1 // SPDX-License-Identifier: GPL-2.0
3 * in-kernel handling for sie intercepts
5 * Copyright IBM Corp. 2008, 2020
7 * Author(s): Carsten Otte <cotte@de.ibm.com>
8 * Christian Borntraeger <borntraeger@de.ibm.com>
11 #include <linux/kvm_host.h>
12 #include <linux/errno.h>
13 #include <linux/pagemap.h>
15 #include <asm/asm-offsets.h>
17 #include <asm/sysinfo.h>
23 #include "trace-s390.h"
25 u8 kvm_s390_get_ilen(struct kvm_vcpu *vcpu)
27 struct kvm_s390_sie_block *sie_block = vcpu->arch.sie_block;
30 switch (vcpu->arch.sie_block->icptcode) {
36 /* instruction only stored for these icptcodes */
37 ilen = insn_length(vcpu->arch.sie_block->ipa >> 8);
38 /* Use the length of the EXECUTE instruction if necessary */
39 if (sie_block->icptstatus & 1) {
40 ilen = (sie_block->icptstatus >> 4) & 0x6;
46 /* bit 1+2 of pgmilc are the ilc, so we directly get ilen */
47 ilen = vcpu->arch.sie_block->pgmilc & 0x6;
53 static int handle_stop(struct kvm_vcpu *vcpu)
55 struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
57 uint8_t flags, stop_pending;
59 vcpu->stat.exit_stop_request++;
61 /* delay the stop if any non-stop irq is pending */
62 if (kvm_s390_vcpu_has_irq(vcpu, 1))
65 /* avoid races with the injection/SIGP STOP code */
67 flags = li->irq.stop.flags;
68 stop_pending = kvm_s390_is_stop_irq_pending(vcpu);
69 spin_unlock(&li->lock);
71 trace_kvm_s390_stop_request(stop_pending, flags);
75 if (flags & KVM_S390_STOP_FLAG_STORE_STATUS) {
76 rc = kvm_s390_vcpu_store_status(vcpu,
77 KVM_S390_STORE_STATUS_NOADDR);
83 * no need to check the return value of vcpu_stop as it can only have
84 * an error for protvirt, but protvirt means user cpu state
86 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
87 kvm_s390_vcpu_stop(vcpu);
91 static int handle_validity(struct kvm_vcpu *vcpu)
93 int viwhy = vcpu->arch.sie_block->ipb >> 16;
95 vcpu->stat.exit_validity++;
96 trace_kvm_s390_intercept_validity(vcpu, viwhy);
97 KVM_EVENT(3, "validity intercept 0x%x for pid %u (kvm 0x%pK)", viwhy,
98 current->pid, vcpu->kvm);
100 /* do not warn on invalid runtime instrumentation mode */
101 WARN_ONCE(viwhy != 0x44, "kvm: unhandled validity intercept 0x%x\n",
106 static int handle_instruction(struct kvm_vcpu *vcpu)
108 vcpu->stat.exit_instruction++;
109 trace_kvm_s390_intercept_instruction(vcpu,
110 vcpu->arch.sie_block->ipa,
111 vcpu->arch.sie_block->ipb);
113 switch (vcpu->arch.sie_block->ipa >> 8) {
115 return kvm_s390_handle_01(vcpu);
117 return kvm_s390_handle_lpsw(vcpu);
119 return kvm_s390_handle_diag(vcpu);
121 return kvm_s390_handle_aa(vcpu);
123 return kvm_s390_handle_sigp(vcpu);
125 return kvm_s390_handle_b2(vcpu);
127 return kvm_s390_handle_stctl(vcpu);
129 return kvm_s390_handle_lctl(vcpu);
