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>
18 #include <asm/mmu_context.h>
26 struct kvm_s390_sie_block scb_s; /* 0x0000 */
27 /* the pinned originial scb */
28 struct kvm_s390_sie_block *scb_o; /* 0x0200 */
29 /* the shadow gmap in use by the vsie_page */
30 struct gmap *gmap; /* 0x0208 */
31 /* address of the last reported fault to guest2 */
32 unsigned long fault_addr; /* 0x0210 */
33 __u8 reserved[0x0700 - 0x0218]; /* 0x0218 */
34 struct kvm_s390_crypto_cb crycb; /* 0x0700 */
35 __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */
38 /* trigger a validity icpt for the given scb */
39 static int set_validity_icpt(struct kvm_s390_sie_block *scb,
43 scb->ipb = ((__u32) reason_code) << 16;
44 scb->icptcode = ICPT_VALIDITY;
48 /* mark the prefix as unmapped, this will block the VSIE */
49 static void prefix_unmapped(struct vsie_page *vsie_page)
51 atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
54 /* mark the prefix as unmapped and wait until the VSIE has been left */
55 static void prefix_unmapped_sync(struct vsie_page *vsie_page)
57 prefix_unmapped(vsie_page);
58 if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
59 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
60 while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
64 /* mark the prefix as mapped, this will allow the VSIE to run */
65 static void prefix_mapped(struct vsie_page *vsie_page)
67 atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
70 /* test if the prefix is mapped into the gmap shadow */
71 static int prefix_is_mapped(struct vsie_page *vsie_page)
73 return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
76 /* copy the updated intervention request bits into the shadow scb */
77 static void update_intervention_requests(struct vsie_page *vsie_page)
79 const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
82 cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
83 atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
84 atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
87 /* shadow (filter and validate) the cpuflags */
88 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
90 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
91 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
92 int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
94 /* we don't allow ESA/390 guests */
95 if (!(cpuflags & CPUSTAT_ZARCH))
96 return set_validity_icpt(scb_s, 0x0001U);
98 if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
99 return set_validity_icpt(scb_s, 0x0001U);
100 else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
101 return set_validity_icpt(scb_s, 0x0007U);
103 /* intervention requests will be set later */
104 newflags = CPUSTAT_ZARCH;
105 if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
106 newflags |= CPUSTAT_GED;
107 if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
108 if (cpuflags & CPUSTAT_GED)
109 return set_validity_icpt(scb_s, 0x0001U);
110 newflags |= CPUSTAT_GED2;
112 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
113 newflags |= cpuflags & CPUSTAT_P;
114 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
115 newflags |= cpuflags & CPUSTAT_SM;
116 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
117 newflags |= cpuflags & CPUSTAT_IBS;
119 atomic_set(&scb_s->cpuflags, newflags);
124 * Create a shadow copy of the crycb block and setup key wrapping, if
125 * requested for guest 3 and enabled for guest 2.
127 * We only accept format-1 (no AP in g2), but convert it into format-2
128 * There is nothing to do for format-0.
