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, &vsie_page->crycb, 56))
160 return set_validity_icpt(scb_s, 0x0035U);
162 scb_s->ecb3 |= ecb3_flags;
163 scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT1 |
166 /* xor both blocks in one run */
167 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
168 b2 = (unsigned long *)
169 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
170 /* as 56%8 == 0, bitmap_xor won't overwrite any data */
171 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
175 /* shadow (round up/down) the ibc to avoid validity icpt */
176 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
178 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
179 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
180 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
183 /* ibc installed in g2 and requested for g3 */
184 if (vcpu->kvm->arch.model.ibc && (scb_o->ibc & 0x0fffU)) {
185 scb_s->ibc = scb_o->ibc & 0x0fffU;
186 /* takte care of the minimum ibc level of the machine */
187 if (scb_s->ibc < min_ibc)
188 scb_s->ibc = min_ibc;
189 /* take care of the maximum ibc level set for the guest */
190 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
191 scb_s->ibc = vcpu->kvm->arch.model.ibc;
195 /* unshadow the scb, copying parameters back to the real scb */
196 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
198 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
199 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
202 scb_o->icptcode = scb_s->icptcode;
203 scb_o->icptstatus = scb_s->icptstatus;
204 scb_o->ipa = scb_s->ipa;
205 scb_o->ipb = scb_s->ipb;
206 scb_o->gbea = scb_s->gbea;
209 scb_o->cputm = scb_s->cputm;
210 scb_o->ckc = scb_s->ckc;
211 scb_o->todpr = scb_s->todpr;
214 scb_o->gpsw = scb_s->gpsw;
215 scb_o->gg14 = scb_s->gg14;
216 scb_o->gg15 = scb_s->gg15;
217 memcpy(scb_o->gcr, scb_s->gcr, 128);
218 scb_o->pp = scb_s->pp;
220 /* interrupt intercept */
221 switch (scb_s->icptcode) {
225 memcpy((void *)((u64)scb_o + 0xc0),
226 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
230 memcpy((void *)((u64)scb_o + 0xc0),
231 (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0);
235 if (scb_s->ihcpu != 0xffffU)
236 scb_o->ihcpu = scb_s->ihcpu;
240 * Setup the shadow scb by copying and checking the relevant parts of the g2
243 * Returns: - 0 if the scb has been shadowed
244 * - > 0 if control has to be given to guest 2
246 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
248 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
249 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
250 bool had_tx = scb_s->ecb & 0x10U;
251 unsigned long new_mso = 0;
254 /* make sure we don't have any leftovers when reusing the scb */
263 rc = prepare_cpuflags(vcpu, vsie_page);
268 scb_s->cputm = scb_o->cputm;
269 scb_s->ckc = scb_o->ckc;
270 scb_s->todpr = scb_o->todpr;
271 scb_s->epoch = scb_o->epoch;
274 scb_s->gpsw = scb_o->gpsw;
275 scb_s->gg14 = scb_o->gg14;
276 scb_s->gg15 = scb_o->gg15;
277 memcpy(scb_s->gcr, scb_o->gcr, 128);
278 scb_s->pp = scb_o->pp;
280 /* interception / execution handling */
281 scb_s->gbea = scb_o->gbea;
282 scb_s->lctl = scb_o->lctl;
283 scb_s->svcc = scb_o->svcc;
284 scb_s->ictl = scb_o->ictl;
286 * SKEY handling functions can't deal with false setting of PTE invalid
287 * bits. Therefore we cannot provide interpretation and would later
288 * have to provide own emulation handlers.
