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;
121 atomic_set(&scb_s->cpuflags, newflags);
126 * Create a shadow copy of the crycb block and setup key wrapping, if
127 * requested for guest 3 and enabled for guest 2.
129 * We only accept format-1 (no AP in g2), but convert it into format-2
130 * There is nothing to do for format-0.
132 * Returns: - 0 if shadowed or nothing to do
133 * - > 0 if control has to be given to guest 2
135 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
137 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
138 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
139 u32 crycb_addr = scb_o->crycbd & 0x7ffffff8U;
140 unsigned long *b1, *b2;
144 if (!(scb_o->crycbd & vcpu->arch.sie_block->crycbd & CRYCB_FORMAT1))
146 /* format-1 is supported with message-security-assist extension 3 */
147 if (!test_kvm_facility(vcpu->kvm, 76))
149 /* we may only allow it if enabled for guest 2 */
150 ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
151 (ECB3_AES | ECB3_DEA);
155 if ((crycb_addr & PAGE_MASK) != ((crycb_addr + 128) & PAGE_MASK))
156 return set_validity_icpt(scb_s, 0x003CU);
157 else if (!crycb_addr)
158 return set_validity_icpt(scb_s, 0x0039U);
160 /* copy only the wrapping keys */
161 if (read_guest_real(vcpu, crycb_addr + 72, &vsie_page->crycb, 56))
162 return set_validity_icpt(scb_s, 0x0035U);
164 scb_s->ecb3 |= ecb3_flags;
165 scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT1 |
168 /* xor both blocks in one run */
169 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
170 b2 = (unsigned long *)
171 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
172 /* as 56%8 == 0, bitmap_xor won't overwrite any data */
173 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
177 /* shadow (round up/down) the ibc to avoid validity icpt */
178 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
180 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
181 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
182 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
185 /* ibc installed in g2 and requested for g3 */
186 if (vcpu->kvm->arch.model.ibc && (scb_o->ibc & 0x0fffU)) {
187 scb_s->ibc = scb_o->ibc & 0x0fffU;
188 /* takte care of the minimum ibc level of the machine */
189 if (scb_s->ibc < min_ibc)
190 scb_s->ibc = min_ibc;
191 /* take care of the maximum ibc level set for the guest */
192 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
193 scb_s->ibc = vcpu->kvm->arch.model.ibc;
197 /* unshadow the scb, copying parameters back to the real scb */
198 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
200 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
201 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
204 scb_o->icptcode = scb_s->icptcode;
205 scb_o->icptstatus = scb_s->icptstatus;
206 scb_o->ipa = scb_s->ipa;
207 scb_o->ipb = scb_s->ipb;
208 scb_o->gbea = scb_s->gbea;
211 scb_o->cputm = scb_s->cputm;
212 scb_o->ckc = scb_s->ckc;
213 scb_o->todpr = scb_s->todpr;
216 scb_o->gpsw = scb_s->gpsw;
217 scb_o->gg14 = scb_s->gg14;
218 scb_o->gg15 = scb_s->gg15;
219 memcpy(scb_o->gcr, scb_s->gcr, 128);
220 scb_o->pp = scb_s->pp;
222 /* interrupt intercept */
223 switch (scb_s->icptcode) {
227 memcpy((void *)((u64)scb_o + 0xc0),
228 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
232 memcpy((void *)((u64)scb_o + 0xc0),
233 (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0);
237 if (scb_s->ihcpu != 0xffffU)
238 scb_o->ihcpu = scb_s->ihcpu;
242 * Setup the shadow scb by copying and checking the relevant parts of the g2
245 * Returns: - 0 if the scb has been shadowed
246 * - > 0 if control has to be given to guest 2
248 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
250 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
251 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
252 bool had_tx = scb_s->ecb & 0x10U;
253 unsigned long new_mso = 0;
256 /* make sure we don't have any leftovers when reusing the scb */
265 rc = prepare_cpuflags(vcpu, vsie_page);
270 scb_s->cputm = scb_o->cputm;
271 scb_s->ckc = scb_o->ckc;
272 scb_s->todpr = scb_o->todpr;
273 scb_s->epoch = scb_o->epoch;
276 scb_s->gpsw = scb_o->gpsw;
277 scb_s->gg14 = scb_o->gg14;
278 scb_s->gg15 = scb_o->gg15;
279 memcpy(scb_s->gcr, scb_o->gcr, 128);
280 scb_s->pp = scb_o->pp;
282 /* interception / execution handling */
283 scb_s->gbea = scb_o->gbea;
284 scb_s->lctl = scb_o->lctl;
285 scb_s->svcc = scb_o->svcc;
286 scb_s->ictl = scb_o->ictl;
288 * SKEY handling functions can't deal with false setting of PTE invalid
289 * bits. Therefore we cannot provide interpretation and would later
290 * have to provide own emulation handlers.
