1 // SPDX-License-Identifier: GPL-2.0
3 * kvm nested virtualization support for s390x
5 * Copyright IBM Corp. 2016, 2018
7 * Author(s): David Hildenbrand <dahi@linux.vnet.ibm.com>
9 #include <linux/vmalloc.h>
10 #include <linux/kvm_host.h>
11 #include <linux/bug.h>
12 #include <linux/list.h>
13 #include <linux/bitmap.h>
14 #include <linux/sched/signal.h>
17 #include <asm/mmu_context.h>
21 #include <asm/fpu/api.h>
26 struct kvm_s390_sie_block scb_s; /* 0x0000 */
28 * the backup info for machine check. ensure it's at
29 * the same offset as that in struct sie_page!
31 struct mcck_volatile_info mcck_info; /* 0x0200 */
33 * The pinned original scb. Be aware that other VCPUs can modify
34 * it while we read from it. Values that are used for conditions or
35 * are reused conditionally, should be accessed via READ_ONCE.
37 struct kvm_s390_sie_block *scb_o; /* 0x0218 */
38 /* the shadow gmap in use by the vsie_page */
39 struct gmap *gmap; /* 0x0220 */
40 /* address of the last reported fault to guest2 */
41 unsigned long fault_addr; /* 0x0228 */
42 /* calculated guest addresses of satellite control blocks */
43 gpa_t sca_gpa; /* 0x0230 */
44 gpa_t itdba_gpa; /* 0x0238 */
45 gpa_t gvrd_gpa; /* 0x0240 */
46 gpa_t riccbd_gpa; /* 0x0248 */
47 gpa_t sdnx_gpa; /* 0x0250 */
48 __u8 reserved[0x0700 - 0x0258]; /* 0x0258 */
49 struct kvm_s390_crypto_cb crycb; /* 0x0700 */
50 __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */
53 /* trigger a validity icpt for the given scb */
54 static int set_validity_icpt(struct kvm_s390_sie_block *scb,
58 scb->ipb = ((__u32) reason_code) << 16;
59 scb->icptcode = ICPT_VALIDITY;
63 /* mark the prefix as unmapped, this will block the VSIE */
64 static void prefix_unmapped(struct vsie_page *vsie_page)
66 atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
69 /* mark the prefix as unmapped and wait until the VSIE has been left */
70 static void prefix_unmapped_sync(struct vsie_page *vsie_page)
72 prefix_unmapped(vsie_page);
73 if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
74 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
75 while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
79 /* mark the prefix as mapped, this will allow the VSIE to run */
80 static void prefix_mapped(struct vsie_page *vsie_page)
82 atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
85 /* test if the prefix is mapped into the gmap shadow */
86 static int prefix_is_mapped(struct vsie_page *vsie_page)
88 return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
91 /* copy the updated intervention request bits into the shadow scb */
92 static void update_intervention_requests(struct vsie_page *vsie_page)
94 const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
97 cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
98 atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
99 atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
102 /* shadow (filter and validate) the cpuflags */
103 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
105 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
106 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
107 int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
109 /* we don't allow ESA/390 guests */
110 if (!(cpuflags & CPUSTAT_ZARCH))
111 return set_validity_icpt(scb_s, 0x0001U);
113 if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
114 return set_validity_icpt(scb_s, 0x0001U);
115 else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
116 return set_validity_icpt(scb_s, 0x0007U);
118 /* intervention requests will be set later */
119 newflags = CPUSTAT_ZARCH;
120 if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
121 newflags |= CPUSTAT_GED;
122 if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
123 if (cpuflags & CPUSTAT_GED)
124 return set_validity_icpt(scb_s, 0x0001U);
125 newflags |= CPUSTAT_GED2;
127 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
128 newflags |= cpuflags & CPUSTAT_P;
129 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
130 newflags |= cpuflags & CPUSTAT_SM;
131 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
132 newflags |= cpuflags & CPUSTAT_IBS;
133 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS))
134 newflags |= cpuflags & CPUSTAT_KSS;
136 atomic_set(&scb_s->cpuflags, newflags);
139 /* Copy to APCB FORMAT1 from APCB FORMAT0 */
140 static int setup_apcb10(struct kvm_vcpu *vcpu, struct kvm_s390_apcb1 *apcb_s,
141 unsigned long apcb_o, struct kvm_s390_apcb1 *apcb_h)
143 struct kvm_s390_apcb0 tmp;
145 if (read_guest_real(vcpu, apcb_o, &tmp, sizeof(struct kvm_s390_apcb0)))
148 apcb_s->apm[0] = apcb_h->apm[0] & tmp.apm[0];
149 apcb_s->aqm[0] = apcb_h->aqm[0] & tmp.aqm[0] & 0xffff000000000000UL;
150 apcb_s->adm[0] = apcb_h->adm[0] & tmp.adm[0] & 0xffff000000000000UL;
157 * setup_apcb00 - Copy to APCB FORMAT0 from APCB FORMAT0
158 * @vcpu: pointer to the virtual CPU
159 * @apcb_s: pointer to start of apcb in the shadow crycb
160 * @apcb_o: pointer to start of original apcb in the guest2
161 * @apcb_h: pointer to start of apcb in the guest1
163 * Returns 0 and -EFAULT on error reading guest apcb
165 static int setup_apcb00(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
166 unsigned long apcb_o, unsigned long *apcb_h)
168 if (read_guest_real(vcpu, apcb_o, apcb_s,
169 sizeof(struct kvm_s390_apcb0)))
172 bitmap_and(apcb_s, apcb_s, apcb_h,
173 BITS_PER_BYTE * sizeof(struct kvm_s390_apcb0));
179 * setup_apcb11 - Copy the FORMAT1 APCB from the guest to the shadow CRYCB
180 * @vcpu: pointer to the virtual CPU
181 * @apcb_s: pointer to start of apcb in the shadow crycb
182 * @apcb_o: pointer to start of original guest apcb
183 * @apcb_h: pointer to start of apcb in the host
185 * Returns 0 and -EFAULT on error reading guest apcb
187 static int setup_apcb11(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
188 unsigned long apcb_o,
189 unsigned long *apcb_h)
191 if (read_guest_real(vcpu, apcb_o, apcb_s,
192 sizeof(struct kvm_s390_apcb1)))
195 bitmap_and(apcb_s, apcb_s, apcb_h,
196 BITS_PER_BYTE * sizeof(struct kvm_s390_apcb1));
202 * setup_apcb - Create a shadow copy of the apcb.
