2 * kvm nested virtualization support for s390x
4 * Copyright IBM Corp. 2016
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): David Hildenbrand <dahi@linux.vnet.ibm.com>
12 #include <linux/vmalloc.h>
13 #include <linux/kvm_host.h>
14 #include <linux/bug.h>
15 #include <linux/list.h>
16 #include <linux/bitmap.h>
17 #include <linux/sched/signal.h>
20 #include <asm/mmu_context.h>
28 struct kvm_s390_sie_block scb_s; /* 0x0000 */
29 /* the pinned originial scb */
30 struct kvm_s390_sie_block *scb_o; /* 0x0200 */
31 /* the shadow gmap in use by the vsie_page */
32 struct gmap *gmap; /* 0x0208 */
33 /* address of the last reported fault to guest2 */
34 unsigned long fault_addr; /* 0x0210 */
35 __u8 reserved[0x0700 - 0x0218]; /* 0x0218 */
36 struct kvm_s390_crypto_cb crycb; /* 0x0700 */
37 __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */
40 /* trigger a validity icpt for the given scb */
41 static int set_validity_icpt(struct kvm_s390_sie_block *scb,
45 scb->ipb = ((__u32) reason_code) << 16;
46 scb->icptcode = ICPT_VALIDITY;
50 /* mark the prefix as unmapped, this will block the VSIE */
51 static void prefix_unmapped(struct vsie_page *vsie_page)
53 atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
56 /* mark the prefix as unmapped and wait until the VSIE has been left */
57 static void prefix_unmapped_sync(struct vsie_page *vsie_page)
59 prefix_unmapped(vsie_page);
60 if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
61 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
62 while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
66 /* mark the prefix as mapped, this will allow the VSIE to run */
67 static void prefix_mapped(struct vsie_page *vsie_page)
69 atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
72 /* test if the prefix is mapped into the gmap shadow */
73 static int prefix_is_mapped(struct vsie_page *vsie_page)
75 return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
78 /* copy the updated intervention request bits into the shadow scb */
79 static void update_intervention_requests(struct vsie_page *vsie_page)
81 const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
84 cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
85 atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
86 atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
89 /* shadow (filter and validate) the cpuflags */
90 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
92 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
93 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
94 int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
96 /* we don't allow ESA/390 guests */
97 if (!(cpuflags & CPUSTAT_ZARCH))
98 return set_validity_icpt(scb_s, 0x0001U);
100 if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
101 return set_validity_icpt(scb_s, 0x0001U);
102 else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
103 return set_validity_icpt(scb_s, 0x0007U);
105 /* intervention requests will be set later */
106 newflags = CPUSTAT_ZARCH;
107 if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
108 newflags |= CPUSTAT_GED;
109 if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
110 if (cpuflags & CPUSTAT_GED)
111 return set_validity_icpt(scb_s, 0x0001U);
112 newflags |= CPUSTAT_GED2;
114 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
115 newflags |= cpuflags & CPUSTAT_P;
116 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
117 newflags |= cpuflags & CPUSTAT_SM;
118 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
119 newflags |= cpuflags & CPUSTAT_IBS;
120 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS))
121 newflags |= cpuflags & CPUSTAT_KSS;
123 atomic_set(&scb_s->cpuflags, newflags);
128 * Create a shadow copy of the crycb block and setup key wrapping, if
129 * requested for guest 3 and enabled for guest 2.
131 * We only accept format-1 (no AP in g2), but convert it into format-2
132 * There is nothing to do for format-0.
