2 * handling kvm guest interrupts
4 * Copyright IBM Corp. 2008, 2015
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.
10 * Author(s): Carsten Otte <cotte@de.ibm.com>
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>
23 #include <linux/uaccess.h>
27 #include <asm/switch_to.h>
31 #include "trace-s390.h"
33 #define PFAULT_INIT 0x0600
34 #define PFAULT_DONE 0x0680
35 #define VIRTIO_PARAM 0x0d00
37 /* handle external calls via sigp interpretation facility */
38 static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
42 if (!(atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_ECALL_PEND))
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;
55 struct bsca_block *sca = vcpu->kvm->arch.sca;
56 union bsca_sigp_ctrl sigp_ctrl =
57 sca->cpu[vcpu->vcpu_id].sigp_ctrl;
62 read_unlock(&vcpu->kvm->arch.sca_lock);
70 static int sca_inject_ext_call(struct kvm_vcpu *vcpu, int src_id)
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;
86 expect = old_val.value;
87 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
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;
98 expect = old_val.value;
99 rc = cmpxchg(&sigp_ctrl->value, old_val.value, new_val.value);
101 read_unlock(&vcpu->kvm->arch.sca_lock);
104 /* another external call is pending */
107 atomic_or(CPUSTAT_ECALL_PEND, &vcpu->arch.sie_block->cpuflags);
111 static void sca_clear_ext_call(struct kvm_vcpu *vcpu)
113 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
116 if (!kvm_s390_use_sca_entries())
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;
127 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
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;
135 rc = cmpxchg(&sigp_ctrl->value, old.value, 0);
137 read_unlock(&vcpu->kvm->arch.sca_lock);
138 WARN_ON(rc != expect); /* cannot clear? */
141 int psw_extint_disabled(struct kvm_vcpu *vcpu)
143 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
146 static int psw_ioint_disabled(struct kvm_vcpu *vcpu)
148 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_IO);
151 static int psw_mchk_disabled(struct kvm_vcpu *vcpu)
153 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK);
156 static int psw_interrupts_disabled(struct kvm_vcpu *vcpu)
158 return psw_extint_disabled(vcpu) &&
159 psw_ioint_disabled(vcpu) &&
160 psw_mchk_disabled(vcpu);
163 static int ckc_interrupts_enabled(struct kvm_vcpu *vcpu)
165 if (psw_extint_disabled(vcpu) ||
166 !(vcpu->arch.sie_block->gcr[0] & 0x800ul))
168 if (guestdbg_enabled(vcpu) && guestdbg_sstep_enabled(vcpu))
169 /* No timer interrupts when single stepping */
174 static int ckc_irq_pending(struct kvm_vcpu *vcpu)
176 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
177 const u64 ckc = vcpu->arch.sie_block->ckc;
179 if (vcpu->arch.sie_block->gcr[0] & 0x0020000000000000ul) {
180 if ((s64)ckc >= (s64)now)
182 } else if (ckc >= now) {
185 return ckc_interrupts_enabled(vcpu);
188 static int cpu_timer_interrupts_enabled(struct kvm_vcpu *vcpu)
190 return !psw_extint_disabled(vcpu) &&
191 (vcpu->arch.sie_block->gcr[0] & 0x400ul);
194 static int cpu_timer_irq_pending(struct kvm_vcpu *vcpu)
196 if (!cpu_timer_interrupts_enabled(vcpu))
198 return kvm_s390_get_cpu_timer(vcpu) >> 63;
201 static inline int is_ioirq(unsigned long irq_type)
203 return ((irq_type >= IRQ_PEND_IO_ISC_0) &&
204 (irq_type <= IRQ_PEND_IO_ISC_7));
207 static uint64_t isc_to_isc_bits(int isc)
209 return (0x80 >> isc) << 24;
212 static inline u8 int_word_to_isc(u32 int_word)
214 return (int_word & 0x38000000) >> 27;
217 static inline unsigned long pending_irqs(struct kvm_vcpu *vcpu)
219 return vcpu->kvm->arch.float_int.pending_irqs |
220 vcpu->arch.local_int.pending_irqs;
223 static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
224 unsigned long active_mask)
228 for (i = 0; i <= MAX_ISC; i++)
229 if (!(vcpu->arch.sie_block->gcr[6] & isc_to_isc_bits(i)))
230 active_mask &= ~(1UL << (IRQ_PEND_IO_ISC_0 + i));
235 static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
237 unsigned long active_mask;
239 active_mask = pending_irqs(vcpu);
243 if (psw_extint_disabled(vcpu))
244 active_mask &= ~IRQ_PEND_EXT_MASK;
245 if (psw_ioint_disabled(vcpu))
246 active_mask &= ~IRQ_PEND_IO_MASK;
248 active_mask = disable_iscs(vcpu, active_mask);
249 if (!(vcpu->arch.sie_block->gcr[0] & 0x2000ul))
250 __clear_bit(IRQ_PEND_EXT_EXTERNAL, &active_mask);
251 if (!(vcpu->arch.sie_block->gcr[0] & 0x4000ul))
252 __clear_bit(IRQ_PEND_EXT_EMERGENCY, &active_mask);
253 if (!(vcpu->arch.sie_block->gcr[0] & 0x800ul))
254 __clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &active_mask);
255 if (!(vcpu->arch.sie_block->gcr[0] & 0x400ul))
256 __clear_bit(IRQ_PEND_EXT_CPU_TIMER, &active_mask);
257 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
258 __clear_bit(IRQ_PEND_EXT_SERVICE, &active_mask);
259 if (psw_mchk_disabled(vcpu))
260 active_mask &= ~IRQ_PEND_MCHK_MASK;
262 * Check both floating and local interrupt's cr14 because
263 * bit IRQ_PEND_MCHK_REP could be set in both cases.
265 if (!(vcpu->arch.sie_block->gcr[14] &
266 (vcpu->kvm->arch.float_int.mchk.cr14 |
267 vcpu->arch.local_int.irq.mchk.cr14)))
268 __clear_bit(IRQ_PEND_MCHK_REP, &active_mask);
271 * STOP irqs will never be actively delivered. They are triggered via
272 * intercept requests and cleared when the stop intercept is performed.
274 __clear_bit(IRQ_PEND_SIGP_STOP, &active_mask);
279 static void __set_cpu_idle(struct kvm_vcpu *vcpu)
281 atomic_or(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
282 set_bit(vcpu->vcpu_id, vcpu->arch.local_int.float_int->idle_mask);
285 static void __unset_cpu_idle(struct kvm_vcpu *vcpu)
287 atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
288 clear_bit(vcpu->vcpu_id, vcpu->arch.local_int.float_int->idle_mask);
291 static void __reset_intercept_indicators(struct kvm_vcpu *vcpu)
293 atomic_andnot(CPUSTAT_IO_INT | CPUSTAT_EXT_INT | CPUSTAT_STOP_INT,
294 &vcpu->arch.sie_block->cpuflags);
295 vcpu->arch.sie_block->lctl = 0x0000;
296 vcpu->arch.sie_block->ictl &= ~(ICTL_LPSW | ICTL_STCTL | ICTL_PINT);
298 if (guestdbg_enabled(vcpu)) {
299 vcpu->arch.sie_block->lctl |= (LCTL_CR0 | LCTL_CR9 |
300 LCTL_CR10 | LCTL_CR11);
301 vcpu->arch.sie_block->ictl |= (ICTL_STCTL | ICTL_PINT);
305 static void __set_cpuflag(struct kvm_vcpu *vcpu, u32 flag)
307 atomic_or(flag, &vcpu->arch.sie_block->cpuflags);
310 static void set_intercept_indicators_io(struct kvm_vcpu *vcpu)
312 if (!(pending_irqs(vcpu) & IRQ_PEND_IO_MASK))
314 else if (psw_ioint_disabled(vcpu))
315 __set_cpuflag(vcpu, CPUSTAT_IO_INT);
317 vcpu->arch.sie_block->lctl |= LCTL_CR6;
320 static void set_intercept_indicators_ext(struct kvm_vcpu *vcpu)
322 if (!(pending_irqs(vcpu) & IRQ_PEND_EXT_MASK))
324 if (psw_extint_disabled(vcpu))
325 __set_cpuflag(vcpu, CPUSTAT_EXT_INT);
327 vcpu->arch.sie_block->lctl |= LCTL_CR0;
330 static void set_intercept_indicators_mchk(struct kvm_vcpu *vcpu)
332 if (!(pending_irqs(vcpu) & IRQ_PEND_MCHK_MASK))
334 if (psw_mchk_disabled(vcpu))
335 vcpu->arch.sie_block->ictl |= ICTL_LPSW;
337 vcpu->arch.sie_block->lctl |= LCTL_CR14;
340 static void set_intercept_indicators_stop(struct kvm_vcpu *vcpu)
342 if (kvm_s390_is_stop_irq_pending(vcpu))
343 __set_cpuflag(vcpu, CPUSTAT_STOP_INT);
346 /* Set interception request for non-deliverable interrupts */
347 static void set_intercept_indicators(struct kvm_vcpu *vcpu)
349 set_intercept_indicators_io(vcpu);
350 set_intercept_indicators_ext(vcpu);
351 set_intercept_indicators_mchk(vcpu);
352 set_intercept_indicators_stop(vcpu);
355 static int __must_check __deliver_cpu_timer(struct kvm_vcpu *vcpu)
357 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
360 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
363 rc = put_guest_lc(vcpu, EXT_IRQ_CPU_TIMER,
364 (u16 *)__LC_EXT_INT_CODE);
365 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
366 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
367 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
368 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
369 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
370 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
371 return rc ? -EFAULT : 0;
374 static int __must_check __deliver_ckc(struct kvm_vcpu *vcpu)
376 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
379 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
382 rc = put_guest_lc(vcpu, EXT_IRQ_CLK_COMP,
383 (u16 __user *)__LC_EXT_INT_CODE);
384 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
385 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
386 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
387 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
388 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
389 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
390 return rc ? -EFAULT : 0;
393 static int __must_check __deliver_pfault_init(struct kvm_vcpu *vcpu)
395 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
396 struct kvm_s390_ext_info ext;
399 spin_lock(&li->lock);
401 clear_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
402 li->irq.ext.ext_params2 = 0;
403 spin_unlock(&li->lock);
405 VCPU_EVENT(vcpu, 4, "deliver: pfault init token 0x%llx",
