xref: /linux/arch/s390/kvm/vsie.c (revision ba199dc909a20fe62270ae4e93f263987bb9d119)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * kvm nested virtualization support for s390x
4  *
5  * Copyright IBM Corp. 2016, 2018
6  *
7  *    Author(s): David Hildenbrand <dahi@linux.vnet.ibm.com>
8  */
9 #include <linux/vmalloc.h>
10 #include <linux/kvm_host.h>
11 #include <linux/bug.h>
12 #include <linux/list.h>
13 #include <linux/bitmap.h>
14 #include <linux/sched/signal.h>
15 #include <linux/io.h>
16 
17 #include <asm/gmap.h>
18 #include <asm/mmu_context.h>
19 #include <asm/sclp.h>
20 #include <asm/nmi.h>
21 #include <asm/dis.h>
22 #include <asm/facility.h>
23 #include "kvm-s390.h"
24 #include "gaccess.h"
25 
26 struct vsie_page {
27 	struct kvm_s390_sie_block scb_s;	/* 0x0000 */
28 	/*
29 	 * the backup info for machine check. ensure it's at
30 	 * the same offset as that in struct sie_page!
31 	 */
32 	struct mcck_volatile_info mcck_info;    /* 0x0200 */
33 	/*
34 	 * The pinned original scb. Be aware that other VCPUs can modify
35 	 * it while we read from it. Values that are used for conditions or
36 	 * are reused conditionally, should be accessed via READ_ONCE.
37 	 */
38 	struct kvm_s390_sie_block *scb_o;	/* 0x0218 */
39 	/* the shadow gmap in use by the vsie_page */
40 	struct gmap *gmap;			/* 0x0220 */
41 	/* address of the last reported fault to guest2 */
42 	unsigned long fault_addr;		/* 0x0228 */
43 	/* calculated guest addresses of satellite control blocks */
44 	gpa_t sca_gpa;				/* 0x0230 */
45 	gpa_t itdba_gpa;			/* 0x0238 */
46 	gpa_t gvrd_gpa;				/* 0x0240 */
47 	gpa_t riccbd_gpa;			/* 0x0248 */
48 	gpa_t sdnx_gpa;				/* 0x0250 */
49 	__u8 reserved[0x0700 - 0x0258];		/* 0x0258 */
50 	struct kvm_s390_crypto_cb crycb;	/* 0x0700 */
51 	__u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE];	/* 0x0800 */
52 };
53 
54 /* trigger a validity icpt for the given scb */
55 static int set_validity_icpt(struct kvm_s390_sie_block *scb,
56 			     __u16 reason_code)
57 {
58 	scb->ipa = 0x1000;
59 	scb->ipb = ((__u32) reason_code) << 16;
60 	scb->icptcode = ICPT_VALIDITY;
61 	return 1;
62 }
63 
64 /* mark the prefix as unmapped, this will block the VSIE */
65 static void prefix_unmapped(struct vsie_page *vsie_page)
66 {
67 	atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
68 }
69 
70 /* mark the prefix as unmapped and wait until the VSIE has been left */
71 static void prefix_unmapped_sync(struct vsie_page *vsie_page)
72 {
73 	prefix_unmapped(vsie_page);
74 	if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
75 		atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
76 	while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
77 		cpu_relax();
78 }
79 
80 /* mark the prefix as mapped, this will allow the VSIE to run */
81 static void prefix_mapped(struct vsie_page *vsie_page)
82 {
83 	atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
84 }
85 
86 /* test if the prefix is mapped into the gmap shadow */
87 static int prefix_is_mapped(struct vsie_page *vsie_page)
88 {
89 	return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
90 }
91 
92 /* copy the updated intervention request bits into the shadow scb */
93 static void update_intervention_requests(struct vsie_page *vsie_page)
94 {
95 	const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
96 	int cpuflags;
97 
98 	cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
99 	atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
100 	atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
101 }
102 
103 /* shadow (filter and validate) the cpuflags  */
104 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
105 {
106 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
107 	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
108 	int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
109 
110 	/* we don't allow ESA/390 guests */
111 	if (!(cpuflags & CPUSTAT_ZARCH))
112 		return set_validity_icpt(scb_s, 0x0001U);
113 
114 	if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
115 		return set_validity_icpt(scb_s, 0x0001U);
116 	else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
117 		return set_validity_icpt(scb_s, 0x0007U);
118 
119 	/* intervention requests will be set later */
120 	newflags = CPUSTAT_ZARCH;
121 	if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
122 		newflags |= CPUSTAT_GED;
123 	if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
124 		if (cpuflags & CPUSTAT_GED)
125 			return set_validity_icpt(scb_s, 0x0001U);
126 		newflags |= CPUSTAT_GED2;
127 	}
128 	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
129 		newflags |= cpuflags & CPUSTAT_P;
130 	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
131 		newflags |= cpuflags & CPUSTAT_SM;
132 	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
133 		newflags |= cpuflags & CPUSTAT_IBS;
134 	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS))
135 		newflags |= cpuflags & CPUSTAT_KSS;
136 
137 	atomic_set(&scb_s->cpuflags, newflags);
138 	return 0;
139 }
140 /* Copy to APCB FORMAT1 from APCB FORMAT0 */
141 static int setup_apcb10(struct kvm_vcpu *vcpu, struct kvm_s390_apcb1 *apcb_s,
142 			unsigned long crycb_gpa, struct kvm_s390_apcb1 *apcb_h)
143 {
144 	struct kvm_s390_apcb0 tmp;
145 	unsigned long apcb_gpa;
146 
147 	apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb0);
148 
149 	if (read_guest_real(vcpu, apcb_gpa, &tmp,
150 			    sizeof(struct kvm_s390_apcb0)))
151 		return -EFAULT;
152 
153 	apcb_s->apm[0] = apcb_h->apm[0] & tmp.apm[0];
154 	apcb_s->aqm[0] = apcb_h->aqm[0] & tmp.aqm[0] & 0xffff000000000000UL;
155 	apcb_s->adm[0] = apcb_h->adm[0] & tmp.adm[0] & 0xffff000000000000UL;
156 
157 	return 0;
158 
159 }
160 
161 /**
162  * setup_apcb00 - Copy to APCB FORMAT0 from APCB FORMAT0
163  * @vcpu: pointer to the virtual CPU
164  * @apcb_s: pointer to start of apcb in the shadow crycb
165  * @crycb_gpa: guest physical address to start of original guest crycb
166  * @apcb_h: pointer to start of apcb in the guest1
167  *
168  * Returns 0 and -EFAULT on error reading guest apcb
169  */
170 static int setup_apcb00(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
171 			unsigned long crycb_gpa, unsigned long *apcb_h)
172 {
173 	unsigned long apcb_gpa;
174 
175 	apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb0);
176 
177 	if (read_guest_real(vcpu, apcb_gpa, apcb_s,
178 			    sizeof(struct kvm_s390_apcb0)))
179 		return -EFAULT;
180 
181 	bitmap_and(apcb_s, apcb_s, apcb_h,
182 		   BITS_PER_BYTE * sizeof(struct kvm_s390_apcb0));
183 
184 	return 0;
185 }
186 
187 /**
188  * setup_apcb11 - Copy the FORMAT1 APCB from the guest to the shadow CRYCB
189  * @vcpu: pointer to the virtual CPU
190  * @apcb_s: pointer to start of apcb in the shadow crycb
191  * @crycb_gpa: guest physical address to start of original guest crycb
192  * @apcb_h: pointer to start of apcb in the host
193  *
194  * Returns 0 and -EFAULT on error reading guest apcb
195  */
196 static int setup_apcb11(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
197 			unsigned long crycb_gpa,
198 			unsigned long *apcb_h)
199 {
200 	unsigned long apcb_gpa;
201 
202 	apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb1);
203 
204 	if (read_guest_real(vcpu, apcb_gpa, apcb_s,
205 			    sizeof(struct kvm_s390_apcb1)))
206 		return -EFAULT;
207 
208 	bitmap_and(apcb_s, apcb_s, apcb_h,
209 		   BITS_PER_BYTE * sizeof(struct kvm_s390_apcb1));
210 
211 	return 0;
212 }
213 
214 /**
215  * setup_apcb - Create a shadow copy of the apcb.