134 * Returns: - 0 if shadowed or nothing to do
135 * - > 0 if control has to be given to guest 2
137 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
139 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
140 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
141 u32 crycb_addr = scb_o->crycbd & 0x7ffffff8U;
142 unsigned long *b1, *b2;
146 if (!(scb_o->crycbd & vcpu->arch.sie_block->crycbd & CRYCB_FORMAT1))
148 /* format-1 is supported with message-security-assist extension 3 */
149 if (!test_kvm_facility(vcpu->kvm, 76))
151 /* we may only allow it if enabled for guest 2 */
152 ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
153 (ECB3_AES | ECB3_DEA);
157 if ((crycb_addr & PAGE_MASK) != ((crycb_addr + 128) & PAGE_MASK))
158 return set_validity_icpt(scb_s, 0x003CU);
159 else if (!crycb_addr)
160 return set_validity_icpt(scb_s, 0x0039U);
162 /* copy only the wrapping keys */
163 if (read_guest_real(vcpu, crycb_addr + 72, &vsie_page->crycb, 56))
164 return set_validity_icpt(scb_s, 0x0035U);
166 scb_s->ecb3 |= ecb3_flags;
167 scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT1 |
170 /* xor both blocks in one run */
171 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
172 b2 = (unsigned long *)
173 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
174 /* as 56%8 == 0, bitmap_xor won't overwrite any data */
175 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
179 /* shadow (round up/down) the ibc to avoid validity icpt */
180 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
182 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
183 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
184 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
187 /* ibc installed in g2 and requested for g3 */
188 if (vcpu->kvm->arch.model.ibc && (scb_o->ibc & 0x0fffU)) {
189 scb_s->ibc = scb_o->ibc & 0x0fffU;
190 /* takte care of the minimum ibc level of the machine */
191 if (scb_s->ibc < min_ibc)
192 scb_s->ibc = min_ibc;
193 /* take care of the maximum ibc level set for the guest */
194 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
195 scb_s->ibc = vcpu->kvm->arch.model.ibc;
199 /* unshadow the scb, copying parameters back to the real scb */
200 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
202 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
203 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
206 scb_o->icptcode = scb_s->icptcode;
207 scb_o->icptstatus = scb_s->icptstatus;
208 scb_o->ipa = scb_s->ipa;
209 scb_o->ipb = scb_s->ipb;
210 scb_o->gbea = scb_s->gbea;
213 scb_o->cputm = scb_s->cputm;
214 scb_o->ckc = scb_s->ckc;
215 scb_o->todpr = scb_s->todpr;
218 scb_o->gpsw = scb_s->gpsw;
219 scb_o->gg14 = scb_s->gg14;
220 scb_o->gg15 = scb_s->gg15;
221 memcpy(scb_o->gcr, scb_s->gcr, 128);
222 scb_o->pp = scb_s->pp;
224 /* interrupt intercept */
225 switch (scb_s->icptcode) {
229 memcpy((void *)((u64)scb_o + 0xc0),
230 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
234 memcpy((void *)((u64)scb_o + 0xc0),
235 (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0);
239 if (scb_s->ihcpu != 0xffffU)
240 scb_o->ihcpu = scb_s->ihcpu;
244 * Setup the shadow scb by copying and checking the relevant parts of the g2
247 * Returns: - 0 if the scb has been shadowed
248 * - > 0 if control has to be given to guest 2
250 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
252 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
253 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
254 bool had_tx = scb_s->ecb & ECB_TE;
255 unsigned long new_mso = 0;
258 /* make sure we don't have any leftovers when reusing the scb */
267 rc = prepare_cpuflags(vcpu, vsie_page);
272 scb_s->cputm = scb_o->cputm;
273 scb_s->ckc = scb_o->ckc;
274 scb_s->todpr = scb_o->todpr;
275 scb_s->epoch = scb_o->epoch;
278 scb_s->gpsw = scb_o->gpsw;
279 scb_s->gg14 = scb_o->gg14;
280 scb_s->gg15 = scb_o->gg15;
281 memcpy(scb_s->gcr, scb_o->gcr, 128);
282 scb_s->pp = scb_o->pp;
284 /* interception / execution handling */
285 scb_s->gbea = scb_o->gbea;
286 scb_s->lctl = scb_o->lctl;
287 scb_s->svcc = scb_o->svcc;
288 scb_s->ictl = scb_o->ictl;
290 * SKEY handling functions can't deal with false setting of PTE invalid
291 * bits. Therefore we cannot provide interpretation and would later
292 * have to provide own emulation handlers.