290 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
291 scb_s->icpua = scb_o->icpua;
293 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
294 new_mso = scb_o->mso & 0xfffffffffff00000UL;
295 /* if the hva of the prefix changes, we have to remap the prefix */
296 if (scb_s->mso != new_mso || scb_s->prefix != scb_o->prefix)
297 prefix_unmapped(vsie_page);
298 /* SIE will do mso/msl validity and exception checks for us */
299 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
300 scb_s->mso = new_mso;
301 scb_s->prefix = scb_o->prefix;
303 /* We have to definetly flush the tlb if this scb never ran */
304 if (scb_s->ihcpu != 0xffffU)
305 scb_s->ihcpu = scb_o->ihcpu;
307 /* MVPG and Protection Exception Interpretation are always available */
308 scb_s->eca |= scb_o->eca & 0x01002000U;
309 /* Host-protection-interruption introduced with ESOP */
310 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
311 scb_s->ecb |= scb_o->ecb & 0x02U;
312 /* transactional execution */
313 if (test_kvm_facility(vcpu->kvm, 73)) {
314 /* remap the prefix is tx is toggled on */
315 if ((scb_o->ecb & 0x10U) && !had_tx)
316 prefix_unmapped(vsie_page);
317 scb_s->ecb |= scb_o->ecb & 0x10U;
320 if (test_kvm_facility(vcpu->kvm, 129)) {
321 scb_s->eca |= scb_o->eca & 0x00020000U;
322 scb_s->ecd |= scb_o->ecd & 0x20000000U;
324 /* Run-time-Instrumentation */
325 if (test_kvm_facility(vcpu->kvm, 64))
326 scb_s->ecb3 |= scb_o->ecb3 & 0x01U;
327 /* Instruction Execution Prevention */
328 if (test_kvm_facility(vcpu->kvm, 130))
329 scb_s->ecb2 |= scb_o->ecb2 & 0x20U;
330 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
331 scb_s->eca |= scb_o->eca & 0x00000001U;
332 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
333 scb_s->eca |= scb_o->eca & 0x40000000U;
334 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
335 scb_s->eca |= scb_o->eca & 0x80000000U;
337 prepare_ibc(vcpu, vsie_page);
338 rc = shadow_crycb(vcpu, vsie_page);
341 unshadow_scb(vcpu, vsie_page);
345 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
348 struct kvm *kvm = gmap->private;
349 struct vsie_page *cur;
350 unsigned long prefix;
354 if (!gmap_is_shadow(gmap))
356 if (start >= 1UL << 31)
357 /* We are only interested in prefix pages */
361 * Only new shadow blocks are added to the list during runtime,
362 * therefore we can safely reference them all the time.
364 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
365 page = READ_ONCE(kvm->arch.vsie.pages[i]);
368 cur = page_to_virt(page);
369 if (READ_ONCE(cur->gmap) != gmap)
371 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
372 /* with mso/msl, the prefix lies at an offset */
373 prefix += cur->scb_s.mso;
374 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
375 prefix_unmapped_sync(cur);
380 * Map the first prefix page and if tx is enabled also the second prefix page.
382 * The prefix will be protected, a gmap notifier will inform about unmaps.
383 * The shadow scb must not be executed until the prefix is remapped, this is
384 * guaranteed by properly handling PROG_REQUEST.
386 * Returns: - 0 on if successfully mapped or already mapped
387 * - > 0 if control has to be given to guest 2
388 * - -EAGAIN if the caller can retry immediately
389 * - -ENOMEM if out of memory
391 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
393 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
394 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
397 if (prefix_is_mapped(vsie_page))
400 /* mark it as mapped so we can catch any concurrent unmappers */
401 prefix_mapped(vsie_page);
403 /* with mso/msl, the prefix lies at offset *mso* */
404 prefix += scb_s->mso;
406 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix);
407 if (!rc && (scb_s->ecb & 0x10U))
408 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
411 * We don't have to mprotect, we will be called for all unshadows.
412 * SIE will detect if protection applies and trigger a validity.
415 prefix_unmapped(vsie_page);
416 if (rc > 0 || rc == -EFAULT)
417 rc = set_validity_icpt(scb_s, 0x0037U);
422 * Pin the guest page given by gpa and set hpa to the pinned host address.
423 * Will always be pinned writable.
425 * Returns: - 0 on success
426 * - -EINVAL if the gpa is not valid guest storage
427 * - -ENOMEM if out of memory
429 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
435 hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
436 if (kvm_is_error_hva(hva))
438 rc = get_user_pages_fast(hva, 1, 1, &page);
443 *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK);
447 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
448 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
452 page = virt_to_page(hpa);
453 set_page_dirty_lock(page);
455 /* mark the page always as dirty for migration */
456 mark_page_dirty(kvm, gpa_to_gfn(gpa));
459 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
460 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
462 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
463 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
467 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
469 gpa = scb_o->scaol & ~0xfUL;
470 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
471 gpa |= (u64) scb_o->scaoh << 32;
472 unpin_guest_page(vcpu->kvm, gpa, hpa);
479 gpa = scb_o->itdba & ~0xffUL;
480 unpin_guest_page(vcpu->kvm, gpa, hpa);
486 gpa = scb_o->gvrd & ~0x1ffUL;
487 unpin_guest_page(vcpu->kvm, gpa, hpa);
493 gpa = scb_o->riccbd & ~0x3fUL;
494 unpin_guest_page(vcpu->kvm, gpa, hpa);
500 * Instead of shadowing some blocks, we can simply forward them because the
501 * addresses in the scb are 64 bit long.
