =
123 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
124 union esca_sigp_ctrl old = *sigp_ctrl;
125
126 expect = old.value;
127 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
128 } else {
129 struct bsca_block *sca = vcpu->kvm->arch.sca;
130 union bsca_sigp_ctrl *sigp_ctrl =
131 &(sca->cpu[vcpu->vcpu_id].sigp_ctrl);
132 union bsca_sigp_ctrl old = *sigp_ctrl;
133
134 expect = old.value;
135 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
136 }
137 read_unlock(&vcpu->kvm->arch.sca_lock);
138 WARN_ON(rc != expect); /* cannot clear? */
139}
140
141int psw_extint_disabled(struct kvm_vcpu *vcpu)
142{
143 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
144}
145
146static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
147{
148 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
149}
150
151static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
152{
153 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
154}
155
156static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
157{
158 return psw_extint_disabled(vcpu) &&
159 psw_ioint_disabled(vcpu) &&
160 psw_mchk_disabled(vcpu);
161}
162
163static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
164{
165 if (psw_extint_disabled(vcpu) ||
166 !(vcpu->arch.sie_block->gcr[0] & 0x800ul))
167 return 0;
168 if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
169 /* No timer interrupts when single stepping */
170 return 0;
171 return 1;
172}
173
174static int ckc_irq_pending(struct kvm_vcpu *vcpu)
175{
176 if (vcpu->arch.sie_block->ckc >= kvm_s390_get_tod_clock_fast(vcpu->kvm))
177 return 0;
178 return ckc_interrupts_enabled(vcpu);
179}
180
181static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
182{
183 return !psw_extint_disabled(vcpu) &&
184 (vcpu->arch.sie_block->gcr[0] & 0x400ul);
185}
186
187static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
188{
189 if (!cpu_timer_interrupts_enabled(vcpu))
190 return 0;
191 return kvm_s390_get_cpu_timer(vcpu) >> 63;
192}
193
194static inline int is_ioirq(unsigned long irq_type)
195{
196 return ((irq_type >= IRQ_PEND_IO_ISC_0) &&
197 (irq_type <= IRQ_PEND_IO_ISC_7));
198}
199
200static uint64_t isc_to_isc_bits(int isc)
201{
202 return (0x80 >> isc) << 24;
203}
204
205static inline u8 int_word_to_isc(u32 int_word)
206{
207 return (int_word & 0x38000000) >> 27;
208}
209
210static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
211{
212 return vcpu->kvm->arch.float_int.pending_irqs |
213 vcpu->arch.local_int.pending_irqs;
214}
215
216static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
217 unsigned long active_mask)
218{
219 int i;
220
221 for (i = 0; i <= MAX_ISC; i++)
222 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
223 active_mask &= ~(1UL << (IRQ_PEND_IO_ISC_0 + i));
224
225 return active_mask;
226}
227
228static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
229{
230 unsigned long active_mask;
231
232 active_mask = pending_irqs(vcpu);
233 if (!active_mask)
234 return 0;
235
236 if (psw_extint_disabled(vcpu))
237 active_mask &= ~IRQ_PEND_EXT_MASK;
238 if (psw_ioint_disabled(vcpu))
239 active_mask &= ~IRQ_PEND_IO_MASK;
240 else
241 active_mask = disable_iscs(vcpu, active_mask);
242 if (!(vcpu->arch.sie_block->gcr[0] & 0x2000ul))
243 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
244 if (!(vcpu->arch.sie_block->gcr[0] & 0x4000ul))
245 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
246 if (!(vcpu->arch.sie_block->gcr[0] & 0x800ul))
247 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
248 if (!(vcpu->arch.sie_block->gcr[0] & 0x400ul))
249 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
250 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
251 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
252 if (psw_mchk_disabled(vcpu))
253 active_mask &= ~IRQ_PEND_MCHK_MASK;
254 if (!(vcpu->arch.sie_block->gcr[14] &
255 vcpu->kvm->arch.float_int.mchk.cr14))
256 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
257
258 /*
259 * STOP irqs will never be actively delivered. They are triggered via
260 * intercept requests and cleared when the stop intercept is performed.
261 */
262 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
263
264 return active_mask;
265}
266
267static void __set_cpu_idle(struct kvm_vcpu *vcpu)
268{
269 atomic_or(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
270 set_bit(vcpu->vcpu_id, vcpu->arch.local_int.float_int->idle_mask);
271}
272
273static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
274{
275 atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
276 clear_bit(vcpu->vcpu_id, vcpu->arch.local_int.float_int->idle_mask);
277}
278
279static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
280{
281 atomic_andnot(CPUSTAT_IO_INT | CPUSTAT_EXT_INT | CPUSTAT_STOP_INT,
282 &vcpu->arch.sie_block->cpuflags);
283 vcpu->arch.sie_block->lctl = 0x0000;
284 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
285
286 if (guestdbg_enabled(vcpu)) {
287 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
288 LCTL_CR10 | LCTL_CR11);
289 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
290 }
291}
292
293static void __set_cpuflag(struct kvm_vcpu *vcpu, u32 flag)
294{
295 atomic_or(flag, &vcpu->arch.sie_block->cpuflags);
296}
297
298static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
299{
300 if (!(pending_irqs(vcpu) & IRQ_PEND_IO_MASK))
301 return;
302 else if (psw_ioint_disabled(vcpu))
303 __set_cpuflag(vcpu, CPUSTAT_IO_INT);
304 else
305 vcpu->arch.sie_block->lctl |= LCTL_CR6;
306}
307
308static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
309{
310 if (!(pending_irqs(vcpu) & IRQ_PEND_EXT_MASK))
311 return;
312 if (psw_extint_disabled(vcpu))
313 __set_cpuflag(vcpu, CPUSTAT_EXT_INT);
314 else
315 vcpu->arch.sie_block->lctl |= LCTL_CR0;
316}
317
318static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
319{
320 if (!(pending_irqs(vcpu) & IRQ_PEND_MCHK_MASK))
321 return;
322 if (psw_mchk_disabled(vcpu))
323 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
324 else
325 vcpu->arch.sie_block->lctl |= LCTL_CR14;
326}
327
328static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
329{
330 if (kvm_s390_is_stop_irq_pending(vcpu))
331 __set_cpuflag(vcpu, CPUSTAT_STOP_INT);
332}
333
334/* Set interception request for non-deliverable interrupts */
335static void set_intercept_indicators(struct kvm_vcpu *vcpu)
336{
337 set_intercept_indicators_io(vcpu);
338 set_intercept_indicators_ext(vcpu);
339 set_intercept_indicators_mchk(vcpu);
340 set_intercept_indicators_stop(vcpu);
341}
342
343static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
344{
345 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
346 int rc;
347
348 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
349 0, 0);
350
351 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
352 (u16 *)__LC_EXT_INT_CODE);
353 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
354 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
355 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
356 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
357 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
358 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
359 return rc ? -EFAULT : 0;
360}
361
362static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
363{
364 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
365 int rc;
366
367 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
368 0, 0);
369
370 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
371 (u16 __user *)__LC_EXT_INT_CODE);
372 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
373 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
374 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
375 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
376 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
377 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
378 return rc ? -EFAULT : 0;
379}
380
381static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
382{
383 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
384 struct kvm_s390_ext_info ext;
385 int rc;
386
387 spin_lock(&li->lock);
388 ext = li->irq.ext;
389 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
390 li->irq.ext.ext_params2 = 0;
391 spin_unlock(&li->lock);
392
393 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
394 ext.ext_params2);
395 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
396 KVM_S390_INT_PFAULT_INIT,
397 0, ext.ext_params2);
398
399 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
400 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
401 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
402 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
403 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
404 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
405 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
406 return rc ? -EFAULT : 0;
407}
408
409static int __write_machine_check(struct kvm_vcpu *vcpu,
410 struct kvm_s390_mchk_info *mchk)
411{
412 unsigned long ext_sa_addr;
413 freg_t fprs[NUM_FPRS];
414 union mci mci;
415 int rc;
416
417 mci.val = mchk->mcic;
418 /* take care of lazy register loading */
419 save_fpu_regs();
420 save_access_regs(vcpu->run->s.regs.acrs);
421
422 /* Extended save area */
423 rc = read_guest_lc(vcpu, __LC_VX_SAVE_AREA_ADDR, &ext_sa_addr,
424 sizeof(unsigned long));
425 /* Only bits 0-53 are used for address formation */
426 ext_sa_addr &= ~0x3ffUL;
427 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
428 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
429 512))
430 mci.vr = 0;
431 } else {
432 mci.vr = 0;
433 }
434
435 /* General interruption information */
436 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
437 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
438 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
439 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
440 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
441 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
442
443 /* Register-save areas */
444 if (MACHINE_HAS_VX) {
445 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
446 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
447 } else {
448 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
449 vcpu->run->s.regs.fprs, 128);
450 }
451 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
452 vcpu->run->s.regs.gprs, 128);
453 rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
454 (u32 __user *) __LC_FP_CREG_SAVE_AREA);
455 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
456 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
457 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
458 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
459 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
460 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
461 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
462 &vcpu->run->s.regs.acrs, 64);
463 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
464 &vcpu->arch.sie_block->gcr, 128);
465
466 /* Extended interruption information */
467 rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
468 (u32 __user *) __LC_EXT_DAMAGE_CODE);
469 rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
470 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
471 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
472 sizeof(mchk->fixed_logout));
473 return rc ? -EFAULT : 0;
474}
475
476static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
477{
478 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
479 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
480 struct kvm_s390_mchk_info mchk = {};
481 int deliver = 0;
482 int rc = 0;
483
484 spin_lock(&fi->lock);
485 spin_lock(&li->lock);
486 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
487 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
488 /*
489 * If there was an exigent machine check pending, then any
490 * repressible machine checks that might have been pending
491 * are indicated along with it, so always clear bits for
492 * repressible and exigent interrupts
493 */
494 mchk = li->irq.mchk;
495 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
496 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
497 memset(&li->irq.mchk, 0, sizeof(mchk));
498 deliver = 1;
499 }
500 /*
501 * We indicate floating repressible conditions along with
502 * other pending conditions. Channel Report Pending and Channel
503 * Subsystem damage are the only two and and are indicated by
504 * bits in mcic and masked in cr14.