130 * Returns: - 0 if shadowed or nothing to do
131 * - > 0 if control has to be given to guest 2
133 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
135 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
136 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
137 u32 crycb_addr = scb_o->crycbd & 0x7ffffff8U;
138 unsigned long *b1, *b2;
142 if (!(scb_o->crycbd & vcpu->arch.sie_block->crycbd & CRYCB_FORMAT1))
144 /* format-1 is supported with message-security-assist extension 3 */
145 if (!test_kvm_facility(vcpu->kvm, 76))
147 /* we may only allow it if enabled for guest 2 */
148 ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
149 (ECB3_AES | ECB3_DEA);
153 if ((crycb_addr & PAGE_MASK) != ((crycb_addr + 128) & PAGE_MASK))
154 return set_validity_icpt(scb_s, 0x003CU);
155 else if (!crycb_addr)
156 return set_validity_icpt(scb_s, 0x0039U);
158 /* copy only the wrapping keys */
159 if (read_guest_real(vcpu, crycb_addr + 72,
160 vsie_page->crycb.dea_wrapping_key_mask, 56))
161 return set_validity_icpt(scb_s, 0x0035U);
163 scb_s->ecb3 |= ecb3_flags;
164 scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT1 |
167 /* xor both blocks in one run */
168 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
169 b2 = (unsigned long *)
170 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
171 /* as 56%8 == 0, bitmap_xor won't overwrite any data */
172 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
176 /* shadow (round up/down) the ibc to avoid validity icpt */
177 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
179 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
180 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
181 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
184 /* ibc installed in g2 and requested for g3 */
185 if (vcpu->kvm->arch.model.ibc && (scb_o->ibc & 0x0fffU)) {
186 scb_s->ibc = scb_o->ibc & 0x0fffU;
187 /* takte care of the minimum ibc level of the machine */
188 if (scb_s->ibc < min_ibc)
189 scb_s->ibc = min_ibc;
190 /* take care of the maximum ibc level set for the guest */
191 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
192 scb_s->ibc = vcpu->kvm->arch.model.ibc;
196 /* unshadow the scb, copying parameters back to the real scb */
197 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
199 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
200 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
203 scb_o->icptcode = scb_s->icptcode;
204 scb_o->icptstatus = scb_s->icptstatus;
205 scb_o->ipa = scb_s->ipa;
206 scb_o->ipb = scb_s->ipb;
207 scb_o->gbea = scb_s->gbea;
210 scb_o->cputm = scb_s->cputm;
211 scb_o->ckc = scb_s->ckc;
212 scb_o->todpr = scb_s->todpr;
215 scb_o->gpsw = scb_s->gpsw;
216 scb_o->gg14 = scb_s->gg14;
217 scb_o->gg15 = scb_s->gg15;
218 memcpy(scb_o->gcr, scb_s->gcr, 128);
219 scb_o->pp = scb_s->pp;
221 /* branch prediction */
222 if (test_kvm_facility(vcpu->kvm, 82)) {
223 scb_o->fpf &= ~FPF_BPBC;
224 scb_o->fpf |= scb_s->fpf & FPF_BPBC;
227 /* interrupt intercept */
228 switch (scb_s->icptcode) {
232 memcpy((void *)((u64)scb_o + 0xc0),
233 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
237 memcpy((void *)((u64)scb_o + 0xc0),
238 (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0);
242 if (scb_s->ihcpu != 0xffffU)
243 scb_o->ihcpu = scb_s->ihcpu;
247 * Setup the shadow scb by copying and checking the relevant parts of the g2
250 * Returns: - 0 if the scb has been shadowed
251 * - > 0 if control has to be given to guest 2
253 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
255 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
256 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
257 bool had_tx = scb_s->ecb & 0x10U;
258 unsigned long new_mso = 0;
261 /* make sure we don't have any leftovers when reusing the scb */
271 rc = prepare_cpuflags(vcpu, vsie_page);
276 scb_s->cputm = scb_o->cputm;
277 scb_s->ckc = scb_o->ckc;
278 scb_s->todpr = scb_o->todpr;
279 scb_s->epoch = scb_o->epoch;
282 scb_s->gpsw = scb_o->gpsw;
283 scb_s->gg14 = scb_o->gg14;
284 scb_s->gg15 = scb_o->gg15;
285 memcpy(scb_s->gcr, scb_o->gcr, 128);
286 scb_s->pp = scb_o->pp;
288 /* interception / execution handling */
289 scb_s->gbea = scb_o->gbea;
290 scb_s->lctl = scb_o->lctl;
291 scb_s->svcc = scb_o->svcc;
292 scb_s->ictl = scb_o->ictl;
294 * SKEY handling functions can't deal with false setting of PTE invalid
295 * bits. Therefore we cannot provide interpretation and would later
296 * have to provide own emulation handlers.