503 * This works as long as the data lies in one page. If blocks ever exceed one
504 * page, we have to fall back to shadowing.
506 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
507 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
509 * Returns: - 0 if all blocks were pinned.
510 * - > 0 if control has to be given to guest 2
511 * - -ENOMEM if out of memory
513 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
515 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
516 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
521 gpa = scb_o->scaol & ~0xfUL;
522 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
523 gpa |= (u64) scb_o->scaoh << 32;
525 if (!(gpa & ~0x1fffUL))
526 rc = set_validity_icpt(scb_s, 0x0038U);
527 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
528 rc = set_validity_icpt(scb_s, 0x0011U);
529 else if ((gpa & PAGE_MASK) !=
530 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
531 rc = set_validity_icpt(scb_s, 0x003bU);
533 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
535 rc = set_validity_icpt(scb_s, 0x0034U);
539 scb_s->scaoh = (u32)((u64)hpa >> 32);
540 scb_s->scaol = (u32)(u64)hpa;
543 gpa = scb_o->itdba & ~0xffUL;
544 if (gpa && (scb_s->ecb & 0x10U)) {
545 if (!(gpa & ~0x1fffU)) {
546 rc = set_validity_icpt(scb_s, 0x0080U);
549 /* 256 bytes cannot cross page boundaries */
550 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
552 rc = set_validity_icpt(scb_s, 0x0080U);
558 gpa = scb_o->gvrd & ~0x1ffUL;
559 if (gpa && (scb_s->eca & 0x00020000U) &&
560 !(scb_s->ecd & 0x20000000U)) {
561 if (!(gpa & ~0x1fffUL)) {
562 rc = set_validity_icpt(scb_s, 0x1310U);
566 * 512 bytes vector registers cannot cross page boundaries
567 * if this block gets bigger, we have to shadow it.
569 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
571 rc = set_validity_icpt(scb_s, 0x1310U);
577 gpa = scb_o->riccbd & ~0x3fUL;
578 if (gpa && (scb_s->ecb3 & 0x01U)) {
579 if (!(gpa & ~0x1fffUL)) {
580 rc = set_validity_icpt(scb_s, 0x0043U);
583 /* 64 bytes cannot cross page boundaries */
584 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
586 rc = set_validity_icpt(scb_s, 0x0043U);
587 /* Validity 0x0044 will be checked by SIE */
594 unpin_blocks(vcpu, vsie_page);
598 /* unpin the scb provided by guest 2, marking it as dirty */
599 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
602 hpa_t hpa = (hpa_t) vsie_page->scb_o;
605 unpin_guest_page(vcpu->kvm, gpa, hpa);
606 vsie_page->scb_o = NULL;
610 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
612 * Returns: - 0 if the scb was pinned.
613 * - > 0 if control has to be given to guest 2
614 * - -ENOMEM if out of memory
616 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
622 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
624 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
629 vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
634 * Inject a fault into guest 2.
636 * Returns: - > 0 if control has to be given to guest 2
637 * < 0 if an error occurred during injection.
639 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
642 struct kvm_s390_pgm_info pgm = {
645 /* 0-51: virtual address */
646 (vaddr & 0xfffffffffffff000UL) |
647 /* 52-53: store / fetch */
648 (((unsigned int) !write_flag) + 1) << 10,
649 /* 62-63: asce id (alway primary == 0) */
650 .exc_access_id = 0, /* always primary */
651 .op_access_id = 0, /* not MVPG */
655 if (code == PGM_PROTECTION)
656 pgm.trans_exc_code |= 0x4UL;
658 rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
663 * Handle a fault during vsie execution on a gmap shadow.
665 * Returns: - 0 if the fault was resolved
666 * - > 0 if control has to be given to guest 2
667 * - < 0 if an error occurred
669 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
673 if (current->thread.gmap_int_code == PGM_PROTECTION)
674 /* we can directly forward all protection exceptions */
675 return inject_fault(vcpu, PGM_PROTECTION,
676 current->thread.gmap_addr, 1);
678 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
679 current->thread.gmap_addr);
681 rc = inject_fault(vcpu, rc,
682 current->thread.gmap_addr,
683 current->thread.gmap_write_flag);
685 vsie_page->fault_addr = current->thread.gmap_addr;
691 * Retry the previous fault that required guest 2 intervention. This avoids
692 * one superfluous SIE re-entry and direct exit.
694 * Will ignore any errors. The next SIE fault will do proper fault handling.