505 */
506 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
507 mchk.mcic |= fi->mchk.mcic;
508 mchk.cr14 |= fi->mchk.cr14;
509 memset(&fi->mchk, 0, sizeof(mchk));
510 deliver = 1;
511 }
512 spin_unlock(&li->lock);
513 spin_unlock(&fi->lock);
514
515 if (deliver) {
516 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
517 mchk.mcic);
518 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
519 KVM_S390_MCHK,
520 mchk.cr14, mchk.mcic);
521 rc = __write_machine_check(vcpu, &mchk);
522 }
523 return rc;
524}
525
526static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
527{
528 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
529 int rc;
530
531 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
532 vcpu->stat.deliver_restart_signal++;
533 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
534
535 rc = write_guest_lc(vcpu,
536 offsetof(struct lowcore, restart_old_psw),
537 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
538 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
539 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
540 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
541 return rc ? -EFAULT : 0;
542}
543
544static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
545{
546 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
547 struct kvm_s390_prefix_info prefix;
548
549 spin_lock(&li->lock);
550 prefix = li->irq.prefix;
551 li->irq.prefix.address = 0;
552 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
553 spin_unlock(&li->lock);
554
555 vcpu->stat.deliver_prefix_signal++;
556 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
557 KVM_S390_SIGP_SET_PREFIX,
558 prefix.address, 0);
559
560 kvm_s390_set_prefix(vcpu, prefix.address);
561 return 0;
562}
563
564static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
565{
566 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
567 int rc;
568 int cpu_addr;
569
570 spin_lock(&li->lock);
571 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
572 clear_bit(cpu_addr, li->sigp_emerg_pending);
573 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
574 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
575 spin_unlock(&li->lock);
576
577 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
578 vcpu->stat.deliver_emergency_signal++;
579 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
580 cpu_addr, 0);
581
582 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
583 (u16 *)__LC_EXT_INT_CODE);
584 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
585 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
586 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
587 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
588 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
589 return rc ? -EFAULT : 0;
590}
591
592static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
593{
594 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
595 struct kvm_s390_extcall_info extcall;
596 int rc;
597
598 spin_lock(&li->lock);
599 extcall = li->irq.extcall;
600 li->irq.extcall.code = 0;
601 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
602 spin_unlock(&li->lock);
603
604 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
605 vcpu->stat.deliver_external_call++;
606 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
607 KVM_S390_INT_EXTERNAL_CALL,
608 extcall.code, 0);
609
610 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
611 (u16 *)__LC_EXT_INT_CODE);
612 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
613 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
614 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
615 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
616 sizeof(psw_t));
617 return rc ? -EFAULT : 0;
618}
619
620static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
621{
622 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
623 struct kvm_s390_pgm_info pgm_info;
624 int rc = 0, nullifying = false;
625 u16 ilen;
626
627 spin_lock(&li->lock);
628 pgm_info = li->irq.pgm;
629 clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
630 memset(&li->irq.pgm, 0, sizeof(pgm_info));
631 spin_unlock(&li->lock);
632
633 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
634 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
635 pgm_info.code, ilen);
636 vcpu->stat.deliver_program_int++;
637 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
638 pgm_info.code, 0);
639
640 switch (pgm_info.code & ~PGM_PER) {
641 case PGM_AFX_TRANSLATION:
642 case PGM_ASX_TRANSLATION:
643 case PGM_EX_TRANSLATION:
644 case PGM_LFX_TRANSLATION:
645 case PGM_LSTE_SEQUENCE:
646 case PGM_LSX_TRANSLATION:
647 case PGM_LX_TRANSLATION:
648 case PGM_PRIMARY_AUTHORITY:
649 case PGM_SECONDARY_AUTHORITY:
650 nullifying = true;
651 /* fall through */
652 case PGM_SPACE_SWITCH:
653 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
654 (u64 *)__LC_TRANS_EXC_CODE);
655 break;
656 case PGM_ALEN_TRANSLATION:
657 case PGM_ALE_SEQUENCE:
658 case PGM_ASTE_INSTANCE:
659 case PGM_ASTE_SEQUENCE:
660 case PGM_ASTE_VALIDITY:
661 case PGM_EXTENDED_AUTHORITY:
662 rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
663 (u8 *)__LC_EXC_ACCESS_ID);
664 nullifying = true;
665 break;
666 case PGM_ASCE_TYPE:
667 case PGM_PAGE_TRANSLATION:
668 case PGM_REGION_FIRST_TRANS:
669 case PGM_REGION_SECOND_TRANS:
670 case PGM_REGION_THIRD_TRANS:
671 case PGM_SEGMENT_TRANSLATION:
672 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
673 (u64 *)__LC_TRANS_EXC_CODE);
674 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
675 (u8 *)__LC_EXC_ACCESS_ID);
676 rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
677 (u8 *)__LC_OP_ACCESS_ID);
678 nullifying = true;
679 break;
680 case PGM_MONITOR:
681 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
682 (u16 *)__LC_MON_CLASS_NR);
683 rc |= put_guest_lc(vcpu, pgm_info.mon_code,
684 (u64 *)__LC_MON_CODE);
685 break;
686 case PGM_VECTOR_PROCESSING:
687 case PGM_DATA:
688 rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
689 (u32 *)__LC_DATA_EXC_CODE);
690 break;
691 case PGM_PROTECTION:
692 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
693 (u64 *)__LC_TRANS_EXC_CODE);
694 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
695 (u8 *)__LC_EXC_ACCESS_ID);
696 break;
697 case PGM_STACK_FULL:
698 case PGM_STACK_EMPTY:
699 case PGM_STACK_SPECIFICATION:
700 case PGM_STACK_TYPE:
701 case PGM_STACK_OPERATION:
702 case PGM_TRACE_TABEL:
703 case PGM_CRYPTO_OPERATION:
704 nullifying = true;
705 break;
706 }
707
708 if (pgm_info.code & PGM_PER) {
709 rc |= put_guest_lc(vcpu, pgm_info.per_code,
710 (u8 *) __LC_PER_CODE);
711 rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
712 (u8 *)__LC_PER_ATMID);
713 rc |= put_guest_lc(vcpu, pgm_info.per_address,
714 (u64 *) __LC_PER_ADDRESS);
715 rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
716 (u8 *) __LC_PER_ACCESS_ID);
717 }
718
719 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
720 kvm_s390_rewind_psw(vcpu, ilen);
721
722 /* bit 1+2 of the target are the ilc, so we can directly use ilen */
723 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
724 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
725 (u64 *) __LC_LAST_BREAK);
726 rc |= put_guest_lc(vcpu, pgm_info.code,
727 (u16 *)__LC_PGM_INT_CODE);
728 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
729 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
730 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
731 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
732 return rc ? -EFAULT : 0;
733}
734
735static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
736{
737 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
738 struct kvm_s390_ext_info ext;
739 int rc = 0;
740
741 spin_lock(&fi->lock);
742 if (!(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
743 spin_unlock(&fi->lock);
744 return 0;
745 }
746 ext = fi->srv_signal;
747 memset(&fi->srv_signal, 0, sizeof(ext));
748 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
749 spin_unlock(&fi->lock);
750
751 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
752 ext.ext_params);
753 vcpu->stat.deliver_service_signal++;
754 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
755 ext.ext_params, 0);
756
757 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
758 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
759 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
760 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
761 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
762 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
763 rc |= put_guest_lc(vcpu, ext.ext_params,