292 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
293 scb_s->icpua = scb_o->icpua;
295 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
296 new_mso = scb_o->mso & 0xfffffffffff00000UL;
297 /* if the hva of the prefix changes, we have to remap the prefix */
298 if (scb_s->mso != new_mso || scb_s->prefix != scb_o->prefix)
299 prefix_unmapped(vsie_page);
300 /* SIE will do mso/msl validity and exception checks for us */
301 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
302 scb_s->mso = new_mso;
303 scb_s->prefix = scb_o->prefix;
305 /* We have to definetly flush the tlb if this scb never ran */
306 if (scb_s->ihcpu != 0xffffU)
307 scb_s->ihcpu = scb_o->ihcpu;
309 /* MVPG and Protection Exception Interpretation are always available */
310 scb_s->eca |= scb_o->eca & 0x01002000U;
311 /* Host-protection-interruption introduced with ESOP */
312 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
313 scb_s->ecb |= scb_o->ecb & 0x02U;
314 /* transactional execution */
315 if (test_kvm_facility(vcpu->kvm, 73)) {
316 /* remap the prefix is tx is toggled on */
317 if ((scb_o->ecb & 0x10U) && !had_tx)
318 prefix_unmapped(vsie_page);
319 scb_s->ecb |= scb_o->ecb & 0x10U;
322 if (test_kvm_facility(vcpu->kvm, 129)) {
323 scb_s->eca |= scb_o->eca & 0x00020000U;
324 scb_s->ecd |= scb_o->ecd & 0x20000000U;
326 /* Run-time-Instrumentation */
327 if (test_kvm_facility(vcpu->kvm, 64))
328 scb_s->ecb3 |= scb_o->ecb3 & 0x01U;
329 /* Instruction Execution Prevention */
330 if (test_kvm_facility(vcpu->kvm, 130))
331 scb_s->ecb2 |= scb_o->ecb2 & 0x20U;
332 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
333 scb_s->eca |= scb_o->eca & 0x00000001U;
334 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
335 scb_s->eca |= scb_o->eca & 0x40000000U;
336 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
337 scb_s->eca |= scb_o->eca & 0x80000000U;
339 prepare_ibc(vcpu, vsie_page);
340 rc = shadow_crycb(vcpu, vsie_page);
343 unshadow_scb(vcpu, vsie_page);
347 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
350 struct kvm *kvm = gmap->private;
351 struct vsie_page *cur;
352 unsigned long prefix;
356 if (!gmap_is_shadow(gmap))
358 if (start >= 1UL << 31)
359 /* We are only interested in prefix pages */
363 * Only new shadow blocks are added to the list during runtime,
364 * therefore we can safely reference them all the time.
366 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
367 page = READ_ONCE(kvm->arch.vsie.pages[i]);
370 cur = page_to_virt(page);
371 if (READ_ONCE(cur->gmap) != gmap)
373 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
374 /* with mso/msl, the prefix lies at an offset */
375 prefix += cur->scb_s.mso;
376 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
377 prefix_unmapped_sync(cur);
382 * Map the first prefix page and if tx is enabled also the second prefix page.
384 * The prefix will be protected, a gmap notifier will inform about unmaps.
