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 */
30 * the backup info for machine check. ensure it's at
31 * the same offset as that in struct sie_page!
33 struct mcck_volatile_info mcck_info; /* 0x0200 */
34 /* the pinned originial scb */
35 struct kvm_s390_sie_block *scb_o; /* 0x0218 */
36 /* the shadow gmap in use by the vsie_page */
37 struct gmap *gmap; /* 0x0220 */
38 /* address of the last reported fault to guest2 */
39 unsigned long fault_addr; /* 0x0228 */
40 __u8 reserved[0x0700 - 0x0230]; /* 0x0230 */
41 struct kvm_s390_crypto_cb crycb; /* 0x0700 */
42 __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */
45 /* trigger a validity icpt for the given scb */
46 static int set_validity_icpt(struct kvm_s390_sie_block *scb,
50 scb->ipb = ((__u32) reason_code) << 16;
51 scb->icptcode = ICPT_VALIDITY;
55 /* mark the prefix as unmapped, this will block the VSIE */
56 static void prefix_unmapped(struct vsie_page *vsie_page)
58 atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
61 /* mark the prefix as unmapped and wait until the VSIE has been left */
62 static void prefix_unmapped_sync(struct vsie_page *vsie_page)
64 prefix_unmapped(vsie_page);
65 if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
66 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
67 while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
71 /* mark the prefix as mapped, this will allow the VSIE to run */
72 static void prefix_mapped(struct vsie_page *vsie_page)
74 atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
77 /* test if the prefix is mapped into the gmap shadow */
78 static int prefix_is_mapped(struct vsie_page *vsie_page)
80 return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
83 /* copy the updated intervention request bits into the shadow scb */
84 static void update_intervention_requests(struct vsie_page *vsie_page)
86 const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
89 cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
90 atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
91 atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
94 /* shadow (filter and validate) the cpuflags */
95 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
97 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
98 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
99 int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
101 /* we don't allow ESA/390 guests */
102 if (!(cpuflags & CPUSTAT_ZARCH))
103 return set_validity_icpt(scb_s, 0x0001U);
105 if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
106 return set_validity_icpt(scb_s, 0x0001U);
107 else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
108 return set_validity_icpt(scb_s, 0x0007U);
110 /* intervention requests will be set later */
111 newflags = CPUSTAT_ZARCH;
112 if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
113 newflags |= CPUSTAT_GED;
114 if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
115 if (cpuflags & CPUSTAT_GED)
116 return set_validity_icpt(scb_s, 0x0001U);
117 newflags |= CPUSTAT_GED2;
119 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
120 newflags |= cpuflags & CPUSTAT_P;
121 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
122 newflags |= cpuflags & CPUSTAT_SM;
123 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
124 newflags |= cpuflags & CPUSTAT_IBS;
125 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS))
126 newflags |= cpuflags & CPUSTAT_KSS;
128 atomic_set(&scb_s->cpuflags, newflags);
133 * Create a shadow copy of the crycb block and setup key wrapping, if
134 * requested for guest 3 and enabled for guest 2.
136 * We only accept format-1 (no AP in g2), but convert it into format-2
137 * There is nothing to do for format-0.