764 (u32 *)__LC_EXT_PARAMS);
765
766 return rc ? -EFAULT : 0;
767}
768
769static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
770{
771 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
772 struct kvm_s390_interrupt_info *inti;
773 int rc = 0;
774
775 spin_lock(&fi->lock);
776 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
777 struct kvm_s390_interrupt_info,
778 list);
779 if (inti) {
780 list_del(&inti->list);
781 fi->counters[FIRQ_CNTR_PFAULT] -= 1;
782 }
783 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
784 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
785 spin_unlock(&fi->lock);
786
787 if (inti) {
788 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
789 KVM_S390_INT_PFAULT_DONE, 0,
790 inti->ext.ext_params2);
791 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
792 inti->ext.ext_params2);
793
794 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
795 (u16 *)__LC_EXT_INT_CODE);
796 rc |= put_guest_lc(vcpu, PFAULT_DONE,
797 (u16 *)__LC_EXT_CPU_ADDR);
798 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
799 &vcpu->arch.sie_block->gpsw,
800 sizeof(psw_t));
801 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
802 &vcpu->arch.sie_block->gpsw,
803 sizeof(psw_t));
804 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
805 (u64 *)__LC_EXT_PARAMS2);
806 kfree(inti);
807 }
808 return rc ? -EFAULT : 0;
809}
810
811static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
812{
813 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
814 struct kvm_s390_interrupt_info *inti;
815 int rc = 0;
816
817 spin_lock(&fi->lock);
818 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
819 struct kvm_s390_interrupt_info,
820 list);
821 if (inti) {
822 VCPU_EVENT(vcpu, 4,
823 "deliver: virtio parm: 0x%x,parm64: 0x%llx",
824 inti->ext.ext_params, inti->ext.ext_params2);
825 vcpu->stat.deliver_virtio_interrupt++;
826 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
827 inti->type,
828 inti->ext.ext_params,
829 inti->ext.ext_params2);
830 list_del(&inti->list);
831 fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
832 }
833 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
834 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
835 spin_unlock(&fi->lock);
836
837 if (inti) {
838 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
839 (u16 *)__LC_EXT_INT_CODE);
840 rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
841 (u16 *)__LC_EXT_CPU_ADDR);
842 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
843 &vcpu->arch.sie_block->gpsw,
844 sizeof(psw_t));
845 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
846 &vcpu->arch.sie_block->gpsw,
847 sizeof(psw_t));
848 rc |= put_guest_lc(vcpu, inti->ext.ext_params,
849 (u32 *)__LC_EXT_PARAMS);
850 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
851 (u64 *)__LC_EXT_PARAMS2);
852 kfree(inti);
853 }
854 return rc ? -EFAULT : 0;
855}
856
857static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
858 unsigned long irq_type)
859{
860 struct list_head *isc_list;
861 struct kvm_s390_float_interrupt *fi;
862 struct kvm_s390_interrupt_info *inti = NULL;
863 int rc = 0;
864
865 fi = &vcpu->kvm->arch.float_int;
866
867 spin_lock(&fi->lock);
868 isc_list = &fi->lists[irq_type - IRQ_PEND_IO_ISC_0];
869 inti = list_first_entry_or_null(isc_list,
870 struct kvm_s390_interrupt_info,
871 list);
872 if (inti) {
873 if (inti->type & KVM_S390_INT_IO_AI_MASK)
874 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
875 else
876 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
877 inti->io.subchannel_id >> 8,
878 inti->io.subchannel_id >> 1 & 0x3,
879 inti->io.subchannel_nr);
880
881 vcpu->stat.deliver_io_int++;
882 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
883 inti->type,
884 ((__u32)inti->io.subchannel_id << 16) |
885 inti->io.subchannel_nr,
886 ((__u64)inti->io.io_int_parm << 32) |
887 inti->io.io_int_word);
888 list_del(&inti->list);
889 fi->counters[FIRQ_CNTR_IO] -= 1;
890 }
891 if (list_empty(isc_list))
892 clear_bit(irq_type, &fi->pending_irqs);
893 spin_unlock(&fi->lock);
894
895 if (inti) {
896 rc = put_guest_lc(vcpu, inti->io.subchannel_id,
897 (u16 *)__LC_SUBCHANNEL_ID);
898 rc |= put_guest_lc(vcpu, inti->io.subchannel_nr,
899 (u16 *)__LC_SUBCHANNEL_NR);
900 rc |= put_guest_lc(vcpu, inti->io.io_int_parm,
901 (u32 *)__LC_IO_INT_PARM);
902 rc |= put_guest_lc(vcpu, inti->io.io_int_word,
903 (u32 *)__LC_IO_INT_WORD);
904 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
905 &vcpu->arch.sie_block->gpsw,
906 sizeof(psw_t));
907 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
908 &vcpu->arch.sie_block->gpsw,
909 sizeof(psw_t));
910 kfree(inti);
911 }
912
913 return rc ? -EFAULT : 0;
914}
915
916typedef int (*deliver_irq_t)(struct kvm_vcpu *vcpu);
917
918static const deliver_irq_t deliver_irq_funcs[] = {
919 [IRQ_PEND_MCHK_EX] = __deliver_machine_check,
920 [IRQ_PEND_MCHK_REP] = __deliver_machine_check,
921 [IRQ_PEND_PROG] = __deliver_prog,
922 [IRQ_PEND_EXT_EMERGENCY] = __deliver_emergency_signal,
923 [IRQ_PEND_EXT_EXTERNAL] = __deliver_external_call,
924 [IRQ_PEND_EXT_CLOCK_COMP] = __deliver_ckc,
925 [IRQ_PEND_EXT_CPU_TIMER] = __deliver_cpu_timer,
926 [IRQ_PEND_RESTART] = __deliver_restart,
927 [IRQ_PEND_SET_PREFIX] = __deliver_set_prefix,
928 [IRQ_PEND_PFAULT_INIT] = __deliver_pfault_init,
929 [IRQ_PEND_EXT_SERVICE] = __deliver_service,
930 [IRQ_PEND_PFAULT_DONE] = __deliver_pfault_done,
931 [IRQ_PEND_VIRTIO] = __deliver_virtio,
932};
933
934/* Check whether an external call is pending (deliverable or not) */
935int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
936{
937 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
938
939 if (!sclp.has_sigpif)
940 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
941
942 return sca_ext_call_pending(vcpu, NULL);
943}
944
945int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
946{
947 if (deliverable_irqs(vcpu))
948 return 1;
949
950 if (kvm_cpu_has_pending_timer(vcpu))
951 return 1;
952
953 /* external call pending and deliverable */
954 if (kvm_s390_ext_call_pending(vcpu) &&
955 !psw_extint_disabled(vcpu) &&
956 (vcpu->arch.sie_block->gcr[0] & 0x2000ul))
957 return 1;
958
959 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
960 return 1;
961 return 0;
962}
963
964int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
965{
966 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
967}
968
969static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
970{
971 u64 now, cputm, sltime = 0;
972
973 if (ckc_interrupts_enabled(vcpu)) {
974 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
975 sltime = tod_to_ns(vcpu->arch.sie_block->ckc - now);
976 /* already expired or overflow? */
977 if (!sltime || vcpu->arch.sie_block->ckc <= now)
978 return 0;
979 if (cpu_timer_interrupts_enabled(vcpu)) {
980 cputm = kvm_s390_get_cpu_timer(vcpu);
981 /* already expired? */
982 if (cputm >> 63)
983 return 0;
984 return min(sltime, tod_to_ns(cputm));
985 }
986 } else if (cpu_timer_interrupts_enabled(vcpu)) {
987 sltime = kvm_s390_get_cpu_timer(vcpu);
988 /* already expired? */
989 if (sltime >> 63)
990 return 0;
991 }
992 return sltime;
993}
994
995int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
996{
997 u64 sltime;
998
999 vcpu->stat.exit_wait_state++;
1000
1001 /* fast path */
1002 if (kvm_arch_vcpu_runnable(vcpu))
1003 return 0;
1004
1005 if (psw_interrupts_disabled(vcpu)) {
1006 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1007 return -EOPNOTSUPP; /* disabled wait */
1008 }
1009
1010 if (!ckc_interrupts_enabled(vcpu) &&
1011 !cpu_timer_interrupts_enabled(vcpu)) {
1012 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1013 __set_cpu_idle(vcpu);
1014 goto no_timer;
1015 }
1016
1017 sltime = __calculate_sltime(vcpu);
1018 if (!sltime)
1019 return 0;
1020
1021 __set_cpu_idle(vcpu);
1022 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1023 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1024no_timer:
1025 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1026 kvm_vcpu_block(vcpu);
1027 __unset_cpu_idle(vcpu);
1028 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1029
1030 hrtimer_cancel(&vcpu->arch.ckc_timer);
1031 return 0;
1032}
1033
1034void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1035{
1036 /*
1037 * We cannot move this into the if, as the CPU might be already
1038 * in kvm_vcpu_block without having the waitqueue set (polling)
1039 */
1040 vcpu->valid_wakeup = true;
1041 if (swait_active(&vcpu->wq)) {
1042 /*
1043 * The vcpu gave up the cpu voluntarily, mark it as a good
1044 * yield-candidate.