298 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
299 scb_s->icpua = scb_o->icpua;
301 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
302 new_mso = scb_o->mso & 0xfffffffffff00000UL;
303 /* if the hva of the prefix changes, we have to remap the prefix */
304 if (scb_s->mso != new_mso || scb_s->prefix != scb_o->prefix)
305 prefix_unmapped(vsie_page);
306 /* SIE will do mso/msl validity and exception checks for us */
307 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
308 scb_s->mso = new_mso;
309 scb_s->prefix = scb_o->prefix;
311 /* We have to definetly flush the tlb if this scb never ran */
312 if (scb_s->ihcpu != 0xffffU)
313 scb_s->ihcpu = scb_o->ihcpu;
315 /* MVPG and Protection Exception Interpretation are always available */
316 scb_s->eca |= scb_o->eca & 0x01002000U;
317 /* Host-protection-interruption introduced with ESOP */
318 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
319 scb_s->ecb |= scb_o->ecb & 0x02U;
320 /* transactional execution */
321 if (test_kvm_facility(vcpu->kvm, 73)) {
322 /* remap the prefix is tx is toggled on */
323 if ((scb_o->ecb & 0x10U) && !had_tx)
324 prefix_unmapped(vsie_page);
325 scb_s->ecb |= scb_o->ecb & 0x10U;
327 /* branch prediction */
328 if (test_kvm_facility(vcpu->kvm, 82))
329 scb_s->fpf |= scb_o->fpf & FPF_BPBC;
331 if (test_kvm_facility(vcpu->kvm, 129)) {
332 scb_s->eca |= scb_o->eca & 0x00020000U;
333 scb_s->ecd |= scb_o->ecd & 0x20000000U;
335 /* Run-time-Instrumentation */
336 if (test_kvm_facility(vcpu->kvm, 64))
337 scb_s->ecb3 |= scb_o->ecb3 & 0x01U;
338 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
339 scb_s->eca |= scb_o->eca & 0x00000001U;
340 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
341 scb_s->eca |= scb_o->eca & 0x40000000U;
342 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
343 scb_s->eca |= scb_o->eca & 0x80000000U;
345 prepare_ibc(vcpu, vsie_page);
346 rc = shadow_crycb(vcpu, vsie_page);
349 unshadow_scb(vcpu, vsie_page);
353 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
356 struct kvm *kvm = gmap->private;
357 struct vsie_page *cur;
358 unsigned long prefix;
362 if (!gmap_is_shadow(gmap))
364 if (start >= 1UL << 31)
365 /* We are only interested in prefix pages */
369 * Only new shadow blocks are added to the list during runtime,
370 * therefore we can safely reference them all the time.
372 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
373 page = READ_ONCE(kvm->arch.vsie.pages[i]);
376 cur = page_to_virt(page);
377 if (READ_ONCE(cur->gmap) != gmap)
379 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
380 /* with mso/msl, the prefix lies at an offset */
381 prefix += cur->scb_s.mso;
382 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
383 prefix_unmapped_sync(cur);
388 * Map the first prefix page and if tx is enabled also the second prefix page.
390 * The prefix will be protected, a gmap notifier will inform about unmaps.
391 * The shadow scb must not be executed until the prefix is remapped, this is
392 * guaranteed by properly handling PROG_REQUEST.
394 * Returns: - 0 on if successfully mapped or already mapped
395 * - > 0 if control has to be given to guest 2
396 * - -EAGAIN if the caller can retry immediately
397 * - -ENOMEM if out of memory
399 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
401 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
402 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
405 if (prefix_is_mapped(vsie_page))
408 /* mark it as mapped so we can catch any concurrent unmappers */
409 prefix_mapped(vsie_page);
411 /* with mso/msl, the prefix lies at offset *mso* */
412 prefix += scb_s->mso;
414 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix);
415 if (!rc && (scb_s->ecb & 0x10U))
416 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
419 * We don't have to mprotect, we will be called for all unshadows.
420 * SIE will detect if protection applies and trigger a validity.