203 * @vcpu: pointer to the virtual CPU
204 * @crycb_s: pointer to shadow crycb
205 * @crycb_o: pointer to original guest crycb
206 * @crycb_h: pointer to the host crycb
207 * @fmt_o: format of the original guest crycb.
208 * @fmt_h: format of the host crycb.
210 * Checks the compatibility between the guest and host crycb and calls the
211 * appropriate copy function.
213 * Return 0 or an error number if the guest and host crycb are incompatible.
215 static int setup_apcb(struct kvm_vcpu *vcpu, struct kvm_s390_crypto_cb *crycb_s,
217 struct kvm_s390_crypto_cb *crycb_h,
218 int fmt_o, int fmt_h)
220 struct kvm_s390_crypto_cb *crycb;
222 crycb = (struct kvm_s390_crypto_cb *) (unsigned long)crycb_o;
226 if ((crycb_o & PAGE_MASK) != ((crycb_o + 256) & PAGE_MASK))
228 if (fmt_h != CRYCB_FORMAT2)
230 return setup_apcb11(vcpu, (unsigned long *)&crycb_s->apcb1,
231 (unsigned long) &crycb->apcb1,
232 (unsigned long *)&crycb_h->apcb1);
236 return setup_apcb10(vcpu, &crycb_s->apcb1,
237 (unsigned long) &crycb->apcb0,
240 return setup_apcb00(vcpu,
241 (unsigned long *) &crycb_s->apcb0,
242 (unsigned long) &crycb->apcb0,
243 (unsigned long *) &crycb_h->apcb0);
247 if ((crycb_o & PAGE_MASK) != ((crycb_o + 32) & PAGE_MASK))
252 return setup_apcb10(vcpu, &crycb_s->apcb1,
253 (unsigned long) &crycb->apcb0,
257 return setup_apcb00(vcpu,
258 (unsigned long *) &crycb_s->apcb0,
259 (unsigned long) &crycb->apcb0,
260 (unsigned long *) &crycb_h->apcb0);
267 * shadow_crycb - Create a shadow copy of the crycb block
268 * @vcpu: a pointer to the virtual CPU
269 * @vsie_page: a pointer to internal date used for the vSIE
271 * Create a shadow copy of the crycb block and setup key wrapping, if
272 * requested for guest 3 and enabled for guest 2.
274 * We accept format-1 or format-2, but we convert format-1 into format-2
275 * in the shadow CRYCB.
276 * Using format-2 enables the firmware to choose the right format when
277 * scheduling the SIE.
278 * There is nothing to do for format-0.
280 * This function centralize the issuing of set_validity_icpt() for all
281 * the subfunctions working on the crycb.