216  * @vcpu: pointer to the virtual CPU
217  * @crycb_s: pointer to shadow crycb
218  * @crycb_gpa: guest physical address of original guest crycb
219  * @crycb_h: pointer to the host crycb
220  * @fmt_o: format of the original guest crycb.
221  * @fmt_h: format of the host crycb.
222  *
223  * Checks the compatibility between the guest and host crycb and calls the
224  * appropriate copy function.
225  *
226  * Return 0 or an error number if the guest and host crycb are incompatible.
227  */
228 static int setup_apcb(struct kvm_vcpu *vcpu, struct kvm_s390_crypto_cb *crycb_s,
229 	       const u32 crycb_gpa,
230 	       struct kvm_s390_crypto_cb *crycb_h,
231 	       int fmt_o, int fmt_h)
232 {
233 	switch (fmt_o) {
234 	case CRYCB_FORMAT2:
235 		if ((crycb_gpa & PAGE_MASK) != ((crycb_gpa + 256) & PAGE_MASK))
236 			return -EACCES;
237 		if (fmt_h != CRYCB_FORMAT2)
238 			return -EINVAL;
239 		return setup_apcb11(vcpu, (unsigned long *)&crycb_s->apcb1,
240 				    crycb_gpa,
241 				    (unsigned long *)&crycb_h->apcb1);
242 	case CRYCB_FORMAT1:
243 		switch (fmt_h) {
244 		case CRYCB_FORMAT2:
245 			return setup_apcb10(vcpu, &crycb_s->apcb1,
246 					    crycb_gpa,
247 					    &crycb_h->apcb1);
248 		case CRYCB_FORMAT1:
249 			return setup_apcb00(vcpu,
250 					    (unsigned long *) &crycb_s->apcb0,
251 					    crycb_gpa,
252 					    (unsigned long *) &crycb_h->apcb0);
253 		}
254 		break;
255 	case CRYCB_FORMAT0:
256 		if ((crycb_gpa & PAGE_MASK) != ((crycb_gpa + 32) & PAGE_MASK))
257 			return -EACCES;
258 
259 		switch (fmt_h) {
260 		case CRYCB_FORMAT2:
261 			return setup_apcb10(vcpu, &crycb_s->apcb1,
262 					    crycb_gpa,
263 					    &crycb_h->apcb1);
264 		case CRYCB_FORMAT1:
265 		case CRYCB_FORMAT0:
266 			return setup_apcb00(vcpu,
267 					    (unsigned long *) &crycb_s->apcb0,
268 					    crycb_gpa,
269 					    (unsigned long *) &crycb_h->apcb0);
270 		}
271 	}
272 	return -EINVAL;
273 }
274 
275 /**
276  * shadow_crycb - Create a shadow copy of the crycb block
277  * @vcpu: a pointer to the virtual CPU
278  * @vsie_page: a pointer to internal date used for the vSIE
279  *
280  * Create a shadow copy of the crycb block and setup key wrapping, if
281  * requested for guest 3 and enabled for guest 2.
282  *
283  * We accept format-1 or format-2, but we convert format-1 into format-2
284  * in the shadow CRYCB.
285  * Using format-2 enables the firmware to choose the right format when
286  * scheduling the SIE.
287  * There is nothing to do for format-0.
288  *
289  * This function centralize the issuing of set_validity_icpt() for all
290  * the subfunctions working on the crycb.
291  *
292  * Returns: - 0 if shadowed or nothing to do
293  *          - > 0 if control has to be given to guest 2
294  */
295 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
296 {
297 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
298 	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
299 	const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd);
300 	const u32 crycb_addr = crycbd_o & 0x7ffffff8U;
301 	unsigned long *b1, *b2;
302 	u8 ecb3_flags;
303 	u32 ecd_flags;
304 	int apie_h;
305 	int apie_s;
306 	int key_msk = test_kvm_facility(vcpu->kvm, 76);
307 	int fmt_o = crycbd_o & CRYCB_FORMAT_MASK;
308 	int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK;
309 	int ret = 0;
310 
311 	scb_s->crycbd = 0;
312 
313 	apie_h = vcpu->arch.sie_block->eca & ECA_APIE;
314 	apie_s = apie_h & scb_o->eca;
315 	if (!apie_s && (!key_msk || (fmt_o == CRYCB_FORMAT0)))
316 		return 0;
317 
318 	if (!crycb_addr)
319 		return set_validity_icpt(scb_s, 0x0039U);
320 
321 	if (fmt_o == CRYCB_FORMAT1)
322 		if ((crycb_addr & PAGE_MASK) !=
323 		    ((crycb_addr + 128) & PAGE_MASK))
324 			return set_validity_icpt(scb_s, 0x003CU);
325 
326 	if (apie_s) {
327 		ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr,
328 				 vcpu->kvm->arch.crypto.crycb,
329 				 fmt_o, fmt_h);
330 		if (ret)
331 			goto end;
332 		scb_s->eca |= scb_o->eca & ECA_APIE;
333 	}
334 
335 	/* we may only allow it if enabled for guest 2 */
336 	ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
337 		     (ECB3_AES | ECB3_DEA);
338 	ecd_flags = scb_o->ecd & vcpu->arch.sie_block->ecd & ECD_ECC;
339 	if (!ecb3_flags && !ecd_flags)
340 		goto end;
341 
342 	/* copy only the wrapping keys */
343 	if (read_guest_real(vcpu, crycb_addr + 72,
344 			    vsie_page->crycb.dea_wrapping_key_mask, 56))
345 		return set_validity_icpt(scb_s, 0x0035U);
346 
347 	scb_s->ecb3 |= ecb3_flags;
348 	scb_s->ecd |= ecd_flags;
349 
350 	/* xor both blocks in one run */
351 	b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