423 prefix_unmapped(vsie_page);
424 if (rc > 0 || rc == -EFAULT)
425 rc = set_validity_icpt(scb_s, 0x0037U);
430 * Pin the guest page given by gpa and set hpa to the pinned host address.
431 * Will always be pinned writable.
433 * Returns: - 0 on success
434 * - -EINVAL if the gpa is not valid guest storage
435 * - -ENOMEM if out of memory
437 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
443 hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
444 if (kvm_is_error_hva(hva))
446 rc = get_user_pages_fast(hva, 1, 1, &page);
451 *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK);
455 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
456 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
460 page = virt_to_page(hpa);
461 set_page_dirty_lock(page);
463 /* mark the page always as dirty for migration */
464 mark_page_dirty(kvm, gpa_to_gfn(gpa));
467 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
468 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
470 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
471 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
475 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
477 gpa = scb_o->scaol & ~0xfUL;
478 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
479 gpa |= (u64) scb_o->scaoh << 32;
480 unpin_guest_page(vcpu->kvm, gpa, hpa);
487 gpa = scb_o->itdba & ~0xffUL;
488 unpin_guest_page(vcpu->kvm, gpa, hpa);
494 gpa = scb_o->gvrd & ~0x1ffUL;
495 unpin_guest_page(vcpu->kvm, gpa, hpa);
501 gpa = scb_o->riccbd & ~0x3fUL;
502 unpin_guest_page(vcpu->kvm, gpa, hpa);
508 * Instead of shadowing some blocks, we can simply forward them because the
509 * addresses in the scb are 64 bit long.
511 * This works as long as the data lies in one page. If blocks ever exceed one
512 * page, we have to fall back to shadowing.
514 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
515 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
517 * Returns: - 0 if all blocks were pinned.
518 * - > 0 if control has to be given to guest 2
519 * - -ENOMEM if out of memory
521 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
523 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
524 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
529 gpa = scb_o->scaol & ~0xfUL;
530 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
531 gpa |= (u64) scb_o->scaoh << 32;
533 if (!(gpa & ~0x1fffUL))
534 rc = set_validity_icpt(scb_s, 0x0038U);
535 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
536 rc = set_validity_icpt(scb_s, 0x0011U);
537 else if ((gpa & PAGE_MASK) !=
538 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
539 rc = set_validity_icpt(scb_s, 0x003bU);
541 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
543 rc = set_validity_icpt(scb_s, 0x0034U);
547 scb_s->scaoh = (u32)((u64)hpa >> 32);
548 scb_s->scaol = (u32)(u64)hpa;
551 gpa = scb_o->itdba & ~0xffUL;
552 if (gpa && (scb_s->ecb & 0x10U)) {
553 if (!(gpa & ~0x1fffUL)) {
554 rc = set_validity_icpt(scb_s, 0x0080U);
557 /* 256 bytes cannot cross page boundaries */
558 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
560 rc = set_validity_icpt(scb_s, 0x0080U);
566 gpa = scb_o->gvrd & ~0x1ffUL;
567 if (gpa && (scb_s->eca & 0x00020000U) &&
568 !(scb_s->ecd & 0x20000000U)) {
569 if (!(gpa & ~0x1fffUL)) {
570 rc = set_validity_icpt(scb_s, 0x1310U);
574 * 512 bytes vector registers cannot cross page boundaries
575 * if this block gets bigger, we have to shadow it.
577 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
579 rc = set_validity_icpt(scb_s, 0x1310U);
585 gpa = scb_o->riccbd & ~0x3fUL;
586 if (gpa && (scb_s->ecb3 & 0x01U)) {
587 if (!(gpa & ~0x1fffUL)) {
588 rc = set_validity_icpt(scb_s, 0x0043U);
591 /* 64 bytes cannot cross page boundaries */
592 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
594 rc = set_validity_icpt(scb_s, 0x0043U);
595 /* Validity 0x0044 will be checked by SIE */
602 unpin_blocks(vcpu, vsie_page);
606 /* unpin the scb provided by guest 2, marking it as dirty */
607 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
610 hpa_t hpa = (hpa_t) vsie_page->scb_o;
613 unpin_guest_page(vcpu->kvm, gpa, hpa);
614 vsie_page->scb_o = NULL;
618 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
620 * Returns: - 0 if the scb was pinned.