131 return kvm_s390_handle_b9(vcpu);
133 return kvm_s390_handle_e3(vcpu);
135 return kvm_s390_handle_e5(vcpu);
137 return kvm_s390_handle_eb(vcpu);
143 static int inject_prog_on_prog_intercept(struct kvm_vcpu *vcpu)
145 struct kvm_s390_pgm_info pgm_info = {
146 .code = vcpu->arch.sie_block->iprcc,
147 /* the PSW has already been rewound */
148 .flags = KVM_S390_PGM_FLAGS_NO_REWIND,
151 switch (vcpu->arch.sie_block->iprcc & ~PGM_PER) {
152 case PGM_AFX_TRANSLATION:
153 case PGM_ASX_TRANSLATION:
154 case PGM_EX_TRANSLATION:
155 case PGM_LFX_TRANSLATION:
156 case PGM_LSTE_SEQUENCE:
157 case PGM_LSX_TRANSLATION:
158 case PGM_LX_TRANSLATION:
159 case PGM_PRIMARY_AUTHORITY:
160 case PGM_SECONDARY_AUTHORITY:
161 case PGM_SPACE_SWITCH:
162 pgm_info.trans_exc_code = vcpu->arch.sie_block->tecmc;
164 case PGM_ALEN_TRANSLATION:
165 case PGM_ALE_SEQUENCE:
166 case PGM_ASTE_INSTANCE:
167 case PGM_ASTE_SEQUENCE:
168 case PGM_ASTE_VALIDITY:
169 case PGM_EXTENDED_AUTHORITY:
170 pgm_info.exc_access_id = vcpu->arch.sie_block->eai;
173 case PGM_PAGE_TRANSLATION:
174 case PGM_REGION_FIRST_TRANS:
175 case PGM_REGION_SECOND_TRANS:
176 case PGM_REGION_THIRD_TRANS:
177 case PGM_SEGMENT_TRANSLATION:
178 pgm_info.trans_exc_code = vcpu->arch.sie_block->tecmc;
179 pgm_info.exc_access_id = vcpu->arch.sie_block->eai;
180 pgm_info.op_access_id = vcpu->arch.sie_block->oai;
183 pgm_info.mon_class_nr = vcpu->arch.sie_block->mcn;
184 pgm_info.mon_code = vcpu->arch.sie_block->tecmc;
186 case PGM_VECTOR_PROCESSING:
188 pgm_info.data_exc_code = vcpu->arch.sie_block->dxc;
191 pgm_info.trans_exc_code = vcpu->arch.sie_block->tecmc;
192 pgm_info.exc_access_id = vcpu->arch.sie_block->eai;
198 if (vcpu->arch.sie_block->iprcc & PGM_PER) {
199 pgm_info.per_code = vcpu->arch.sie_block->perc;
200 pgm_info.per_atmid = vcpu->arch.sie_block->peratmid;
201 pgm_info.per_address = vcpu->arch.sie_block->peraddr;
202 pgm_info.per_access_id = vcpu->arch.sie_block->peraid;
204 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
208 * restore ITDB to program-interruption TDB in guest lowcore
209 * and set TX abort indication if required
211 static int handle_itdb(struct kvm_vcpu *vcpu)
213 struct kvm_s390_itdb *itdb;
216 if (!IS_TE_ENABLED(vcpu) || !IS_ITDB_VALID(vcpu))
218 if (current->thread.per_flags & PER_FLAG_NO_TE)
220 itdb = (struct kvm_s390_itdb *)vcpu->arch.sie_block->itdba;
221 rc = write_guest_lc(vcpu, __LC_PGM_TDB, itdb, sizeof(*itdb));
224 memset(itdb, 0, sizeof(*itdb));
229 #define per_event(vcpu) (vcpu->arch.sie_block->iprcc & PGM_PER)
231 static int handle_prog(struct kvm_vcpu *vcpu)
236 vcpu->stat.exit_program_interruption++;
239 * Intercept 8 indicates a loop of specification exceptions
240 * for protected guests.