352 	b2 = (unsigned long *)
353 			    vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
354 	/* as 56%8 == 0, bitmap_xor won't overwrite any data */
355 	bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
356 end:
357 	switch (ret) {
358 	case -EINVAL:
359 		return set_validity_icpt(scb_s, 0x0022U);
360 	case -EFAULT:
361 		return set_validity_icpt(scb_s, 0x0035U);
362 	case -EACCES:
363 		return set_validity_icpt(scb_s, 0x003CU);
364 	}
365 	scb_s->crycbd = (u32)virt_to_phys(&vsie_page->crycb) | CRYCB_FORMAT2;
366 	return 0;
367 }
368 
369 /* shadow (round up/down) the ibc to avoid validity icpt */
370 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
371 {
372 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
373 	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
374 	/* READ_ONCE does not work on bitfields - use a temporary variable */
375 	const uint32_t __new_ibc = scb_o->ibc;
376 	const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU;
377 	__u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
378 
379 	scb_s->ibc = 0;
380 	/* ibc installed in g2 and requested for g3 */
381 	if (vcpu->kvm->arch.model.ibc && new_ibc) {
382 		scb_s->ibc = new_ibc;
383 		/* takte care of the minimum ibc level of the machine */
384 		if (scb_s->ibc < min_ibc)
385 			scb_s->ibc = min_ibc;
386 		/* take care of the maximum ibc level set for the guest */
387 		if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
388 			scb_s->ibc = vcpu->kvm->arch.model.ibc;
389 	}
390 }
391 
392 /* unshadow the scb, copying parameters back to the real scb */
393 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
394 {
395 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
396 	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
397 
398 	/* interception */
399 	scb_o->icptcode = scb_s->icptcode;
400 	scb_o->icptstatus = scb_s->icptstatus;
401 	scb_o->ipa = scb_s->ipa;
402 	scb_o->ipb = scb_s->ipb;
403 	scb_o->gbea = scb_s->gbea;
404 
405 	/* timer */
406 	scb_o->cputm = scb_s->cputm;
407 	scb_o->ckc = scb_s->ckc;
408 	scb_o->todpr = scb_s->todpr;
409 
410 	/* guest state */
411 	scb_o->gpsw = scb_s->gpsw;
412 	scb_o->gg14 = scb_s->gg14;
413 	scb_o->gg15 = scb_s->gg15;
414 	memcpy(scb_o->gcr, scb_s->gcr, 128);
415 	scb_o->pp = scb_s->pp;
416 
417 	/* branch prediction */
418 	if (test_kvm_facility(vcpu->kvm, 82)) {
419 		scb_o->fpf &= ~FPF_BPBC;
420 		scb_o->fpf |= scb_s->fpf & FPF_BPBC;
421 	}
422 
423 	/* interrupt intercept */
424 	switch (scb_s->icptcode) {
425 	case ICPT_PROGI:
426 	case ICPT_INSTPROGI:
427 	case ICPT_EXTINT:
428 		memcpy((void *)((u64)scb_o + 0xc0),
429 		       (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
430 		break;
431 	}
432 
433 	if (scb_s->ihcpu != 0xffffU)
434 		scb_o->ihcpu = scb_s->ihcpu;
435 }
436 
437 /*
438  * Setup the shadow scb by copying and checking the relevant parts of the g2
439  * provided scb.
440  *
441  * Returns: - 0 if the scb has been shadowed
442  *          - > 0 if control has to be given to guest 2
443  */
444 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
445 {
446 	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
447 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
448 	/* READ_ONCE does not work on bitfields - use a temporary variable */
449 	const uint32_t __new_prefix = scb_o->prefix;
450 	const uint32_t new_prefix = READ_ONCE(__new_prefix);
451 	const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE;
452 	bool had_tx = scb_s->ecb & ECB_TE;
453 	unsigned long new_mso = 0;
454 	int rc;
455 
456 	/* make sure we don't have any leftovers when reusing the scb */
457 	scb_s->icptcode = 0;
458 	scb_s->eca = 0;
459 	scb_s->ecb = 0;
460 	scb_s->ecb2 = 0;
461 	scb_s->ecb3 = 0;
462 	scb_s->ecd = 0;
463 	scb_s->fac = 0;
464 	scb_s->fpf = 0;
465 
466 	rc = prepare_cpuflags(vcpu, vsie_page);
467 	if (rc)
468 		goto out;
469 
470 	/* timer */
471 	scb_s->cputm = scb_o->cputm;
472 	scb_s->ckc = scb_o->ckc;
473 	scb_s->todpr = scb_o->todpr;
474 	scb_s->epoch = scb_o->epoch;
475 
476 	/* guest state */
477 	scb_s->gpsw = scb_o->gpsw;
478 	scb_s->gg14 = scb_o->gg14;
479 	scb_s->gg15 = scb_o->gg15;
480 	memcpy(scb_s->gcr, scb_o->gcr, 128);
481 	scb_s->pp = scb_o->pp;
482 
483 	/* interception / execution handling */
484 	scb_s->gbea = scb_o->gbea;
485 	scb_s->lctl = scb_o->lctl;
486 	scb_s->svcc = scb_o->svcc;
487 	scb_s->ictl = scb_o->ictl;
488 	/*
489 	 * SKEY handling functions can't deal with false setting of PTE invalid
490 	 * bits. Therefore we cannot provide interpretation and would later
491 	 * have to provide own emulation handlers.