385 * The shadow scb must not be executed until the prefix is remapped, this is
386 * guaranteed by properly handling PROG_REQUEST.
388 * Returns: - 0 on if successfully mapped or already mapped
389 * - > 0 if control has to be given to guest 2
390 * - -EAGAIN if the caller can retry immediately
391 * - -ENOMEM if out of memory
393 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
395 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
396 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
399 if (prefix_is_mapped(vsie_page))
402 /* mark it as mapped so we can catch any concurrent unmappers */
403 prefix_mapped(vsie_page);
405 /* with mso/msl, the prefix lies at offset *mso* */
406 prefix += scb_s->mso;
408 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix);
409 if (!rc && (scb_s->ecb & 0x10U))
410 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
413 * We don't have to mprotect, we will be called for all unshadows.
414 * SIE will detect if protection applies and trigger a validity.
417 prefix_unmapped(vsie_page);
418 if (rc > 0 || rc == -EFAULT)
419 rc = set_validity_icpt(scb_s, 0x0037U);
424 * Pin the guest page given by gpa and set hpa to the pinned host address.
425 * Will always be pinned writable.
427 * Returns: - 0 on success
428 * - -EINVAL if the gpa is not valid guest storage
429 * - -ENOMEM if out of memory
431 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
437 hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
438 if (kvm_is_error_hva(hva))
440 rc = get_user_pages_fast(hva, 1, 1, &page);
445 *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK);
449 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
450 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
454 page = virt_to_page(hpa);
455 set_page_dirty_lock(page);
457 /* mark the page always as dirty for migration */
458 mark_page_dirty(kvm, gpa_to_gfn(gpa));
461 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
462 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
464 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
465 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
469 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
471 gpa = scb_o->scaol & ~0xfUL;
472 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
473 gpa |= (u64) scb_o->scaoh << 32;
474 unpin_guest_page(vcpu->kvm, gpa, hpa);
481 gpa = scb_o->itdba & ~0xffUL;
482 unpin_guest_page(vcpu->kvm, gpa, hpa);
488 gpa = scb_o->gvrd & ~0x1ffUL;
489 unpin_guest_page(vcpu->kvm, gpa, hpa);
495 gpa = scb_o->riccbd & ~0x3fUL;
496 unpin_guest_page(vcpu->kvm, gpa, hpa);
502 * Instead of shadowing some blocks, we can simply forward them because the
503 * addresses in the scb are 64 bit long.
505 * This works as long as the data lies in one page. If blocks ever exceed one
506 * page, we have to fall back to shadowing.
508 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
509 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
511 * Returns: - 0 if all blocks were pinned.
512 * - > 0 if control has to be given to guest 2
513 * - -ENOMEM if out of memory
515 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
517 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
518 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
523 gpa = scb_o->scaol & ~0xfUL;
524 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
525 gpa |= (u64) scb_o->scaoh << 32;
527 if (!(gpa & ~0x1fffUL))
528 rc = set_validity_icpt(scb_s, 0x0038U);
529 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
530 rc = set_validity_icpt(scb_s, 0x0011U);
531 else if ((gpa & PAGE_MASK) !=
532 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
533 rc = set_validity_icpt(scb_s, 0x003bU);
535 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
537 rc = set_validity_icpt(scb_s, 0x0034U);
541 scb_s->scaoh = (u32)((u64)hpa >> 32);
542 scb_s->scaol = (u32)(u64)hpa;
545 gpa = scb_o->itdba & ~0xffUL;
546 if (gpa && (scb_s->ecb & 0x10U)) {
547 if (!(gpa & ~0x1fffU)) {
548 rc = set_validity_icpt(scb_s, 0x0080U);
551 /* 256 bytes cannot cross page boundaries */
552 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
554 rc = set_validity_icpt(scb_s, 0x0080U);
560 gpa = scb_o->gvrd & ~0x1ffUL;
561 if (gpa && (scb_s->eca & 0x00020000U) &&
562 !(scb_s->ecd & 0x20000000U)) {
563 if (!(gpa & ~0x1fffUL)) {
564 rc = set_validity_icpt(scb_s, 0x1310U);
568 * 512 bytes vector registers cannot cross page boundaries
569 * if this block gets bigger, we have to shadow it.