283 * Returns: - 0 if shadowed or nothing to do
284 * - > 0 if control has to be given to guest 2
286 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
288 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
289 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
290 const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd);
291 const u32 crycb_addr = crycbd_o & 0x7ffffff8U;
292 unsigned long *b1, *b2;
297 int key_msk = test_kvm_facility(vcpu->kvm, 76);
298 int fmt_o = crycbd_o & CRYCB_FORMAT_MASK;
299 int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK;
304 apie_h = vcpu->arch.sie_block->eca & ECA_APIE;
305 apie_s = apie_h & scb_o->eca;
306 if (!apie_s && (!key_msk || (fmt_o == CRYCB_FORMAT0)))
310 return set_validity_icpt(scb_s, 0x0039U);
312 if (fmt_o == CRYCB_FORMAT1)
313 if ((crycb_addr & PAGE_MASK) !=
314 ((crycb_addr + 128) & PAGE_MASK))
315 return set_validity_icpt(scb_s, 0x003CU);
318 ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr,
319 vcpu->kvm->arch.crypto.crycb,
323 scb_s->eca |= scb_o->eca & ECA_APIE;
326 /* we may only allow it if enabled for guest 2 */
327 ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
328 (ECB3_AES | ECB3_DEA);
329 ecd_flags = scb_o->ecd & vcpu->arch.sie_block->ecd & ECD_ECC;
330 if (!ecb3_flags && !ecd_flags)
333 /* copy only the wrapping keys */
334 if (read_guest_real(vcpu, crycb_addr + 72,
335 vsie_page->crycb.dea_wrapping_key_mask, 56))
336 return set_validity_icpt(scb_s, 0x0035U);
338 scb_s->ecb3 |= ecb3_flags;
339 scb_s->ecd |= ecd_flags;
341 /* xor both blocks in one run */
342 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
343 b2 = (unsigned long *)
344 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
345 /* as 56%8 == 0, bitmap_xor won't overwrite any data */
346 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
350 return set_validity_icpt(scb_s, 0x0022U);
352 return set_validity_icpt(scb_s, 0x0035U);
354 return set_validity_icpt(scb_s, 0x003CU);
356 scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT2;
360 /* shadow (round up/down) the ibc to avoid validity icpt */
361 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
363 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
364 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
365 /* READ_ONCE does not work on bitfields - use a temporary variable */
366 const uint32_t __new_ibc = scb_o->ibc;
367 const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU;
368 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
371 /* ibc installed in g2 and requested for g3 */
372 if (vcpu->kvm->arch.model.ibc && new_ibc) {
373 scb_s->ibc = new_ibc;
374 /* takte care of the minimum ibc level of the machine */
375 if (scb_s->ibc < min_ibc)
376 scb_s->ibc = min_ibc;
377 /* take care of the maximum ibc level set for the guest */
378 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
379 scb_s->ibc = vcpu->kvm->arch.model.ibc;
383 /* unshadow the scb, copying parameters back to the real scb */
384 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
386 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
387 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
390 scb_o->icptcode = scb_s->icptcode;
391 scb_o->icptstatus = scb_s->icptstatus;
392 scb_o->ipa = scb_s->ipa;
393 scb_o->ipb = scb_s->ipb;
394 scb_o->gbea = scb_s->gbea;
397 scb_o->cputm = scb_s->cputm;
398 scb_o->ckc = scb_s->ckc;
399 scb_o->todpr = scb_s->todpr;
402 scb_o->gpsw = scb_s->gpsw;
403 scb_o->gg14 = scb_s->gg14;
404 scb_o->gg15 = scb_s->gg15;
405 memcpy(scb_o->gcr, scb_s->gcr, 128);
406 scb_o->pp = scb_s->pp;
408 /* branch prediction */
409 if (test_kvm_facility(vcpu->kvm, 82)) {
410 scb_o->fpf &= ~FPF_BPBC;
411 scb_o->fpf |= scb_s->fpf & FPF_BPBC;
414 /* interrupt intercept */
415 switch (scb_s->icptcode) {
419 memcpy((void *)((u64)scb_o + 0xc0),
420 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
424 if (scb_s->ihcpu != 0xffffU)
425 scb_o->ihcpu = scb_s->ihcpu;
429 * Setup the shadow scb by copying and checking the relevant parts of the g2
432 * Returns: - 0 if the scb has been shadowed
433 * - > 0 if control has to be given to guest 2
435 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
437 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
438 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
439 /* READ_ONCE does not work on bitfields - use a temporary variable */
440 const uint32_t __new_prefix = scb_o->prefix;
441 const uint32_t new_prefix = READ_ONCE(__new_prefix);
442 const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE;
443 bool had_tx = scb_s->ecb & ECB_TE;
444 unsigned long new_mso = 0;
447 /* make sure we don't have any leftovers when reusing the scb */
457 rc = prepare_cpuflags(vcpu, vsie_page);
462 scb_s->cputm = scb_o->cputm;
463 scb_s->ckc = scb_o->ckc;
464 scb_s->todpr = scb_o->todpr;
465 scb_s->epoch = scb_o->epoch;
468 scb_s->gpsw = scb_o->gpsw;
469 scb_s->gg14 = scb_o->gg14;
470 scb_s->gg15 = scb_o->gg15;
471 memcpy(scb_s->gcr, scb_o->gcr, 128);
472 scb_s->pp = scb_o->pp;
474 /* interception / execution handling */
475 scb_s->gbea = scb_o->gbea;
476 scb_s->lctl = scb_o->lctl;
477 scb_s->svcc = scb_o->svcc;
478 scb_s->ictl = scb_o->ictl;
480 * SKEY handling functions can't deal with false setting of PTE invalid
481 * bits. Therefore we cannot provide interpretation and would later
482 * have to provide own emulation handlers.