407 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
408 KVM_S390_INT_PFAULT_INIT,
411 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE, (u16 *) __LC_EXT_INT_CODE);
412 rc |= put_guest_lc(vcpu, PFAULT_INIT, (u16 *) __LC_EXT_CPU_ADDR);
413 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
414 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
415 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
416 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
417 rc |= put_guest_lc(vcpu, ext.ext_params2, (u64 *) __LC_EXT_PARAMS2);
418 return rc ? -EFAULT : 0;
421 static int __write_machine_check(struct kvm_vcpu *vcpu,
422 struct kvm_s390_mchk_info *mchk)
424 unsigned long ext_sa_addr;
426 freg_t fprs[NUM_FPRS];
430 mci.val = mchk->mcic;
431 /* take care of lazy register loading */
433 save_access_regs(vcpu->run->s.regs.acrs);
434 if (MACHINE_HAS_GS && vcpu->arch.gs_enabled)
435 save_gs_cb(current->thread.gs_cb);
437 /* Extended save area */
438 rc = read_guest_lc(vcpu, __LC_MCESAD, &ext_sa_addr,
439 sizeof(unsigned long));
440 /* Only bits 0 through 63-LC are used for address formation */
441 lc = ext_sa_addr & MCESA_LC_MASK;
442 if (test_kvm_facility(vcpu->kvm, 133)) {
446 ext_sa_addr &= ~0x3ffUL;
449 ext_sa_addr &= ~0x7ffUL;
452 ext_sa_addr &= ~0xfffUL;
459 ext_sa_addr &= ~0x3ffUL;
462 if (!rc && mci.vr && ext_sa_addr && test_kvm_facility(vcpu->kvm, 129)) {
463 if (write_guest_abs(vcpu, ext_sa_addr, vcpu->run->s.regs.vrs,
469 if (!rc && mci.gs && ext_sa_addr && test_kvm_facility(vcpu->kvm, 133)
470 && (lc == 11 || lc == 12)) {
471 if (write_guest_abs(vcpu, ext_sa_addr + 1024,
472 &vcpu->run->s.regs.gscb, 32))
478 /* General interruption information */
479 rc |= put_guest_lc(vcpu, 1, (u8 __user *) __LC_AR_MODE_ID);
480 rc |= write_guest_lc(vcpu, __LC_MCK_OLD_PSW,
481 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
482 rc |= read_guest_lc(vcpu, __LC_MCK_NEW_PSW,
483 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
484 rc |= put_guest_lc(vcpu, mci.val, (u64 __user *) __LC_MCCK_CODE);
486 /* Register-save areas */
487 if (MACHINE_HAS_VX) {
488 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
489 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA, fprs, 128);
491 rc |= write_guest_lc(vcpu, __LC_FPREGS_SAVE_AREA,
492 vcpu->run->s.regs.fprs, 128);
494 rc |= write_guest_lc(vcpu, __LC_GPREGS_SAVE_AREA,
495 vcpu->run->s.regs.gprs, 128);
496 rc |= put_guest_lc(vcpu, current->thread.fpu.fpc,
497 (u32 __user *) __LC_FP_CREG_SAVE_AREA);
498 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->todpr,
499 (u32 __user *) __LC_TOD_PROGREG_SAVE_AREA);
500 rc |= put_guest_lc(vcpu, kvm_s390_get_cpu_timer(vcpu),
501 (u64 __user *) __LC_CPU_TIMER_SAVE_AREA);
502 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->ckc >> 8,
503 (u64 __user *) __LC_CLOCK_COMP_SAVE_AREA);
504 rc |= write_guest_lc(vcpu, __LC_AREGS_SAVE_AREA,
505 &vcpu->run->s.regs.acrs, 64);
506 rc |= write_guest_lc(vcpu, __LC_CREGS_SAVE_AREA,
507 &vcpu->arch.sie_block->gcr, 128);
509 /* Extended interruption information */
510 rc |= put_guest_lc(vcpu, mchk->ext_damage_code,
511 (u32 __user *) __LC_EXT_DAMAGE_CODE);
512 rc |= put_guest_lc(vcpu, mchk->failing_storage_address,
513 (u64 __user *) __LC_MCCK_FAIL_STOR_ADDR);
514 rc |= write_guest_lc(vcpu, __LC_PSW_SAVE_AREA, &mchk->fixed_logout,
515 sizeof(mchk->fixed_logout));
516 return rc ? -EFAULT : 0;
519 static int __must_check __deliver_machine_check(struct kvm_vcpu *vcpu)
521 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
522 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
523 struct kvm_s390_mchk_info mchk = {};
527 spin_lock(&fi->lock);
528 spin_lock(&li->lock);
529 if (test_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs) ||
530 test_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs)) {
532 * If there was an exigent machine check pending, then any
533 * repressible machine checks that might have been pending
534 * are indicated along with it, so always clear bits for
535 * repressible and exigent interrupts
538 clear_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
539 clear_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
540 memset(&li->irq.mchk, 0, sizeof(mchk));
544 * We indicate floating repressible conditions along with
545 * other pending conditions. Channel Report Pending and Channel
546 * Subsystem damage are the only two and and are indicated by
547 * bits in mcic and masked in cr14.
549 if (test_and_clear_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
550 mchk.mcic |= fi->mchk.mcic;
551 mchk.cr14 |= fi->mchk.cr14;
552 memset(&fi->mchk, 0, sizeof(mchk));
555 spin_unlock(&li->lock);
556 spin_unlock(&fi->lock);
559 VCPU_EVENT(vcpu, 3, "deliver: machine check mcic 0x%llx",
561 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
563 mchk.cr14, mchk.mcic);
564 rc = __write_machine_check(vcpu, &mchk);
569 static int __must_check __deliver_restart(struct kvm_vcpu *vcpu)
571 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
574 VCPU_EVENT(vcpu, 3, "%s", "deliver: cpu restart");
575 vcpu->stat.deliver_restart_signal++;
576 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
578 rc = write_guest_lc(vcpu,
579 offsetof(struct lowcore, restart_old_psw),
580 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
581 rc |= read_guest_lc(vcpu, offsetof(struct lowcore, restart_psw),
582 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
583 clear_bit(IRQ_PEND_RESTART, &li->pending_irqs);
584 return rc ? -EFAULT : 0;
587 static int __must_check __deliver_set_prefix(struct kvm_vcpu *vcpu)
589 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
590 struct kvm_s390_prefix_info prefix;
592 spin_lock(&li->lock);
593 prefix = li->irq.prefix;
594 li->irq.prefix.address = 0;
595 clear_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
596 spin_unlock(&li->lock);
598 vcpu->stat.deliver_prefix_signal++;
599 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
600 KVM_S390_SIGP_SET_PREFIX,
603 kvm_s390_set_prefix(vcpu, prefix.address);
607 static int __must_check __deliver_emergency_signal(struct kvm_vcpu *vcpu)
609 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
613 spin_lock(&li->lock);
614 cpu_addr = find_first_bit(li->sigp_emerg_pending, KVM_MAX_VCPUS);
615 clear_bit(cpu_addr, li->sigp_emerg_pending);
616 if (bitmap_empty(li->sigp_emerg_pending, KVM_MAX_VCPUS))
617 clear_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
618 spin_unlock(&li->lock);
620 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp emerg");
621 vcpu->stat.deliver_emergency_signal++;
622 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
625 rc = put_guest_lc(vcpu, EXT_IRQ_EMERGENCY_SIG,
626 (u16 *)__LC_EXT_INT_CODE);
627 rc |= put_guest_lc(vcpu, cpu_addr, (u16 *)__LC_EXT_CPU_ADDR);
628 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
629 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
630 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
631 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
632 return rc ? -EFAULT : 0;
635 static int __must_check __deliver_external_call(struct kvm_vcpu *vcpu)
637 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
638 struct kvm_s390_extcall_info extcall;
641 spin_lock(&li->lock);
642 extcall = li->irq.extcall;
643 li->irq.extcall.code = 0;
644 clear_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
645 spin_unlock(&li->lock);
647 VCPU_EVENT(vcpu, 4, "%s", "deliver: sigp ext call");
648 vcpu->stat.deliver_external_call++;
649 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
650 KVM_S390_INT_EXTERNAL_CALL,
653 rc = put_guest_lc(vcpu, EXT_IRQ_EXTERNAL_CALL,
654 (u16 *)__LC_EXT_INT_CODE);
655 rc |= put_guest_lc(vcpu, extcall.code, (u16 *)__LC_EXT_CPU_ADDR);
656 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
657 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
658 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &vcpu->arch.sie_block->gpsw,
660 return rc ? -EFAULT : 0;
663 static int __must_check __deliver_prog(struct kvm_vcpu *vcpu)
665 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
666 struct kvm_s390_pgm_info pgm_info;
667 int rc = 0, nullifying = false;
670 spin_lock(&li->lock);
671 pgm_info = li->irq.pgm;
672 clear_bit(IRQ_PEND_PROG, &li->pending_irqs);
673 memset(&li->irq.pgm, 0, sizeof(pgm_info));
674 spin_unlock(&li->lock);
676 ilen = pgm_info.flags & KVM_S390_PGM_FLAGS_ILC_MASK;
677 VCPU_EVENT(vcpu, 3, "deliver: program irq code 0x%x, ilen:%d",
678 pgm_info.code, ilen);
679 vcpu->stat.deliver_program_int++;
680 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
683 switch (pgm_info.code & ~PGM_PER) {
684 case PGM_AFX_TRANSLATION:
685 case PGM_ASX_TRANSLATION:
686 case PGM_EX_TRANSLATION:
687 case PGM_LFX_TRANSLATION:
688 case PGM_LSTE_SEQUENCE:
689 case PGM_LSX_TRANSLATION:
690 case PGM_LX_TRANSLATION:
691 case PGM_PRIMARY_AUTHORITY:
692 case PGM_SECONDARY_AUTHORITY:
695 case PGM_SPACE_SWITCH:
696 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
697 (u64 *)__LC_TRANS_EXC_CODE);
699 case PGM_ALEN_TRANSLATION:
700 case PGM_ALE_SEQUENCE:
701 case PGM_ASTE_INSTANCE:
702 case PGM_ASTE_SEQUENCE:
703 case PGM_ASTE_VALIDITY:
704 case PGM_EXTENDED_AUTHORITY:
705 rc = put_guest_lc(vcpu, pgm_info.exc_access_id,
706 (u8 *)__LC_EXC_ACCESS_ID);
710 case PGM_PAGE_TRANSLATION:
711 case PGM_REGION_FIRST_TRANS:
712 case PGM_REGION_SECOND_TRANS:
713 case PGM_REGION_THIRD_TRANS:
714 case PGM_SEGMENT_TRANSLATION:
715 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
716 (u64 *)__LC_TRANS_EXC_CODE);
717 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
718 (u8 *)__LC_EXC_ACCESS_ID);
719 rc |= put_guest_lc(vcpu, pgm_info.op_access_id,
720 (u8 *)__LC_OP_ACCESS_ID);
724 rc = put_guest_lc(vcpu, pgm_info.mon_class_nr,
725 (u16 *)__LC_MON_CLASS_NR);
726 rc |= put_guest_lc(vcpu, pgm_info.mon_code,
727 (u64 *)__LC_MON_CODE);
729 case PGM_VECTOR_PROCESSING:
731 rc = put_guest_lc(vcpu, pgm_info.data_exc_code,
732 (u32 *)__LC_DATA_EXC_CODE);
735 rc = put_guest_lc(vcpu, pgm_info.trans_exc_code,
736 (u64 *)__LC_TRANS_EXC_CODE);
737 rc |= put_guest_lc(vcpu, pgm_info.exc_access_id,
738 (u8 *)__LC_EXC_ACCESS_ID);
741 case PGM_STACK_EMPTY:
742 case PGM_STACK_SPECIFICATION:
744 case PGM_STACK_OPERATION:
745 case PGM_TRACE_TABEL:
746 case PGM_CRYPTO_OPERATION:
751 if (pgm_info.code & PGM_PER) {
752 rc |= put_guest_lc(vcpu, pgm_info.per_code,
753 (u8 *) __LC_PER_CODE);
754 rc |= put_guest_lc(vcpu, pgm_info.per_atmid,
755 (u8 *)__LC_PER_ATMID);
756 rc |= put_guest_lc(vcpu, pgm_info.per_address,
757 (u64 *) __LC_PER_ADDRESS);
758 rc |= put_guest_lc(vcpu, pgm_info.per_access_id,
759 (u8 *) __LC_PER_ACCESS_ID);
762 if (nullifying && !(pgm_info.flags & KVM_S390_PGM_FLAGS_NO_REWIND))
763 kvm_s390_rewind_psw(vcpu, ilen);
765 /* bit 1+2 of the target are the ilc, so we can directly use ilen */
766 rc |= put_guest_lc(vcpu, ilen, (u16 *) __LC_PGM_ILC);
767 rc |= put_guest_lc(vcpu, vcpu->arch.sie_block->gbea,
768 (u64 *) __LC_LAST_BREAK);
769 rc |= put_guest_lc(vcpu, pgm_info.code,
770 (u16 *)__LC_PGM_INT_CODE);
771 rc |= write_guest_lc(vcpu, __LC_PGM_OLD_PSW,
772 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
773 rc |= read_guest_lc(vcpu, __LC_PGM_NEW_PSW,
774 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
775 return rc ? -EFAULT : 0;
778 static int __must_check __deliver_service(struct kvm_vcpu *vcpu)
780 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
781 struct kvm_s390_ext_info ext;
784 spin_lock(&fi->lock);
785 if (!(test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs))) {
786 spin_unlock(&fi->lock);
789 ext = fi->srv_signal;
790 memset(&fi->srv_signal, 0, sizeof(ext));
791 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
792 spin_unlock(&fi->lock);
794 VCPU_EVENT(vcpu, 4, "deliver: sclp parameter 0x%x",
796 vcpu->stat.deliver_service_signal++;
797 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, KVM_S390_INT_SERVICE,
800 rc = put_guest_lc(vcpu, EXT_IRQ_SERVICE_SIG, (u16 *)__LC_EXT_INT_CODE);
801 rc |= put_guest_lc(vcpu, 0, (u16 *)__LC_EXT_CPU_ADDR);
802 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
803 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
804 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
805 &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
806 rc |= put_guest_lc(vcpu, ext.ext_params,
807 (u32 *)__LC_EXT_PARAMS);
809 return rc ? -EFAULT : 0;
812 static int __must_check __deliver_pfault_done(struct kvm_vcpu *vcpu)
814 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
815 struct kvm_s390_interrupt_info *inti;
818 spin_lock(&fi->lock);
819 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_PFAULT],
820 struct kvm_s390_interrupt_info,
823 list_del(&inti->list);
824 fi->counters[FIRQ_CNTR_PFAULT] -= 1;
826 if (list_empty(&fi->lists[FIRQ_LIST_PFAULT]))
827 clear_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
828 spin_unlock(&fi->lock);
831 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
832 KVM_S390_INT_PFAULT_DONE, 0,
833 inti->ext.ext_params2);
834 VCPU_EVENT(vcpu, 4, "deliver: pfault done token 0x%llx",
835 inti->ext.ext_params2);
837 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
838 (u16 *)__LC_EXT_INT_CODE);
839 rc |= put_guest_lc(vcpu, PFAULT_DONE,
840 (u16 *)__LC_EXT_CPU_ADDR);
841 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
842 &vcpu->arch.sie_block->gpsw,
844 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
845 &vcpu->arch.sie_block->gpsw,
847 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
848 (u64 *)__LC_EXT_PARAMS2);
851 return rc ? -EFAULT : 0;
854 static int __must_check __deliver_virtio(struct kvm_vcpu *vcpu)
856 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
857 struct kvm_s390_interrupt_info *inti;
860 spin_lock(&fi->lock);
861 inti = list_first_entry_or_null(&fi->lists[FIRQ_LIST_VIRTIO],
862 struct kvm_s390_interrupt_info,
866 "deliver: virtio parm: 0x%x,parm64: 0x%llx",
867 inti->ext.ext_params, inti->ext.ext_params2);
868 vcpu->stat.deliver_virtio_interrupt++;
869 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
871 inti->ext.ext_params,
872 inti->ext.ext_params2);
873 list_del(&inti->list);
874 fi->counters[FIRQ_CNTR_VIRTIO] -= 1;
876 if (list_empty(&fi->lists[FIRQ_LIST_VIRTIO]))
877 clear_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
878 spin_unlock(&fi->lock);
881 rc = put_guest_lc(vcpu, EXT_IRQ_CP_SERVICE,
882 (u16 *)__LC_EXT_INT_CODE);
883 rc |= put_guest_lc(vcpu, VIRTIO_PARAM,
884 (u16 *)__LC_EXT_CPU_ADDR);
885 rc |= write_guest_lc(vcpu, __LC_EXT_OLD_PSW,
886 &vcpu->arch.sie_block->gpsw,
888 rc |= read_guest_lc(vcpu, __LC_EXT_NEW_PSW,
889 &vcpu->arch.sie_block->gpsw,
891 rc |= put_guest_lc(vcpu, inti->ext.ext_params,
892 (u32 *)__LC_EXT_PARAMS);
893 rc |= put_guest_lc(vcpu, inti->ext.ext_params2,
894 (u64 *)__LC_EXT_PARAMS2);
897 return rc ? -EFAULT : 0;
900 static int __must_check __deliver_io(struct kvm_vcpu *vcpu,
901 unsigned long irq_type)
903 struct list_head *isc_list;
904 struct kvm_s390_float_interrupt *fi;
905 struct kvm_s390_interrupt_info *inti = NULL;
908 fi = &vcpu->kvm->arch.float_int;
910 spin_lock(&fi->lock);
911 isc_list = &fi->lists[irq_type - IRQ_PEND_IO_ISC_0];
912 inti = list_first_entry_or_null(isc_list,
913 struct kvm_s390_interrupt_info,
916 if (inti->type & KVM_S390_INT_IO_AI_MASK)
917 VCPU_EVENT(vcpu, 4, "%s", "deliver: I/O (AI)");
919 VCPU_EVENT(vcpu, 4, "deliver: I/O %x ss %x schid %04x",
920 inti->io.subchannel_id >> 8,
921 inti->io.subchannel_id >> 1 & 0x3,
922 inti->io.subchannel_nr);
924 vcpu->stat.deliver_io_int++;
925 trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id,
927 ((__u32)inti->io.subchannel_id << 16) |
928 inti->io.subchannel_nr,
929 ((__u64)inti->io.io_int_parm << 32) |
930 inti->io.io_int_word);
931 list_del(&inti->list);
932 fi->counters[FIRQ_CNTR_IO] -= 1;
934 if (list_empty(isc_list))
935 clear_bit(irq_type, &fi->pending_irqs);
936 spin_unlock(&fi->lock);
939 rc = put_guest_lc(vcpu, inti->io.subchannel_id,
940 (u16 *)__LC_SUBCHANNEL_ID);
941 rc |= put_guest_lc(vcpu, inti->io.subchannel_nr,
942 (u16 *)__LC_SUBCHANNEL_NR);
943 rc |= put_guest_lc(vcpu, inti->io.io_int_parm,
944 (u32 *)__LC_IO_INT_PARM);
945 rc |= put_guest_lc(vcpu, inti->io.io_int_word,
946 (u32 *)__LC_IO_INT_WORD);
947 rc |= write_guest_lc(vcpu, __LC_IO_OLD_PSW,
948 &vcpu->arch.sie_block->gpsw,
950 rc |= read_guest_lc(vcpu, __LC_IO_NEW_PSW,
951 &vcpu->arch.sie_block->gpsw,
956 return rc ? -EFAULT : 0;
959 typedef int (*deliver_irq_t)(struct kvm_vcpu *vcpu);
961 static const deliver_irq_t deliver_irq_funcs[] = {
962 [IRQ_PEND_MCHK_EX] = __deliver_machine_check,
963 [IRQ_PEND_MCHK_REP] = __deliver_machine_check,
964 [IRQ_PEND_PROG] = __deliver_prog,
965 [IRQ_PEND_EXT_EMERGENCY] = __deliver_emergency_signal,
966 [IRQ_PEND_EXT_EXTERNAL] = __deliver_external_call,
967 [IRQ_PEND_EXT_CLOCK_COMP] = __deliver_ckc,
968 [IRQ_PEND_EXT_CPU_TIMER] = __deliver_cpu_timer,
969 [IRQ_PEND_RESTART] = __deliver_restart,
970 [IRQ_PEND_SET_PREFIX] = __deliver_set_prefix,
971 [IRQ_PEND_PFAULT_INIT] = __deliver_pfault_init,
972 [IRQ_PEND_EXT_SERVICE] = __deliver_service,
973 [IRQ_PEND_PFAULT_DONE] = __deliver_pfault_done,
974 [IRQ_PEND_VIRTIO] = __deliver_virtio,
977 /* Check whether an external call is pending (deliverable or not) */
978 int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
980 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
982 if (!sclp.has_sigpif)
983 return test_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs);
985 return sca_ext_call_pending(vcpu, NULL);
988 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
990 if (deliverable_irqs(vcpu))
993 if (kvm_cpu_has_pending_timer(vcpu))
996 /* external call pending and deliverable */
997 if (kvm_s390_ext_call_pending(vcpu) &&
998 !psw_extint_disabled(vcpu) &&
999 (vcpu->arch.sie_block->gcr[0] & 0x2000ul))
1002 if (!exclude_stop && kvm_s390_is_stop_irq_pending(vcpu))
1007 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
1009 return ckc_irq_pending(vcpu) || cpu_timer_irq_pending(vcpu);
1012 static u64 __calculate_sltime(struct kvm_vcpu *vcpu)
1014 const u64 now = kvm_s390_get_tod_clock_fast(vcpu->kvm);
1015 const u64 ckc = vcpu->arch.sie_block->ckc;
1016 u64 cputm, sltime = 0;
1018 if (ckc_interrupts_enabled(vcpu)) {
1019 if (vcpu->arch.sie_block->gcr[0] & 0x0020000000000000ul) {
1020 if ((s64)now < (s64)ckc)
1021 sltime = tod_to_ns((s64)ckc - (s64)now);