621 * - > 0 if control has to be given to guest 2
622 * - -ENOMEM if out of memory
624 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
630 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
632 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
637 vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
642 * Inject a fault into guest 2.
644 * Returns: - > 0 if control has to be given to guest 2
645 * < 0 if an error occurred during injection.
647 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
650 struct kvm_s390_pgm_info pgm = {
653 /* 0-51: virtual address */
654 (vaddr & 0xfffffffffffff000UL) |
655 /* 52-53: store / fetch */
656 (((unsigned int) !write_flag) + 1) << 10,
657 /* 62-63: asce id (alway primary == 0) */
658 .exc_access_id = 0, /* always primary */
659 .op_access_id = 0, /* not MVPG */
663 if (code == PGM_PROTECTION)
664 pgm.trans_exc_code |= 0x4UL;
666 rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
671 * Handle a fault during vsie execution on a gmap shadow.
673 * Returns: - 0 if the fault was resolved
674 * - > 0 if control has to be given to guest 2
675 * - < 0 if an error occurred
677 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
681 if (current->thread.gmap_int_code == PGM_PROTECTION)
682 /* we can directly forward all protection exceptions */
683 return inject_fault(vcpu, PGM_PROTECTION,
684 current->thread.gmap_addr, 1);
686 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
687 current->thread.gmap_addr);
689 rc = inject_fault(vcpu, rc,
690 current->thread.gmap_addr,
691 current->thread.gmap_write_flag);
693 vsie_page->fault_addr = current->thread.gmap_addr;
699 * Retry the previous fault that required guest 2 intervention. This avoids
700 * one superfluous SIE re-entry and direct exit.
702 * Will ignore any errors. The next SIE fault will do proper fault handling.
704 static void handle_last_fault(struct kvm_vcpu *vcpu,
705 struct vsie_page *vsie_page)
707 if (vsie_page->fault_addr)
708 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
709 vsie_page->fault_addr);
710 vsie_page->fault_addr = 0;
713 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
715 vsie_page->scb_s.icptcode = 0;
718 /* rewind the psw and clear the vsie icpt, so we can retry execution */
719 static void retry_vsie_icpt(struct vsie_page *vsie_page)
721 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
722 int ilen = insn_length(scb_s->ipa >> 8);
724 /* take care of EXECUTE instructions */
725 if (scb_s->icptstatus & 1) {
726 ilen = (scb_s->icptstatus >> 4) & 0x6;
730 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
731 clear_vsie_icpt(vsie_page);
735 * Try to shadow + enable the guest 2 provided facility list.
736 * Retry instruction execution if enabled for and provided by guest 2.
738 * Returns: - 0 if handled (retry or guest 2 icpt)
739 * - > 0 if control has to be given to guest 2
741 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
743 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
744 __u32 fac = vsie_page->scb_o->fac & 0x7ffffff8U;
746 if (fac && test_kvm_facility(vcpu->kvm, 7)) {
747 retry_vsie_icpt(vsie_page);
748 if (read_guest_real(vcpu, fac, &vsie_page->fac,
749 sizeof(vsie_page->fac)))
750 return set_validity_icpt(scb_s, 0x1090U);
751 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
757 * Run the vsie on a shadow scb and a shadow gmap, without any further
758 * sanity checks, handling SIE faults.