1045 */
1046 vcpu->preempted = true;
1047 swake_up(&vcpu->wq);
1048 vcpu->stat.halt_wakeup++;
1049 }
1050 /*
1051 * The VCPU might not be sleeping but is executing the VSIE. Let's
1052 * kick it, so it leaves the SIE to process the request.
1053 */
1054 kvm_s390_vsie_kick(vcpu);
1055}
1056
1057enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1058{
1059 struct kvm_vcpu *vcpu;
1060 u64 sltime;
1061
1062 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1063 sltime = __calculate_sltime(vcpu);
1064
1065 /*
1066 * If the monotonic clock runs faster than the tod clock we might be
1067 * woken up too early and have to go back to sleep to avoid deadlocks.
1068 */
1069 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1070 return HRTIMER_RESTART;
1071 kvm_s390_vcpu_wakeup(vcpu);
1072 return HRTIMER_NORESTART;
1073}
1074
1075void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1076{
1077 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1078
1079 spin_lock(&li->lock);
1080 li->pending_irqs = 0;
1081 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1082 memset(&li->irq, 0, sizeof(li->irq));
1083 spin_unlock(&li->lock);
1084
1085 sca_clear_ext_call(vcpu);
1086}
1087
1088int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1089{
1090 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1091 deliver_irq_t func;
1092 int rc = 0;
1093 unsigned long irq_type;
1094 unsigned long irqs;
1095
1096 __reset_intercept_indicators(vcpu);
1097
1098 /* pending ckc conditions might have been invalidated */
1099 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1100 if (ckc_irq_pending(vcpu))
1101 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1102
1103 /* pending cpu timer conditions might have been invalidated */
1104 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1105 if (cpu_timer_irq_pending(vcpu))
1106 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1107
1108 while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1109 /* bits are in the order of interrupt priority */
1110 irq_type = find_first_bit(&irqs, IRQ_PEND_COUNT);
1111 if (is_ioirq(irq_type)) {
1112 rc = __deliver_io(vcpu, irq_type);
1113 } else {
1114 func = deliver_irq_funcs[irq_type];
1115 if (!func) {
1116 WARN_ON_ONCE(func == NULL);
1117 clear_bit(irq_type, &li->pending_irqs);
1118 continue;
1119 }
1120 rc = func(vcpu);
1121 }
1122 }
1123
1124 set_intercept_indicators(vcpu);
1125
1126 return rc;
1127}
1128
1129static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1130{
1131 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1132
1133 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1134 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1135 irq->u.pgm.code, 0);
1136
1137 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1138 /* auto detection if no valid ILC was given */
1139 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1140 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1141 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1142 }
1143
1144 if (irq->u.pgm.code == PGM_PER) {
1145 li->irq.pgm.code |= PGM_PER;
1146 li->irq.pgm.flags = irq->u.pgm.flags;
1147 /* only modify PER related information */
1148 li->irq.pgm.per_address = irq->u.pgm.per_address;
1149 li->irq.pgm.per_code = irq->u.pgm.per_code;
1150 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1151 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1152 } else if (!(irq->u.pgm.code & PGM_PER)) {
1153 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1154 irq->u.pgm.code;
1155 li->irq.pgm.flags = irq->u.pgm.flags;
1156 /* only modify non-PER information */
1157 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1158 li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1159 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1160 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1161 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1162 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1163 } else {
1164 li->irq.pgm = irq->u.pgm;
1165 }
1166 set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1167 return 0;
1168}
1169
1170static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1171{
1172 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1173
1174 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1175 irq->u.ext.ext_params2);
1176 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1177 irq->u.ext.ext_params,
1178 irq->u.ext.ext_params2);
1179
1180 li->irq.ext = irq->u.ext;
1181 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1182 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1183 return 0;
1184}
1185
1186static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1187{
1188 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1189 struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1190 uint16_t src_id = irq->u.extcall.code;
1191
1192 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1193 src_id);
1194 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1195 src_id, 0);
1196
1197 /* sending vcpu invalid */
1198 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1199 return -EINVAL;
1200
1201 if (sclp.has_sigpif)
1202 return sca_inject_ext_call(vcpu, src_id);
1203
1204 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1205 return -EBUSY;
1206 *extcall = irq->u.extcall;
1207 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1208 return 0;
1209}
1210
1211static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1212{
1213 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1214 struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1215
1216 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1217 irq->u.prefix.address);
1218 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1219 irq->u.prefix.address, 0);
1220
1221 if (!is_vcpu_stopped(vcpu))
1222 return -EBUSY;
1223
1224 *prefix = irq->u.prefix;
1225 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1226 return 0;
1227}
1228
1229#define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1230static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1231{
1232 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1233 struct kvm_s390_stop_info *stop = &li->irq.stop;
1234 int rc = 0;
1235
1236 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1237
1238 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1239 return -EINVAL;
1240
1241 if (is_vcpu_stopped(vcpu)) {
1242 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1243 rc = kvm_s390_store_status_unloaded(vcpu,
1244 KVM_S390_STORE_STATUS_NOADDR);
1245 return rc;
1246 }
1247
1248 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1249 return -EBUSY;
1250 stop->flags = irq->u.stop.flags;
1251 __set_cpuflag(vcpu, CPUSTAT_STOP_INT);
1252 return 0;
1253}
1254
1255static int __inject_sigp_restart(struct kvm_vcpu *vcpu,
1256 struct kvm_s390_irq *irq)
1257{
1258 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1259
1260 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1261 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1262
1263 set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1264 return 0;
1265}
1266
1267static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1268 struct kvm_s390_irq *irq)
1269{
1270 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1271
1272 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1273 irq->u.emerg.code);
1274 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1275 irq->u.emerg.code, 0);
1276
1277 /* sending vcpu invalid */
1278 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1279 return -EINVAL;
1280
1281 set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1282 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1283 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1284 return 0;
1285}
1286
1287static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1288{
1289 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1290 struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1291
1292 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1293 irq->u.mchk.mcic);
1294 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1295 irq->u.mchk.mcic);
1296
1297 /*
1298 * Because repressible machine checks can be indicated along with
1299 * exigent machine checks (PoP, Chapter 11, Interruption action)
1300 * we need to combine cr14, mcic and external damage code.
1301 * Failing storage address and the logout area should not be or'ed
1302 * together, we just indicate the last occurrence of the corresponding
1303 * machine check
1304 */
1305 mchk->cr14 |= irq->u.mchk.cr14;
1306 mchk->mcic |= irq->u.mchk.mcic;
1307 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1308 mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1309 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1310 sizeof(mchk->fixed_logout));
1311 if (mchk->mcic & MCHK_EX_MASK)
1312 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1313 else if (mchk->mcic & MCHK_REP_MASK)
1314 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
1315 return 0;
1316}
1317
1318static int __inject_ckc(struct kvm_vcpu *vcpu)
1319{
1320 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1321
1322 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1323 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1324 0, 0);
1325
1326 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1327 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1328 return 0;
1329}
1330
1331static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1332{
1333 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1334
1335 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1336 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1337 0, 0);
1338
1339 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1340 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1341 return 0;
1342}
1343
1344static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1345 int isc, u32 schid)
1346{
1347 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1348 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1349 struct kvm_s390_interrupt_info *iter;
1350 u16 id = (schid & 0xffff0000U) >> 16;
1351 u16 nr = schid & 0x0000ffffU;
1352
1353 spin_lock(&fi->lock);
1354 list_for_each_entry(iter, isc_list, list) {
1355 if (schid && (id != iter->io.subchannel_id ||
1356 nr != iter->io.subchannel_nr))
1357 continue;
1358 /* found an appropriate entry */
1359 list_del_init(&iter->list);
1360 fi->counters[FIRQ_CNTR_IO] -= 1;
1361 if (list_empty(isc_list))
1362 clear_bit(IRQ_PEND_IO_ISC_0 + isc, &fi->pending_irqs);
1363 spin_unlock(&fi->lock);
1364 return iter;
1365 }
1366 spin_unlock(&fi->lock);
1367 return NULL;
1368}
1369
1370/*
1371 * Dequeue and return an I/O interrupt matching any of the interruption
1372 * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1373 */
1374struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1375 u64 isc_mask, u32 schid)
1376{
1377 struct kvm_s390_interrupt_info *inti = NULL;
1378 int isc;
1379
1380 for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1381 if (isc_mask & isc_to_isc_bits(isc))
1382 inti = get_io_int(kvm, isc, schid);
1383 }
1384 return inti;
1385}
1386
1387#define SCCB_MASK 0xFFFFFFF8
1388#define SCCB_EVENT_PENDING 0x3
1389
1390static int __inject_service(struct kvm *kvm,
1391 struct kvm_s390_interrupt_info *inti)
1392{
1393 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1394
1395 spin_lock(&fi->lock);
1396 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1397 /*
1398 * Early versions of the QEMU s390 bios will inject several
1399 * service interrupts after another without handling a
1400 * condition code indicating busy.