696 static void handle_last_fault(struct kvm_vcpu *vcpu,
697 struct vsie_page *vsie_page)
699 if (vsie_page->fault_addr)
700 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
701 vsie_page->fault_addr);
702 vsie_page->fault_addr = 0;
705 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
707 vsie_page->scb_s.icptcode = 0;
710 /* rewind the psw and clear the vsie icpt, so we can retry execution */
711 static void retry_vsie_icpt(struct vsie_page *vsie_page)
713 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
714 int ilen = insn_length(scb_s->ipa >> 8);
716 /* take care of EXECUTE instructions */
717 if (scb_s->icptstatus & 1) {
718 ilen = (scb_s->icptstatus >> 4) & 0x6;
722 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
723 clear_vsie_icpt(vsie_page);
727 * Try to shadow + enable the guest 2 provided facility list.
728 * Retry instruction execution if enabled for and provided by guest 2.
730 * Returns: - 0 if handled (retry or guest 2 icpt)
731 * - > 0 if control has to be given to guest 2
733 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
735 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
736 __u32 fac = vsie_page->scb_o->fac & 0x7ffffff8U;
738 if (fac && test_kvm_facility(vcpu->kvm, 7)) {
739 retry_vsie_icpt(vsie_page);
740 if (read_guest_real(vcpu, fac, &vsie_page->fac,
741 sizeof(vsie_page->fac)))
742 return set_validity_icpt(scb_s, 0x1090U);
743 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
749 * Run the vsie on a shadow scb and a shadow gmap, without any further
750 * sanity checks, handling SIE faults.
752 * Returns: - 0 everything went fine
753 * - > 0 if control has to be given to guest 2
754 * - < 0 if an error occurred
756 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
758 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
759 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
762 handle_last_fault(vcpu, vsie_page);
766 if (test_cpu_flag(CIF_MCCK_PENDING))
769 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
771 guest_enter_irqoff();
774 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
779 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
782 rc = 0; /* we could still have an icpt */
783 else if (rc == -EFAULT)
784 return handle_fault(vcpu, vsie_page);
786 switch (scb_s->icptcode) {
788 if (scb_s->ipa == 0xb2b0)
789 rc = handle_stfle(vcpu, vsie_page);
792 /* stop not requested by g2 - must have been a kick */
793 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
794 clear_vsie_icpt(vsie_page);
797 if ((scb_s->ipa & 0xf000) != 0xf000)
798 scb_s->ipa += 0x1000;
804 static void release_gmap_shadow(struct vsie_page *vsie_page)
807 gmap_put(vsie_page->gmap);
808 WRITE_ONCE(vsie_page->gmap, NULL);
809 prefix_unmapped(vsie_page);
812 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
813 struct vsie_page *vsie_page)
820 asce = vcpu->arch.sie_block->gcr[1];
821 cr0.val = vcpu->arch.sie_block->gcr[0];
822 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
823 edat += edat && test_kvm_facility(vcpu->kvm, 78);
826 * ASCE or EDAT could have changed since last icpt, or the gmap
827 * we're holding has been unshadowed. If the gmap is still valid,
828 * we can safely reuse it.
830 if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat))
833 /* release the old shadow - if any, and mark the prefix as unmapped */
834 release_gmap_shadow(vsie_page);
835 gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
837 return PTR_ERR(gmap);
838 gmap->private = vcpu->kvm;
839 WRITE_ONCE(vsie_page->gmap, gmap);
844 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
846 static void register_shadow_scb(struct kvm_vcpu *vcpu,
847 struct vsie_page *vsie_page)
849 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
851 WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
853 * External calls have to lead to a kick of the vcpu and
854 * therefore the vsie -> Simulate Wait state.
856 atomic_or(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
858 * We have to adjust the g3 epoch by the g2 epoch. The epoch will
859 * automatically be adjusted on tod clock changes via kvm_sync_clock.
862 scb_s->epoch += vcpu->kvm->arch.epoch;
867 * Unregister a shadow scb from a VCPU.
869 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
871 atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
872 WRITE_ONCE(vcpu->arch.vsie_block, NULL);
876 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
877 * prefix pages and faults.
879 * Returns: - 0 if no errors occurred
880 * - > 0 if control has to be given to guest 2
881 * - -ENOMEM if out of memory
883 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
885 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
889 rc = acquire_gmap_shadow(vcpu, vsie_page);
891 rc = map_prefix(vcpu, vsie_page);
893 gmap_enable(vsie_page->gmap);
894 update_intervention_requests(vsie_page);
895 rc = do_vsie_run(vcpu, vsie_page);
896 gmap_enable(vcpu->arch.gmap);
898 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
902 if (rc || scb_s->icptcode || signal_pending(current) ||
903 kvm_s390_vcpu_has_irq(vcpu, 0))
909 * Addressing exceptions are always presentes as intercepts.