760 * Returns: - 0 everything went fine
761 * - > 0 if control has to be given to guest 2
762 * - < 0 if an error occurred
764 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
766 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
767 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
768 int guest_bp_isolation;
771 handle_last_fault(vcpu, vsie_page);
775 if (test_cpu_flag(CIF_MCCK_PENDING))
778 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
780 /* save current guest state of bp isolation override */
781 guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
784 * The guest is running with BPBC, so we have to force it on for our
785 * nested guest. This is done by enabling BPBC globally, so the BPBC
786 * control in the SCB (which the nested guest can modify) is simply
789 if (test_kvm_facility(vcpu->kvm, 82) &&
790 vcpu->arch.sie_block->fpf & FPF_BPBC)
791 set_thread_flag(TIF_ISOLATE_BP_GUEST);
794 guest_enter_irqoff();
797 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
803 /* restore guest state for bp isolation override */
804 if (!guest_bp_isolation)
805 clear_thread_flag(TIF_ISOLATE_BP_GUEST);
807 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
810 rc = 0; /* we could still have an icpt */
811 else if (rc == -EFAULT)
812 return handle_fault(vcpu, vsie_page);
814 switch (scb_s->icptcode) {
816 if (scb_s->ipa == 0xb2b0)
817 rc = handle_stfle(vcpu, vsie_page);
820 /* stop not requested by g2 - must have been a kick */
821 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
822 clear_vsie_icpt(vsie_page);
825 if ((scb_s->ipa & 0xf000) != 0xf000)
826 scb_s->ipa += 0x1000;
832 static void release_gmap_shadow(struct vsie_page *vsie_page)
835 gmap_put(vsie_page->gmap);
836 WRITE_ONCE(vsie_page->gmap, NULL);
837 prefix_unmapped(vsie_page);
840 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
841 struct vsie_page *vsie_page)
848 asce = vcpu->arch.sie_block->gcr[1];
849 cr0.val = vcpu->arch.sie_block->gcr[0];
850 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
851 edat += edat && test_kvm_facility(vcpu->kvm, 78);
854 * ASCE or EDAT could have changed since last icpt, or the gmap
855 * we're holding has been unshadowed. If the gmap is still valid,
856 * we can safely reuse it.
858 if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat))
861 /* release the old shadow - if any, and mark the prefix as unmapped */
862 release_gmap_shadow(vsie_page);
863 gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
865 return PTR_ERR(gmap);
866 gmap->private = vcpu->kvm;
867 WRITE_ONCE(vsie_page->gmap, gmap);
872 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
874 static void register_shadow_scb(struct kvm_vcpu *vcpu,
875 struct vsie_page *vsie_page)
877 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
879 WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
881 * External calls have to lead to a kick of the vcpu and
882 * therefore the vsie -> Simulate Wait state.
884 atomic_or(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
886 * We have to adjust the g3 epoch by the g2 epoch. The epoch will
887 * automatically be adjusted on tod clock changes via kvm_sync_clock.
890 scb_s->epoch += vcpu->kvm->arch.epoch;
895 * Unregister a shadow scb from a VCPU.
897 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
899 atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
900 WRITE_ONCE(vcpu->arch.vsie_block, NULL);
904 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
905 * prefix pages and faults.
907 * Returns: - 0 if no errors occurred
908 * - > 0 if control has to be given to guest 2
909 * - -ENOMEM if out of memory
911 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
913 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
917 rc = acquire_gmap_shadow(vcpu, vsie_page);
919 rc = map_prefix(vcpu, vsie_page);
921 gmap_enable(vsie_page->gmap);
922 update_intervention_requests(vsie_page);
923 rc = do_vsie_run(vcpu, vsie_page);
924 gmap_enable(vcpu->arch.gmap);
926 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
930 if (rc || scb_s->icptcode || signal_pending(current) ||
931 kvm_s390_vcpu_has_irq(vcpu, 0))
937 * Addressing exceptions are always presentes as intercepts.
938 * As addressing exceptions are suppressing and our guest 3 PSW
939 * points at the responsible instruction, we have to
940 * forward the PSW and set the ilc. If we can't read guest 3
941 * instruction, we can use an arbitrary ilc. Let's always use
942 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
943 * memory. (we could also fake the shadow so the hardware
946 scb_s->icptcode = ICPT_PROGI;
947 scb_s->iprcc = PGM_ADDRESSING;
949 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
956 * Get or create a vsie page for a scb address.