571 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
573 rc = set_validity_icpt(scb_s, 0x1310U);
579 gpa = scb_o->riccbd & ~0x3fUL;
580 if (gpa && (scb_s->ecb3 & 0x01U)) {
581 if (!(gpa & ~0x1fffUL)) {
582 rc = set_validity_icpt(scb_s, 0x0043U);
585 /* 64 bytes cannot cross page boundaries */
586 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
588 rc = set_validity_icpt(scb_s, 0x0043U);
589 /* Validity 0x0044 will be checked by SIE */
596 unpin_blocks(vcpu, vsie_page);
600 /* unpin the scb provided by guest 2, marking it as dirty */
601 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
604 hpa_t hpa = (hpa_t) vsie_page->scb_o;
607 unpin_guest_page(vcpu->kvm, gpa, hpa);
608 vsie_page->scb_o = NULL;
612 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
614 * Returns: - 0 if the scb was pinned.
615 * - > 0 if control has to be given to guest 2
616 * - -ENOMEM if out of memory
618 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
624 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
626 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
631 vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
636 * Inject a fault into guest 2.
638 * Returns: - > 0 if control has to be given to guest 2
639 * < 0 if an error occurred during injection.
641 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
644 struct kvm_s390_pgm_info pgm = {
647 /* 0-51: virtual address */
648 (vaddr & 0xfffffffffffff000UL) |
649 /* 52-53: store / fetch */
650 (((unsigned int) !write_flag) + 1) << 10,
651 /* 62-63: asce id (alway primary == 0) */
652 .exc_access_id = 0, /* always primary */
653 .op_access_id = 0, /* not MVPG */
657 if (code == PGM_PROTECTION)
658 pgm.trans_exc_code |= 0x4UL;
660 rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
665 * Handle a fault during vsie execution on a gmap shadow.
667 * Returns: - 0 if the fault was resolved
668 * - > 0 if control has to be given to guest 2
669 * - < 0 if an error occurred
671 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
675 if (current->thread.gmap_int_code == PGM_PROTECTION)
676 /* we can directly forward all protection exceptions */
677 return inject_fault(vcpu, PGM_PROTECTION,
678 current->thread.gmap_addr, 1);
680 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
681 current->thread.gmap_addr);
683 rc = inject_fault(vcpu, rc,
684 current->thread.gmap_addr,
685 current->thread.gmap_write_flag);
687 vsie_page->fault_addr = current->thread.gmap_addr;
693 * Retry the previous fault that required guest 2 intervention. This avoids
694 * one superfluous SIE re-entry and direct exit.
696 * Will ignore any errors. The next SIE fault will do proper fault handling.
698 static void handle_last_fault(struct kvm_vcpu *vcpu,
699 struct vsie_page *vsie_page)
701 if (vsie_page->fault_addr)
702 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
703 vsie_page->fault_addr);
704 vsie_page->fault_addr = 0;
707 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
709 vsie_page->scb_s.icptcode = 0;
712 /* rewind the psw and clear the vsie icpt, so we can retry execution */
713 static void retry_vsie_icpt(struct vsie_page *vsie_page)
715 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
716 int ilen = insn_length(scb_s->ipa >> 8);
718 /* take care of EXECUTE instructions */
719 if (scb_s->icptstatus & 1) {
720 ilen = (scb_s->icptstatus >> 4) & 0x6;
724 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
725 clear_vsie_icpt(vsie_page);
729 * Try to shadow + enable the guest 2 provided facility list.
730 * Retry instruction execution if enabled for and provided by guest 2.