294 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
295 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
297 scb_s->icpua = scb_o->icpua;
299 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
300 new_mso = scb_o->mso & 0xfffffffffff00000UL;
301 /* if the hva of the prefix changes, we have to remap the prefix */
302 if (scb_s->mso != new_mso || scb_s->prefix != scb_o->prefix)
303 prefix_unmapped(vsie_page);
304 /* SIE will do mso/msl validity and exception checks for us */
305 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
306 scb_s->mso = new_mso;
307 scb_s->prefix = scb_o->prefix;
309 /* We have to definetly flush the tlb if this scb never ran */
310 if (scb_s->ihcpu != 0xffffU)
311 scb_s->ihcpu = scb_o->ihcpu;
313 /* MVPG and Protection Exception Interpretation are always available */
314 scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
315 /* Host-protection-interruption introduced with ESOP */
316 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
317 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
318 /* transactional execution */
319 if (test_kvm_facility(vcpu->kvm, 73)) {
320 /* remap the prefix is tx is toggled on */
321 if ((scb_o->ecb & ECB_TE) && !had_tx)
322 prefix_unmapped(vsie_page);
323 scb_s->ecb |= scb_o->ecb & ECB_TE;
326 if (test_kvm_facility(vcpu->kvm, 129)) {
327 scb_s->eca |= scb_o->eca & ECA_VX;
328 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
330 /* Run-time-Instrumentation */
331 if (test_kvm_facility(vcpu->kvm, 64))
332 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
333 /* Instruction Execution Prevention */
334 if (test_kvm_facility(vcpu->kvm, 130))
335 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
336 /* Guarded Storage */
337 if (test_kvm_facility(vcpu->kvm, 133)) {
338 scb_s->ecb |= scb_o->ecb & ECB_GS;
339 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
341 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
342 scb_s->eca |= scb_o->eca & ECA_SII;
343 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
344 scb_s->eca |= scb_o->eca & ECA_IB;
345 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
346 scb_s->eca |= scb_o->eca & ECA_CEI;
348 prepare_ibc(vcpu, vsie_page);
349 rc = shadow_crycb(vcpu, vsie_page);
352 unshadow_scb(vcpu, vsie_page);
356 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
359 struct kvm *kvm = gmap->private;
360 struct vsie_page *cur;
361 unsigned long prefix;
365 if (!gmap_is_shadow(gmap))
367 if (start >= 1UL << 31)
368 /* We are only interested in prefix pages */
372 * Only new shadow blocks are added to the list during runtime,
373 * therefore we can safely reference them all the time.
375 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
376 page = READ_ONCE(kvm->arch.vsie.pages[i]);
379 cur = page_to_virt(page);
380 if (READ_ONCE(cur->gmap) != gmap)
382 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
383 /* with mso/msl, the prefix lies at an offset */
384 prefix += cur->scb_s.mso;
385 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
386 prefix_unmapped_sync(cur);
391 * Map the first prefix page and if tx is enabled also the second prefix page.
393 * The prefix will be protected, a gmap notifier will inform about unmaps.
394 * The shadow scb must not be executed until the prefix is remapped, this is
395 * guaranteed by properly handling PROG_REQUEST.
397 * Returns: - 0 on if successfully mapped or already mapped
398 * - > 0 if control has to be given to guest 2
399 * - -EAGAIN if the caller can retry immediately
400 * - -ENOMEM if out of memory
402 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
404 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
405 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
408 if (prefix_is_mapped(vsie_page))
411 /* mark it as mapped so we can catch any concurrent unmappers */
412 prefix_mapped(vsie_page);
414 /* with mso/msl, the prefix lies at offset *mso* */
415 prefix += scb_s->mso;
417 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix);
418 if (!rc && (scb_s->ecb & ECB_TE))
419 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
422 * We don't have to mprotect, we will be called for all unshadows.
423 * SIE will detect if protection applies and trigger a validity.