242 if (kvm_s390_pv_cpu_is_protected(vcpu))
245 if (guestdbg_enabled(vcpu) && per_event(vcpu)) {
246 rc = kvm_s390_handle_per_event(vcpu);
249 /* the interrupt might have been filtered out completely */
250 if (vcpu->arch.sie_block->iprcc == 0)
254 trace_kvm_s390_intercept_prog(vcpu, vcpu->arch.sie_block->iprcc);
255 if (vcpu->arch.sie_block->iprcc == PGM_SPECIFICATION) {
256 rc = read_guest_lc(vcpu, __LC_PGM_NEW_PSW, &psw, sizeof(psw_t));
259 /* Avoid endless loops of specification exceptions */
260 if (!is_valid_psw(&psw))
263 rc = handle_itdb(vcpu);
267 return inject_prog_on_prog_intercept(vcpu);
271 * handle_external_interrupt - used for external interruption interceptions
274 * This interception occurs if:
275 * - the CPUSTAT_EXT_INT bit was already set when the external interrupt
276 * occurred. In this case, the interrupt needs to be injected manually to
277 * preserve interrupt priority.
278 * - the external new PSW has external interrupts enabled, which will cause an
279 * interruption loop. We drop to userspace in this case.
281 * The latter case can be detected by inspecting the external mask bit in the
284 * Under PV, only the latter case can occur, since interrupt priorities are
285 * handled in the ultravisor.
287 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
289 u16 eic = vcpu->arch.sie_block->eic;
290 struct kvm_s390_irq irq;
294 vcpu->stat.exit_external_interrupt++;
296 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
297 newpsw = vcpu->arch.sie_block->gpsw;
299 rc = read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &newpsw, sizeof(psw_t));
305 * Clock comparator or timer interrupt with external interrupt enabled
306 * will cause interrupt loop. Drop to userspace.
308 if ((eic == EXT_IRQ_CLK_COMP || eic == EXT_IRQ_CPU_TIMER) &&
309 (newpsw.mask & PSW_MASK_EXT))
313 case EXT_IRQ_CLK_COMP:
314 irq.type = KVM_S390_INT_CLOCK_COMP;
316 case EXT_IRQ_CPU_TIMER:
317 irq.type = KVM_S390_INT_CPU_TIMER;
319 case EXT_IRQ_EXTERNAL_CALL:
320 irq.type = KVM_S390_INT_EXTERNAL_CALL;
321 irq.u.extcall.code = vcpu->arch.sie_block->extcpuaddr;
322 rc = kvm_s390_inject_vcpu(vcpu, &irq);
323 /* ignore if another external call is already pending */
331 return kvm_s390_inject_vcpu(vcpu, &irq);
335 * handle_mvpg_pei - Handle MOVE PAGE partial execution interception.
338 * This interception can only happen for guests with DAT disabled and
339 * addresses that are currently not mapped in the host. Thus we try to
340 * set up the mappings for the corresponding user pages here (or throw
341 * addressing exceptions in case of illegal guest addresses).
343 static int handle_mvpg_pei(struct kvm_vcpu *vcpu)
345 unsigned long srcaddr, dstaddr;
348 kvm_s390_get_regs_rre(vcpu, ®1, ®2);
350 /* Ensure that the source is paged-in, no actual access -> no key checking */
351 rc = guest_translate_address_with_key(vcpu, vcpu->run->s.regs.gprs[reg2],
352 reg2, &srcaddr, GACC_FETCH, 0);
354 return kvm_s390_inject_prog_cond(vcpu, rc);
355 rc = kvm_arch_fault_in_page(vcpu, srcaddr, 0);
359 /* Ensure that the source is paged-in, no actual access -> no key checking */
360 rc = guest_translate_address_with_key(vcpu, vcpu->run->s.regs.gprs[reg1],
361 reg1, &dstaddr, GACC_STORE, 0);
363 return kvm_s390_inject_prog_cond(vcpu, rc);
364 rc = kvm_arch_fault_in_page(vcpu, dstaddr, 1);
368 kvm_s390_retry_instr(vcpu);