484 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
485 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
487 scb_s->icpua = scb_o->icpua;
489 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
490 new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL;
491 /* if the hva of the prefix changes, we have to remap the prefix */
492 if (scb_s->mso != new_mso || scb_s->prefix != new_prefix)
493 prefix_unmapped(vsie_page);
494 /* SIE will do mso/msl validity and exception checks for us */
495 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
496 scb_s->mso = new_mso;
497 scb_s->prefix = new_prefix;
499 /* We have to definetly flush the tlb if this scb never ran */
500 if (scb_s->ihcpu != 0xffffU)
501 scb_s->ihcpu = scb_o->ihcpu;
503 /* MVPG and Protection Exception Interpretation are always available */
504 scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
505 /* Host-protection-interruption introduced with ESOP */
506 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
507 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
510 * This facility only uses the utility field of the SCA and none of
511 * the cpu entries that are problematic with the other interpretation
512 * facilities so we can pass it through
514 if (test_kvm_facility(vcpu->kvm, 11))
515 scb_s->ecb |= scb_o->ecb & ECB_PTF;
516 /* transactional execution */
517 if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) {
518 /* remap the prefix is tx is toggled on */
520 prefix_unmapped(vsie_page);
521 scb_s->ecb |= ECB_TE;
523 /* specification exception interpretation */
524 scb_s->ecb |= scb_o->ecb & ECB_SPECI;
525 /* branch prediction */
526 if (test_kvm_facility(vcpu->kvm, 82))
527 scb_s->fpf |= scb_o->fpf & FPF_BPBC;
529 if (test_kvm_facility(vcpu->kvm, 129)) {
530 scb_s->eca |= scb_o->eca & ECA_VX;
531 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
533 /* Run-time-Instrumentation */
534 if (test_kvm_facility(vcpu->kvm, 64))
535 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
536 /* Instruction Execution Prevention */
537 if (test_kvm_facility(vcpu->kvm, 130))
538 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
539 /* Guarded Storage */
540 if (test_kvm_facility(vcpu->kvm, 133)) {
541 scb_s->ecb |= scb_o->ecb & ECB_GS;
542 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
544 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
545 scb_s->eca |= scb_o->eca & ECA_SII;
546 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
547 scb_s->eca |= scb_o->eca & ECA_IB;
548 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
549 scb_s->eca |= scb_o->eca & ECA_CEI;
550 /* Epoch Extension */
551 if (test_kvm_facility(vcpu->kvm, 139)) {
552 scb_s->ecd |= scb_o->ecd & ECD_MEF;
553 scb_s->epdx = scb_o->epdx;
557 if (test_kvm_facility(vcpu->kvm, 156))
558 scb_s->ecd |= scb_o->ecd & ECD_ETOKENF;
560 scb_s->hpid = HPID_VSIE;
561 scb_s->cpnc = scb_o->cpnc;
563 prepare_ibc(vcpu, vsie_page);
564 rc = shadow_crycb(vcpu, vsie_page);
567 unshadow_scb(vcpu, vsie_page);
571 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
574 struct kvm *kvm = gmap->private;
575 struct vsie_page *cur;
576 unsigned long prefix;
580 if (!gmap_is_shadow(gmap))
582 if (start >= 1UL << 31)
583 /* We are only interested in prefix pages */
587 * Only new shadow blocks are added to the list during runtime,
588 * therefore we can safely reference them all the time.
590 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
591 page = READ_ONCE(kvm->arch.vsie.pages[i]);
594 cur = page_to_virt(page);
595 if (READ_ONCE(cur->gmap) != gmap)
597 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
598 /* with mso/msl, the prefix lies at an offset */
599 prefix += cur->scb_s.mso;
600 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
601 prefix_unmapped_sync(cur);
606 * Map the first prefix page and if tx is enabled also the second prefix page.
608 * The prefix will be protected, a gmap notifier will inform about unmaps.
609 * The shadow scb must not be executed until the prefix is remapped, this is
610 * guaranteed by properly handling PROG_REQUEST.
612 * Returns: - 0 on if successfully mapped or already mapped
613 * - > 0 if control has to be given to guest 2
614 * - -EAGAIN if the caller can retry immediately
615 * - -ENOMEM if out of memory
617 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
619 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
620 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
623 if (prefix_is_mapped(vsie_page))
626 /* mark it as mapped so we can catch any concurrent unmappers */
627 prefix_mapped(vsie_page);
629 /* with mso/msl, the prefix lies at offset *mso* */
630 prefix += scb_s->mso;
632 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix, NULL);
633 if (!rc && (scb_s->ecb & ECB_TE))
634 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
635 prefix + PAGE_SIZE, NULL);
637 * We don't have to mprotect, we will be called for all unshadows.
638 * SIE will detect if protection applies and trigger a validity.
641 prefix_unmapped(vsie_page);
642 if (rc > 0 || rc == -EFAULT)
643 rc = set_validity_icpt(scb_s, 0x0037U);
648 * Pin the guest page given by gpa and set hpa to the pinned host address.
649 * Will always be pinned writable.