426 prefix_unmapped(vsie_page);
427 if (rc > 0 || rc == -EFAULT)
428 rc = set_validity_icpt(scb_s, 0x0037U);
433 * Pin the guest page given by gpa and set hpa to the pinned host address.
434 * Will always be pinned writable.
436 * Returns: - 0 on success
437 * - -EINVAL if the gpa is not valid guest storage
438 * - -ENOMEM if out of memory
440 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
446 hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
447 if (kvm_is_error_hva(hva))
449 rc = get_user_pages_fast(hva, 1, 1, &page);
454 *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK);
458 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
459 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
463 page = virt_to_page(hpa);
464 set_page_dirty_lock(page);
466 /* mark the page always as dirty for migration */
467 mark_page_dirty(kvm, gpa_to_gfn(gpa));
470 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
471 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
473 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
474 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
478 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
480 gpa = scb_o->scaol & ~0xfUL;
481 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
482 gpa |= (u64) scb_o->scaoh << 32;
483 unpin_guest_page(vcpu->kvm, gpa, hpa);
490 gpa = scb_o->itdba & ~0xffUL;
491 unpin_guest_page(vcpu->kvm, gpa, hpa);
497 gpa = scb_o->gvrd & ~0x1ffUL;
498 unpin_guest_page(vcpu->kvm, gpa, hpa);
504 gpa = scb_o->riccbd & ~0x3fUL;
505 unpin_guest_page(vcpu->kvm, gpa, hpa);
512 unpin_guest_page(vcpu->kvm, gpa, hpa);
518 * Instead of shadowing some blocks, we can simply forward them because the
519 * addresses in the scb are 64 bit long.
521 * This works as long as the data lies in one page. If blocks ever exceed one
522 * page, we have to fall back to shadowing.
524 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
525 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
527 * Returns: - 0 if all blocks were pinned.
528 * - > 0 if control has to be given to guest 2
529 * - -ENOMEM if out of memory
531 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
533 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
534 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
539 gpa = scb_o->scaol & ~0xfUL;
540 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
541 gpa |= (u64) scb_o->scaoh << 32;
543 if (!(gpa & ~0x1fffUL))
544 rc = set_validity_icpt(scb_s, 0x0038U);
545 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
546 rc = set_validity_icpt(scb_s, 0x0011U);
547 else if ((gpa & PAGE_MASK) !=
548 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
549 rc = set_validity_icpt(scb_s, 0x003bU);
551 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
553 rc = set_validity_icpt(scb_s, 0x0034U);
557 scb_s->scaoh = (u32)((u64)hpa >> 32);
558 scb_s->scaol = (u32)(u64)hpa;
561 gpa = scb_o->itdba & ~0xffUL;
562 if (gpa && (scb_s->ecb & ECB_TE)) {
563 if (!(gpa & ~0x1fffU)) {
564 rc = set_validity_icpt(scb_s, 0x0080U);
567 /* 256 bytes cannot cross page boundaries */
568 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
570 rc = set_validity_icpt(scb_s, 0x0080U);
576 gpa = scb_o->gvrd & ~0x1ffUL;
577 if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
578 if (!(gpa & ~0x1fffUL)) {
579 rc = set_validity_icpt(scb_s, 0x1310U);
583 * 512 bytes vector registers cannot cross page boundaries
584 * if this block gets bigger, we have to shadow it.