492 	 */
493 	if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
494 		scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
495 
496 	scb_s->icpua = scb_o->icpua;
497 
498 	if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
499 		new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL;
500 	/* if the hva of the prefix changes, we have to remap the prefix */
501 	if (scb_s->mso != new_mso || scb_s->prefix != new_prefix)
502 		prefix_unmapped(vsie_page);
503 	 /* SIE will do mso/msl validity and exception checks for us */
504 	scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
505 	scb_s->mso = new_mso;
506 	scb_s->prefix = new_prefix;
507 
508 	/* We have to definitely flush the tlb if this scb never ran */
509 	if (scb_s->ihcpu != 0xffffU)
510 		scb_s->ihcpu = scb_o->ihcpu;
511 
512 	/* MVPG and Protection Exception Interpretation are always available */
513 	scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
514 	/* Host-protection-interruption introduced with ESOP */
515 	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
516 		scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
517 	/*
518 	 * CPU Topology
519 	 * This facility only uses the utility field of the SCA and none of
520 	 * the cpu entries that are problematic with the other interpretation
521 	 * facilities so we can pass it through
522 	 */
523 	if (test_kvm_facility(vcpu->kvm, 11))
524 		scb_s->ecb |= scb_o->ecb & ECB_PTF;
525 	/* transactional execution */
526 	if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) {
527 		/* remap the prefix is tx is toggled on */
528 		if (!had_tx)
529 			prefix_unmapped(vsie_page);
530 		scb_s->ecb |= ECB_TE;
531 	}
532 	/* specification exception interpretation */
533 	scb_s->ecb |= scb_o->ecb & ECB_SPECI;
534 	/* branch prediction */
535 	if (test_kvm_facility(vcpu->kvm, 82))
536 		scb_s->fpf |= scb_o->fpf & FPF_BPBC;
537 	/* SIMD */
538 	if (test_kvm_facility(vcpu->kvm, 129)) {
539 		scb_s->eca |= scb_o->eca & ECA_VX;
540 		scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
541 	}
542 	/* Run-time-Instrumentation */
543 	if (test_kvm_facility(vcpu->kvm, 64))
544 		scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
545 	/* Instruction Execution Prevention */
546 	if (test_kvm_facility(vcpu->kvm, 130))
547 		scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
548 	/* Guarded Storage */
549 	if (test_kvm_facility(vcpu->kvm, 133)) {
550 		scb_s->ecb |= scb_o->ecb & ECB_GS;
551 		scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
552 	}
553 	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
554 		scb_s->eca |= scb_o->eca & ECA_SII;
555 	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
556 		scb_s->eca |= scb_o->eca & ECA_IB;
557 	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
558 		scb_s->eca |= scb_o->eca & ECA_CEI;
559 	/* Epoch Extension */
560 	if (test_kvm_facility(vcpu->kvm, 139)) {
561 		scb_s->ecd |= scb_o->ecd & ECD_MEF;
562 		scb_s->epdx = scb_o->epdx;
563 	}
564 
565 	/* etoken */
566 	if (test_kvm_facility(vcpu->kvm, 156))
567 		scb_s->ecd |= scb_o->ecd & ECD_ETOKENF;
568 
569 	scb_s->hpid = HPID_VSIE;
570 	scb_s->cpnc = scb_o->cpnc;
571 
572 	prepare_ibc(vcpu, vsie_page);
573 	rc = shadow_crycb(vcpu, vsie_page);
574 out:
575 	if (rc)
576 		unshadow_scb(vcpu, vsie_page);
577 	return rc;
578 }
579 
580 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
581 				 unsigned long end)
582 {
583 	struct kvm *kvm = gmap->private;
584 	struct vsie_page *cur;
585 	unsigned long prefix;
586 	struct page *page;
587 	int i;
588 
589 	if (!gmap_is_shadow(gmap))
590 		return;
591 	/*
592 	 * Only new shadow blocks are added to the list during runtime,
593 	 * therefore we can safely reference them all the time.
594 	 */
595 	for (i = 0; i < kvm->arch.vsie.page_count; i++) {
596 		page = READ_ONCE(kvm->arch.vsie.pages[i]);
597 		if (!page)
598 			continue;
599 		cur = page_to_virt(page);
600 		if (READ_ONCE(cur->gmap) != gmap)
601 			continue;
602 		prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
603 		/* with mso/msl, the prefix lies at an offset */
604 		prefix += cur->scb_s.mso;
605 		if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
606 			prefix_unmapped_sync(cur);
607 	}
608 }
609 
610 /*
611  * Map the first prefix page and if tx is enabled also the second prefix page.
612  *
613  * The prefix will be protected, a gmap notifier will inform about unmaps.
614  * The shadow scb must not be executed until the prefix is remapped, this is
615  * guaranteed by properly handling PROG_REQUEST.
616  *
617  * Returns: - 0 on if successfully mapped or already mapped
618  *          - > 0 if control has to be given to guest 2
619  *          - -EAGAIN if the caller can retry immediately
620  *          - -ENOMEM if out of memory
621  */
622 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
623 {
624 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
625 	u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
626 	int rc;
627 
628 	if (prefix_is_mapped(vsie_page))
629 		return 0;
630 
631 	/* mark it as mapped so we can catch any concurrent unmappers */
632 	prefix_mapped(vsie_page);
633 
634 	/* with mso/msl, the prefix lies at offset *mso* */
635 	prefix += scb_s->mso;
636 
637 	rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix, NULL);
638 	if (!rc && (scb_s->ecb & ECB_TE))
639 		rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
640 					   prefix + PAGE_SIZE, NULL);
641 	/*
642 	 * We don't have to mprotect, we will be called for all unshadows.
643 	 * SIE will detect if protection applies and trigger a validity.
644 	 */
645 	if (rc)
646 		prefix_unmapped(vsie_page);
647 	if (rc > 0 || rc == -EFAULT)
648 		rc = set_validity_icpt(scb_s, 0x0037U);
649 	return rc;
650 }
651 
652 /*
653  * Pin the guest page given by gpa and set hpa to the pinned host address.
654  * Will always be pinned writable.
655  *
656  * Returns: - 0 on success
657  *          - -EINVAL if the gpa is not valid guest storage
658  */
659 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
660 {
661 	struct page *page;
662 
663 	page = gfn_to_page(kvm, gpa_to_gfn(gpa));
664 	if (is_error_page(page))
665 		return -EINVAL;
666 	*hpa = (hpa_t)page_to_phys(page) + (gpa & ~PAGE_MASK);
667 	return 0;
668 }
669 
670 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
671 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
672 {
673 	kvm_release_pfn_dirty(hpa >> PAGE_SHIFT);
674 	/* mark the page always as dirty for migration */
675 	mark_page_dirty(kvm, gpa_to_gfn(gpa));
676 }
677 
678 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
679 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
680 {
681 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
682 	hpa_t hpa;
683 
684 	hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
685 	if (hpa) {
686 		unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa);
687 		vsie_page->sca_gpa = 0;
688 		scb_s->scaol = 0;
689 		scb_s->scaoh = 0;
690 	}
691 
692 	hpa = scb_s->itdba;
693 	if (hpa) {
694 		unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa);
695 		vsie_page->itdba_gpa = 0;
696 		scb_s->itdba = 0;
697 	}
698 
699 	hpa = scb_s->gvrd;
700 	if (hpa) {
701 		unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa);
702 		vsie_page->gvrd_gpa = 0;
703 		scb_s->gvrd = 0;
704 	}
705 
706 	hpa = scb_s->riccbd;
707 	if (hpa) {
708 		unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa);
709 		vsie_page->riccbd_gpa = 0;
710 		scb_s->riccbd = 0;
711 	}
712 
713 	hpa = scb_s->sdnxo;
714 	if (hpa) {
715 		unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa);
716 		vsie_page->sdnx_gpa = 0;
717 		scb_s->sdnxo = 0;
718 	}
719 }
720 
721 /*
722  * Instead of shadowing some blocks, we can simply forward them because the
723  * addresses in the scb are 64 bit long.
724  *
725  * This works as long as the data lies in one page. If blocks ever exceed one
726  * page, we have to fall back to shadowing.
727  *
728  * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
729  * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
730  *
731  * Returns: - 0 if all blocks were pinned.