651 * Returns: - 0 on success
652 * - -EINVAL if the gpa is not valid guest storage
654 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
658 page = gfn_to_page(kvm, gpa_to_gfn(gpa));
659 if (is_error_page(page))
661 *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK);
665 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
666 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
668 kvm_release_pfn_dirty(hpa >> PAGE_SHIFT);
669 /* mark the page always as dirty for migration */
670 mark_page_dirty(kvm, gpa_to_gfn(gpa));
673 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
674 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
676 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
679 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
681 unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa);
682 vsie_page->sca_gpa = 0;
689 unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa);
690 vsie_page->itdba_gpa = 0;
696 unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa);
697 vsie_page->gvrd_gpa = 0;
703 unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa);
704 vsie_page->riccbd_gpa = 0;
710 unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa);
711 vsie_page->sdnx_gpa = 0;
717 * Instead of shadowing some blocks, we can simply forward them because the
718 * addresses in the scb are 64 bit long.
720 * This works as long as the data lies in one page. If blocks ever exceed one
721 * page, we have to fall back to shadowing.
723 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
724 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
726 * Returns: - 0 if all blocks were pinned.
727 * - > 0 if control has to be given to guest 2
728 * - -ENOMEM if out of memory
730 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
732 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
733 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
738 gpa = READ_ONCE(scb_o->scaol) & ~0xfUL;
739 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
740 gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32;
742 if (gpa < 2 * PAGE_SIZE)
743 rc = set_validity_icpt(scb_s, 0x0038U);
744 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
745 rc = set_validity_icpt(scb_s, 0x0011U);
746 else if ((gpa & PAGE_MASK) !=
747 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
748 rc = set_validity_icpt(scb_s, 0x003bU);
750 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
752 rc = set_validity_icpt(scb_s, 0x0034U);
756 vsie_page->sca_gpa = gpa;
757 scb_s->scaoh = (u32)((u64)hpa >> 32);
758 scb_s->scaol = (u32)(u64)hpa;
761 gpa = READ_ONCE(scb_o->itdba) & ~0xffUL;
762 if (gpa && (scb_s->ecb & ECB_TE)) {
763 if (gpa < 2 * PAGE_SIZE) {
764 rc = set_validity_icpt(scb_s, 0x0080U);
767 /* 256 bytes cannot cross page boundaries */
768 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
770 rc = set_validity_icpt(scb_s, 0x0080U);
773 vsie_page->itdba_gpa = gpa;
777 gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL;
778 if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
779 if (gpa < 2 * PAGE_SIZE) {
780 rc = set_validity_icpt(scb_s, 0x1310U);
784 * 512 bytes vector registers cannot cross page boundaries
785 * if this block gets bigger, we have to shadow it.
787 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
789 rc = set_validity_icpt(scb_s, 0x1310U);
792 vsie_page->gvrd_gpa = gpa;
796 gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL;
797 if (gpa && (scb_s->ecb3 & ECB3_RI)) {
798 if (gpa < 2 * PAGE_SIZE) {
799 rc = set_validity_icpt(scb_s, 0x0043U);
802 /* 64 bytes cannot cross page boundaries */
803 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
805 rc = set_validity_icpt(scb_s, 0x0043U);
808 /* Validity 0x0044 will be checked by SIE */
809 vsie_page->riccbd_gpa = gpa;
812 if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) ||
813 (scb_s->ecd & ECD_ETOKENF)) {
816 gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL;
817 sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL;
818 if (!gpa || gpa < 2 * PAGE_SIZE) {
819 rc = set_validity_icpt(scb_s, 0x10b0U);
822 if (sdnxc < 6 || sdnxc > 12) {
823 rc = set_validity_icpt(scb_s, 0x10b1U);
826 if (gpa & ((1 << sdnxc) - 1)) {
827 rc = set_validity_icpt(scb_s, 0x10b2U);
830 /* Due to alignment rules (checked above) this cannot
831 * cross page boundaries
833 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
835 rc = set_validity_icpt(scb_s, 0x10b0U);
838 vsie_page->sdnx_gpa = gpa;
839 scb_s->sdnxo = hpa | sdnxc;
843 unpin_blocks(vcpu, vsie_page);
847 /* unpin the scb provided by guest 2, marking it as dirty */
848 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
851 hpa_t hpa = (hpa_t) vsie_page->scb_o;
854 unpin_guest_page(vcpu->kvm, gpa, hpa);
855 vsie_page->scb_o = NULL;
859 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
861 * Returns: - 0 if the scb was pinned.
862 * - > 0 if control has to be given to guest 2
864 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
870 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
872 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
876 vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
881 * Inject a fault into guest 2.
883 * Returns: - > 0 if control has to be given to guest 2
884 * < 0 if an error occurred during injection.
886 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
889 struct kvm_s390_pgm_info pgm = {
892 /* 0-51: virtual address */
893 (vaddr & 0xfffffffffffff000UL) |
894 /* 52-53: store / fetch */
895 (((unsigned int) !write_flag) + 1) << 10,
896 /* 62-63: asce id (alway primary == 0) */
897 .exc_access_id = 0, /* always primary */
898 .op_access_id = 0, /* not MVPG */
902 if (code == PGM_PROTECTION)
903 pgm.trans_exc_code |= 0x4UL;
905 rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
910 * Handle a fault during vsie execution on a gmap shadow.