1022 } else if (now < ckc) {
1023 sltime = tod_to_ns(ckc - now);
1025 /* already expired */
1028 if (cpu_timer_interrupts_enabled(vcpu)) {
1029 cputm = kvm_s390_get_cpu_timer(vcpu);
1030 /* already expired? */
1033 return min(sltime, tod_to_ns(cputm));
1035 } else if (cpu_timer_interrupts_enabled(vcpu)) {
1036 sltime = kvm_s390_get_cpu_timer(vcpu);
1037 /* already expired? */
1044 int kvm_s390_handle_wait(struct kvm_vcpu *vcpu)
1048 vcpu->stat.exit_wait_state++;
1051 if (kvm_arch_vcpu_runnable(vcpu))
1054 if (psw_interrupts_disabled(vcpu)) {
1055 VCPU_EVENT(vcpu, 3, "%s", "disabled wait");
1056 return -EOPNOTSUPP; /* disabled wait */
1059 if (!ckc_interrupts_enabled(vcpu) &&
1060 !cpu_timer_interrupts_enabled(vcpu)) {
1061 VCPU_EVENT(vcpu, 3, "%s", "enabled wait w/o timer");
1062 __set_cpu_idle(vcpu);
1066 sltime = __calculate_sltime(vcpu);
1070 __set_cpu_idle(vcpu);
1071 hrtimer_start(&vcpu->arch.ckc_timer, sltime, HRTIMER_MODE_REL);
1072 VCPU_EVENT(vcpu, 4, "enabled wait: %llu ns", sltime);
1074 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1075 kvm_vcpu_block(vcpu);
1076 __unset_cpu_idle(vcpu);
1077 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1079 hrtimer_cancel(&vcpu->arch.ckc_timer);
1083 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
1086 * We cannot move this into the if, as the CPU might be already
1087 * in kvm_vcpu_block without having the waitqueue set (polling)
1089 vcpu->valid_wakeup = true;
1090 if (swait_active(&vcpu->wq)) {
1092 * The vcpu gave up the cpu voluntarily, mark it as a good
1095 vcpu->preempted = true;
1096 swake_up(&vcpu->wq);
1097 vcpu->stat.halt_wakeup++;
1100 * The VCPU might not be sleeping but is executing the VSIE. Let's
1101 * kick it, so it leaves the SIE to process the request.
1103 kvm_s390_vsie_kick(vcpu);
1106 enum hrtimer_restart kvm_s390_idle_wakeup(struct hrtimer *timer)
1108 struct kvm_vcpu *vcpu;
1111 vcpu = container_of(timer, struct kvm_vcpu, arch.ckc_timer);
1112 sltime = __calculate_sltime(vcpu);
1115 * If the monotonic clock runs faster than the tod clock we might be
1116 * woken up too early and have to go back to sleep to avoid deadlocks.
1118 if (sltime && hrtimer_forward_now(timer, ns_to_ktime(sltime)))
1119 return HRTIMER_RESTART;
1120 kvm_s390_vcpu_wakeup(vcpu);
1121 return HRTIMER_NORESTART;
1124 void kvm_s390_clear_local_irqs(struct kvm_vcpu *vcpu)
1126 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1128 spin_lock(&li->lock);
1129 li->pending_irqs = 0;
1130 bitmap_zero(li->sigp_emerg_pending, KVM_MAX_VCPUS);
1131 memset(&li->irq, 0, sizeof(li->irq));
1132 spin_unlock(&li->lock);
1134 sca_clear_ext_call(vcpu);
1137 int __must_check kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
1139 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1142 unsigned long irq_type;
1145 __reset_intercept_indicators(vcpu);
1147 /* pending ckc conditions might have been invalidated */
1148 clear_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1149 if (ckc_irq_pending(vcpu))
1150 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1152 /* pending cpu timer conditions might have been invalidated */
1153 clear_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1154 if (cpu_timer_irq_pending(vcpu))
1155 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1157 while ((irqs = deliverable_irqs(vcpu)) && !rc) {
1158 /* bits are in the order of interrupt priority */
1159 irq_type = find_first_bit(&irqs, IRQ_PEND_COUNT);
1160 if (is_ioirq(irq_type)) {
1161 rc = __deliver_io(vcpu, irq_type);
1163 func = deliver_irq_funcs[irq_type];
1165 WARN_ON_ONCE(func == NULL);
1166 clear_bit(irq_type, &li->pending_irqs);
1173 set_intercept_indicators(vcpu);
1178 static int __inject_prog(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1180 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1182 VCPU_EVENT(vcpu, 3, "inject: program irq code 0x%x", irq->u.pgm.code);
1183 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_PROGRAM_INT,
1184 irq->u.pgm.code, 0);
1186 if (!(irq->u.pgm.flags & KVM_S390_PGM_FLAGS_ILC_VALID)) {
1187 /* auto detection if no valid ILC was given */
1188 irq->u.pgm.flags &= ~KVM_S390_PGM_FLAGS_ILC_MASK;
1189 irq->u.pgm.flags |= kvm_s390_get_ilen(vcpu);
1190 irq->u.pgm.flags |= KVM_S390_PGM_FLAGS_ILC_VALID;
1193 if (irq->u.pgm.code == PGM_PER) {
1194 li->irq.pgm.code |= PGM_PER;
1195 li->irq.pgm.flags = irq->u.pgm.flags;
1196 /* only modify PER related information */
1197 li->irq.pgm.per_address = irq->u.pgm.per_address;
1198 li->irq.pgm.per_code = irq->u.pgm.per_code;
1199 li->irq.pgm.per_atmid = irq->u.pgm.per_atmid;
1200 li->irq.pgm.per_access_id = irq->u.pgm.per_access_id;
1201 } else if (!(irq->u.pgm.code & PGM_PER)) {
1202 li->irq.pgm.code = (li->irq.pgm.code & PGM_PER) |
1204 li->irq.pgm.flags = irq->u.pgm.flags;
1205 /* only modify non-PER information */
1206 li->irq.pgm.trans_exc_code = irq->u.pgm.trans_exc_code;
1207 li->irq.pgm.mon_code = irq->u.pgm.mon_code;
1208 li->irq.pgm.data_exc_code = irq->u.pgm.data_exc_code;
1209 li->irq.pgm.mon_class_nr = irq->u.pgm.mon_class_nr;
1210 li->irq.pgm.exc_access_id = irq->u.pgm.exc_access_id;
1211 li->irq.pgm.op_access_id = irq->u.pgm.op_access_id;
1213 li->irq.pgm = irq->u.pgm;
1215 set_bit(IRQ_PEND_PROG, &li->pending_irqs);
1219 static int __inject_pfault_init(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1221 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1223 VCPU_EVENT(vcpu, 4, "inject: pfault init parameter block at 0x%llx",
1224 irq->u.ext.ext_params2);
1225 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_PFAULT_INIT,
1226 irq->u.ext.ext_params,
1227 irq->u.ext.ext_params2);
1229 li->irq.ext = irq->u.ext;
1230 set_bit(IRQ_PEND_PFAULT_INIT, &li->pending_irqs);
1231 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1235 static int __inject_extcall(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1237 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1238 struct kvm_s390_extcall_info *extcall = &li->irq.extcall;
1239 uint16_t src_id = irq->u.extcall.code;
1241 VCPU_EVENT(vcpu, 4, "inject: external call source-cpu:%u",
1243 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EXTERNAL_CALL,
1246 /* sending vcpu invalid */
1247 if (kvm_get_vcpu_by_id(vcpu->kvm, src_id) == NULL)
1250 if (sclp.has_sigpif)
1251 return sca_inject_ext_call(vcpu, src_id);
1253 if (test_and_set_bit(IRQ_PEND_EXT_EXTERNAL, &li->pending_irqs))
1255 *extcall = irq->u.extcall;
1256 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1260 static int __inject_set_prefix(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1262 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1263 struct kvm_s390_prefix_info *prefix = &li->irq.prefix;
1265 VCPU_EVENT(vcpu, 3, "inject: set prefix to %x",
1266 irq->u.prefix.address);
1267 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_SET_PREFIX,
1268 irq->u.prefix.address, 0);
1270 if (!is_vcpu_stopped(vcpu))
1273 *prefix = irq->u.prefix;
1274 set_bit(IRQ_PEND_SET_PREFIX, &li->pending_irqs);
1278 #define KVM_S390_STOP_SUPP_FLAGS (KVM_S390_STOP_FLAG_STORE_STATUS)
1279 static int __inject_sigp_stop(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1281 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1282 struct kvm_s390_stop_info *stop = &li->irq.stop;
1285 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_SIGP_STOP, 0, 0);
1287 if (irq->u.stop.flags & ~KVM_S390_STOP_SUPP_FLAGS)
1290 if (is_vcpu_stopped(vcpu)) {
1291 if (irq->u.stop.flags & KVM_S390_STOP_FLAG_STORE_STATUS)
1292 rc = kvm_s390_store_status_unloaded(vcpu,
1293 KVM_S390_STORE_STATUS_NOADDR);
1297 if (test_and_set_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs))
1299 stop->flags = irq->u.stop.flags;
1300 __set_cpuflag(vcpu, CPUSTAT_STOP_INT);
1304 static int __inject_sigp_restart(struct kvm_vcpu *vcpu,
1305 struct kvm_s390_irq *irq)
1307 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1309 VCPU_EVENT(vcpu, 3, "%s", "inject: restart int");
1310 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_RESTART, 0, 0);
1312 set_bit(IRQ_PEND_RESTART, &li->pending_irqs);
1316 static int __inject_sigp_emergency(struct kvm_vcpu *vcpu,
1317 struct kvm_s390_irq *irq)
1319 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1321 VCPU_EVENT(vcpu, 4, "inject: emergency from cpu %u",
1323 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_EMERGENCY,
1324 irq->u.emerg.code, 0);
1326 /* sending vcpu invalid */
1327 if (kvm_get_vcpu_by_id(vcpu->kvm, irq->u.emerg.code) == NULL)
1330 set_bit(irq->u.emerg.code, li->sigp_emerg_pending);
1331 set_bit(IRQ_PEND_EXT_EMERGENCY, &li->pending_irqs);
1332 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1336 static int __inject_mchk(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1338 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1339 struct kvm_s390_mchk_info *mchk = &li->irq.mchk;
1341 VCPU_EVENT(vcpu, 3, "inject: machine check mcic 0x%llx",
1343 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_MCHK, 0,
1347 * Because repressible machine checks can be indicated along with
1348 * exigent machine checks (PoP, Chapter 11, Interruption action)
1349 * we need to combine cr14, mcic and external damage code.