139 * Returns: - 0 if shadowed or nothing to do
140 * - > 0 if control has to be given to guest 2
142 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
144 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
145 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
146 u32 crycb_addr = scb_o->crycbd & 0x7ffffff8U;
147 unsigned long *b1, *b2;
151 if (!(scb_o->crycbd & vcpu->arch.sie_block->crycbd & CRYCB_FORMAT1))
153 /* format-1 is supported with message-security-assist extension 3 */
154 if (!test_kvm_facility(vcpu->kvm, 76))
156 /* we may only allow it if enabled for guest 2 */
157 ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
158 (ECB3_AES | ECB3_DEA);
162 if ((crycb_addr & PAGE_MASK) != ((crycb_addr + 128) & PAGE_MASK))
163 return set_validity_icpt(scb_s, 0x003CU);
164 else if (!crycb_addr)
165 return set_validity_icpt(scb_s, 0x0039U);
167 /* copy only the wrapping keys */
168 if (read_guest_real(vcpu, crycb_addr + 72, &vsie_page->crycb, 56))
169 return set_validity_icpt(scb_s, 0x0035U);
171 scb_s->ecb3 |= ecb3_flags;
172 scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT1 |
175 /* xor both blocks in one run */
176 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
177 b2 = (unsigned long *)
178 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
179 /* as 56%8 == 0, bitmap_xor won't overwrite any data */
180 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
184 /* shadow (round up/down) the ibc to avoid validity icpt */
185 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
187 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
188 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
189 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
192 /* ibc installed in g2 and requested for g3 */
193 if (vcpu->kvm->arch.model.ibc && (scb_o->ibc & 0x0fffU)) {
194 scb_s->ibc = scb_o->ibc & 0x0fffU;
195 /* takte care of the minimum ibc level of the machine */
196 if (scb_s->ibc < min_ibc)
197 scb_s->ibc = min_ibc;
198 /* take care of the maximum ibc level set for the guest */
199 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
200 scb_s->ibc = vcpu->kvm->arch.model.ibc;
204 /* unshadow the scb, copying parameters back to the real scb */
205 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
207 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
208 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
211 scb_o->icptcode = scb_s->icptcode;
212 scb_o->icptstatus = scb_s->icptstatus;
213 scb_o->ipa = scb_s->ipa;
214 scb_o->ipb = scb_s->ipb;
215 scb_o->gbea = scb_s->gbea;
218 scb_o->cputm = scb_s->cputm;
219 scb_o->ckc = scb_s->ckc;
220 scb_o->todpr = scb_s->todpr;
223 scb_o->gpsw = scb_s->gpsw;
224 scb_o->gg14 = scb_s->gg14;
225 scb_o->gg15 = scb_s->gg15;
226 memcpy(scb_o->gcr, scb_s->gcr, 128);
227 scb_o->pp = scb_s->pp;
229 /* interrupt intercept */
230 switch (scb_s->icptcode) {
234 memcpy((void *)((u64)scb_o + 0xc0),
235 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
239 memcpy((void *)((u64)scb_o + 0xc0),
240 (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0);
244 if (scb_s->ihcpu != 0xffffU)
245 scb_o->ihcpu = scb_s->ihcpu;
249 * Setup the shadow scb by copying and checking the relevant parts of the g2
252 * Returns: - 0 if the scb has been shadowed
253 * - > 0 if control has to be given to guest 2
255 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
257 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
258 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
259 bool had_tx = scb_s->ecb & ECB_TE;
260 unsigned long new_mso = 0;
263 /* make sure we don't have any leftovers when reusing the scb */
272 rc = prepare_cpuflags(vcpu, vsie_page);
277 scb_s->cputm = scb_o->cputm;
278 scb_s->ckc = scb_o->ckc;
279 scb_s->todpr = scb_o->todpr;
280 scb_s->epoch = scb_o->epoch;
283 scb_s->gpsw = scb_o->gpsw;
284 scb_s->gg14 = scb_o->gg14;
285 scb_s->gg15 = scb_o->gg15;
286 memcpy(scb_s->gcr, scb_o->gcr, 128);
287 scb_s->pp = scb_o->pp;
289 /* interception / execution handling */
290 scb_s->gbea = scb_o->gbea;
291 scb_s->lctl = scb_o->lctl;
292 scb_s->svcc = scb_o->svcc;
293 scb_s->ictl = scb_o->ictl;
295 * SKEY handling functions can't deal with false setting of PTE invalid
296 * bits. Therefore we cannot provide interpretation and would later
297 * have to provide own emulation handlers.