586 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
588 rc = set_validity_icpt(scb_s, 0x1310U);
594 gpa = scb_o->riccbd & ~0x3fUL;
595 if (gpa && (scb_s->ecb3 & ECB3_RI)) {
596 if (!(gpa & ~0x1fffUL)) {
597 rc = set_validity_icpt(scb_s, 0x0043U);
600 /* 64 bytes cannot cross page boundaries */
601 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
603 rc = set_validity_icpt(scb_s, 0x0043U);
604 /* Validity 0x0044 will be checked by SIE */
609 if ((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
612 gpa = scb_o->sdnxo & ~0xfUL;
613 sdnxc = scb_o->sdnxo & 0xfUL;
614 if (!gpa || !(gpa & ~0x1fffUL)) {
615 rc = set_validity_icpt(scb_s, 0x10b0U);
618 if (sdnxc < 6 || sdnxc > 12) {
619 rc = set_validity_icpt(scb_s, 0x10b1U);
622 if (gpa & ((1 << sdnxc) - 1)) {
623 rc = set_validity_icpt(scb_s, 0x10b2U);
626 /* Due to alignment rules (checked above) this cannot
627 * cross page boundaries
629 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
631 rc = set_validity_icpt(scb_s, 0x10b0U);
634 scb_s->sdnxo = hpa | sdnxc;
638 unpin_blocks(vcpu, vsie_page);
642 /* unpin the scb provided by guest 2, marking it as dirty */
643 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
646 hpa_t hpa = (hpa_t) vsie_page->scb_o;
649 unpin_guest_page(vcpu->kvm, gpa, hpa);
650 vsie_page->scb_o = NULL;
654 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
656 * Returns: - 0 if the scb was pinned.
657 * - > 0 if control has to be given to guest 2
658 * - -ENOMEM if out of memory
660 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
666 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
668 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
673 vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
678 * Inject a fault into guest 2.
680 * Returns: - > 0 if control has to be given to guest 2
681 * < 0 if an error occurred during injection.
683 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
686 struct kvm_s390_pgm_info pgm = {
689 /* 0-51: virtual address */
690 (vaddr & 0xfffffffffffff000UL) |
691 /* 52-53: store / fetch */
692 (((unsigned int) !write_flag) + 1) << 10,
693 /* 62-63: asce id (alway primary == 0) */
694 .exc_access_id = 0, /* always primary */
695 .op_access_id = 0, /* not MVPG */
699 if (code == PGM_PROTECTION)
700 pgm.trans_exc_code |= 0x4UL;
702 rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
707 * Handle a fault during vsie execution on a gmap shadow.
709 * Returns: - 0 if the fault was resolved
710 * - > 0 if control has to be given to guest 2
711 * - < 0 if an error occurred
713 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
717 if (current->thread.gmap_int_code == PGM_PROTECTION)
718 /* we can directly forward all protection exceptions */
719 return inject_fault(vcpu, PGM_PROTECTION,
720 current->thread.gmap_addr, 1);
722 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
723 current->thread.gmap_addr);
725 rc = inject_fault(vcpu, rc,
726 current->thread.gmap_addr,
727 current->thread.gmap_write_flag);
729 vsie_page->fault_addr = current->thread.gmap_addr;
735 * Retry the previous fault that required guest 2 intervention. This avoids
736 * one superfluous SIE re-entry and direct exit.
738 * Will ignore any errors. The next SIE fault will do proper fault handling.
740 static void handle_last_fault(struct kvm_vcpu *vcpu,
741 struct vsie_page *vsie_page)
743 if (vsie_page->fault_addr)
744 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
745 vsie_page->fault_addr);
746 vsie_page->fault_addr = 0;
749 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
751 vsie_page->scb_s.icptcode = 0;
754 /* rewind the psw and clear the vsie icpt, so we can retry execution */
755 static void retry_vsie_icpt(struct vsie_page *vsie_page)
757 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
758 int ilen = insn_length(scb_s->ipa >> 8);
760 /* take care of EXECUTE instructions */
761 if (scb_s->icptstatus & 1) {
762 ilen = (scb_s->icptstatus >> 4) & 0x6;
766 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
767 clear_vsie_icpt(vsie_page);
771 * Try to shadow + enable the guest 2 provided facility list.
772 * Retry instruction execution if enabled for and provided by guest 2.
774 * Returns: - 0 if handled (retry or guest 2 icpt)
775 * - > 0 if control has to be given to guest 2
777 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
779 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
780 __u32 fac = vsie_page->scb_o->fac & 0x7ffffff8U;
782 if (fac && test_kvm_facility(vcpu->kvm, 7)) {
783 retry_vsie_icpt(vsie_page);
784 if (read_guest_real(vcpu, fac, &vsie_page->fac,
785 sizeof(vsie_page->fac)))
786 return set_validity_icpt(scb_s, 0x1090U);
787 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
793 * Run the vsie on a shadow scb and a shadow gmap, without any further
794 * sanity checks, handling SIE faults.