1401 * We will silently ignore those superfluous sccb values.
1402 * A future version of QEMU will take care of serialization
1403 * of servc requests
1404 */
1405 if (fi->srv_signal.ext_params & SCCB_MASK)
1406 goto out;
1407 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1408 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1409out:
1410 spin_unlock(&fi->lock);
1411 kfree(inti);
1412 return 0;
1413}
1414
1415static int __inject_virtio(struct kvm *kvm,
1416 struct kvm_s390_interrupt_info *inti)
1417{
1418 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1419
1420 spin_lock(&fi->lock);
1421 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1422 spin_unlock(&fi->lock);
1423 return -EBUSY;
1424 }
1425 fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1426 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1427 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1428 spin_unlock(&fi->lock);
1429 return 0;
1430}
1431
1432static int __inject_pfault_done(struct kvm *kvm,
1433 struct kvm_s390_interrupt_info *inti)
1434{
1435 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1436
1437 spin_lock(&fi->lock);
1438 if (fi->counters[FIRQ_CNTR_PFAULT] >=
1439 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1440 spin_unlock(&fi->lock);
1441 return -EBUSY;
1442 }
1443 fi->counters[FIRQ_CNTR_PFAULT] += 1;
1444 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1445 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1446 spin_unlock(&fi->lock);
1447 return 0;
1448}
1449
1450#define CR_PENDING_SUBCLASS 28
1451static int __inject_float_mchk(struct kvm *kvm,
1452 struct kvm_s390_interrupt_info *inti)
1453{
1454 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1455
1456 spin_lock(&fi->lock);
1457 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1458 fi->mchk.mcic |= inti->mchk.mcic;
1459 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1460 spin_unlock(&fi->lock);
1461 kfree(inti);
1462 return 0;
1463}
1464
1465static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1466{
1467 struct kvm_s390_float_interrupt *fi;
1468 struct list_head *list;
1469 int isc;
1470
1471 fi = &kvm->arch.float_int;
1472 spin_lock(&fi->lock);
1473 if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1474 spin_unlock(&fi->lock);
1475 return -EBUSY;
1476 }
1477 fi->counters[FIRQ_CNTR_IO] += 1;
1478
1479 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1480 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1481 else
1482 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1483 inti->io.subchannel_id >> 8,
1484 inti->io.subchannel_id >> 1 & 0x3,
1485 inti->io.subchannel_nr);
1486 isc = int_word_to_isc(inti->io.io_int_word);
1487 list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1488 list_add_tail(&inti->list, list);
1489 set_bit(IRQ_PEND_IO_ISC_0 + isc, &fi->pending_irqs);
1490 spin_unlock(&fi->lock);
1491 return 0;
1492}
1493
1494/*
1495 * Find a destination VCPU for a floating irq and kick it.
1496 */
1497static void __floating_irq_kick(struct kvm *kvm, u64 type)
1498{
1499 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1500 struct kvm_s390_local_interrupt *li;
1501 struct kvm_vcpu *dst_vcpu;
1502 int sigcpu, online_vcpus, nr_tries = 0;
1503
1504 online_vcpus = atomic_read(&kvm->online_vcpus);
1505 if (!online_vcpus)
1506 return;
1507
1508 /* find idle VCPUs first, then round robin */
1509 sigcpu = find_first_bit(fi->idle_mask, online_vcpus);
1510 if (sigcpu == online_vcpus) {
1511 do {
1512 sigcpu = fi->next_rr_cpu;
1513 fi->next_rr_cpu = (fi->next_rr_cpu + 1) % online_vcpus;
1514 /* avoid endless loops if all vcpus are stopped */
1515 if (nr_tries++ >= online_vcpus)
1516 return;
1517 } while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1518 }
1519 dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1520
1521 /* make the VCPU drop out of the SIE, or wake it up if sleeping */
1522 li = &dst_vcpu->arch.local_int;
1523 spin_lock(&li->lock);
1524 switch (type) {
1525 case KVM_S390_MCHK:
1526 atomic_or(CPUSTAT_STOP_INT, li->cpuflags);
1527 break;
1528 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1529 atomic_or(CPUSTAT_IO_INT, li->cpuflags);
1530 break;
1531 default:
1532 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1533 break;
1534 }
1535 spin_unlock(&li->lock);
1536 kvm_s390_vcpu_wakeup(dst_vcpu);
1537}
1538
1539static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1540{
1541 u64 type = READ_ONCE(inti->type);
1542 int rc;
1543
1544 switch (type) {
1545 case KVM_S390_MCHK:
1546 rc = __inject_float_mchk(kvm, inti);
1547 break;
1548 case KVM_S390_INT_VIRTIO:
1549 rc = __inject_virtio(kvm, inti);
1550 break;
1551 case KVM_S390_INT_SERVICE:
1552 rc = __inject_service(kvm, inti);
1553 break;
1554 case KVM_S390_INT_PFAULT_DONE:
1555 rc = __inject_pfault_done(kvm, inti);
1556 break;
1557 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1558 rc = __inject_io(kvm, inti);
1559 break;
1560 default:
1561 rc = -EINVAL;
1562 }
1563 if (rc)
1564 return rc;
1565
1566 __floating_irq_kick(kvm, type);
1567 return 0;
1568}
1569
1570int kvm_s390_inject_vm(struct kvm *kvm,
1571 struct kvm_s390_interrupt *s390int)
1572{
1573 struct kvm_s390_interrupt_info *inti;
1574 int rc;
1575
1576 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1577 if (!inti)
1578 return -ENOMEM;
1579
1580 inti->type = s390int->type;
1581 switch (inti->type) {
1582 case KVM_S390_INT_VIRTIO:
1583 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
1584 s390int->parm, s390int->parm64);
1585 inti->ext.ext_params = s390int->parm;
1586 inti->ext.ext_params2 = s390int->parm64;
1587 break;
1588 case KVM_S390_INT_SERVICE:
1589 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
1590 inti->ext.ext_params = s390int->parm;
1591 break;
1592 case KVM_S390_INT_PFAULT_DONE:
1593 inti->ext.ext_params2 = s390int->parm64;
1594 break;
1595 case KVM_S390_MCHK:
1596 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
1597 s390int->parm64);
1598 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
1599 inti->mchk.mcic = s390int->parm64;
1600 break;
1601 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1602 inti->io.subchannel_id = s390int->parm >> 16;
1603 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
1604 inti->io.io_int_parm = s390int->parm64 >> 32;
1605 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
1606 break;
1607 default:
1608 kfree(inti);
1609 return -EINVAL;
1610 }
1611 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
1612 2);
1613
1614 rc = __inject_vm(kvm, inti);
1615 if (rc)
1616 kfree(inti);
1617 return rc;
1618}
1619
1620int kvm_s390_reinject_io_int(struct kvm *kvm,
1621 struct kvm_s390_interrupt_info *inti)
1622{
1623 return __inject_vm(kvm, inti);
1624}
1625