732  *          - > 0 if control has to be given to guest 2
733  *          - -ENOMEM if out of memory
734  */
735 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
736 {
737 	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
738 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
739 	hpa_t hpa;
740 	gpa_t gpa;
741 	int rc = 0;
742 
743 	gpa = READ_ONCE(scb_o->scaol) & ~0xfUL;
744 	if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
745 		gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32;
746 	if (gpa) {
747 		if (gpa < 2 * PAGE_SIZE)
748 			rc = set_validity_icpt(scb_s, 0x0038U);
749 		else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
750 			rc = set_validity_icpt(scb_s, 0x0011U);
751 		else if ((gpa & PAGE_MASK) !=
752 			 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
753 			rc = set_validity_icpt(scb_s, 0x003bU);
754 		if (!rc) {
755 			rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
756 			if (rc)
757 				rc = set_validity_icpt(scb_s, 0x0034U);
758 		}
759 		if (rc)
760 			goto unpin;
761 		vsie_page->sca_gpa = gpa;
762 		scb_s->scaoh = (u32)((u64)hpa >> 32);
763 		scb_s->scaol = (u32)(u64)hpa;
764 	}
765 
766 	gpa = READ_ONCE(scb_o->itdba) & ~0xffUL;
767 	if (gpa && (scb_s->ecb & ECB_TE)) {
768 		if (gpa < 2 * PAGE_SIZE) {
769 			rc = set_validity_icpt(scb_s, 0x0080U);
770 			goto unpin;
771 		}
772 		/* 256 bytes cannot cross page boundaries */
773 		rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
774 		if (rc) {
775 			rc = set_validity_icpt(scb_s, 0x0080U);
776 			goto unpin;
777 		}
778 		vsie_page->itdba_gpa = gpa;
779 		scb_s->itdba = hpa;
780 	}
781 
782 	gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL;
783 	if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
784 		if (gpa < 2 * PAGE_SIZE) {
785 			rc = set_validity_icpt(scb_s, 0x1310U);
786 			goto unpin;
787 		}
788 		/*
789 		 * 512 bytes vector registers cannot cross page boundaries
790 		 * if this block gets bigger, we have to shadow it.
791 		 */
792 		rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
793 		if (rc) {
794 			rc = set_validity_icpt(scb_s, 0x1310U);
795 			goto unpin;
796 		}
797 		vsie_page->gvrd_gpa = gpa;
798 		scb_s->gvrd = hpa;
799 	}
800 
801 	gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL;
802 	if (gpa && (scb_s->ecb3 & ECB3_RI)) {
803 		if (gpa < 2 * PAGE_SIZE) {
804 			rc = set_validity_icpt(scb_s, 0x0043U);
805 			goto unpin;
806 		}
807 		/* 64 bytes cannot cross page boundaries */
808 		rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
809 		if (rc) {
810 			rc = set_validity_icpt(scb_s, 0x0043U);
811 			goto unpin;
812 		}
813 		/* Validity 0x0044 will be checked by SIE */
814 		vsie_page->riccbd_gpa = gpa;
815 		scb_s->riccbd = hpa;
816 	}
817 	if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) ||
818 	    (scb_s->ecd & ECD_ETOKENF)) {
819 		unsigned long sdnxc;
820 
821 		gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL;
822 		sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL;
823 		if (!gpa || gpa < 2 * PAGE_SIZE) {
824 			rc = set_validity_icpt(scb_s, 0x10b0U);
825 			goto unpin;
826 		}
827 		if (sdnxc < 6 || sdnxc > 12) {
828 			rc = set_validity_icpt(scb_s, 0x10b1U);
829 			goto unpin;
830 		}
831 		if (gpa & ((1 << sdnxc) - 1)) {
832 			rc = set_validity_icpt(scb_s, 0x10b2U);
833 			goto unpin;
834 		}
835 		/* Due to alignment rules (checked above) this cannot
836 		 * cross page boundaries
837 		 */
838 		rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
839 		if (rc) {
840 			rc = set_validity_icpt(scb_s, 0x10b0U);
841 			goto unpin;
842 		}
843 		vsie_page->sdnx_gpa = gpa;
844 		scb_s->sdnxo = hpa | sdnxc;
845 	}
846 	return 0;
847 unpin:
848 	unpin_blocks(vcpu, vsie_page);
849 	return rc;
850 }
851 
852 /* unpin the scb provided by guest 2, marking it as dirty */
853 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
854 		      gpa_t gpa)
855 {
856 	hpa_t hpa = (hpa_t) vsie_page->scb_o;
857 
858 	if (hpa)
859 		unpin_guest_page(vcpu->kvm, gpa, hpa);
860 	vsie_page->scb_o = NULL;
861 }
862 
863 /*
864  * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
865  *
866  * Returns: - 0 if the scb was pinned.
867  *          - > 0 if control has to be given to guest 2
868  */
869 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
870 		   gpa_t gpa)
871 {
872 	hpa_t hpa;
873 	int rc;
874 
875 	rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
876 	if (rc) {
877 		rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
878 		WARN_ON_ONCE(rc);
879 		return 1;
880 	}
881 	vsie_page->scb_o = phys_to_virt(hpa);
882 	return 0;
883 }
884 
885 /*
886  * Inject a fault into guest 2.
887  *
888  * Returns: - > 0 if control has to be given to guest 2
889  *            < 0 if an error occurred during injection.
890  */
891 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
892 			bool write_flag)
893 {
894 	struct kvm_s390_pgm_info pgm = {
895 		.code = code,
896 		.trans_exc_code =
897 			/* 0-51: virtual address */
898 			(vaddr & 0xfffffffffffff000UL) |
899 			/* 52-53: store / fetch */
900 			(((unsigned int) !write_flag) + 1) << 10,
901 			/* 62-63: asce id (always primary == 0) */
902 		.exc_access_id = 0, /* always primary */
903 		.op_access_id = 0, /* not MVPG */
904 	};
905 	int rc;
906 
907 	if (code == PGM_PROTECTION)
908 		pgm.trans_exc_code |= 0x4UL;
909 
910 	rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
911 	return rc ? rc : 1;
912 }
913 
914 /*
915  * Handle a fault during vsie execution on a gmap shadow.
916  *
917  * Returns: - 0 if the fault was resolved
918  *          - > 0 if control has to be given to guest 2
919  *          - < 0 if an error occurred
920  */
921 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
922 {
923 	int rc;
924 
925 	if (current->thread.gmap_int_code == PGM_PROTECTION)
926 		/* we can directly forward all protection exceptions */
927 		return inject_fault(vcpu, PGM_PROTECTION,
928 				    current->thread.gmap_addr, 1);
929 
930 	rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
931 				   current->thread.gmap_addr, NULL);
932 	if (rc > 0) {
933 		rc = inject_fault(vcpu, rc,
934 				  current->thread.gmap_addr,
935 				  current->thread.gmap_write_flag);
936 		if (rc >= 0)
937 			vsie_page->fault_addr = current->thread.gmap_addr;
938 	}
939 	return rc;
940 }
941 
942 /*
943  * Retry the previous fault that required guest 2 intervention. This avoids
944  * one superfluous SIE re-entry and direct exit.
945  *
946  * Will ignore any errors. The next SIE fault will do proper fault handling.