796 * Returns: - 0 everything went fine
797 * - > 0 if control has to be given to guest 2
798 * - < 0 if an error occurred
800 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
802 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
803 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
806 handle_last_fault(vcpu, vsie_page);
810 if (test_cpu_flag(CIF_MCCK_PENDING))
813 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
815 guest_enter_irqoff();
818 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
823 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
826 rc = 0; /* we could still have an icpt */
827 else if (rc == -EFAULT)
828 return handle_fault(vcpu, vsie_page);
830 switch (scb_s->icptcode) {
832 if (scb_s->ipa == 0xb2b0)
833 rc = handle_stfle(vcpu, vsie_page);
836 /* stop not requested by g2 - must have been a kick */
837 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
838 clear_vsie_icpt(vsie_page);
841 if ((scb_s->ipa & 0xf000) != 0xf000)
842 scb_s->ipa += 0x1000;
848 static void release_gmap_shadow(struct vsie_page *vsie_page)
851 gmap_put(vsie_page->gmap);
852 WRITE_ONCE(vsie_page->gmap, NULL);
853 prefix_unmapped(vsie_page);
856 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
857 struct vsie_page *vsie_page)
864 asce = vcpu->arch.sie_block->gcr[1];
865 cr0.val = vcpu->arch.sie_block->gcr[0];
866 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
867 edat += edat && test_kvm_facility(vcpu->kvm, 78);
870 * ASCE or EDAT could have changed since last icpt, or the gmap
871 * we're holding has been unshadowed. If the gmap is still valid,
872 * we can safely reuse it.
874 if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat))
877 /* release the old shadow - if any, and mark the prefix as unmapped */
878 release_gmap_shadow(vsie_page);
879 gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
881 return PTR_ERR(gmap);
882 gmap->private = vcpu->kvm;
883 WRITE_ONCE(vsie_page->gmap, gmap);
888 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
890 static void register_shadow_scb(struct kvm_vcpu *vcpu,
891 struct vsie_page *vsie_page)
893 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
895 WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
897 * External calls have to lead to a kick of the vcpu and
898 * therefore the vsie -> Simulate Wait state.
900 atomic_or(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
902 * We have to adjust the g3 epoch by the g2 epoch. The epoch will
903 * automatically be adjusted on tod clock changes via kvm_sync_clock.
906 scb_s->epoch += vcpu->kvm->arch.epoch;
911 * Unregister a shadow scb from a VCPU.
913 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
915 atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
916 WRITE_ONCE(vcpu->arch.vsie_block, NULL);
920 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
921 * prefix pages and faults.
923 * Returns: - 0 if no errors occurred
924 * - > 0 if control has to be given to guest 2
925 * - -ENOMEM if out of memory
927 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
929 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
933 rc = acquire_gmap_shadow(vcpu, vsie_page);
935 rc = map_prefix(vcpu, vsie_page);
937 gmap_enable(vsie_page->gmap);
938 update_intervention_requests(vsie_page);
939 rc = do_vsie_run(vcpu, vsie_page);
940 gmap_enable(vcpu->arch.gmap);
942 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
946 if (rc || scb_s->icptcode || signal_pending(current) ||
947 kvm_s390_vcpu_has_irq(vcpu, 0))
953 * Addressing exceptions are always presentes as intercepts.
954 * As addressing exceptions are suppressing and our guest 3 PSW
955 * points at the responsible instruction, we have to
956 * forward the PSW and set the ilc. If we can't read guest 3
957 * instruction, we can use an arbitrary ilc. Let's always use
958 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
959 * memory. (we could also fake the shadow so the hardware
962 scb_s->icptcode = ICPT_PROGI;
963 scb_s->iprcc = PGM_ADDRESSING;
965 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
971 * Get or create a vsie page for a scb address.