373 static int handle_partial_execution(struct kvm_vcpu *vcpu)
375 vcpu->stat.exit_pei++;
377 if (vcpu->arch.sie_block->ipa == 0xb254) /* MVPG */
378 return handle_mvpg_pei(vcpu);
379 if (vcpu->arch.sie_block->ipa >> 8 == 0xae) /* SIGP */
380 return kvm_s390_handle_sigp_pei(vcpu);
386 * Handle the sthyi instruction that provides the guest with system
387 * information, like current CPU resources available at each level of
390 int handle_sthyi(struct kvm_vcpu *vcpu)
392 int reg1, reg2, cc = 0, r = 0;
393 u64 code, addr, rc = 0;
394 struct sthyi_sctns *sctns = NULL;
396 if (!test_kvm_facility(vcpu->kvm, 74))
397 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
399 kvm_s390_get_regs_rre(vcpu, ®1, ®2);
400 code = vcpu->run->s.regs.gprs[reg1];
401 addr = vcpu->run->s.regs.gprs[reg2];
403 vcpu->stat.instruction_sthyi++;
404 VCPU_EVENT(vcpu, 3, "STHYI: fc: %llu addr: 0x%016llx", code, addr);
405 trace_kvm_s390_handle_sthyi(vcpu, code, addr);
407 if (reg1 == reg2 || reg1 & 1 || reg2 & 1)
408 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
416 if (!kvm_s390_pv_cpu_is_protected(vcpu) && (addr & ~PAGE_MASK))
417 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
419 sctns = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
423 cc = sthyi_fill(sctns, &rc);
425 free_page((unsigned long)sctns);
430 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
431 memcpy((void *)(sida_origin(vcpu->arch.sie_block)),
434 r = write_guest(vcpu, addr, reg2, sctns, PAGE_SIZE);
436 free_page((unsigned long)sctns);
437 return kvm_s390_inject_prog_cond(vcpu, r);
442 free_page((unsigned long)sctns);
443 vcpu->run->s.regs.gprs[reg2 + 1] = rc;
444 kvm_s390_set_psw_cc(vcpu, cc);
448 static int handle_operexc(struct kvm_vcpu *vcpu)
450 psw_t oldpsw, newpsw;
453 vcpu->stat.exit_operation_exception++;
454 trace_kvm_s390_handle_operexc(vcpu, vcpu->arch.sie_block->ipa,
455 vcpu->arch.sie_block->ipb);
457 if (vcpu->arch.sie_block->ipa == 0xb256)
458 return handle_sthyi(vcpu);
460 if (vcpu->arch.sie_block->ipa == 0 && vcpu->kvm->arch.user_instr0)
462 rc = read_guest_lc(vcpu, __LC_PGM_NEW_PSW, &newpsw, sizeof(psw_t));
466 * Avoid endless loops of operation exceptions, if the pgm new
467 * PSW will cause a new operation exception.
468 * The heuristic checks if the pgm new psw is within 6 bytes before
469 * the faulting psw address (with same DAT, AS settings) and the
470 * new psw is not a wait psw and the fault was not triggered by
473 oldpsw = vcpu->arch.sie_block->gpsw;
474 if (oldpsw.addr - newpsw.addr <= 6 &&
475 !(newpsw.mask & PSW_MASK_WAIT) &&
476 !(oldpsw.mask & PSW_MASK_PSTATE) &&
477 (newpsw.mask & PSW_MASK_ASC) == (oldpsw.mask & PSW_MASK_ASC) &&
478 (newpsw.mask & PSW_MASK_DAT) == (oldpsw.mask & PSW_MASK_DAT))
481 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
484 static int handle_pv_spx(struct kvm_vcpu *vcpu)
486 u32 pref = *(u32 *)vcpu->arch.sie_block->sidad;
488 kvm_s390_set_prefix(vcpu, pref);
489 trace_kvm_s390_handle_prefix(vcpu, 1, pref);
493 static int handle_pv_sclp(struct kvm_vcpu *vcpu)
495 struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int;
497 spin_lock(&fi->lock);
500 * a: an sccb answering interrupt was already pending or in flight.
501 * As the sccb value is not known we can simply set some value to
502 * trigger delivery of a saved SCCB. UV will then use its saved
503 * copy of the SCCB value.