1350 * Failing storage address and the logout area should not be or'ed
1351 * together, we just indicate the last occurrence of the corresponding
1354 mchk->cr14 |= irq->u.mchk.cr14;
1355 mchk->mcic |= irq->u.mchk.mcic;
1356 mchk->ext_damage_code |= irq->u.mchk.ext_damage_code;
1357 mchk->failing_storage_address = irq->u.mchk.failing_storage_address;
1358 memcpy(&mchk->fixed_logout, &irq->u.mchk.fixed_logout,
1359 sizeof(mchk->fixed_logout));
1360 if (mchk->mcic & MCHK_EX_MASK)
1361 set_bit(IRQ_PEND_MCHK_EX, &li->pending_irqs);
1362 else if (mchk->mcic & MCHK_REP_MASK)
1363 set_bit(IRQ_PEND_MCHK_REP, &li->pending_irqs);
1367 static int __inject_ckc(struct kvm_vcpu *vcpu)
1369 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1371 VCPU_EVENT(vcpu, 3, "%s", "inject: clock comparator external");
1372 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CLOCK_COMP,
1375 set_bit(IRQ_PEND_EXT_CLOCK_COMP, &li->pending_irqs);
1376 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1380 static int __inject_cpu_timer(struct kvm_vcpu *vcpu)
1382 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1384 VCPU_EVENT(vcpu, 3, "%s", "inject: cpu timer external");
1385 trace_kvm_s390_inject_vcpu(vcpu->vcpu_id, KVM_S390_INT_CPU_TIMER,
1388 set_bit(IRQ_PEND_EXT_CPU_TIMER, &li->pending_irqs);
1389 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1393 static struct kvm_s390_interrupt_info *get_io_int(struct kvm *kvm,
1396 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1397 struct list_head *isc_list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1398 struct kvm_s390_interrupt_info *iter;
1399 u16 id = (schid & 0xffff0000U) >> 16;
1400 u16 nr = schid & 0x0000ffffU;
1402 spin_lock(&fi->lock);
1403 list_for_each_entry(iter, isc_list, list) {
1404 if (schid && (id != iter->io.subchannel_id ||
1405 nr != iter->io.subchannel_nr))
1407 /* found an appropriate entry */
1408 list_del_init(&iter->list);
1409 fi->counters[FIRQ_CNTR_IO] -= 1;
1410 if (list_empty(isc_list))
1411 clear_bit(IRQ_PEND_IO_ISC_0 + isc, &fi->pending_irqs);
1412 spin_unlock(&fi->lock);
1415 spin_unlock(&fi->lock);
1420 * Dequeue and return an I/O interrupt matching any of the interruption
1421 * subclasses as designated by the isc mask in cr6 and the schid (if != 0).
1423 struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
1424 u64 isc_mask, u32 schid)
1426 struct kvm_s390_interrupt_info *inti = NULL;
1429 for (isc = 0; isc <= MAX_ISC && !inti; isc++) {
1430 if (isc_mask & isc_to_isc_bits(isc))
1431 inti = get_io_int(kvm, isc, schid);
1436 #define SCCB_MASK 0xFFFFFFF8
1437 #define SCCB_EVENT_PENDING 0x3
1439 static int __inject_service(struct kvm *kvm,
1440 struct kvm_s390_interrupt_info *inti)
1442 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1444 spin_lock(&fi->lock);
1445 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_EVENT_PENDING;
1447 * Early versions of the QEMU s390 bios will inject several
1448 * service interrupts after another without handling a
1449 * condition code indicating busy.
1450 * We will silently ignore those superfluous sccb values.
1451 * A future version of QEMU will take care of serialization
1454 if (fi->srv_signal.ext_params & SCCB_MASK)
1456 fi->srv_signal.ext_params |= inti->ext.ext_params & SCCB_MASK;
1457 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
1459 spin_unlock(&fi->lock);
1464 static int __inject_virtio(struct kvm *kvm,
1465 struct kvm_s390_interrupt_info *inti)
1467 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1469 spin_lock(&fi->lock);
1470 if (fi->counters[FIRQ_CNTR_VIRTIO] >= KVM_S390_MAX_VIRTIO_IRQS) {
1471 spin_unlock(&fi->lock);
1474 fi->counters[FIRQ_CNTR_VIRTIO] += 1;
1475 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_VIRTIO]);
1476 set_bit(IRQ_PEND_VIRTIO, &fi->pending_irqs);
1477 spin_unlock(&fi->lock);
1481 static int __inject_pfault_done(struct kvm *kvm,
1482 struct kvm_s390_interrupt_info *inti)
1484 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1486 spin_lock(&fi->lock);
1487 if (fi->counters[FIRQ_CNTR_PFAULT] >=
1488 (ASYNC_PF_PER_VCPU * KVM_MAX_VCPUS)) {
1489 spin_unlock(&fi->lock);
1492 fi->counters[FIRQ_CNTR_PFAULT] += 1;
1493 list_add_tail(&inti->list, &fi->lists[FIRQ_LIST_PFAULT]);
1494 set_bit(IRQ_PEND_PFAULT_DONE, &fi->pending_irqs);
1495 spin_unlock(&fi->lock);
1499 #define CR_PENDING_SUBCLASS 28
1500 static int __inject_float_mchk(struct kvm *kvm,
1501 struct kvm_s390_interrupt_info *inti)
1503 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1505 spin_lock(&fi->lock);
1506 fi->mchk.cr14 |= inti->mchk.cr14 & (1UL << CR_PENDING_SUBCLASS);
1507 fi->mchk.mcic |= inti->mchk.mcic;
1508 set_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs);
1509 spin_unlock(&fi->lock);
1514 static int __inject_io(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1516 struct kvm_s390_float_interrupt *fi;
1517 struct list_head *list;
1520 fi = &kvm->arch.float_int;
1521 spin_lock(&fi->lock);
1522 if (fi->counters[FIRQ_CNTR_IO] >= KVM_S390_MAX_FLOAT_IRQS) {
1523 spin_unlock(&fi->lock);
1526 fi->counters[FIRQ_CNTR_IO] += 1;
1528 if (inti->type & KVM_S390_INT_IO_AI_MASK)
1529 VM_EVENT(kvm, 4, "%s", "inject: I/O (AI)");
1531 VM_EVENT(kvm, 4, "inject: I/O %x ss %x schid %04x",
1532 inti->io.subchannel_id >> 8,
1533 inti->io.subchannel_id >> 1 & 0x3,
1534 inti->io.subchannel_nr);
1535 isc = int_word_to_isc(inti->io.io_int_word);
1536 list = &fi->lists[FIRQ_LIST_IO_ISC_0 + isc];
1537 list_add_tail(&inti->list, list);
1538 set_bit(IRQ_PEND_IO_ISC_0 + isc, &fi->pending_irqs);
1539 spin_unlock(&fi->lock);
1544 * Find a destination VCPU for a floating irq and kick it.