299 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
300 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
302 scb_s->icpua = scb_o->icpua;
304 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
305 new_mso = scb_o->mso & 0xfffffffffff00000UL;
306 /* if the hva of the prefix changes, we have to remap the prefix */
307 if (scb_s->mso != new_mso || scb_s->prefix != scb_o->prefix)
308 prefix_unmapped(vsie_page);
309 /* SIE will do mso/msl validity and exception checks for us */
310 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
311 scb_s->mso = new_mso;
312 scb_s->prefix = scb_o->prefix;
314 /* We have to definetly flush the tlb if this scb never ran */
315 if (scb_s->ihcpu != 0xffffU)
316 scb_s->ihcpu = scb_o->ihcpu;
318 /* MVPG and Protection Exception Interpretation are always available */
319 scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
320 /* Host-protection-interruption introduced with ESOP */
321 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
322 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
323 /* transactional execution */
324 if (test_kvm_facility(vcpu->kvm, 73)) {
325 /* remap the prefix is tx is toggled on */
326 if ((scb_o->ecb & ECB_TE) && !had_tx)
327 prefix_unmapped(vsie_page);
328 scb_s->ecb |= scb_o->ecb & ECB_TE;
331 if (test_kvm_facility(vcpu->kvm, 129)) {
332 scb_s->eca |= scb_o->eca & ECA_VX;
333 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
335 /* Run-time-Instrumentation */
336 if (test_kvm_facility(vcpu->kvm, 64))
337 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
338 /* Instruction Execution Prevention */
339 if (test_kvm_facility(vcpu->kvm, 130))
340 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
341 /* Guarded Storage */
342 if (test_kvm_facility(vcpu->kvm, 133)) {
343 scb_s->ecb |= scb_o->ecb & ECB_GS;
344 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
346 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
347 scb_s->eca |= scb_o->eca & ECA_SII;
348 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
349 scb_s->eca |= scb_o->eca & ECA_IB;
350 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
351 scb_s->eca |= scb_o->eca & ECA_CEI;
352 /* Epoch Extension */
353 if (test_kvm_facility(vcpu->kvm, 139))
354 scb_s->ecd |= scb_o->ecd & ECD_MEF;
356 prepare_ibc(vcpu, vsie_page);
357 rc = shadow_crycb(vcpu, vsie_page);
360 unshadow_scb(vcpu, vsie_page);
364 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
367 struct kvm *kvm = gmap->private;
368 struct vsie_page *cur;
369 unsigned long prefix;
373 if (!gmap_is_shadow(gmap))
375 if (start >= 1UL << 31)
376 /* We are only interested in prefix pages */
380 * Only new shadow blocks are added to the list during runtime,
381 * therefore we can safely reference them all the time.
383 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
384 page = READ_ONCE(kvm->arch.vsie.pages[i]);
387 cur = page_to_virt(page);
388 if (READ_ONCE(cur->gmap) != gmap)
390 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
391 /* with mso/msl, the prefix lies at an offset */
392 prefix += cur->scb_s.mso;
393 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
394 prefix_unmapped_sync(cur);
399 * Map the first prefix page and if tx is enabled also the second prefix page.
401 * The prefix will be protected, a gmap notifier will inform about unmaps.
402 * The shadow scb must not be executed until the prefix is remapped, this is
403 * guaranteed by properly handling PROG_REQUEST.
405 * Returns: - 0 on if successfully mapped or already mapped
406 * - > 0 if control has to be given to guest 2
407 * - -EAGAIN if the caller can retry immediately
408 * - -ENOMEM if out of memory
410 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
412 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
413 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
416 if (prefix_is_mapped(vsie_page))
419 /* mark it as mapped so we can catch any concurrent unmappers */
420 prefix_mapped(vsie_page);
422 /* with mso/msl, the prefix lies at offset *mso* */
423 prefix += scb_s->mso;
425 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix);
426 if (!rc && (scb_s->ecb & ECB_TE))
427 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
430 * We don't have to mprotect, we will be called for all unshadows.
431 * SIE will detect if protection applies and trigger a validity.
434 prefix_unmapped(vsie_page);
435 if (rc > 0 || rc == -EFAULT)
436 rc = set_validity_icpt(scb_s, 0x0037U);
441 * Pin the guest page given by gpa and set hpa to the pinned host address.
442 * Will always be pinned writable.
444 * Returns: - 0 on success
445 * - -EINVAL if the gpa is not valid guest storage
447 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
451 page = gfn_to_page(kvm, gpa_to_gfn(gpa));
452 if (is_error_page(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)
461 kvm_release_pfn_dirty(hpa >> PAGE_SHIFT);
462 /* mark the page always as dirty for migration */
463 mark_page_dirty(kvm, gpa_to_gfn(gpa));
466 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
467 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
469 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
470 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
474 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
476 gpa = scb_o->scaol & ~0xfUL;
477 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
478 gpa |= (u64) scb_o->scaoh << 32;
479 unpin_guest_page(vcpu->kvm, gpa, hpa);
486 gpa = scb_o->itdba & ~0xffUL;
487 unpin_guest_page(vcpu->kvm, gpa, hpa);
493 gpa = scb_o->gvrd & ~0x1ffUL;
494 unpin_guest_page(vcpu->kvm, gpa, hpa);
500 gpa = scb_o->riccbd & ~0x3fUL;
501 unpin_guest_page(vcpu->kvm, gpa, hpa);
508 unpin_guest_page(vcpu->kvm, gpa, hpa);
514 * Instead of shadowing some blocks, we can simply forward them because the
515 * addresses in the scb are 64 bit long.