947  */
948 static void handle_last_fault(struct kvm_vcpu *vcpu,
949 			      struct vsie_page *vsie_page)
950 {
951 	if (vsie_page->fault_addr)
952 		kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
953 				      vsie_page->fault_addr, NULL);
954 	vsie_page->fault_addr = 0;
955 }
956 
957 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
958 {
959 	vsie_page->scb_s.icptcode = 0;
960 }
961 
962 /* rewind the psw and clear the vsie icpt, so we can retry execution */
963 static void retry_vsie_icpt(struct vsie_page *vsie_page)
964 {
965 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
966 	int ilen = insn_length(scb_s->ipa >> 8);
967 
968 	/* take care of EXECUTE instructions */
969 	if (scb_s->icptstatus & 1) {
970 		ilen = (scb_s->icptstatus >> 4) & 0x6;
971 		if (!ilen)
972 			ilen = 4;
973 	}
974 	scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
975 	clear_vsie_icpt(vsie_page);
976 }
977 
978 /*
979  * Try to shadow + enable the guest 2 provided facility list.
980  * Retry instruction execution if enabled for and provided by guest 2.
981  *
982  * Returns: - 0 if handled (retry or guest 2 icpt)
983  *          - > 0 if control has to be given to guest 2
984  */
985 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
986 {
987 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
988 	__u32 fac = READ_ONCE(vsie_page->scb_o->fac);
989 
990 	/*
991 	 * Alternate-STFLE-Interpretive-Execution facilities are not supported
992 	 * -> format-0 flcb
993 	 */
994 	if (fac && test_kvm_facility(vcpu->kvm, 7)) {
995 		retry_vsie_icpt(vsie_page);
996 		/*
997 		 * The facility list origin (FLO) is in bits 1 - 28 of the FLD
998 		 * so we need to mask here before reading.
999 		 */
1000 		fac = fac & 0x7ffffff8U;
1001 		/*
1002 		 * format-0 -> size of nested guest's facility list == guest's size
1003 		 * guest's size == host's size, since STFLE is interpretatively executed
1004 		 * using a format-0 for the guest, too.
1005 		 */
1006 		if (read_guest_real(vcpu, fac, &vsie_page->fac,
1007 				    stfle_size() * sizeof(u64)))
1008 			return set_validity_icpt(scb_s, 0x1090U);
1009 		scb_s->fac = (u32)virt_to_phys(&vsie_page->fac);
1010 	}
1011 	return 0;
1012 }
1013 
1014 /*
1015  * Get a register for a nested guest.
1016  * @vcpu the vcpu of the guest
1017  * @vsie_page the vsie_page for the nested guest
1018  * @reg the register number, the upper 4 bits are ignored.
1019  * returns: the value of the register.
1020  */
1021 static u64 vsie_get_register(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, u8 reg)
1022 {
1023 	/* no need to validate the parameter and/or perform error handling */
1024 	reg &= 0xf;
1025 	switch (reg) {
1026 	case 15:
1027 		return vsie_page->scb_s.gg15;
1028 	case 14:
1029 		return vsie_page->scb_s.gg14;
1030 	default:
1031 		return vcpu->run->s.regs.gprs[reg];
1032 	}
1033 }
1034 
1035 static int vsie_handle_mvpg(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1036 {
1037 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1038 	unsigned long pei_dest, pei_src, src, dest, mask, prefix;
1039 	u64 *pei_block = &vsie_page->scb_o->mcic;
1040 	int edat, rc_dest, rc_src;
1041 	union ctlreg0 cr0;
1042 
1043 	cr0.val = vcpu->arch.sie_block->gcr[0];
1044 	edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1045 	mask = _kvm_s390_logical_to_effective(&scb_s->gpsw, PAGE_MASK);
1046 	prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
1047 
1048 	dest = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 20) & mask;
1049 	dest = _kvm_s390_real_to_abs(prefix, dest) + scb_s->mso;
1050 	src = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 16) & mask;
1051 	src = _kvm_s390_real_to_abs(prefix, src) + scb_s->mso;
1052 
1053 	rc_dest = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, dest, &pei_dest);
1054 	rc_src = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, src, &pei_src);
1055 	/*
1056 	 * Either everything went well, or something non-critical went wrong
1057 	 * e.g. because of a race. In either case, simply retry.
1058 	 */
1059 	if (rc_dest == -EAGAIN || rc_src == -EAGAIN || (!rc_dest && !rc_src)) {
1060 		retry_vsie_icpt(vsie_page);
1061 		return -EAGAIN;
1062 	}
1063 	/* Something more serious went wrong, propagate the error */
1064 	if (rc_dest < 0)
1065 		return rc_dest;
1066 	if (rc_src < 0)
1067 		return rc_src;
1068 
1069 	/* The only possible suppressing exception: just deliver it */
1070 	if (rc_dest == PGM_TRANSLATION_SPEC || rc_src == PGM_TRANSLATION_SPEC) {
1071 		clear_vsie_icpt(vsie_page);
1072 		rc_dest = kvm_s390_inject_program_int(vcpu, PGM_TRANSLATION_SPEC);
1073 		WARN_ON_ONCE(rc_dest);
1074 		return 1;
1075 	}
1076 
1077 	/*
1078 	 * Forward the PEI intercept to the guest if it was a page fault, or
1079 	 * also for segment and region table faults if EDAT applies.
1080 	 */
1081 	if (edat) {
1082 		rc_dest = rc_dest == PGM_ASCE_TYPE ? rc_dest : 0;
1083 		rc_src = rc_src == PGM_ASCE_TYPE ? rc_src : 0;
1084 	} else {
1085 		rc_dest = rc_dest != PGM_PAGE_TRANSLATION ? rc_dest : 0;
1086 		rc_src = rc_src != PGM_PAGE_TRANSLATION ? rc_src : 0;
1087 	}
1088 	if (!rc_dest && !rc_src) {
1089 		pei_block[0] = pei_dest;
1090 		pei_block[1] = pei_src;
1091 		return 1;
1092 	}
1093 
1094 	retry_vsie_icpt(vsie_page);
1095 
1096 	/*
1097 	 * The host has edat, and the guest does not, or it was an ASCE type
1098 	 * exception. The host needs to inject the appropriate DAT interrupts
1099 	 * into the guest.
1100 	 */
1101 	if (rc_dest)
1102 		return inject_fault(vcpu, rc_dest, dest, 1);
1103 	return inject_fault(vcpu, rc_src, src, 0);
1104 }
1105 
1106 /*
1107  * Run the vsie on a shadow scb and a shadow gmap, without any further
1108  * sanity checks, handling SIE faults.
1109  *
1110  * Returns: - 0 everything went fine
1111  *          - > 0 if control has to be given to guest 2
1112  *          - < 0 if an error occurred
1113  */
1114 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1115 	__releases(vcpu->kvm->srcu)
1116 	__acquires(vcpu->kvm->srcu)
1117 {
1118 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1119 	struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
1120 	int guest_bp_isolation;
1121 	int rc = 0;
1122 
1123 	handle_last_fault(vcpu, vsie_page);
1124 
1125 	kvm_vcpu_srcu_read_unlock(vcpu);
1126 
1127 	/* save current guest state of bp isolation override */
1128 	guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
1129 
1130 	/*
1131 	 * The guest is running with BPBC, so we have to force it on for our
1132 	 * nested guest. This is done by enabling BPBC globally, so the BPBC
1133 	 * control in the SCB (which the nested guest can modify) is simply
1134 	 * ignored.