958 * Returns: - address of a vsie page (cached or new one)
959 * - NULL if the same scb address is already used by another VCPU
960 * - ERR_PTR(-ENOMEM) if out of memory
962 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
964 struct vsie_page *vsie_page;
969 page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
972 if (page_ref_inc_return(page) == 2)
973 return page_to_virt(page);
978 * We want at least #online_vcpus shadows, so every VCPU can execute
979 * the VSIE in parallel.
981 nr_vcpus = atomic_read(&kvm->online_vcpus);
983 mutex_lock(&kvm->arch.vsie.mutex);
984 if (kvm->arch.vsie.page_count < nr_vcpus) {
985 page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA);
987 mutex_unlock(&kvm->arch.vsie.mutex);
988 return ERR_PTR(-ENOMEM);
991 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
992 kvm->arch.vsie.page_count++;
994 /* reuse an existing entry that belongs to nobody */
996 page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
997 if (page_ref_inc_return(page) == 2)
1000 kvm->arch.vsie.next++;
1001 kvm->arch.vsie.next %= nr_vcpus;
1003 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1006 /* double use of the same address */
1007 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1009 mutex_unlock(&kvm->arch.vsie.mutex);
1012 mutex_unlock(&kvm->arch.vsie.mutex);
1014 vsie_page = page_to_virt(page);
1015 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1016 release_gmap_shadow(vsie_page);
1017 vsie_page->fault_addr = 0;
1018 vsie_page->scb_s.ihcpu = 0xffffU;
1022 /* put a vsie page acquired via get_vsie_page */
1023 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1025 struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1030 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1032 struct vsie_page *vsie_page;
1033 unsigned long scb_addr;
1036 vcpu->stat.instruction_sie++;
1037 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1039 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1040 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1042 BUILD_BUG_ON(sizeof(struct vsie_page) != 4096);
1043 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1045 /* 512 byte alignment */
1046 if (unlikely(scb_addr & 0x1ffUL))
1047 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1049 if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0))
1052 vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1053 if (IS_ERR(vsie_page))
1054 return PTR_ERR(vsie_page);
1055 else if (!vsie_page)
1056 /* double use of sie control block - simply do nothing */
1059 rc = pin_scb(vcpu, vsie_page, scb_addr);
1062 rc = shadow_scb(vcpu, vsie_page);
1065 rc = pin_blocks(vcpu, vsie_page);
1068 register_shadow_scb(vcpu, vsie_page);
1069 rc = vsie_run(vcpu, vsie_page);
1070 unregister_shadow_scb(vcpu);
1071 unpin_blocks(vcpu, vsie_page);
1073 unshadow_scb(vcpu, vsie_page);
1075 unpin_scb(vcpu, vsie_page, scb_addr);
1077 put_vsie_page(vcpu->kvm, vsie_page);
1079 return rc < 0 ? rc : 0;
1082 /* Init the vsie data structures. To be called when a vm is initialized. */
1083 void kvm_s390_vsie_init(struct kvm *kvm)
1085 mutex_init(&kvm->arch.vsie.mutex);
1086 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL);
1089 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1090 void kvm_s390_vsie_destroy(struct kvm *kvm)
1092 struct vsie_page *vsie_page;
1096 mutex_lock(&kvm->arch.vsie.mutex);
1097 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1098 page = kvm->arch.vsie.pages[i];
1099 kvm->arch.vsie.pages[i] = NULL;
1100 vsie_page = page_to_virt(page);
1101 release_gmap_shadow(vsie_page);
1102 /* free the radix tree entry */
1103 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1106 kvm->arch.vsie.page_count = 0;
1107 mutex_unlock(&kvm->arch.vsie.mutex);
1110 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1112 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1115 * Even if the VCPU lets go of the shadow sie block reference, it is
1116 * still valid in the cache. So we can safely kick it.
1119 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1120 if (scb->prog0c & PROG_IN_SIE)
1121 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);