1626int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
1627 struct kvm_s390_irq *irq)
1628{
1629 irq->type = s390int->type;
1630 switch (irq->type) {
1631 case KVM_S390_PROGRAM_INT:
1632 if (s390int->parm & 0xffff0000)
1633 return -EINVAL;
1634 irq->u.pgm.code = s390int->parm;
1635 break;
1636 case KVM_S390_SIGP_SET_PREFIX:
1637 irq->u.prefix.address = s390int->parm;
1638 break;
1639 case KVM_S390_SIGP_STOP:
1640 irq->u.stop.flags = s390int->parm;
1641 break;
1642 case KVM_S390_INT_EXTERNAL_CALL:
1643 if (s390int->parm & 0xffff0000)
1644 return -EINVAL;
1645 irq->u.extcall.code = s390int->parm;
1646 break;
1647 case KVM_S390_INT_EMERGENCY:
1648 if (s390int->parm & 0xffff0000)
1649 return -EINVAL;
1650 irq->u.emerg.code = s390int->parm;
1651 break;
1652 case KVM_S390_MCHK:
1653 irq->u.mchk.mcic = s390int->parm64;
1654 break;
1655 }
1656 return 0;
1657}
1658
1659int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
1660{
1661 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1662
1663 return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1664}
1665
1666void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
1667{
1668 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1669
1670 spin_lock(&li->lock);
1671 li->irq.stop.flags = 0;
1672 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1673 spin_unlock(&li->lock);
1674}
1675
1676static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1677{
1678 int rc;
1679
1680 switch (irq->type) {
1681 case KVM_S390_PROGRAM_INT:
1682 rc = __inject_prog(vcpu, irq);
1683 break;
1684 case KVM_S390_SIGP_SET_PREFIX:
1685 rc = __inject_set_prefix(vcpu, irq);
1686 break;
1687 case KVM_S390_SIGP_STOP:
1688 rc = __inject_sigp_stop(vcpu, irq);
1689 break;
1690 case KVM_S390_RESTART:
1691 rc = __inject_sigp_restart(vcpu, irq);
1692 break;
1693 case KVM_S390_INT_CLOCK_COMP:
1694 rc = __inject_ckc(vcpu);
1695 break;
1696 case KVM_S390_INT_CPU_TIMER:
1697 rc = __inject_cpu_timer(vcpu);
1698 break;
1699 case KVM_S390_INT_EXTERNAL_CALL:
1700 rc = __inject_extcall(vcpu, irq);
1701 break;
1702 case KVM_S390_INT_EMERGENCY:
1703 rc = __inject_sigp_emergency(vcpu, irq);
1704 break;
1705 case KVM_S390_MCHK:
1706 rc = __inject_mchk(vcpu, irq);
1707 break;
1708 case KVM_S390_INT_PFAULT_INIT:
1709 rc = __inject_pfault_init(vcpu, irq);
1710 break;
1711 case KVM_S390_INT_VIRTIO:
1712 case KVM_S390_INT_SERVICE:
1713 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1714 default:
1715 rc = -EINVAL;
1716 }
1717
1718 return rc;
1719}
1720
1721int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1722{
1723 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1724 int rc;
1725
1726 spin_lock(&li->lock);
1727 rc = do_inject_vcpu(vcpu, irq);
1728 spin_unlock(&li->lock);
1729 if (!rc)
1730 kvm_s390_vcpu_wakeup(vcpu);
1731 return rc;
1732}
1733
1734static inline void clear_irq_list(struct list_head *_list)
1735{
1736 struct kvm_s390_interrupt_info *inti, *n;
1737
1738 list_for_each_entry_safe(inti, n, _list, list) {
1739 list_del(&inti->list);
1740 kfree(inti);
1741 }
1742}
1743
1744static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
1745 struct kvm_s390_irq *irq)
1746{
1747 irq->type = inti->type;
1748 switch (inti->type) {
1749 case KVM_S390_INT_PFAULT_INIT:
1750 case KVM_S390_INT_PFAULT_DONE:
1751 case KVM_S390_INT_VIRTIO:
1752 irq->u.ext = inti->ext;
1753 break;
1754 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1755 irq->u.io = inti->io;
1756 break;
1757 }
1758}
1759
1760void kvm_s390_clear_float_irqs(struct kvm *kvm)
1761{
1762 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1763 int i;
1764
1765 spin_lock(&fi->lock);
1766 fi->pending_irqs = 0;
1767 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
1768 memset(&fi->mchk, 0, sizeof(fi->mchk));
1769 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1770 clear_irq_list(&fi->lists[i]);
1771 for (i = 0; i < FIRQ_MAX_COUNT; i++)
1772 fi->counters[i] = 0;
1773 spin_unlock(&fi->lock);
1774};
1775
1776static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
1777{
1778 struct kvm_s390_interrupt_info *inti;
1779 struct kvm_s390_float_interrupt *fi;
1780 struct kvm_s390_irq *buf;
1781 struct kvm_s390_irq *irq;
1782 int max_irqs;
1783 int ret = 0;
1784 int n = 0;
1785 int i;
1786
1787 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
1788 return -EINVAL;
1789
1790 /*
1791 * We are already using -ENOMEM to signal
1792 * userspace it may retry with a bigger buffer,
1793 * so we need to use something else for this case
1794 */
1795 buf = vzalloc(len);
1796 if (!buf)
1797 return -ENOBUFS;
1798
1799 max_irqs = len / sizeof(struct kvm_s390_irq);
1800
1801 fi = &kvm->arch.float_int;
1802 spin_lock(&fi->lock);
1803 for (i = 0; i < FIRQ_LIST_COUNT; i++) {
1804 list_for_each_entry(inti, &fi->lists[i], list) {
1805 if (n == max_irqs) {
1806 /* signal userspace to try again */
1807 ret = -ENOMEM;
1808 goto out;
1809 }
1810 inti_to_irq(inti, &buf[n]);
1811 n++;
1812 }
1813 }
1814 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs)) {
1815 if (n == max_irqs) {
1816 /* signal userspace to try again */
1817 ret = -ENOMEM;
1818 goto out;
1819 }
1820 irq = (struct kvm_s390_irq *) &buf[n];
1821 irq->type = KVM_S390_INT_SERVICE;
1822 irq->u.ext = fi->srv_signal;
1823 n++;
1824 }
1825 if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
1826 if (n == max_irqs) {
1827 /* signal userspace to try again */
1828 ret = -ENOMEM;
1829 goto out;
1830 }
1831 irq = (struct kvm_s390_irq *) &buf[n];
1832 irq->type = KVM_S390_MCHK;
1833 irq->u.mchk = fi->mchk;
1834 n++;
1835}
1836
1837out:
1838 spin_unlock(&fi->lock);
1839 if (!ret && n > 0) {
1840 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
1841 ret = -EFAULT;
1842 }
1843 vfree(buf);
1844
1845 return ret < 0 ? ret : n;
1846}
1847
1848static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1849{
1850 int r;
1851
1852 switch (attr->group) {
1853 case KVM_DEV_FLIC_GET_ALL_IRQS:
1854 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
1855 attr->attr);
1856 break;
1857 default:
1858 r = -EINVAL;
1859 }
1860
1861 return r;
1862}
1863
1864static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
1865 u64 addr)
1866{
1867 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
1868 void *target = NULL;
1869 void __user *source;
1870 u64 size;
1871
1872 if (get_user(inti->type, (u64 __user *)addr))
1873 return -EFAULT;
1874
1875 switch (inti->type) {
1876 case KVM_S390_INT_PFAULT_INIT:
1877 case KVM_S390_INT_PFAULT_DONE:
1878 case KVM_S390_INT_VIRTIO:
1879 case KVM_S390_INT_SERVICE:
1880 target = (void *) &inti->ext;
1881 source = &uptr->u.ext;
1882 size = sizeof(inti->ext);
1883 break;
1884 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1885 target = (void *) &inti->io;
1886 source = &uptr->u.io;