912 * Returns: - 0 if the fault was resolved
913 * - > 0 if control has to be given to guest 2
914 * - < 0 if an error occurred
916 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
920 if (current->thread.gmap_int_code == PGM_PROTECTION)
921 /* we can directly forward all protection exceptions */
922 return inject_fault(vcpu, PGM_PROTECTION,
923 current->thread.gmap_addr, 1);
925 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
926 current->thread.gmap_addr, NULL);
928 rc = inject_fault(vcpu, rc,
929 current->thread.gmap_addr,
930 current->thread.gmap_write_flag);
932 vsie_page->fault_addr = current->thread.gmap_addr;
938 * Retry the previous fault that required guest 2 intervention. This avoids
939 * one superfluous SIE re-entry and direct exit.
941 * Will ignore any errors. The next SIE fault will do proper fault handling.
943 static void handle_last_fault(struct kvm_vcpu *vcpu,
944 struct vsie_page *vsie_page)
946 if (vsie_page->fault_addr)
947 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
948 vsie_page->fault_addr, NULL);
949 vsie_page->fault_addr = 0;
952 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
954 vsie_page->scb_s.icptcode = 0;
957 /* rewind the psw and clear the vsie icpt, so we can retry execution */
958 static void retry_vsie_icpt(struct vsie_page *vsie_page)
960 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
961 int ilen = insn_length(scb_s->ipa >> 8);
963 /* take care of EXECUTE instructions */
964 if (scb_s->icptstatus & 1) {
965 ilen = (scb_s->icptstatus >> 4) & 0x6;
969 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
970 clear_vsie_icpt(vsie_page);
974 * Try to shadow + enable the guest 2 provided facility list.
975 * Retry instruction execution if enabled for and provided by guest 2.
977 * Returns: - 0 if handled (retry or guest 2 icpt)
978 * - > 0 if control has to be given to guest 2
980 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
982 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
983 __u32 fac = READ_ONCE(vsie_page->scb_o->fac) & 0x7ffffff8U;
985 if (fac && test_kvm_facility(vcpu->kvm, 7)) {
986 retry_vsie_icpt(vsie_page);
987 if (read_guest_real(vcpu, fac, &vsie_page->fac,
988 sizeof(vsie_page->fac)))
989 return set_validity_icpt(scb_s, 0x1090U);
990 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
996 * Get a register for a nested guest.
997 * @vcpu the vcpu of the guest
998 * @vsie_page the vsie_page for the nested guest
999 * @reg the register number, the upper 4 bits are ignored.
1000 * returns: the value of the register.
1002 static u64 vsie_get_register(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, u8 reg)
1004 /* no need to validate the parameter and/or perform error handling */
1008 return vsie_page->scb_s.gg15;
1010 return vsie_page->scb_s.gg14;
1012 return vcpu->run->s.regs.gprs[reg];
1016 static int vsie_handle_mvpg(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1018 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1019 unsigned long pei_dest, pei_src, src, dest, mask, prefix;
1020 u64 *pei_block = &vsie_page->scb_o->mcic;
1021 int edat, rc_dest, rc_src;
1024 cr0.val = vcpu->arch.sie_block->gcr[0];
1025 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1026 mask = _kvm_s390_logical_to_effective(&scb_s->gpsw, PAGE_MASK);
1027 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
1029 dest = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 20) & mask;
1030 dest = _kvm_s390_real_to_abs(prefix, dest) + scb_s->mso;
1031 src = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 16) & mask;
1032 src = _kvm_s390_real_to_abs(prefix, src) + scb_s->mso;
1034 rc_dest = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, dest, &pei_dest);
1035 rc_src = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, src, &pei_src);
1037 * Either everything went well, or something non-critical went wrong
1038 * e.g. because of a race. In either case, simply retry.
1040 if (rc_dest == -EAGAIN || rc_src == -EAGAIN || (!rc_dest && !rc_src)) {
1041 retry_vsie_icpt(vsie_page);
1044 /* Something more serious went wrong, propagate the error */
1050 /* The only possible suppressing exception: just deliver it */
1051 if (rc_dest == PGM_TRANSLATION_SPEC || rc_src == PGM_TRANSLATION_SPEC) {
1052 clear_vsie_icpt(vsie_page);
1053 rc_dest = kvm_s390_inject_program_int(vcpu, PGM_TRANSLATION_SPEC);
1054 WARN_ON_ONCE(rc_dest);
1059 * Forward the PEI intercept to the guest if it was a page fault, or
1060 * also for segment and region table faults if EDAT applies.
1063 rc_dest = rc_dest == PGM_ASCE_TYPE ? rc_dest : 0;
1064 rc_src = rc_src == PGM_ASCE_TYPE ? rc_src : 0;
1066 rc_dest = rc_dest != PGM_PAGE_TRANSLATION ? rc_dest : 0;
1067 rc_src = rc_src != PGM_PAGE_TRANSLATION ? rc_src : 0;
1069 if (!rc_dest && !rc_src) {
1070 pei_block[0] = pei_dest;
1071 pei_block[1] = pei_src;
1075 retry_vsie_icpt(vsie_page);
1078 * The host has edat, and the guest does not, or it was an ASCE type
1079 * exception. The host needs to inject the appropriate DAT interrupts
1083 return inject_fault(vcpu, rc_dest, dest, 1);
1084 return inject_fault(vcpu, rc_src, src, 0);
1088 * Run the vsie on a shadow scb and a shadow gmap, without any further
1089 * sanity checks, handling SIE faults.