1135 	 */
1136 	if (test_kvm_facility(vcpu->kvm, 82) &&
1137 	    vcpu->arch.sie_block->fpf & FPF_BPBC)
1138 		set_thread_flag(TIF_ISOLATE_BP_GUEST);
1139 
1140 	local_irq_disable();
1141 	guest_enter_irqoff();
1142 	local_irq_enable();
1143 
1144 	/*
1145 	 * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking
1146 	 * and VCPU requests also hinder the vSIE from running and lead
1147 	 * to an immediate exit. kvm_s390_vsie_kick() has to be used to
1148 	 * also kick the vSIE.
1149 	 */
1150 	vcpu->arch.sie_block->prog0c |= PROG_IN_SIE;
1151 	barrier();
1152 	if (!kvm_s390_vcpu_sie_inhibited(vcpu))
1153 		rc = sie64a(scb_s, vcpu->run->s.regs.gprs, gmap_get_enabled()->asce);
1154 	barrier();
1155 	vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE;
1156 
1157 	local_irq_disable();
1158 	guest_exit_irqoff();
1159 	local_irq_enable();
1160 
1161 	/* restore guest state for bp isolation override */
1162 	if (!guest_bp_isolation)
1163 		clear_thread_flag(TIF_ISOLATE_BP_GUEST);
1164 
1165 	kvm_vcpu_srcu_read_lock(vcpu);
1166 
1167 	if (rc == -EINTR) {
1168 		VCPU_EVENT(vcpu, 3, "%s", "machine check");
1169 		kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
1170 		return 0;
1171 	}
1172 
1173 	if (rc > 0)
1174 		rc = 0; /* we could still have an icpt */
1175 	else if (rc == -EFAULT)
1176 		return handle_fault(vcpu, vsie_page);
1177 
1178 	switch (scb_s->icptcode) {
1179 	case ICPT_INST:
1180 		if (scb_s->ipa == 0xb2b0)
1181 			rc = handle_stfle(vcpu, vsie_page);
1182 		break;
1183 	case ICPT_STOP:
1184 		/* stop not requested by g2 - must have been a kick */
1185 		if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
1186 			clear_vsie_icpt(vsie_page);
1187 		break;
1188 	case ICPT_VALIDITY:
1189 		if ((scb_s->ipa & 0xf000) != 0xf000)
1190 			scb_s->ipa += 0x1000;
1191 		break;
1192 	case ICPT_PARTEXEC:
1193 		if (scb_s->ipa == 0xb254)
1194 			rc = vsie_handle_mvpg(vcpu, vsie_page);
1195 		break;
1196 	}
1197 	return rc;
1198 }
1199 
1200 static void release_gmap_shadow(struct vsie_page *vsie_page)
1201 {
1202 	if (vsie_page->gmap)
1203 		gmap_put(vsie_page->gmap);
1204 	WRITE_ONCE(vsie_page->gmap, NULL);
1205 	prefix_unmapped(vsie_page);
1206 }
1207 
1208 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
1209 			       struct vsie_page *vsie_page)
1210 {
1211 	unsigned long asce;
1212 	union ctlreg0 cr0;
1213 	struct gmap *gmap;
1214 	int edat;
1215 
1216 	asce = vcpu->arch.sie_block->gcr[1];
1217 	cr0.val = vcpu->arch.sie_block->gcr[0];
1218 	edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1219 	edat += edat && test_kvm_facility(vcpu->kvm, 78);
1220 
1221 	/*
1222 	 * ASCE or EDAT could have changed since last icpt, or the gmap
1223 	 * we're holding has been unshadowed. If the gmap is still valid,
1224 	 * we can safely reuse it.
1225 	 */
1226 	if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat)) {
1227 		vcpu->kvm->stat.gmap_shadow_reuse++;
1228 		return 0;
1229 	}
1230 
1231 	/* release the old shadow - if any, and mark the prefix as unmapped */
1232 	release_gmap_shadow(vsie_page);
1233 	gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
1234 	if (IS_ERR(gmap))
1235 		return PTR_ERR(gmap);
1236 	vcpu->kvm->stat.gmap_shadow_create++;
1237 	WRITE_ONCE(vsie_page->gmap, gmap);
1238 	return 0;
1239 }
1240 
1241 /*
1242  * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
1243  */
1244 static void register_shadow_scb(struct kvm_vcpu *vcpu,
1245 				struct vsie_page *vsie_page)
1246 {
1247 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1248 
1249 	WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
1250 	/*
1251 	 * External calls have to lead to a kick of the vcpu and
1252 	 * therefore the vsie -> Simulate Wait state.
1253 	 */
1254 	kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
1255 	/*
1256 	 * We have to adjust the g3 epoch by the g2 epoch. The epoch will
1257 	 * automatically be adjusted on tod clock changes via kvm_sync_clock.
1258 	 */
1259 	preempt_disable();
1260 	scb_s->epoch += vcpu->kvm->arch.epoch;
1261 
1262 	if (scb_s->ecd & ECD_MEF) {
1263 		scb_s->epdx += vcpu->kvm->arch.epdx;
1264 		if (scb_s->epoch < vcpu->kvm->arch.epoch)
1265 			scb_s->epdx += 1;
1266 	}
1267 
1268 	preempt_enable();
1269 }
1270 
1271 /*
1272  * Unregister a shadow scb from a VCPU.
1273  */
1274 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
1275 {
1276 	kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
1277 	WRITE_ONCE(vcpu->arch.vsie_block, NULL);
1278 }
1279 
1280 /*
1281  * Run the vsie on a shadowed scb, managing the gmap shadow, handling
1282  * prefix pages and faults.
1283  *
1284  * Returns: - 0 if no errors occurred
1285  *          - > 0 if control has to be given to guest 2
1286  *          - -ENOMEM if out of memory
1287  */
1288 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1289 {
1290 	struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1291 	int rc = 0;
1292 
1293 	while (1) {
1294 		rc = acquire_gmap_shadow(vcpu, vsie_page);
1295 		if (!rc)
1296 			rc = map_prefix(vcpu, vsie_page);
1297 		if (!rc) {
1298 			gmap_enable(vsie_page->gmap);
1299 			update_intervention_requests(vsie_page);
1300 			rc = do_vsie_run(vcpu, vsie_page);
1301 			gmap_enable(vcpu->arch.gmap);
1302 		}
1303 		atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
1304 
1305 		if (rc == -EAGAIN)
1306 			rc = 0;
1307 
1308 		/*
1309 		 * Exit the loop if the guest needs to process the intercept
1310 		 */
1311 		if (rc || scb_s->icptcode)
1312 			break;
1313 
1314 		/*
1315 		 * Exit the loop if the host needs to process an intercept,
1316 		 * but rewind the PSW to re-enter SIE once that's completed
1317 		 * instead of passing a "no action" intercept to the guest.