732 * Returns: - 0 if handled (retry or guest 2 icpt)
733 * - > 0 if control has to be given to guest 2
735 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
737 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
738 __u32 fac = vsie_page->scb_o->fac & 0x7ffffff8U;
740 if (fac && test_kvm_facility(vcpu->kvm, 7)) {
741 retry_vsie_icpt(vsie_page);
742 if (read_guest_real(vcpu, fac, &vsie_page->fac,
743 sizeof(vsie_page->fac)))
744 return set_validity_icpt(scb_s, 0x1090U);
745 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
751 * Run the vsie on a shadow scb and a shadow gmap, without any further
752 * sanity checks, handling SIE faults.
754 * Returns: - 0 everything went fine
755 * - > 0 if control has to be given to guest 2
756 * - < 0 if an error occurred
758 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
760 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
761 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
764 handle_last_fault(vcpu, vsie_page);
768 if (test_cpu_flag(CIF_MCCK_PENDING))
771 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
773 guest_enter_irqoff();
776 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
781 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
784 rc = 0; /* we could still have an icpt */
785 else if (rc == -EFAULT)
786 return handle_fault(vcpu, vsie_page);
788 switch (scb_s->icptcode) {
790 if (scb_s->ipa == 0xb2b0)
791 rc = handle_stfle(vcpu, vsie_page);
794 /* stop not requested by g2 - must have been a kick */
795 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
796 clear_vsie_icpt(vsie_page);
799 if ((scb_s->ipa & 0xf000) != 0xf000)
800 scb_s->ipa += 0x1000;
806 static void release_gmap_shadow(struct vsie_page *vsie_page)
809 gmap_put(vsie_page->gmap);
810 WRITE_ONCE(vsie_page->gmap, NULL);
811 prefix_unmapped(vsie_page);
814 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
815 struct vsie_page *vsie_page)
822 asce = vcpu->arch.sie_block->gcr[1];
823 cr0.val = vcpu->arch.sie_block->gcr[0];
824 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
825 edat += edat && test_kvm_facility(vcpu->kvm, 78);
828 * ASCE or EDAT could have changed since last icpt, or the gmap
829 * we're holding has been unshadowed. If the gmap is still valid,
830 * we can safely reuse it.
832 if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat))
835 /* release the old shadow - if any, and mark the prefix as unmapped */
836 release_gmap_shadow(vsie_page);
837 gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
839 return PTR_ERR(gmap);
840 gmap->private = vcpu->kvm;
841 WRITE_ONCE(vsie_page->gmap, gmap);
846 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
848 static void register_shadow_scb(struct kvm_vcpu *vcpu,
849 struct vsie_page *vsie_page)
851 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
853 WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
855 * External calls have to lead to a kick of the vcpu and
856 * therefore the vsie -> Simulate Wait state.
858 atomic_or(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
860 * We have to adjust the g3 epoch by the g2 epoch. The epoch will
861 * automatically be adjusted on tod clock changes via kvm_sync_clock.
864 scb_s->epoch += vcpu->kvm->arch.epoch;
869 * Unregister a shadow scb from a VCPU.
871 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
873 atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
874 WRITE_ONCE(vcpu->arch.vsie_block, NULL);
878 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
879 * prefix pages and faults.
881 * Returns: - 0 if no errors occurred
882 * - > 0 if control has to be given to guest 2
883 * - -ENOMEM if out of memory
885 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
887 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
891 rc = acquire_gmap_shadow(vcpu, vsie_page);
893 rc = map_prefix(vcpu, vsie_page);
895 gmap_enable(vsie_page->gmap);
896 update_intervention_requests(vsie_page);
897 rc = do_vsie_run(vcpu, vsie_page);
898 gmap_enable(vcpu->arch.gmap);
900 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
904 if (rc || scb_s->icptcode || signal_pending(current) ||
905 kvm_s390_vcpu_has_irq(vcpu, 0))
911 * Addressing exceptions are always presentes as intercepts.
912 * As addressing exceptions are suppressing and our guest 3 PSW
913 * points at the responsible instruction, we have to
914 * forward the PSW and set the ilc. If we can't read guest 3
915 * instruction, we can use an arbitrary ilc. Let's always use
916 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
917 * memory. (we could also fake the shadow so the hardware
920 scb_s->icptcode = ICPT_PROGI;
921 scb_s->iprcc = PGM_ADDRESSING;
923 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
929 * Get or create a vsie page for a scb address.