1546 static void __floating_irq_kick(struct kvm *kvm, u64 type)
1548 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1549 struct kvm_s390_local_interrupt *li;
1550 struct kvm_vcpu *dst_vcpu;
1551 int sigcpu, online_vcpus, nr_tries = 0;
1553 online_vcpus = atomic_read(&kvm->online_vcpus);
1557 /* find idle VCPUs first, then round robin */
1558 sigcpu = find_first_bit(fi->idle_mask, online_vcpus);
1559 if (sigcpu == online_vcpus) {
1561 sigcpu = fi->next_rr_cpu;
1562 fi->next_rr_cpu = (fi->next_rr_cpu + 1) % online_vcpus;
1563 /* avoid endless loops if all vcpus are stopped */
1564 if (nr_tries++ >= online_vcpus)
1566 } while (is_vcpu_stopped(kvm_get_vcpu(kvm, sigcpu)));
1568 dst_vcpu = kvm_get_vcpu(kvm, sigcpu);
1570 /* make the VCPU drop out of the SIE, or wake it up if sleeping */
1571 li = &dst_vcpu->arch.local_int;
1572 spin_lock(&li->lock);
1575 atomic_or(CPUSTAT_STOP_INT, li->cpuflags);
1577 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1578 atomic_or(CPUSTAT_IO_INT, li->cpuflags);
1581 atomic_or(CPUSTAT_EXT_INT, li->cpuflags);
1584 spin_unlock(&li->lock);
1585 kvm_s390_vcpu_wakeup(dst_vcpu);
1588 static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
1590 u64 type = READ_ONCE(inti->type);
1595 rc = __inject_float_mchk(kvm, inti);
1597 case KVM_S390_INT_VIRTIO:
1598 rc = __inject_virtio(kvm, inti);
1600 case KVM_S390_INT_SERVICE:
1601 rc = __inject_service(kvm, inti);
1603 case KVM_S390_INT_PFAULT_DONE:
1604 rc = __inject_pfault_done(kvm, inti);
1606 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1607 rc = __inject_io(kvm, inti);
1615 __floating_irq_kick(kvm, type);
1619 int kvm_s390_inject_vm(struct kvm *kvm,
1620 struct kvm_s390_interrupt *s390int)
1622 struct kvm_s390_interrupt_info *inti;
1625 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
1629 inti->type = s390int->type;
1630 switch (inti->type) {
1631 case KVM_S390_INT_VIRTIO:
1632 VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
1633 s390int->parm, s390int->parm64);
1634 inti->ext.ext_params = s390int->parm;
1635 inti->ext.ext_params2 = s390int->parm64;
1637 case KVM_S390_INT_SERVICE:
1638 VM_EVENT(kvm, 4, "inject: sclp parm:%x", s390int->parm);
1639 inti->ext.ext_params = s390int->parm;
1641 case KVM_S390_INT_PFAULT_DONE:
1642 inti->ext.ext_params2 = s390int->parm64;
1645 VM_EVENT(kvm, 3, "inject: machine check mcic 0x%llx",
1647 inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
1648 inti->mchk.mcic = s390int->parm64;
1650 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1651 inti->io.subchannel_id = s390int->parm >> 16;
1652 inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
1653 inti->io.io_int_parm = s390int->parm64 >> 32;
1654 inti->io.io_int_word = s390int->parm64 & 0x00000000ffffffffull;
1660 trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
1663 rc = __inject_vm(kvm, inti);
1669 int kvm_s390_reinject_io_int(struct kvm *kvm,
1670 struct kvm_s390_interrupt_info *inti)
1672 return __inject_vm(kvm, inti);
1675 int s390int_to_s390irq(struct kvm_s390_interrupt *s390int,
1676 struct kvm_s390_irq *irq)
1678 irq->type = s390int->type;
1679 switch (irq->type) {
1680 case KVM_S390_PROGRAM_INT:
1681 if (s390int->parm & 0xffff0000)
1683 irq->u.pgm.code = s390int->parm;
1685 case KVM_S390_SIGP_SET_PREFIX:
1686 irq->u.prefix.address = s390int->parm;
1688 case KVM_S390_SIGP_STOP:
1689 irq->u.stop.flags = s390int->parm;
1691 case KVM_S390_INT_EXTERNAL_CALL:
1692 if (s390int->parm & 0xffff0000)
1694 irq->u.extcall.code = s390int->parm;
1696 case KVM_S390_INT_EMERGENCY:
1697 if (s390int->parm & 0xffff0000)
1699 irq->u.emerg.code = s390int->parm;
1702 irq->u.mchk.mcic = s390int->parm64;
1704 case KVM_S390_INT_PFAULT_INIT:
1705 irq->u.ext.ext_params = s390int->parm;
1706 irq->u.ext.ext_params2 = s390int->parm64;
1708 case KVM_S390_RESTART:
1709 case KVM_S390_INT_CLOCK_COMP:
1710 case KVM_S390_INT_CPU_TIMER:
1718 int kvm_s390_is_stop_irq_pending(struct kvm_vcpu *vcpu)
1720 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1722 return test_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1725 void kvm_s390_clear_stop_irq(struct kvm_vcpu *vcpu)
1727 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1729 spin_lock(&li->lock);
1730 li->irq.stop.flags = 0;
1731 clear_bit(IRQ_PEND_SIGP_STOP, &li->pending_irqs);
1732 spin_unlock(&li->lock);
1735 static int do_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1739 switch (irq->type) {
1740 case KVM_S390_PROGRAM_INT:
1741 rc = __inject_prog(vcpu, irq);
1743 case KVM_S390_SIGP_SET_PREFIX:
1744 rc = __inject_set_prefix(vcpu, irq);
1746 case KVM_S390_SIGP_STOP:
1747 rc = __inject_sigp_stop(vcpu, irq);
1749 case KVM_S390_RESTART:
1750 rc = __inject_sigp_restart(vcpu, irq);
1752 case KVM_S390_INT_CLOCK_COMP:
1753 rc = __inject_ckc(vcpu);
1755 case KVM_S390_INT_CPU_TIMER:
1756 rc = __inject_cpu_timer(vcpu);
1758 case KVM_S390_INT_EXTERNAL_CALL:
1759 rc = __inject_extcall(vcpu, irq);
1761 case KVM_S390_INT_EMERGENCY:
1762 rc = __inject_sigp_emergency(vcpu, irq);
1765 rc = __inject_mchk(vcpu, irq);
1767 case KVM_S390_INT_PFAULT_INIT:
1768 rc = __inject_pfault_init(vcpu, irq);
1770 case KVM_S390_INT_VIRTIO:
1771 case KVM_S390_INT_SERVICE:
1772 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1780 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu, struct kvm_s390_irq *irq)
1782 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
1785 spin_lock(&li->lock);
1786 rc = do_inject_vcpu(vcpu, irq);
1787 spin_unlock(&li->lock);
1789 kvm_s390_vcpu_wakeup(vcpu);
1793 static inline void clear_irq_list(struct list_head *_list)
1795 struct kvm_s390_interrupt_info *inti, *n;
1797 list_for_each_entry_safe(inti, n, _list, list) {
1798 list_del(&inti->list);
1803 static void inti_to_irq(struct kvm_s390_interrupt_info *inti,
1804 struct kvm_s390_irq *irq)
1806 irq->type = inti->type;
1807 switch (inti->type) {
1808 case KVM_S390_INT_PFAULT_INIT:
1809 case KVM_S390_INT_PFAULT_DONE:
1810 case KVM_S390_INT_VIRTIO:
1811 irq->u.ext = inti->ext;
1813 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1814 irq->u.io = inti->io;
1819 void kvm_s390_clear_float_irqs(struct kvm *kvm)
1821 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1824 spin_lock(&fi->lock);
1825 fi->pending_irqs = 0;
1826 memset(&fi->srv_signal, 0, sizeof(fi->srv_signal));
1827 memset(&fi->mchk, 0, sizeof(fi->mchk));
1828 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1829 clear_irq_list(&fi->lists[i]);
1830 for (i = 0; i < FIRQ_MAX_COUNT; i++)
1831 fi->counters[i] = 0;
1832 spin_unlock(&fi->lock);
1835 static int get_all_floating_irqs(struct kvm *kvm, u8 __user *usrbuf, u64 len)
1837 struct kvm_s390_interrupt_info *inti;
1838 struct kvm_s390_float_interrupt *fi;
1839 struct kvm_s390_irq *buf;
1840 struct kvm_s390_irq *irq;
1846 if (len > KVM_S390_FLIC_MAX_BUFFER || len == 0)
1850 * We are already using -ENOMEM to signal
1851 * userspace it may retry with a bigger buffer,
1852 * so we need to use something else for this case
1858 max_irqs = len / sizeof(struct kvm_s390_irq);
1860 fi = &kvm->arch.float_int;
1861 spin_lock(&fi->lock);
1862 for (i = 0; i < FIRQ_LIST_COUNT; i++) {
1863 list_for_each_entry(inti, &fi->lists[i], list) {
1864 if (n == max_irqs) {
1865 /* signal userspace to try again */
1869 inti_to_irq(inti, &buf[n]);
1873 if (test_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs)) {
1874 if (n == max_irqs) {
1875 /* signal userspace to try again */
1879 irq = (struct kvm_s390_irq *) &buf[n];
1880 irq->type = KVM_S390_INT_SERVICE;
1881 irq->u.ext = fi->srv_signal;
1884 if (test_bit(IRQ_PEND_MCHK_REP, &fi->pending_irqs)) {
1885 if (n == max_irqs) {
1886 /* signal userspace to try again */
1890 irq = (struct kvm_s390_irq *) &buf[n];
1891 irq->type = KVM_S390_MCHK;
1892 irq->u.mchk = fi->mchk;
1897 spin_unlock(&fi->lock);
1898 if (!ret && n > 0) {
1899 if (copy_to_user(usrbuf, buf, sizeof(struct kvm_s390_irq) * n))
1904 return ret < 0 ? ret : n;
1907 static int flic_ais_mode_get_all(struct kvm *kvm, struct kvm_device_attr *attr)
1909 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
1910 struct kvm_s390_ais_all ais;
1912 if (attr->attr < sizeof(ais))
1915 if (!test_kvm_facility(kvm, 72))
1918 mutex_lock(&fi->ais_lock);
1919 ais.simm = fi->simm;
1920 ais.nimm = fi->nimm;
1921 mutex_unlock(&fi->ais_lock);
1923 if (copy_to_user((void __user *)attr->addr, &ais, sizeof(ais)))