1318 		 */
1319 		if (signal_pending(current) ||
1320 		    kvm_s390_vcpu_has_irq(vcpu, 0) ||
1321 		    kvm_s390_vcpu_sie_inhibited(vcpu)) {
1322 			kvm_s390_rewind_psw(vcpu, 4);
1323 			break;
1324 		}
1325 		cond_resched();
1326 	}
1327 
1328 	if (rc == -EFAULT) {
1329 		/*
1330 		 * Addressing exceptions are always presentes as intercepts.
1331 		 * As addressing exceptions are suppressing and our guest 3 PSW
1332 		 * points at the responsible instruction, we have to
1333 		 * forward the PSW and set the ilc. If we can't read guest 3
1334 		 * instruction, we can use an arbitrary ilc. Let's always use
1335 		 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
1336 		 * memory. (we could also fake the shadow so the hardware
1337 		 * handles it).
1338 		 */
1339 		scb_s->icptcode = ICPT_PROGI;
1340 		scb_s->iprcc = PGM_ADDRESSING;
1341 		scb_s->pgmilc = 4;
1342 		scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
1343 		rc = 1;
1344 	}
1345 	return rc;
1346 }
1347 
1348 /*
1349  * Get or create a vsie page for a scb address.
1350  *
1351  * Returns: - address of a vsie page (cached or new one)
1352  *          - NULL if the same scb address is already used by another VCPU
1353  *          - ERR_PTR(-ENOMEM) if out of memory
1354  */
1355 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
1356 {
1357 	struct vsie_page *vsie_page;
1358 	struct page *page;
1359 	int nr_vcpus;
1360 
1361 	rcu_read_lock();
1362 	page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
1363 	rcu_read_unlock();
1364 	if (page) {
1365 		if (page_ref_inc_return(page) == 2)
1366 			return page_to_virt(page);
1367 		page_ref_dec(page);
1368 	}
1369 
1370 	/*
1371 	 * We want at least #online_vcpus shadows, so every VCPU can execute
1372 	 * the VSIE in parallel.
1373 	 */
1374 	nr_vcpus = atomic_read(&kvm->online_vcpus);
1375 
1376 	mutex_lock(&kvm->arch.vsie.mutex);
1377 	if (kvm->arch.vsie.page_count < nr_vcpus) {
1378 		page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO | GFP_DMA);
1379 		if (!page) {
1380 			mutex_unlock(&kvm->arch.vsie.mutex);
1381 			return ERR_PTR(-ENOMEM);
1382 		}
1383 		page_ref_inc(page);
1384 		kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1385 		kvm->arch.vsie.page_count++;
1386 	} else {
1387 		/* reuse an existing entry that belongs to nobody */
1388 		while (true) {
1389 			page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1390 			if (page_ref_inc_return(page) == 2)
1391 				break;
1392 			page_ref_dec(page);
1393 			kvm->arch.vsie.next++;
1394 			kvm->arch.vsie.next %= nr_vcpus;
1395 		}
1396 		radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1397 	}
1398 	page->index = addr;
1399 	/* double use of the same address */
1400 	if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1401 		page_ref_dec(page);
1402 		mutex_unlock(&kvm->arch.vsie.mutex);
1403 		return NULL;
1404 	}
1405 	mutex_unlock(&kvm->arch.vsie.mutex);
1406 
1407 	vsie_page = page_to_virt(page);
1408 	memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1409 	release_gmap_shadow(vsie_page);
1410 	vsie_page->fault_addr = 0;
1411 	vsie_page->scb_s.ihcpu = 0xffffU;
1412 	return vsie_page;
1413 }
1414 
1415 /* put a vsie page acquired via get_vsie_page */
1416 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1417 {
1418 	struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1419 
1420 	page_ref_dec(page);
1421 }
1422 
1423 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1424 {
1425 	struct vsie_page *vsie_page;
1426 	unsigned long scb_addr;
1427 	int rc;
1428 
1429 	vcpu->stat.instruction_sie++;
1430 	if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1431 		return -EOPNOTSUPP;
1432 	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1433 		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1434 
1435 	BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1436 	scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1437 
1438 	/* 512 byte alignment */
1439 	if (unlikely(scb_addr & 0x1ffUL))
1440 		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1441 
1442 	if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0) ||
1443 	    kvm_s390_vcpu_sie_inhibited(vcpu)) {
1444 		kvm_s390_rewind_psw(vcpu, 4);
1445 		return 0;
1446 	}
1447 
1448 	vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1449 	if (IS_ERR(vsie_page))
1450 		return PTR_ERR(vsie_page);
1451 	else if (!vsie_page)
1452 		/* double use of sie control block - simply do nothing */
1453 		return 0;
1454 
1455 	rc = pin_scb(vcpu, vsie_page, scb_addr);
1456 	if (rc)
1457 		goto out_put;
1458 	rc = shadow_scb(vcpu, vsie_page);
1459 	if (rc)
1460 		goto out_unpin_scb;
1461 	rc = pin_blocks(vcpu, vsie_page);
1462 	if (rc)
1463 		goto out_unshadow;
1464 	register_shadow_scb(vcpu, vsie_page);
1465 	rc = vsie_run(vcpu, vsie_page);
1466 	unregister_shadow_scb(vcpu);
1467 	unpin_blocks(vcpu, vsie_page);
1468 out_unshadow:
1469 	unshadow_scb(vcpu, vsie_page);
1470 out_unpin_scb:
1471 	unpin_scb(vcpu, vsie_page, scb_addr);
1472 out_put:
1473 	put_vsie_page(vcpu->kvm, vsie_page);
1474 
1475 	return rc < 0 ? rc : 0;
1476 }
1477 
1478 /* Init the vsie data structures. To be called when a vm is initialized. */
1479 void kvm_s390_vsie_init(struct kvm *kvm)
1480 {
1481 	mutex_init(&kvm->arch.vsie.mutex);
1482 	INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL_ACCOUNT);
1483 }
1484 
1485 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1486 void kvm_s390_vsie_destroy(struct kvm *kvm)
1487 {
1488 	struct vsie_page *vsie_page;
1489 	struct page *page;
1490 	int i;
1491 
1492 	mutex_lock(&kvm->arch.vsie.mutex);
1493 	for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1494 		page = kvm->arch.vsie.pages[i];
1495 		kvm->arch.vsie.pages[i] = NULL;
1496 		vsie_page = page_to_virt(page);
1497 		release_gmap_shadow(vsie_page);
1498 		/* free the radix tree entry */
1499 		radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1500 		__free_page(page);
1501 	}
1502 	kvm->arch.vsie.page_count = 0;
1503 	mutex_unlock(&kvm->arch.vsie.mutex);
1504 }
1505 
1506 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1507 {
1508 	struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1509 
1510 	/*
1511 	 * Even if the VCPU lets go of the shadow sie block reference, it is
1512 	 * still valid in the cache. So we can safely kick it.
1513 	 */
1514 	if (scb) {
1515 		atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1516 		if (scb->prog0c & PROG_IN_SIE)
1517 			atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);
1518 	}
1519 }
1520