517 * This works as long as the data lies in one page. If blocks ever exceed one
518 * page, we have to fall back to shadowing.
520 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
521 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
523 * Returns: - 0 if all blocks were pinned.
524 * - > 0 if control has to be given to guest 2
525 * - -ENOMEM if out of memory
527 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
529 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
530 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
535 gpa = scb_o->scaol & ~0xfUL;
536 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
537 gpa |= (u64) scb_o->scaoh << 32;
539 if (!(gpa & ~0x1fffUL))
540 rc = set_validity_icpt(scb_s, 0x0038U);
541 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
542 rc = set_validity_icpt(scb_s, 0x0011U);
543 else if ((gpa & PAGE_MASK) !=
544 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
545 rc = set_validity_icpt(scb_s, 0x003bU);
547 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
549 rc = set_validity_icpt(scb_s, 0x0034U);
553 scb_s->scaoh = (u32)((u64)hpa >> 32);
554 scb_s->scaol = (u32)(u64)hpa;
557 gpa = scb_o->itdba & ~0xffUL;
558 if (gpa && (scb_s->ecb & ECB_TE)) {
559 if (!(gpa & ~0x1fffU)) {
560 rc = set_validity_icpt(scb_s, 0x0080U);
563 /* 256 bytes cannot cross page boundaries */
564 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
566 rc = set_validity_icpt(scb_s, 0x0080U);
572 gpa = scb_o->gvrd & ~0x1ffUL;
573 if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
574 if (!(gpa & ~0x1fffUL)) {
575 rc = set_validity_icpt(scb_s, 0x1310U);
579 * 512 bytes vector registers cannot cross page boundaries
580 * if this block gets bigger, we have to shadow it.
582 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
584 rc = set_validity_icpt(scb_s, 0x1310U);
590 gpa = scb_o->riccbd & ~0x3fUL;
591 if (gpa && (scb_s->ecb3 & ECB3_RI)) {
592 if (!(gpa & ~0x1fffUL)) {
593 rc = set_validity_icpt(scb_s, 0x0043U);
596 /* 64 bytes cannot cross page boundaries */
597 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
599 rc = set_validity_icpt(scb_s, 0x0043U);
602 /* Validity 0x0044 will be checked by SIE */
605 if ((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
608 gpa = scb_o->sdnxo & ~0xfUL;
609 sdnxc = scb_o->sdnxo & 0xfUL;
610 if (!gpa || !(gpa & ~0x1fffUL)) {
611 rc = set_validity_icpt(scb_s, 0x10b0U);
614 if (sdnxc < 6 || sdnxc > 12) {
615 rc = set_validity_icpt(scb_s, 0x10b1U);
618 if (gpa & ((1 << sdnxc) - 1)) {
619 rc = set_validity_icpt(scb_s, 0x10b2U);
622 /* Due to alignment rules (checked above) this cannot
623 * cross page boundaries
625 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
627 rc = set_validity_icpt(scb_s, 0x10b0U);
630 scb_s->sdnxo = hpa | sdnxc;
634 unpin_blocks(vcpu, vsie_page);
638 /* unpin the scb provided by guest 2, marking it as dirty */
639 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
642 hpa_t hpa = (hpa_t) vsie_page->scb_o;
645 unpin_guest_page(vcpu->kvm, gpa, hpa);
646 vsie_page->scb_o = NULL;
650 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
652 * Returns: - 0 if the scb was pinned.
653 * - > 0 if control has to be given to guest 2
655 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
661 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
663 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
667 vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
672 * Inject a fault into guest 2.
674 * Returns: - > 0 if control has to be given to guest 2
675 * < 0 if an error occurred during injection.