1887 size = sizeof(inti->io);
1888 break;
1889 case KVM_S390_MCHK:
1890 target = (void *) &inti->mchk;
1891 source = &uptr->u.mchk;
1892 size = sizeof(inti->mchk);
1893 break;
1894 default:
1895 return -EINVAL;
1896 }
1897
1898 if (copy_from_user(target, source, size))
1899 return -EFAULT;
1900
1901 return 0;
1902}
1903
1904static int enqueue_floating_irq(struct kvm_device *dev,
1905 struct kvm_device_attr *attr)
1906{
1907 struct kvm_s390_interrupt_info *inti = NULL;
1908 int r = 0;
1909 int len = attr->attr;
1910
1911 if (len % sizeof(struct kvm_s390_irq) != 0)
1912 return -EINVAL;
1913 else if (len > KVM_S390_FLIC_MAX_BUFFER)
1914 return -EINVAL;
1915
1916 while (len >= sizeof(struct kvm_s390_irq)) {
1917 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1918 if (!inti)
1919 return -ENOMEM;
1920
1921 r = copy_irq_from_user(inti, attr->addr);
1922 if (r) {
1923 kfree(inti);
1924 return r;
1925 }
1926 r = __inject_vm(dev->kvm, inti);
1927 if (r) {
1928 kfree(inti);
1929 return r;
1930 }
1931 len -= sizeof(struct kvm_s390_irq);
1932 attr->addr += sizeof(struct kvm_s390_irq);
1933 }
1934
1935 return r;
1936}
1937
1938static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
1939{
1940 if (id >= MAX_S390_IO_ADAPTERS)
1941 return NULL;
1942 return kvm->arch.adapters[id];
1943}
1944
1945static int register_io_adapter(struct kvm_device *dev,
1946 struct kvm_device_attr *attr)
1947{
1948 struct s390_io_adapter *adapter;
1949 struct kvm_s390_io_adapter adapter_info;
1950
1951 if (copy_from_user(&adapter_info,
1952 (void __user *)attr->addr, sizeof(adapter_info)))
1953 return -EFAULT;
1954
1955 if ((adapter_info.id >= MAX_S390_IO_ADAPTERS) ||
1956 (dev->kvm->arch.adapters[adapter_info.id] != NULL))
1957 return -EINVAL;
1958
1959 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
1960 if (!adapter)
1961 return -ENOMEM;
1962
1963 INIT_LIST_HEAD(&adapter->maps);
1964 init_rwsem(&adapter->maps_lock);
1965 atomic_set(&adapter->nr_maps, 0);
1966 adapter->id = adapter_info.id;
1967 adapter->isc = adapter_info.isc;
1968 adapter->maskable = adapter_info.maskable;
1969 adapter->masked = false;
1970 adapter->swap = adapter_info.swap;
1971 dev->kvm->arch.adapters[adapter->id] = adapter;
1972
1973 return 0;
1974}
1975
1976int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
1977{
1978 int ret;
1979 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
1980
1981 if (!adapter || !adapter->maskable)
1982 return -EINVAL;
1983 ret = adapter->masked;
1984 adapter->masked = masked;
1985 return ret;
1986}
1987
1988static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr)
1989{
1990 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
1991 struct s390_map_info *map;
1992 int ret;
1993
1994 if (!adapter || !addr)
1995 return -EINVAL;
1996
1997 map = kzalloc(sizeof(*map), GFP_KERNEL);
1998 if (!map) {
1999 ret = -ENOMEM;
2000 goto out;
2001 }
2002 INIT_LIST_HEAD(&map->list);
2003 map->guest_addr = addr;
2004 map->addr = gmap_translate(kvm->arch.gmap, addr);
2005 if (map->addr == -EFAULT) {
2006 ret = -EFAULT;
2007 goto out;
2008 }
2009 ret = get_user_pages_fast(map->addr, 1, 1, &map->page);
2010 if (ret < 0)
2011 goto out;
2012 BUG_ON(ret != 1);
2013 down_write(&adapter->maps_lock);
2014 if (atomic_inc_return(&adapter->nr_maps) < MAX_S390_ADAPTER_MAPS) {
2015 list_add_tail(&map->list, &adapter->maps);
2016 ret = 0;
2017 } else {
2018 put_page(map->page);
2019 ret = -EINVAL;
2020 }
2021 up_write(&adapter->maps_lock);
2022out:
2023 if (ret)
2024 kfree(map);
2025 return ret;
2026}
2027
2028static int kvm_s390_adapter_unmap(struct kvm *kvm, unsigned int id, __u64 addr)
2029{
2030 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2031 struct s390_map_info *map, *tmp;
2032 int found = 0;
2033
2034 if (!adapter || !addr)
2035 return -EINVAL;
2036
2037 down_write(&adapter->maps_lock);
2038 list_for_each_entry_safe(map, tmp, &adapter->maps, list) {
2039 if (map->guest_addr == addr) {
2040 found = 1;
2041 atomic_dec(&adapter->nr_maps);
2042 list_del(&map->list);
2043 put_page(map->page);
2044 kfree(map);
2045 break;
2046 }
2047 }
2048 up_write(&adapter->maps_lock);
2049
2050 return found ? 0 : -EINVAL;
2051}
2052
2053void kvm_s390_destroy_adapters(struct kvm *kvm)
2054{
2055 int i;
2056 struct s390_map_info *map, *tmp;
2057
2058 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) {
2059 if (!kvm->arch.adapters[i])
2060 continue;
2061 list_for_each_entry_safe(map, tmp,
2062 &kvm->arch.adapters[i]->maps, list) {
2063 list_del(&map->list);
2064 put_page(map->page);
2065 kfree(map);
2066 }
2067 kfree(kvm->arch.adapters[i]);
2068 }
2069}
2070
2071static int modify_io_adapter(struct kvm_device *dev,
2072 struct kvm_device_attr *attr)
2073{
2074 struct kvm_s390_io_adapter_req req;
2075 struct s390_io_adapter *adapter;
2076 int ret;
2077
2078 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2079 return -EFAULT;
2080
2081 adapter = get_io_adapter(dev->kvm, req.id);
2082 if (!adapter)
2083 return -EINVAL;
2084 switch (req.type) {
2085 case KVM_S390_IO_ADAPTER_MASK:
2086 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2087 if (ret > 0)
2088 ret = 0;
2089 break;
2090 case KVM_S390_IO_ADAPTER_MAP:
2091 ret = kvm_s390_adapter_map(dev->kvm, req.id, req.addr);
2092 break;
2093 case KVM_S390_IO_ADAPTER_UNMAP:
2094 ret = kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr);
2095 break;
2096 default:
2097 ret = -EINVAL;
2098 }
2099
2100 return ret;
2101}
2102
2103static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2104
2105{
2106 const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2107 u32 schid;
2108
2109 if (attr->flags)
2110 return -EINVAL;
2111 if (attr->attr != sizeof(schid))
2112 return -EINVAL;
2113 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2114 return -EFAULT;
2115 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2116 /*
2117 * If userspace is conforming to the architecture, we can have at most
2118 * one pending I/O interrupt per subchannel, so this is effectively a
2119 * clear all.
2120 */
2121 return 0;
2122}
2123
2124static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2125{
2126 int r = 0;
2127 unsigned int i;
2128 struct kvm_vcpu *vcpu;
2129
2130 switch (attr->group) {
2131 case KVM_DEV_FLIC_ENQUEUE:
2132 r = enqueue_floating_irq(dev, attr);
2133 break;
2134 case KVM_DEV_FLIC_CLEAR_IRQS:
2135 kvm_s390_clear_float_irqs(dev->kvm);
2136 break;
2137 case KVM_DEV_FLIC_APF_ENABLE:
2138 dev->kvm->arch.gmap->pfault_enabled = 1;
2139 break;
2140 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2141 dev->kvm->arch.gmap->pfault_enabled = 0;
2142 /*
2143 * Make sure no async faults are in transition when
2144 * clearing the queues. So we don't need to worry
2145 * about late coming workers.