910 * As addressing exceptions are suppressing and our guest 3 PSW
911 * points at the responsible instruction, we have to
912 * forward the PSW and set the ilc. If we can't read guest 3
913 * instruction, we can use an arbitrary ilc. Let's always use
914 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
915 * memory. (we could also fake the shadow so the hardware
918 scb_s->icptcode = ICPT_PROGI;
919 scb_s->iprcc = PGM_ADDRESSING;
921 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
927 * Get or create a vsie page for a scb address.
929 * Returns: - address of a vsie page (cached or new one)
930 * - NULL if the same scb address is already used by another VCPU
931 * - ERR_PTR(-ENOMEM) if out of memory
933 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
935 struct vsie_page *vsie_page;
940 page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
943 if (page_ref_inc_return(page) == 2)
944 return page_to_virt(page);
949 * We want at least #online_vcpus shadows, so every VCPU can execute
950 * the VSIE in parallel.
952 nr_vcpus = atomic_read(&kvm->online_vcpus);
954 mutex_lock(&kvm->arch.vsie.mutex);
955 if (kvm->arch.vsie.page_count < nr_vcpus) {
956 page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA);
958 mutex_unlock(&kvm->arch.vsie.mutex);
959 return ERR_PTR(-ENOMEM);
962 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
963 kvm->arch.vsie.page_count++;
965 /* reuse an existing entry that belongs to nobody */
967 page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
968 if (page_ref_inc_return(page) == 2)
971 kvm->arch.vsie.next++;
972 kvm->arch.vsie.next %= nr_vcpus;
974 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
977 /* double use of the same address */
978 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
980 mutex_unlock(&kvm->arch.vsie.mutex);
983 mutex_unlock(&kvm->arch.vsie.mutex);
985 vsie_page = page_to_virt(page);
986 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
987 release_gmap_shadow(vsie_page);
988 vsie_page->fault_addr = 0;
989 vsie_page->scb_s.ihcpu = 0xffffU;
993 /* put a vsie page acquired via get_vsie_page */
994 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
996 struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1001 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1003 struct vsie_page *vsie_page;
1004 unsigned long scb_addr;
1007 vcpu->stat.instruction_sie++;
1008 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1010 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1011 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1013 BUILD_BUG_ON(sizeof(struct vsie_page) != 4096);
1014 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1016 /* 512 byte alignment */
1017 if (unlikely(scb_addr & 0x1ffUL))
1018 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1020 if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0))
1023 vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1024 if (IS_ERR(vsie_page))
1025 return PTR_ERR(vsie_page);
1026 else if (!vsie_page)
1027 /* double use of sie control block - simply do nothing */
1030 rc = pin_scb(vcpu, vsie_page, scb_addr);
1033 rc = shadow_scb(vcpu, vsie_page);
1036 rc = pin_blocks(vcpu, vsie_page);
1039 register_shadow_scb(vcpu, vsie_page);
1040 rc = vsie_run(vcpu, vsie_page);
1041 unregister_shadow_scb(vcpu);
1042 unpin_blocks(vcpu, vsie_page);
1044 unshadow_scb(vcpu, vsie_page);
1046 unpin_scb(vcpu, vsie_page, scb_addr);
1048 put_vsie_page(vcpu->kvm, vsie_page);
1050 return rc < 0 ? rc : 0;
1053 /* Init the vsie data structures. To be called when a vm is initialized. */
1054 void kvm_s390_vsie_init(struct kvm *kvm)
1056 mutex_init(&kvm->arch.vsie.mutex);
1057 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL);
1060 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1061 void kvm_s390_vsie_destroy(struct kvm *kvm)
1063 struct vsie_page *vsie_page;
1067 mutex_lock(&kvm->arch.vsie.mutex);
1068 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1069 page = kvm->arch.vsie.pages[i];
1070 kvm->arch.vsie.pages[i] = NULL;
1071 vsie_page = page_to_virt(page);
1072 release_gmap_shadow(vsie_page);
1073 /* free the radix tree entry */
1074 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1077 kvm->arch.vsie.page_count = 0;
1078 mutex_unlock(&kvm->arch.vsie.mutex);
1081 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1083 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1086 * Even if the VCPU lets go of the shadow sie block reference, it is
1087 * still valid in the cache. So we can safely kick it.
1090 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1091 if (scb->prog0c & PROG_IN_SIE)
1092 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);