931 * Returns: - address of a vsie page (cached or new one)
932 * - NULL if the same scb address is already used by another VCPU
933 * - ERR_PTR(-ENOMEM) if out of memory
935 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
937 struct vsie_page *vsie_page;
942 page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
945 if (page_ref_inc_return(page) == 2)
946 return page_to_virt(page);
951 * We want at least #online_vcpus shadows, so every VCPU can execute
952 * the VSIE in parallel.
954 nr_vcpus = atomic_read(&kvm->online_vcpus);
956 mutex_lock(&kvm->arch.vsie.mutex);
957 if (kvm->arch.vsie.page_count < nr_vcpus) {
958 page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA);
960 mutex_unlock(&kvm->arch.vsie.mutex);
961 return ERR_PTR(-ENOMEM);
964 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
965 kvm->arch.vsie.page_count++;
967 /* reuse an existing entry that belongs to nobody */
969 page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
970 if (page_ref_inc_return(page) == 2)
973 kvm->arch.vsie.next++;
974 kvm->arch.vsie.next %= nr_vcpus;
976 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
979 /* double use of the same address */
980 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
982 mutex_unlock(&kvm->arch.vsie.mutex);
985 mutex_unlock(&kvm->arch.vsie.mutex);
987 vsie_page = page_to_virt(page);
988 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
989 release_gmap_shadow(vsie_page);
990 vsie_page->fault_addr = 0;
991 vsie_page->scb_s.ihcpu = 0xffffU;
995 /* put a vsie page acquired via get_vsie_page */
996 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
998 struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1003 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1005 struct vsie_page *vsie_page;
1006 unsigned long scb_addr;
1009 vcpu->stat.instruction_sie++;
1010 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1012 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1013 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1015 BUILD_BUG_ON(sizeof(struct vsie_page) != 4096);
1016 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1018 /* 512 byte alignment */
1019 if (unlikely(scb_addr & 0x1ffUL))
1020 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1022 if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0))
1025 vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1026 if (IS_ERR(vsie_page))
1027 return PTR_ERR(vsie_page);
1028 else if (!vsie_page)
1029 /* double use of sie control block - simply do nothing */
1032 rc = pin_scb(vcpu, vsie_page, scb_addr);
1035 rc = shadow_scb(vcpu, vsie_page);
1038 rc = pin_blocks(vcpu, vsie_page);
1041 register_shadow_scb(vcpu, vsie_page);
1042 rc = vsie_run(vcpu, vsie_page);
1043 unregister_shadow_scb(vcpu);
1044 unpin_blocks(vcpu, vsie_page);
1046 unshadow_scb(vcpu, vsie_page);
1048 unpin_scb(vcpu, vsie_page, scb_addr);
1050 put_vsie_page(vcpu->kvm, vsie_page);
1052 return rc < 0 ? rc : 0;
1055 /* Init the vsie data structures. To be called when a vm is initialized. */
1056 void kvm_s390_vsie_init(struct kvm *kvm)
1058 mutex_init(&kvm->arch.vsie.mutex);
1059 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL);
1062 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1063 void kvm_s390_vsie_destroy(struct kvm *kvm)
1065 struct vsie_page *vsie_page;
1069 mutex_lock(&kvm->arch.vsie.mutex);
1070 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1071 page = kvm->arch.vsie.pages[i];
1072 kvm->arch.vsie.pages[i] = NULL;
1073 vsie_page = page_to_virt(page);
1074 release_gmap_shadow(vsie_page);
1075 /* free the radix tree entry */
1076 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1079 kvm->arch.vsie.page_count = 0;
1080 mutex_unlock(&kvm->arch.vsie.mutex);
1083 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1085 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1088 * Even if the VCPU lets go of the shadow sie block reference, it is
1089 * still valid in the cache. So we can safely kick it.
1092 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1093 if (scb->prog0c & PROG_IN_SIE)
1094 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);