1929 static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
1933 switch (attr->group) {
1934 case KVM_DEV_FLIC_GET_ALL_IRQS:
1935 r = get_all_floating_irqs(dev->kvm, (u8 __user *) attr->addr,
1938 case KVM_DEV_FLIC_AISM_ALL:
1939 r = flic_ais_mode_get_all(dev->kvm, attr);
1948 static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
1951 struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
1952 void *target = NULL;
1953 void __user *source;
1956 if (get_user(inti->type, (u64 __user *)addr))
1959 switch (inti->type) {
1960 case KVM_S390_INT_PFAULT_INIT:
1961 case KVM_S390_INT_PFAULT_DONE:
1962 case KVM_S390_INT_VIRTIO:
1963 case KVM_S390_INT_SERVICE:
1964 target = (void *) &inti->ext;
1965 source = &uptr->u.ext;
1966 size = sizeof(inti->ext);
1968 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1969 target = (void *) &inti->io;
1970 source = &uptr->u.io;
1971 size = sizeof(inti->io);
1974 target = (void *) &inti->mchk;
1975 source = &uptr->u.mchk;
1976 size = sizeof(inti->mchk);
1982 if (copy_from_user(target, source, size))
1988 static int enqueue_floating_irq(struct kvm_device *dev,
1989 struct kvm_device_attr *attr)
1991 struct kvm_s390_interrupt_info *inti = NULL;
1993 int len = attr->attr;
1995 if (len % sizeof(struct kvm_s390_irq) != 0)
1997 else if (len > KVM_S390_FLIC_MAX_BUFFER)
2000 while (len >= sizeof(struct kvm_s390_irq)) {
2001 inti = kzalloc(sizeof(*inti), GFP_KERNEL);
2005 r = copy_irq_from_user(inti, attr->addr);
2010 r = __inject_vm(dev->kvm, inti);
2015 len -= sizeof(struct kvm_s390_irq);
2016 attr->addr += sizeof(struct kvm_s390_irq);
2022 static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
2024 if (id >= MAX_S390_IO_ADAPTERS)
2026 return kvm->arch.adapters[id];
2029 static int register_io_adapter(struct kvm_device *dev,
2030 struct kvm_device_attr *attr)
2032 struct s390_io_adapter *adapter;
2033 struct kvm_s390_io_adapter adapter_info;
2035 if (copy_from_user(&adapter_info,
2036 (void __user *)attr->addr, sizeof(adapter_info)))
2039 if ((adapter_info.id >= MAX_S390_IO_ADAPTERS) ||
2040 (dev->kvm->arch.adapters[adapter_info.id] != NULL))
2043 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2047 INIT_LIST_HEAD(&adapter->maps);
2048 init_rwsem(&adapter->maps_lock);
2049 atomic_set(&adapter->nr_maps, 0);
2050 adapter->id = adapter_info.id;
2051 adapter->isc = adapter_info.isc;
2052 adapter->maskable = adapter_info.maskable;
2053 adapter->masked = false;
2054 adapter->swap = adapter_info.swap;
2055 adapter->suppressible = (adapter_info.flags) &
2056 KVM_S390_ADAPTER_SUPPRESSIBLE;
2057 dev->kvm->arch.adapters[adapter->id] = adapter;
2062 int kvm_s390_mask_adapter(struct kvm *kvm, unsigned int id, bool masked)
2065 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2067 if (!adapter || !adapter->maskable)
2069 ret = adapter->masked;
2070 adapter->masked = masked;
2074 static int kvm_s390_adapter_map(struct kvm *kvm, unsigned int id, __u64 addr)
2076 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2077 struct s390_map_info *map;
2080 if (!adapter || !addr)
2083 map = kzalloc(sizeof(*map), GFP_KERNEL);
2088 INIT_LIST_HEAD(&map->list);
2089 map->guest_addr = addr;
2090 map->addr = gmap_translate(kvm->arch.gmap, addr);
2091 if (map->addr == -EFAULT) {
2095 ret = get_user_pages_fast(map->addr, 1, 1, &map->page);
2099 down_write(&adapter->maps_lock);
2100 if (atomic_inc_return(&adapter->nr_maps) < MAX_S390_ADAPTER_MAPS) {
2101 list_add_tail(&map->list, &adapter->maps);
2104 put_page(map->page);
2107 up_write(&adapter->maps_lock);
2114 static int kvm_s390_adapter_unmap(struct kvm *kvm, unsigned int id, __u64 addr)
2116 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2117 struct s390_map_info *map, *tmp;
2120 if (!adapter || !addr)
2123 down_write(&adapter->maps_lock);
2124 list_for_each_entry_safe(map, tmp, &adapter->maps, list) {
2125 if (map->guest_addr == addr) {
2127 atomic_dec(&adapter->nr_maps);
2128 list_del(&map->list);
2129 put_page(map->page);
2134 up_write(&adapter->maps_lock);
2136 return found ? 0 : -EINVAL;
2139 void kvm_s390_destroy_adapters(struct kvm *kvm)
2142 struct s390_map_info *map, *tmp;
2144 for (i = 0; i < MAX_S390_IO_ADAPTERS; i++) {
2145 if (!kvm->arch.adapters[i])
2147 list_for_each_entry_safe(map, tmp,
2148 &kvm->arch.adapters[i]->maps, list) {
2149 list_del(&map->list);
2150 put_page(map->page);
2153 kfree(kvm->arch.adapters[i]);
2157 static int modify_io_adapter(struct kvm_device *dev,
2158 struct kvm_device_attr *attr)
2160 struct kvm_s390_io_adapter_req req;
2161 struct s390_io_adapter *adapter;
2164 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2167 adapter = get_io_adapter(dev->kvm, req.id);
2171 case KVM_S390_IO_ADAPTER_MASK:
2172 ret = kvm_s390_mask_adapter(dev->kvm, req.id, req.mask);
2176 case KVM_S390_IO_ADAPTER_MAP:
2177 ret = kvm_s390_adapter_map(dev->kvm, req.id, req.addr);
2179 case KVM_S390_IO_ADAPTER_UNMAP:
2180 ret = kvm_s390_adapter_unmap(dev->kvm, req.id, req.addr);
2189 static int clear_io_irq(struct kvm *kvm, struct kvm_device_attr *attr)
2192 const u64 isc_mask = 0xffUL << 24; /* all iscs set */
2197 if (attr->attr != sizeof(schid))
2199 if (copy_from_user(&schid, (void __user *) attr->addr, sizeof(schid)))
2201 kfree(kvm_s390_get_io_int(kvm, isc_mask, schid));
2203 * If userspace is conforming to the architecture, we can have at most
2204 * one pending I/O interrupt per subchannel, so this is effectively a
2210 static int modify_ais_mode(struct kvm *kvm, struct kvm_device_attr *attr)
2212 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2213 struct kvm_s390_ais_req req;
2216 if (!test_kvm_facility(kvm, 72))
2219 if (copy_from_user(&req, (void __user *)attr->addr, sizeof(req)))
2222 if (req.isc > MAX_ISC)
2225 trace_kvm_s390_modify_ais_mode(req.isc,
2226 (fi->simm & AIS_MODE_MASK(req.isc)) ?
2227 (fi->nimm & AIS_MODE_MASK(req.isc)) ?
2228 2 : KVM_S390_AIS_MODE_SINGLE :
2229 KVM_S390_AIS_MODE_ALL, req.mode);
2231 mutex_lock(&fi->ais_lock);
2233 case KVM_S390_AIS_MODE_ALL:
2234 fi->simm &= ~AIS_MODE_MASK(req.isc);
2235 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2237 case KVM_S390_AIS_MODE_SINGLE:
2238 fi->simm |= AIS_MODE_MASK(req.isc);
2239 fi->nimm &= ~AIS_MODE_MASK(req.isc);
2244 mutex_unlock(&fi->ais_lock);
2249 static int kvm_s390_inject_airq(struct kvm *kvm,
2250 struct s390_io_adapter *adapter)
2252 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2253 struct kvm_s390_interrupt s390int = {
2254 .type = KVM_S390_INT_IO(1, 0, 0, 0),
2256 .parm64 = (adapter->isc << 27) | 0x80000000,
2260 if (!test_kvm_facility(kvm, 72) || !adapter->suppressible)
2261 return kvm_s390_inject_vm(kvm, &s390int);
2263 mutex_lock(&fi->ais_lock);
2264 if (fi->nimm & AIS_MODE_MASK(adapter->isc)) {
2265 trace_kvm_s390_airq_suppressed(adapter->id, adapter->isc);
2269 ret = kvm_s390_inject_vm(kvm, &s390int);
2270 if (!ret && (fi->simm & AIS_MODE_MASK(adapter->isc))) {
2271 fi->nimm |= AIS_MODE_MASK(adapter->isc);
2272 trace_kvm_s390_modify_ais_mode(adapter->isc,
2273 KVM_S390_AIS_MODE_SINGLE, 2);
2276 mutex_unlock(&fi->ais_lock);
2280 static int flic_inject_airq(struct kvm *kvm, struct kvm_device_attr *attr)
2282 unsigned int id = attr->attr;
2283 struct s390_io_adapter *adapter = get_io_adapter(kvm, id);
2288 return kvm_s390_inject_airq(kvm, adapter);
2291 static int flic_ais_mode_set_all(struct kvm *kvm, struct kvm_device_attr *attr)
2293 struct kvm_s390_float_interrupt *fi = &kvm->arch.float_int;
2294 struct kvm_s390_ais_all ais;
2296 if (!test_kvm_facility(kvm, 72))
2299 if (copy_from_user(&ais, (void __user *)attr->addr, sizeof(ais)))
2302 mutex_lock(&fi->ais_lock);
2303 fi->simm = ais.simm;
2304 fi->nimm = ais.nimm;
2305 mutex_unlock(&fi->ais_lock);
2310 static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
2314 struct kvm_vcpu *vcpu;
2316 switch (attr->group) {
2317 case KVM_DEV_FLIC_ENQUEUE:
2318 r = enqueue_floating_irq(dev, attr);
2320 case KVM_DEV_FLIC_CLEAR_IRQS:
2321 kvm_s390_clear_float_irqs(dev->kvm);
2323 case KVM_DEV_FLIC_APF_ENABLE:
2324 dev->kvm->arch.gmap->pfault_enabled = 1;
2326 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2327 dev->kvm->arch.gmap->pfault_enabled = 0;
2329 * Make sure no async faults are in transition when
2330 * clearing the queues. So we don't need to worry
2331 * about late coming workers.