677 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
680 struct kvm_s390_pgm_info pgm = {
683 /* 0-51: virtual address */
684 (vaddr & 0xfffffffffffff000UL) |
685 /* 52-53: store / fetch */
686 (((unsigned int) !write_flag) + 1) << 10,
687 /* 62-63: asce id (alway primary == 0) */
688 .exc_access_id = 0, /* always primary */
689 .op_access_id = 0, /* not MVPG */
693 if (code == PGM_PROTECTION)
694 pgm.trans_exc_code |= 0x4UL;
696 rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
701 * Handle a fault during vsie execution on a gmap shadow.
703 * Returns: - 0 if the fault was resolved
704 * - > 0 if control has to be given to guest 2
705 * - < 0 if an error occurred
707 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
711 if (current->thread.gmap_int_code == PGM_PROTECTION)
712 /* we can directly forward all protection exceptions */
713 return inject_fault(vcpu, PGM_PROTECTION,
714 current->thread.gmap_addr, 1);
716 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
717 current->thread.gmap_addr);
719 rc = inject_fault(vcpu, rc,
720 current->thread.gmap_addr,
721 current->thread.gmap_write_flag);
723 vsie_page->fault_addr = current->thread.gmap_addr;
729 * Retry the previous fault that required guest 2 intervention. This avoids
730 * one superfluous SIE re-entry and direct exit.
732 * Will ignore any errors. The next SIE fault will do proper fault handling.
734 static void handle_last_fault(struct kvm_vcpu *vcpu,
735 struct vsie_page *vsie_page)
737 if (vsie_page->fault_addr)
738 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
739 vsie_page->fault_addr);
740 vsie_page->fault_addr = 0;
743 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
745 vsie_page->scb_s.icptcode = 0;
748 /* rewind the psw and clear the vsie icpt, so we can retry execution */
749 static void retry_vsie_icpt(struct vsie_page *vsie_page)
751 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
752 int ilen = insn_length(scb_s->ipa >> 8);
754 /* take care of EXECUTE instructions */
755 if (scb_s->icptstatus & 1) {
756 ilen = (scb_s->icptstatus >> 4) & 0x6;
760 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
761 clear_vsie_icpt(vsie_page);
765 * Try to shadow + enable the guest 2 provided facility list.
766 * Retry instruction execution if enabled for and provided by guest 2.
768 * Returns: - 0 if handled (retry or guest 2 icpt)
769 * - > 0 if control has to be given to guest 2
771 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
773 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
774 __u32 fac = vsie_page->scb_o->fac & 0x7ffffff8U;
776 if (fac && test_kvm_facility(vcpu->kvm, 7)) {
777 retry_vsie_icpt(vsie_page);
778 if (read_guest_real(vcpu, fac, &vsie_page->fac,
779 sizeof(vsie_page->fac)))
780 return set_validity_icpt(scb_s, 0x1090U);
781 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
787 * Run the vsie on a shadow scb and a shadow gmap, without any further
788 * sanity checks, handling SIE faults.
790 * Returns: - 0 everything went fine
791 * - > 0 if control has to be given to guest 2
792 * - < 0 if an error occurred
794 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
796 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
797 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
800 handle_last_fault(vcpu, vsie_page);
804 if (test_cpu_flag(CIF_MCCK_PENDING))
807 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
809 guest_enter_irqoff();
812 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
817 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
820 VCPU_EVENT(vcpu, 3, "%s", "machine check");
821 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
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;
908 if (scb_s->ecd & ECD_MEF) {
909 scb_s->epdx += vcpu->kvm->arch.epdx;
910 if (scb_s->epoch < vcpu->kvm->arch.epoch)
918 * Unregister a shadow scb from a VCPU.
920 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
922 atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags);
923 WRITE_ONCE(vcpu->arch.vsie_block, NULL);
927 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
928 * prefix pages and faults.
930 * Returns: - 0 if no errors occurred
931 * - > 0 if control has to be given to guest 2
932 * - -ENOMEM if out of memory
934 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
936 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
940 rc = acquire_gmap_shadow(vcpu, vsie_page);
942 rc = map_prefix(vcpu, vsie_page);
944 gmap_enable(vsie_page->gmap);
945 update_intervention_requests(vsie_page);
946 rc = do_vsie_run(vcpu, vsie_page);
947 gmap_enable(vcpu->arch.gmap);
949 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
953 if (rc || scb_s->icptcode || signal_pending(current) ||
954 kvm_s390_vcpu_has_irq(vcpu, 0))
960 * Addressing exceptions are always presentes as intercepts.