504 * b: an error SCCB interrupt needs to be injected so we also inject
505 * a fake SCCB address. Firmware will use the proper one.
506 * This makes sure, that both errors and real sccb returns will only
507 * be delivered after a notification intercept (instruction has
508 * finished) but not after others.
510 fi->srv_signal.ext_params |= 0x43000;
511 set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs);
512 clear_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs);
513 spin_unlock(&fi->lock);
517 static int handle_pv_uvc(struct kvm_vcpu *vcpu)
519 struct uv_cb_share *guest_uvcb = (void *)vcpu->arch.sie_block->sidad;
520 struct uv_cb_cts uvcb = {
521 .header.cmd = UVC_CMD_UNPIN_PAGE_SHARED,
522 .header.len = sizeof(uvcb),
523 .guest_handle = kvm_s390_pv_get_handle(vcpu->kvm),
524 .gaddr = guest_uvcb->paddr,
528 if (guest_uvcb->header.cmd != UVC_CMD_REMOVE_SHARED_ACCESS) {
529 WARN_ONCE(1, "Unexpected notification intercept for UVC 0x%x\n",
530 guest_uvcb->header.cmd);
533 rc = gmap_make_secure(vcpu->arch.gmap, uvcb.gaddr, &uvcb);
535 * If the unpin did not succeed, the guest will exit again for the UVC
536 * and we will retry the unpin.
541 * If we got -EAGAIN here, we simply return it. It will eventually
542 * get propagated all the way to userspace, which should then try
548 static int handle_pv_notification(struct kvm_vcpu *vcpu)
552 if (vcpu->arch.sie_block->ipa == 0xb210)
553 return handle_pv_spx(vcpu);
554 if (vcpu->arch.sie_block->ipa == 0xb220)
555 return handle_pv_sclp(vcpu);
556 if (vcpu->arch.sie_block->ipa == 0xb9a4)
557 return handle_pv_uvc(vcpu);
558 if (vcpu->arch.sie_block->ipa >> 8 == 0xae) {
560 * Besides external call, other SIGP orders also cause a
561 * 108 (pv notify) intercept. In contrast to external call,
562 * these orders need to be emulated and hence the appropriate
563 * place to handle them is in handle_instruction().
564 * So first try kvm_s390_handle_sigp_pei() and if that isn't
565 * successful, go on with handle_instruction().
567 ret = kvm_s390_handle_sigp_pei(vcpu);
572 return handle_instruction(vcpu);
575 int kvm_handle_sie_intercept(struct kvm_vcpu *vcpu)
579 if (kvm_is_ucontrol(vcpu->kvm))
582 switch (vcpu->arch.sie_block->icptcode) {
584 vcpu->stat.exit_external_request++;
587 vcpu->stat.exit_io_request++;
590 rc = handle_instruction(vcpu);
593 return handle_prog(vcpu);
595 return handle_external_interrupt(vcpu);
597 return kvm_s390_handle_wait(vcpu);
599 return handle_validity(vcpu);
601 return handle_stop(vcpu);
603 rc = handle_operexc(vcpu);
606 rc = handle_partial_execution(vcpu);
609 rc = kvm_s390_skey_check_enable(vcpu);
612 case ICPT_INT_ENABLE:
614 * PSW bit 13 or a CR (0, 6, 14) changed and we might
615 * now be able to deliver interrupts. The pre-run code
616 * will take care of this.
621 rc = handle_instruction(vcpu);
624 rc = handle_pv_notification(vcpu);
628 gmap_convert_to_secure(vcpu->arch.gmap,
629 kvm_s390_get_prefix(vcpu));
630 gmap_convert_to_secure(vcpu->arch.gmap,
631 kvm_s390_get_prefix(vcpu) + PAGE_SIZE);
637 /* process PER, also if the instrution is processed in user space */
638 if (vcpu->arch.sie_block->icptstatus & 0x02 &&
639 (!rc || rc == -EOPNOTSUPP))
640 per_rc = kvm_s390_handle_per_ifetch_icpt(vcpu);
641 return per_rc ? per_rc : rc;