2333 synchronize_srcu(&dev->kvm->srcu);
2334 kvm_for_each_vcpu(i, vcpu, dev->kvm)
2335 kvm_clear_async_pf_completion_queue(vcpu);
2337 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2338 r = register_io_adapter(dev, attr);
2340 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2341 r = modify_io_adapter(dev, attr);
2343 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2344 r = clear_io_irq(dev->kvm, attr);
2346 case KVM_DEV_FLIC_AISM:
2347 r = modify_ais_mode(dev->kvm, attr);
2349 case KVM_DEV_FLIC_AIRQ_INJECT:
2350 r = flic_inject_airq(dev->kvm, attr);
2352 case KVM_DEV_FLIC_AISM_ALL:
2353 r = flic_ais_mode_set_all(dev->kvm, attr);
2362 static int flic_has_attr(struct kvm_device *dev,
2363 struct kvm_device_attr *attr)
2365 switch (attr->group) {
2366 case KVM_DEV_FLIC_GET_ALL_IRQS:
2367 case KVM_DEV_FLIC_ENQUEUE:
2368 case KVM_DEV_FLIC_CLEAR_IRQS:
2369 case KVM_DEV_FLIC_APF_ENABLE:
2370 case KVM_DEV_FLIC_APF_DISABLE_WAIT:
2371 case KVM_DEV_FLIC_ADAPTER_REGISTER:
2372 case KVM_DEV_FLIC_ADAPTER_MODIFY:
2373 case KVM_DEV_FLIC_CLEAR_IO_IRQ:
2374 case KVM_DEV_FLIC_AISM:
2375 case KVM_DEV_FLIC_AIRQ_INJECT:
2376 case KVM_DEV_FLIC_AISM_ALL:
2382 static int flic_create(struct kvm_device *dev, u32 type)
2386 if (dev->kvm->arch.flic)
2388 dev->kvm->arch.flic = dev;
2392 static void flic_destroy(struct kvm_device *dev)
2394 dev->kvm->arch.flic = NULL;
2398 /* s390 floating irq controller (flic) */
2399 struct kvm_device_ops kvm_flic_ops = {
2401 .get_attr = flic_get_attr,
2402 .set_attr = flic_set_attr,
2403 .has_attr = flic_has_attr,
2404 .create = flic_create,
2405 .destroy = flic_destroy,
2408 static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap)
2412 bit = bit_nr + (addr % PAGE_SIZE) * 8;
2414 return swap ? (bit ^ (BITS_PER_LONG - 1)) : bit;
2417 static struct s390_map_info *get_map_info(struct s390_io_adapter *adapter,
2420 struct s390_map_info *map;
2425 list_for_each_entry(map, &adapter->maps, list) {
2426 if (map->guest_addr == addr)
2432 static int adapter_indicators_set(struct kvm *kvm,
2433 struct s390_io_adapter *adapter,
2434 struct kvm_s390_adapter_int *adapter_int)
2437 int summary_set, idx;
2438 struct s390_map_info *info;
2441 info = get_map_info(adapter, adapter_int->ind_addr);
2444 map = page_address(info->page);
2445 bit = get_ind_bit(info->addr, adapter_int->ind_offset, adapter->swap);
2447 idx = srcu_read_lock(&kvm->srcu);
2448 mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2449 set_page_dirty_lock(info->page);
2450 info = get_map_info(adapter, adapter_int->summary_addr);
2452 srcu_read_unlock(&kvm->srcu, idx);
2455 map = page_address(info->page);
2456 bit = get_ind_bit(info->addr, adapter_int->summary_offset,
2458 summary_set = test_and_set_bit(bit, map);
2459 mark_page_dirty(kvm, info->guest_addr >> PAGE_SHIFT);
2460 set_page_dirty_lock(info->page);
2461 srcu_read_unlock(&kvm->srcu, idx);
2462 return summary_set ? 0 : 1;
2466 * < 0 - not injected due to error
2467 * = 0 - coalesced, summary indicator already active
2468 * > 0 - injected interrupt
2470 static int set_adapter_int(struct kvm_kernel_irq_routing_entry *e,
2471 struct kvm *kvm, int irq_source_id, int level,
2475 struct s390_io_adapter *adapter;
2477 /* We're only interested in the 0->1 transition. */
2480 adapter = get_io_adapter(kvm, e->adapter.adapter_id);
2483 down_read(&adapter->maps_lock);
2484 ret = adapter_indicators_set(kvm, adapter, &e->adapter);
2485 up_read(&adapter->maps_lock);
2486 if ((ret > 0) && !adapter->masked) {
2487 ret = kvm_s390_inject_airq(kvm, adapter);
2495 * Inject the machine check to the guest.
2497 void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
2498 struct mcck_volatile_info *mcck_info)
2500 struct kvm_s390_interrupt_info inti;
2501 struct kvm_s390_irq irq;
2502 struct kvm_s390_mchk_info *mchk;
2504 __u64 cr14 = 0; /* upper bits are not used */
2507 mci.val = mcck_info->mcic;
2509 cr14 |= MCCK_CR14_RECOVERY_SUB_MASK;
2511 cr14 |= MCCK_CR14_DEGRAD_SUB_MASK;
2513 cr14 |= MCCK_CR14_WARN_SUB_MASK;
2515 mchk = mci.ck ? &inti.mchk : &irq.u.mchk;
2517 mchk->mcic = mcck_info->mcic;
2518 mchk->ext_damage_code = mcck_info->ext_damage_code;
2519 mchk->failing_storage_address = mcck_info->failing_storage_address;
2521 /* Inject the floating machine check */
2522 inti.type = KVM_S390_MCHK;
2523 rc = __inject_vm(vcpu->kvm, &inti);
2525 /* Inject the machine check to specified vcpu */
2526 irq.type = KVM_S390_MCHK;
2527 rc = kvm_s390_inject_vcpu(vcpu, &irq);
2532 int kvm_set_routing_entry(struct kvm *kvm,
2533 struct kvm_kernel_irq_routing_entry *e,
2534 const struct kvm_irq_routing_entry *ue)
2539 case KVM_IRQ_ROUTING_S390_ADAPTER:
2540 e->set = set_adapter_int;
2541 e->adapter.summary_addr = ue->u.adapter.summary_addr;
2542 e->adapter.ind_addr = ue->u.adapter.ind_addr;
2543 e->adapter.summary_offset = ue->u.adapter.summary_offset;
2544 e->adapter.ind_offset = ue->u.adapter.ind_offset;
2545 e->adapter.adapter_id = ue->u.adapter.adapter_id;
2555 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
2556 int irq_source_id, int level, bool line_status)
2561 int kvm_s390_set_irq_state(struct kvm_vcpu *vcpu, void __user *irqstate, int len)
2563 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2564 struct kvm_s390_irq *buf;
2572 if (copy_from_user((void *) buf, irqstate, len)) {
2578 * Don't allow setting the interrupt state
2579 * when there are already interrupts pending
2581 spin_lock(&li->lock);
2582 if (li->pending_irqs) {
2587 for (n = 0; n < len / sizeof(*buf); n++) {
2588 r = do_inject_vcpu(vcpu, &buf[n]);
2594 spin_unlock(&li->lock);
2601 static void store_local_irq(struct kvm_s390_local_interrupt *li,
2602 struct kvm_s390_irq *irq,
2603 unsigned long irq_type)
2606 case IRQ_PEND_MCHK_EX:
2607 case IRQ_PEND_MCHK_REP:
2608 irq->type = KVM_S390_MCHK;
2609 irq->u.mchk = li->irq.mchk;
2612 irq->type = KVM_S390_PROGRAM_INT;
2613 irq->u.pgm = li->irq.pgm;
2615 case IRQ_PEND_PFAULT_INIT:
2616 irq->type = KVM_S390_INT_PFAULT_INIT;
2617 irq->u.ext = li->irq.ext;
2619 case IRQ_PEND_EXT_EXTERNAL:
2620 irq->type = KVM_S390_INT_EXTERNAL_CALL;
2621 irq->u.extcall = li->irq.extcall;
2623 case IRQ_PEND_EXT_CLOCK_COMP:
2624 irq->type = KVM_S390_INT_CLOCK_COMP;
2626 case IRQ_PEND_EXT_CPU_TIMER:
2627 irq->type = KVM_S390_INT_CPU_TIMER;
2629 case IRQ_PEND_SIGP_STOP:
2630 irq->type = KVM_S390_SIGP_STOP;
2631 irq->u.stop = li->irq.stop;
2633 case IRQ_PEND_RESTART:
2634 irq->type = KVM_S390_RESTART;
2636 case IRQ_PEND_SET_PREFIX:
2637 irq->type = KVM_S390_SIGP_SET_PREFIX;
2638 irq->u.prefix = li->irq.prefix;
2643 int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu, __u8 __user *buf, int len)
2646 unsigned long sigp_emerg_pending[BITS_TO_LONGS(KVM_MAX_VCPUS)];
2647 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
2648 unsigned long pending_irqs;
2649 struct kvm_s390_irq irq;
2650 unsigned long irq_type;
2654 spin_lock(&li->lock);
2655 pending_irqs = li->pending_irqs;
2656 memcpy(&sigp_emerg_pending, &li->sigp_emerg_pending,
2657 sizeof(sigp_emerg_pending));
2658 spin_unlock(&li->lock);
2660 for_each_set_bit(irq_type, &pending_irqs, IRQ_PEND_COUNT) {
2661 memset(&irq, 0, sizeof(irq));
2662 if (irq_type == IRQ_PEND_EXT_EMERGENCY)
2664 if (n + sizeof(irq) > len)
2666 store_local_irq(&vcpu->arch.local_int, &irq, irq_type);
2667 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2672 if (test_bit(IRQ_PEND_EXT_EMERGENCY, &pending_irqs)) {
2673 for_each_set_bit(cpuaddr, sigp_emerg_pending, KVM_MAX_VCPUS) {
2674 memset(&irq, 0, sizeof(irq));
2675 if (n + sizeof(irq) > len)
2677 irq.type = KVM_S390_INT_EMERGENCY;
2678 irq.u.emerg.code = cpuaddr;
2679 if (copy_to_user(&buf[n], &irq, sizeof(irq)))
2685 if (sca_ext_call_pending(vcpu, &scn)) {
2686 if (n + sizeof(irq) > len)
2688 memset(&irq, 0, sizeof(irq));
2689 irq.type = KVM_S390_INT_EXTERNAL_CALL;
2690 irq.u.extcall.code = scn;
2691 if (copy_to_user(&buf[n], &irq, sizeof(irq)))