961 * As addressing exceptions are suppressing and our guest 3 PSW
962 * points at the responsible instruction, we have to
963 * forward the PSW and set the ilc. If we can't read guest 3
964 * instruction, we can use an arbitrary ilc. Let's always use
965 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
966 * memory. (we could also fake the shadow so the hardware
969 scb_s->icptcode = ICPT_PROGI;
970 scb_s->iprcc = PGM_ADDRESSING;
972 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
978 * Get or create a vsie page for a scb address.
980 * Returns: - address of a vsie page (cached or new one)
981 * - NULL if the same scb address is already used by another VCPU
982 * - ERR_PTR(-ENOMEM) if out of memory
984 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
986 struct vsie_page *vsie_page;
991 page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
994 if (page_ref_inc_return(page) == 2)
995 return page_to_virt(page);
1000 * We want at least #online_vcpus shadows, so every VCPU can execute
1001 * the VSIE in parallel.
1003 nr_vcpus = atomic_read(&kvm->online_vcpus);
1005 mutex_lock(&kvm->arch.vsie.mutex);
1006 if (kvm->arch.vsie.page_count < nr_vcpus) {
1007 page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA);
1009 mutex_unlock(&kvm->arch.vsie.mutex);
1010 return ERR_PTR(-ENOMEM);
1013 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1014 kvm->arch.vsie.page_count++;
1016 /* reuse an existing entry that belongs to nobody */
1018 page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1019 if (page_ref_inc_return(page) == 2)
1022 kvm->arch.vsie.next++;
1023 kvm->arch.vsie.next %= nr_vcpus;
1025 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1028 /* double use of the same address */
1029 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1031 mutex_unlock(&kvm->arch.vsie.mutex);
1034 mutex_unlock(&kvm->arch.vsie.mutex);
1036 vsie_page = page_to_virt(page);
1037 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1038 release_gmap_shadow(vsie_page);
1039 vsie_page->fault_addr = 0;
1040 vsie_page->scb_s.ihcpu = 0xffffU;
1044 /* put a vsie page acquired via get_vsie_page */
1045 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1047 struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1052 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1054 struct vsie_page *vsie_page;
1055 unsigned long scb_addr;
1058 vcpu->stat.instruction_sie++;
1059 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1061 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1062 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1064 BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1065 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1067 /* 512 byte alignment */
1068 if (unlikely(scb_addr & 0x1ffUL))
1069 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1071 if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0))
1074 vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1075 if (IS_ERR(vsie_page))
1076 return PTR_ERR(vsie_page);
1077 else if (!vsie_page)
1078 /* double use of sie control block - simply do nothing */
1081 rc = pin_scb(vcpu, vsie_page, scb_addr);
1084 rc = shadow_scb(vcpu, vsie_page);
1087 rc = pin_blocks(vcpu, vsie_page);
1090 register_shadow_scb(vcpu, vsie_page);
1091 rc = vsie_run(vcpu, vsie_page);
1092 unregister_shadow_scb(vcpu);
1093 unpin_blocks(vcpu, vsie_page);
1095 unshadow_scb(vcpu, vsie_page);
1097 unpin_scb(vcpu, vsie_page, scb_addr);
1099 put_vsie_page(vcpu->kvm, vsie_page);
1101 return rc < 0 ? rc : 0;
1104 /* Init the vsie data structures. To be called when a vm is initialized. */
1105 void kvm_s390_vsie_init(struct kvm *kvm)
1107 mutex_init(&kvm->arch.vsie.mutex);
1108 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL);
1111 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1112 void kvm_s390_vsie_destroy(struct kvm *kvm)
1114 struct vsie_page *vsie_page;
1118 mutex_lock(&kvm->arch.vsie.mutex);
1119 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1120 page = kvm->arch.vsie.pages[i];
1121 kvm->arch.vsie.pages[i] = NULL;
1122 vsie_page = page_to_virt(page);
1123 release_gmap_shadow(vsie_page);
1124 /* free the radix tree entry */
1125 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1128 kvm->arch.vsie.page_count = 0;
1129 mutex_unlock(&kvm->arch.vsie.mutex);
1132 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1134 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1137 * Even if the VCPU lets go of the shadow sie block reference, it is
1138 * still valid in the cache. So we can safely kick it.
1141 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1142 if (scb->prog0c & PROG_IN_SIE)
1143 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);