973 * Returns: - address of a vsie page (cached or new one)
974 * - NULL if the same scb address is already used by another VCPU
975 * - ERR_PTR(-ENOMEM) if out of memory
977 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
979 struct vsie_page *vsie_page;
984 page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
987 if (page_ref_inc_return(page) == 2)
988 return page_to_virt(page);
993 * We want at least #online_vcpus shadows, so every VCPU can execute
994 * the VSIE in parallel.
996 nr_vcpus = atomic_read(&kvm->online_vcpus);
998 mutex_lock(&kvm->arch.vsie.mutex);
999 if (kvm->arch.vsie.page_count < nr_vcpus) {
1000 page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA);
1002 mutex_unlock(&kvm->arch.vsie.mutex);
1003 return ERR_PTR(-ENOMEM);
1006 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1007 kvm->arch.vsie.page_count++;
1009 /* reuse an existing entry that belongs to nobody */
1011 page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1012 if (page_ref_inc_return(page) == 2)
1015 kvm->arch.vsie.next++;
1016 kvm->arch.vsie.next %= nr_vcpus;
1018 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1021 /* double use of the same address */
1022 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1024 mutex_unlock(&kvm->arch.vsie.mutex);
1027 mutex_unlock(&kvm->arch.vsie.mutex);
1029 vsie_page = page_to_virt(page);
1030 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1031 release_gmap_shadow(vsie_page);
1032 vsie_page->fault_addr = 0;
1033 vsie_page->scb_s.ihcpu = 0xffffU;
1037 /* put a vsie page acquired via get_vsie_page */
1038 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1040 struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1045 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1047 struct vsie_page *vsie_page;
1048 unsigned long scb_addr;
1051 vcpu->stat.instruction_sie++;
1052 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1054 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1055 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1057 BUILD_BUG_ON(sizeof(struct vsie_page) != 4096);
1058 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1060 /* 512 byte alignment */
1061 if (unlikely(scb_addr & 0x1ffUL))
1062 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1064 if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0))
1067 vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1068 if (IS_ERR(vsie_page))
1069 return PTR_ERR(vsie_page);
1070 else if (!vsie_page)
1071 /* double use of sie control block - simply do nothing */
1074 rc = pin_scb(vcpu, vsie_page, scb_addr);
1077 rc = shadow_scb(vcpu, vsie_page);
1080 rc = pin_blocks(vcpu, vsie_page);
1083 register_shadow_scb(vcpu, vsie_page);
1084 rc = vsie_run(vcpu, vsie_page);
1085 unregister_shadow_scb(vcpu);
1086 unpin_blocks(vcpu, vsie_page);
1088 unshadow_scb(vcpu, vsie_page);
1090 unpin_scb(vcpu, vsie_page, scb_addr);
1092 put_vsie_page(vcpu->kvm, vsie_page);
1094 return rc < 0 ? rc : 0;
1097 /* Init the vsie data structures. To be called when a vm is initialized. */
1098 void kvm_s390_vsie_init(struct kvm *kvm)
1100 mutex_init(&kvm->arch.vsie.mutex);
1101 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL);
1104 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1105 void kvm_s390_vsie_destroy(struct kvm *kvm)
1107 struct vsie_page *vsie_page;
1111 mutex_lock(&kvm->arch.vsie.mutex);
1112 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1113 page = kvm->arch.vsie.pages[i];
1114 kvm->arch.vsie.pages[i] = NULL;
1115 vsie_page = page_to_virt(page);
1116 release_gmap_shadow(vsie_page);
1117 /* free the radix tree entry */
1118 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1121 kvm->arch.vsie.page_count = 0;
1122 mutex_unlock(&kvm->arch.vsie.mutex);
1125 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1127 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1130 * Even if the VCPU lets go of the shadow sie block reference, it is
1131 * still valid in the cache. So we can safely kick it.
1134 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1135 if (scb->prog0c & PROG_IN_SIE)
1136 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);