2146 */
2147 synchronize_srcu(&dev->kvm->srcu);
2148 kvm_for_each_vcpu(i, vcpu, dev->kvm)
2149 kvm_clear_async_pf_completion_queue(vcpu);
2150 break;
2151 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2152 r = register_io_adapter(dev, attr);
2153 break;
2154 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2155 r = modify_io_adapter(dev, attr);
2156 break;
2157 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2158 r = clear_io_irq(dev->kvm, attr);
2159 break;
2160 default:
2161 r = -EINVAL;
2162 }
2163
2164 return r;
2165}
2166
2167static int flic_has_attr(struct kvm_device *dev,
2168 struct kvm_device_attr *attr)
2169{
2170 switch (attr->group) {
2171 case KVM_DEV_FLIC_GET_ALL_IRQS:
2172 case KVM_DEV_FLIC_ENQUEUE:
2173 case KVM_DEV_FLIC_CLEAR_IRQS:
2174 case KVM_DEV_FLIC_APF_ENABLE:
2175 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2176 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2177 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2178 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2179 return 0;
2180 }
2181 return -ENXIO;
2182}
2183
2184static int flic_create(struct kvm_device *dev, u32 type)
2185{
2186 if (!dev)
2187 return -EINVAL;
2188 if (dev->kvm->arch.flic)
2189 return -EINVAL;
2190 dev->kvm->arch.flic = dev;
2191 return 0;
2192}
2193
2194static void flic_destroy(struct kvm_device *dev)
2195{
2196 dev->kvm->arch.flic = NULL;
2197 kfree(dev);
2198}
2199
2200/* s390 floating irq controller (flic) */
2201struct kvm_device_ops kvm_flic_ops = {
2202 .name = "kvm-flic",
2203 .get_attr = flic_get_attr,
2204 .set_attr = flic_set_attr,
2205 .has_attr = flic_has_attr,
2206 .create = flic_create,
2207 .destroy = flic_destroy,
2208};
2209
2210static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2211{
2212 unsigned long bit;
2213
2214 bit = bit_nr + (addr % PAGE_SIZE) * 8;
2215
2216 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2217}
2218
2219static struct s390_map_info *get_map_info(struct s390_io_adapter *adapter,
2220 u64 addr)
2221{
2222 struct s390_map_info *map;
2223
2224 if (!adapter)
2225 return NULL;
2226
2227 list_for_each_entry(map, &adapter->maps, list) {
2228 if (map->guest_addr == addr)
2229 return map;
2230 }
2231 return NULL;
2232}
2233
2234static int adapter_indicators_set(struct kvm *kvm,
2235 struct s390_io_adapter *adapter,
2236 struct kvm_s390_adapter_int *adapter_int)
2237{
2238 unsigned long bit;
2239 int summary_set, idx;
2240 struct s390_map_info *info;
2241 void *map;
2242
2243 info = get_map_info(adapter, adapter_int->ind_addr);
2244 if (!info)
2245 return -1;
2246 map = page_address(info->page);
2247 bit = get_ind_bit(info->addr, adapter_int->ind_offset, adapter->swap);
2248 set_bit(bit, map);
2249 idx = srcu_read_lock(&kvm->srcu);
2250 mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2251 set_page_dirty_lock(info->page);
2252 info = get_map_info(adapter, adapter_int->summary_addr);
2253 if (!info) {
2254 srcu_read_unlock(&kvm->srcu, idx);
2255 return -1;
2256 }
2257 map = page_address(info->page);
2258 bit = get_ind_bit(info->addr, adapter_int->summary_offset,
2259 adapter->swap);
2260 summary_set = test_and_set_bit(bit, map);
2261 mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2262 set_page_dirty_lock(info->page);
2263 srcu_read_unlock(&kvm->srcu, idx);
2264 return summary_set ? 0 : 1;
2265}
2266
2267/*
2268 * < 0 - not injected due to error
2269 * = 0 - coalesced, summary indicator already active
2270 * > 0 - injected interrupt
2271 */
2272static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2273 struct kvm *kvm, int irq_source_id, int level,
2274 bool line_status)
2275{
2276 int ret;
2277 struct s390_io_adapter *adapter;
2278
2279 /* We're only interested in the 0->1 transition. */
2280 if (!level)
2281 return 0;
2282 adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2283 if (!adapter)
2284 return -1;
2285 down_read(&adapter->maps_lock);
2286 ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2287 up_read(&adapter->maps_lock);
2288 if ((ret > 0) && !adapter->masked) {
2289 struct kvm_s390_interrupt s390int = {
2290 .type = KVM_S390_INT_IO(1, 0, 0, 0),
2291 .parm = 0,
2292 .parm64 = (adapter->isc << 27) | 0x80000000,
2293 };
2294 ret = kvm_s390_inject_vm(kvm, &s390int);
2295 if (ret == 0)
2296 ret = 1;
2297 }
2298 return ret;
2299}
2300
2301int kvm_set_routing_entry(struct kvm *kvm,
2302 struct kvm_kernel_irq_routing_entry *e,
2303 const struct kvm_irq_routing_entry *ue)
2304{
2305 int ret;
2306
2307 switch (ue->type) {
2308 case KVM_IRQ_ROUTING_S390_ADAPTER:
2309 e->set = set_adapter_int;
2310 e->adapter.summary_addr = ue->u.adapter.summary_addr;
2311 e->adapter.ind_addr = ue->u.adapter.ind_addr;
2312 e->adapter.summary_offset = ue->u.adapter.summary_offset;
2313 e->adapter.ind_offset = ue->u.adapter.ind_offset;
2314 e->adapter.adapter_id = ue->u.adapter.adapter_id;
2315 ret = 0;
2316 break;
2317 default:
2318 ret = -EINVAL;
2319 }
2320
2321 return ret;
2322}
2323
2324int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2325 int irq_source_id, int level, bool line_status)
2326{
2327 return -EINVAL;
2328}
2329
2330int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2331{
2332 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2333 struct kvm_s390_irq *buf;
2334 int r = 0;
2335 int n;
2336
2337 buf = vmalloc(len);
2338 if (!buf)
2339 return -ENOMEM;
2340
2341 if (copy_from_user((void *) buf, irqstate, len)) {
2342 r = -EFAULT;
2343 goto out_free;
2344 }
2345
2346 /*
2347 * Don't allow setting the interrupt state
2348 * when there are already interrupts pending
2349 */
2350 spin_lock(&li->lock);
2351 if (li->pending_irqs) {
2352 r = -EBUSY;
2353 goto out_unlock;
2354 }
2355
2356 for (n = 0; n < len / sizeof(*buf); n++) {
2357 r = do_inject_vcpu(vcpu, &buf[n]);
2358 if (r)
2359 break;
2360 }
2361
2362out_unlock:
2363 spin_unlock(&li->lock);
2364out_free:
2365 vfree(buf);
2366
2367 return r;
2368}
2369
2370static void store_local_irq(struct kvm_s390_local_interrupt *li,
2371 struct kvm_s390_irq *irq,
2372 unsigned long irq_type)
2373{
2374 switch (irq_type) {
2375 case IRQ_PEND_MCHK_EX:
2376 case IRQ_PEND_MCHK_REP:
2377 irq->type = KVM_S390_MCHK;
2378 irq->u.mchk = li->irq.mchk;
2379 break;
2380 case IRQ_PEND_PROG:
2381 irq->type = KVM_S390_PROGRAM_INT;
2382 irq->u.pgm = li->irq.pgm;
2383 break;
2384 case IRQ_PEND_PFAULT_INIT:
2385 irq->type = KVM_S390_INT_PFAULT_INIT;
2386 irq->u.ext = li->irq.ext;
2387 break;
2388 case IRQ_PEND_EXT_EXTERNAL:
2389 irq->type = KVM_S390_INT_EXTERNAL_CALL;
2390 irq->u.extcall = li->irq.extcall;
2391 break;
2392 case IRQ_PEND_EXT_CLOCK_COMP:
2393 irq->type = KVM_S390_INT_CLOCK_COMP;
2394 break;
2395 case IRQ_PEND_EXT_CPU_TIMER:
2396 irq->type = KVM_S390_INT_CPU_TIMER;
2397 break;
2398 case IRQ_PEND_SIGP_STOP:
2399 irq->type = KVM_S390_SIGP_STOP;
2400 irq->u.stop = li->irq.stop;
2401 break;
2402 case IRQ_PEND_RESTART:
2403 irq->type = KVM_S390_RESTART;
2404 break;
2405 case IRQ_PEND_SET_PREFIX:
2406 irq->type = KVM_S390_SIGP_SET_PREFIX;
2407 irq->u.prefix = li->irq.prefix;
2408 break;
2409 }
2410}
2411
2412int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
2413{
2414 int scn;
2415 unsigned long sigp_emerg_pending[BITS_TO_LONGS(KVM_MAX_VCPUS)];
2416 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2417 unsigned long pending_irqs;
2418 struct kvm_s390_irq irq;
2419 unsigned long irq_type;
2420 int cpuaddr;
2421 int n = 0;
2422
2423 spin_lock(&li->lock);
2424 pending_irqs = li->pending_irqs;
2425 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
2426 sizeof(sigp_emerg_pending));
2427 spin_unlock(&li->lock);
2428
2429 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
2430 memset(&irq, 0, sizeof(irq));
2431 if (irq_type == IRQ_PEND_EXT_EMERGENCY)
2432 continue;
2433 if (n + sizeof(irq) > len)
2434 return -ENOBUFS;
2435 store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
2436 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2437 return -EFAULT;
2438 n += sizeof(irq);
2439 }
2440
2441 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
2442 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
2443 memset(&irq, 0, sizeof(irq));
2444 if (n + sizeof(irq) > len)
2445 return -ENOBUFS;
2446 irq.type = KVM_S390_INT_EMERGENCY;
2447 irq.u.emerg.code = cpuaddr;
2448 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2449 return -EFAULT;
2450 n += sizeof(irq);
2451 }
2452 }
2453
2454 if (sca_ext_call_pending(vcpu, &scn)) {
2455 if (n + sizeof(irq) > len)
2456 return -ENOBUFS;
2457 memset(&irq, 0, sizeof(irq));
2458 irq.type = KVM_S390_INT_EXTERNAL_CALL;
2459 irq.u.extcall.code = scn;
2460 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2461 return -EFAULT;
2462 n += sizeof(irq);
2463 }
2464
2465 return n;
2466}