1091 * Returns: - 0 everything went fine
1092 * - > 0 if control has to be given to guest 2
1093 * - < 0 if an error occurred
1095 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1096 __releases(vcpu->kvm->srcu)
1097 __acquires(vcpu->kvm->srcu)
1099 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1100 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
1101 int guest_bp_isolation;
1104 handle_last_fault(vcpu, vsie_page);
1106 kvm_vcpu_srcu_read_unlock(vcpu);
1108 /* save current guest state of bp isolation override */
1109 guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
1112 * The guest is running with BPBC, so we have to force it on for our
1113 * nested guest. This is done by enabling BPBC globally, so the BPBC
1114 * control in the SCB (which the nested guest can modify) is simply
1117 if (test_kvm_facility(vcpu->kvm, 82) &&
1118 vcpu->arch.sie_block->fpf & FPF_BPBC)
1119 set_thread_flag(TIF_ISOLATE_BP_GUEST);
1121 local_irq_disable();
1122 guest_enter_irqoff();
1126 * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking
1127 * and VCPU requests also hinder the vSIE from running and lead
1128 * to an immediate exit. kvm_s390_vsie_kick() has to be used to
1129 * also kick the vSIE.
1131 vcpu->arch.sie_block->prog0c |= PROG_IN_SIE;
1133 if (test_cpu_flag(CIF_FPU))
1135 if (!kvm_s390_vcpu_sie_inhibited(vcpu))
1136 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
1138 vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE;
1140 local_irq_disable();
1141 guest_exit_irqoff();
1144 /* restore guest state for bp isolation override */
1145 if (!guest_bp_isolation)
1146 clear_thread_flag(TIF_ISOLATE_BP_GUEST);
1148 kvm_vcpu_srcu_read_lock(vcpu);
1151 VCPU_EVENT(vcpu, 3, "%s", "machine check");
1152 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
1157 rc = 0; /* we could still have an icpt */
1158 else if (rc == -EFAULT)
1159 return handle_fault(vcpu, vsie_page);
1161 switch (scb_s->icptcode) {
1163 if (scb_s->ipa == 0xb2b0)
1164 rc = handle_stfle(vcpu, vsie_page);
1167 /* stop not requested by g2 - must have been a kick */
1168 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
1169 clear_vsie_icpt(vsie_page);
1172 if ((scb_s->ipa & 0xf000) != 0xf000)
1173 scb_s->ipa += 0x1000;
1176 if (scb_s->ipa == 0xb254)
1177 rc = vsie_handle_mvpg(vcpu, vsie_page);
1183 static void release_gmap_shadow(struct vsie_page *vsie_page)
1185 if (vsie_page->gmap)
1186 gmap_put(vsie_page->gmap);
1187 WRITE_ONCE(vsie_page->gmap, NULL);
1188 prefix_unmapped(vsie_page);
1191 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
1192 struct vsie_page *vsie_page)
1199 asce = vcpu->arch.sie_block->gcr[1];
1200 cr0.val = vcpu->arch.sie_block->gcr[0];
1201 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1202 edat += edat && test_kvm_facility(vcpu->kvm, 78);
1205 * ASCE or EDAT could have changed since last icpt, or the gmap
1206 * we're holding has been unshadowed. If the gmap is still valid,
1207 * we can safely reuse it.
1209 if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat)) {
1210 vcpu->kvm->stat.gmap_shadow_reuse++;
1214 /* release the old shadow - if any, and mark the prefix as unmapped */
1215 release_gmap_shadow(vsie_page);
1216 gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
1218 return PTR_ERR(gmap);
1219 vcpu->kvm->stat.gmap_shadow_create++;
1220 WRITE_ONCE(vsie_page->gmap, gmap);
1225 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
1227 static void register_shadow_scb(struct kvm_vcpu *vcpu,
1228 struct vsie_page *vsie_page)
1230 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1232 WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
1234 * External calls have to lead to a kick of the vcpu and
1235 * therefore the vsie -> Simulate Wait state.
1237 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
1239 * We have to adjust the g3 epoch by the g2 epoch. The epoch will
1240 * automatically be adjusted on tod clock changes via kvm_sync_clock.
1243 scb_s->epoch += vcpu->kvm->arch.epoch;
1245 if (scb_s->ecd & ECD_MEF) {
1246 scb_s->epdx += vcpu->kvm->arch.epdx;
1247 if (scb_s->epoch < vcpu->kvm->arch.epoch)
1255 * Unregister a shadow scb from a VCPU.
1257 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
1259 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
1260 WRITE_ONCE(vcpu->arch.vsie_block, NULL);
1264 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
1265 * prefix pages and faults.
1267 * Returns: - 0 if no errors occurred
1268 * - > 0 if control has to be given to guest 2
1269 * - -ENOMEM if out of memory
1271 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1273 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1277 rc = acquire_gmap_shadow(vcpu, vsie_page);
1279 rc = map_prefix(vcpu, vsie_page);
1281 gmap_enable(vsie_page->gmap);
1282 update_intervention_requests(vsie_page);
1283 rc = do_vsie_run(vcpu, vsie_page);
1284 gmap_enable(vcpu->arch.gmap);
1286 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
1290 if (rc || scb_s->icptcode || signal_pending(current) ||
1291 kvm_s390_vcpu_has_irq(vcpu, 0) ||
1292 kvm_s390_vcpu_sie_inhibited(vcpu))
1297 if (rc == -EFAULT) {
1299 * Addressing exceptions are always presentes as intercepts.
1300 * As addressing exceptions are suppressing and our guest 3 PSW
1301 * points at the responsible instruction, we have to
1302 * forward the PSW and set the ilc. If we can't read guest 3
1303 * instruction, we can use an arbitrary ilc. Let's always use
1304 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
1305 * memory. (we could also fake the shadow so the hardware
1308 scb_s->icptcode = ICPT_PROGI;
1309 scb_s->iprcc = PGM_ADDRESSING;
1311 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
1318 * Get or create a vsie page for a scb address.
1320 * Returns: - address of a vsie page (cached or new one)
1321 * - NULL if the same scb address is already used by another VCPU
1322 * - ERR_PTR(-ENOMEM) if out of memory
1324 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
1326 struct vsie_page *vsie_page;
1331 page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
1334 if (page_ref_inc_return(page) == 2)
1335 return page_to_virt(page);
1340 * We want at least #online_vcpus shadows, so every VCPU can execute
1341 * the VSIE in parallel.
1343 nr_vcpus = atomic_read(&kvm->online_vcpus);
1345 mutex_lock(&kvm->arch.vsie.mutex);
1346 if (kvm->arch.vsie.page_count < nr_vcpus) {
1347 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO | GFP_DMA);
1349 mutex_unlock(&kvm->arch.vsie.mutex);
1350 return ERR_PTR(-ENOMEM);
1353 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1354 kvm->arch.vsie.page_count++;
1356 /* reuse an existing entry that belongs to nobody */
1358 page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1359 if (page_ref_inc_return(page) == 2)
1362 kvm->arch.vsie.next++;
1363 kvm->arch.vsie.next %= nr_vcpus;
1365 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1368 /* double use of the same address */
1369 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1371 mutex_unlock(&kvm->arch.vsie.mutex);
1374 mutex_unlock(&kvm->arch.vsie.mutex);
1376 vsie_page = page_to_virt(page);
1377 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1378 release_gmap_shadow(vsie_page);
1379 vsie_page->fault_addr = 0;
1380 vsie_page->scb_s.ihcpu = 0xffffU;
1384 /* put a vsie page acquired via get_vsie_page */
1385 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1387 struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1392 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1394 struct vsie_page *vsie_page;
1395 unsigned long scb_addr;
1398 vcpu->stat.instruction_sie++;
1399 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1401 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1402 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1404 BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1405 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1407 /* 512 byte alignment */
1408 if (unlikely(scb_addr & 0x1ffUL))
1409 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1411 if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0) ||
1412 kvm_s390_vcpu_sie_inhibited(vcpu))
1415 vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1416 if (IS_ERR(vsie_page))
1417 return PTR_ERR(vsie_page);
1418 else if (!vsie_page)
1419 /* double use of sie control block - simply do nothing */
1422 rc = pin_scb(vcpu, vsie_page, scb_addr);
1425 rc = shadow_scb(vcpu, vsie_page);
1428 rc = pin_blocks(vcpu, vsie_page);
1431 register_shadow_scb(vcpu, vsie_page);
1432 rc = vsie_run(vcpu, vsie_page);
1433 unregister_shadow_scb(vcpu);
1434 unpin_blocks(vcpu, vsie_page);
1436 unshadow_scb(vcpu, vsie_page);
1438 unpin_scb(vcpu, vsie_page, scb_addr);
1440 put_vsie_page(vcpu->kvm, vsie_page);
1442 return rc < 0 ? rc : 0;
1445 /* Init the vsie data structures. To be called when a vm is initialized. */
1446 void kvm_s390_vsie_init(struct kvm *kvm)
1448 mutex_init(&kvm->arch.vsie.mutex);
1449 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL_ACCOUNT);
1452 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1453 void kvm_s390_vsie_destroy(struct kvm *kvm)
1455 struct vsie_page *vsie_page;
1459 mutex_lock(&kvm->arch.vsie.mutex);
1460 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1461 page = kvm->arch.vsie.pages[i];
1462 kvm->arch.vsie.pages[i] = NULL;
1463 vsie_page = page_to_virt(page);
1464 release_gmap_shadow(vsie_page);
1465 /* free the radix tree entry */
1466 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1469 kvm->arch.vsie.page_count = 0;
1470 mutex_unlock(&kvm->arch.vsie.mutex);
1473 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1475 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1478 * Even if the VCPU lets go of the shadow sie block reference, it is
1479 * still valid in the cache. So we can safely kick it.
1482 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1483 if (scb->prog0c & PROG_IN_SIE)
1484 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);