xref: /linux/arch/s390/kvm/gaccess.c (revision 0678df8271820bcf8fb4f877129f05d68a237de4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * guest access functions
4  *
5  * Copyright IBM Corp. 2014
6  *
7  */
8 
9 #include <linux/vmalloc.h>
10 #include <linux/mm_types.h>
11 #include <linux/err.h>
12 #include <linux/pgtable.h>
13 #include <linux/bitfield.h>
14 #include <asm/fault.h>
15 #include <asm/gmap.h>
16 #include "kvm-s390.h"
17 #include "gaccess.h"
18 #include <asm/switch_to.h>
19 
20 union asce {
21 	unsigned long val;
22 	struct {
23 		unsigned long origin : 52; /* Region- or Segment-Table Origin */
24 		unsigned long	 : 2;
25 		unsigned long g  : 1; /* Subspace Group Control */
26 		unsigned long p  : 1; /* Private Space Control */
27 		unsigned long s  : 1; /* Storage-Alteration-Event Control */
28 		unsigned long x  : 1; /* Space-Switch-Event Control */
29 		unsigned long r  : 1; /* Real-Space Control */
30 		unsigned long	 : 1;
31 		unsigned long dt : 2; /* Designation-Type Control */
32 		unsigned long tl : 2; /* Region- or Segment-Table Length */
33 	};
34 };
35 
36 enum {
37 	ASCE_TYPE_SEGMENT = 0,
38 	ASCE_TYPE_REGION3 = 1,
39 	ASCE_TYPE_REGION2 = 2,
40 	ASCE_TYPE_REGION1 = 3
41 };
42 
43 union region1_table_entry {
44 	unsigned long val;
45 	struct {
46 		unsigned long rto: 52;/* Region-Table Origin */
47 		unsigned long	 : 2;
48 		unsigned long p  : 1; /* DAT-Protection Bit */
49 		unsigned long	 : 1;
50 		unsigned long tf : 2; /* Region-Second-Table Offset */
51 		unsigned long i  : 1; /* Region-Invalid Bit */
52 		unsigned long	 : 1;
53 		unsigned long tt : 2; /* Table-Type Bits */
54 		unsigned long tl : 2; /* Region-Second-Table Length */
55 	};
56 };
57 
58 union region2_table_entry {
59 	unsigned long val;
60 	struct {
61 		unsigned long rto: 52;/* Region-Table Origin */
62 		unsigned long	 : 2;
63 		unsigned long p  : 1; /* DAT-Protection Bit */
64 		unsigned long	 : 1;
65 		unsigned long tf : 2; /* Region-Third-Table Offset */
66 		unsigned long i  : 1; /* Region-Invalid Bit */
67 		unsigned long	 : 1;
68 		unsigned long tt : 2; /* Table-Type Bits */
69 		unsigned long tl : 2; /* Region-Third-Table Length */
70 	};
71 };
72 
73 struct region3_table_entry_fc0 {
74 	unsigned long sto: 52;/* Segment-Table Origin */
75 	unsigned long	 : 1;
76 	unsigned long fc : 1; /* Format-Control */
77 	unsigned long p  : 1; /* DAT-Protection Bit */
78 	unsigned long	 : 1;
79 	unsigned long tf : 2; /* Segment-Table Offset */
80 	unsigned long i  : 1; /* Region-Invalid Bit */
81 	unsigned long cr : 1; /* Common-Region Bit */
82 	unsigned long tt : 2; /* Table-Type Bits */
83 	unsigned long tl : 2; /* Segment-Table Length */
84 };
85 
86 struct region3_table_entry_fc1 {
87 	unsigned long rfaa : 33; /* Region-Frame Absolute Address */
88 	unsigned long	 : 14;
89 	unsigned long av : 1; /* ACCF-Validity Control */
90 	unsigned long acc: 4; /* Access-Control Bits */
91 	unsigned long f  : 1; /* Fetch-Protection Bit */
92 	unsigned long fc : 1; /* Format-Control */
93 	unsigned long p  : 1; /* DAT-Protection Bit */
94 	unsigned long iep: 1; /* Instruction-Execution-Protection */
95 	unsigned long	 : 2;
96 	unsigned long i  : 1; /* Region-Invalid Bit */
97 	unsigned long cr : 1; /* Common-Region Bit */
98 	unsigned long tt : 2; /* Table-Type Bits */
99 	unsigned long	 : 2;
100 };
101 
102 union region3_table_entry {
103 	unsigned long val;
104 	struct region3_table_entry_fc0 fc0;
105 	struct region3_table_entry_fc1 fc1;
106 	struct {
107 		unsigned long	 : 53;
108 		unsigned long fc : 1; /* Format-Control */
109 		unsigned long	 : 4;
110 		unsigned long i  : 1; /* Region-Invalid Bit */
111 		unsigned long cr : 1; /* Common-Region Bit */
112 		unsigned long tt : 2; /* Table-Type Bits */
113 		unsigned long	 : 2;
114 	};
115 };
116 
117 struct segment_entry_fc0 {
118 	unsigned long pto: 53;/* Page-Table Origin */
119 	unsigned long fc : 1; /* Format-Control */
120 	unsigned long p  : 1; /* DAT-Protection Bit */
121 	unsigned long	 : 3;
122 	unsigned long i  : 1; /* Segment-Invalid Bit */
123 	unsigned long cs : 1; /* Common-Segment Bit */
124 	unsigned long tt : 2; /* Table-Type Bits */
125 	unsigned long	 : 2;
126 };
127 
128 struct segment_entry_fc1 {
129 	unsigned long sfaa : 44; /* Segment-Frame Absolute Address */
130 	unsigned long	 : 3;
131 	unsigned long av : 1; /* ACCF-Validity Control */
132 	unsigned long acc: 4; /* Access-Control Bits */
133 	unsigned long f  : 1; /* Fetch-Protection Bit */
134 	unsigned long fc : 1; /* Format-Control */
135 	unsigned long p  : 1; /* DAT-Protection Bit */
136 	unsigned long iep: 1; /* Instruction-Execution-Protection */
137 	unsigned long	 : 2;
138 	unsigned long i  : 1; /* Segment-Invalid Bit */
139 	unsigned long cs : 1; /* Common-Segment Bit */
140 	unsigned long tt : 2; /* Table-Type Bits */
141 	unsigned long	 : 2;
142 };
143 
144 union segment_table_entry {
145 	unsigned long val;
146 	struct segment_entry_fc0 fc0;
147 	struct segment_entry_fc1 fc1;
148 	struct {
149 		unsigned long	 : 53;
150 		unsigned long fc : 1; /* Format-Control */
151 		unsigned long	 : 4;
152 		unsigned long i  : 1; /* Segment-Invalid Bit */
153 		unsigned long cs : 1; /* Common-Segment Bit */
154 		unsigned long tt : 2; /* Table-Type Bits */
155 		unsigned long	 : 2;
156 	};
157 };
158 
159 enum {
160 	TABLE_TYPE_SEGMENT = 0,
161 	TABLE_TYPE_REGION3 = 1,
162 	TABLE_TYPE_REGION2 = 2,
163 	TABLE_TYPE_REGION1 = 3
164 };
165 
166 union page_table_entry {
167 	unsigned long val;
168 	struct {
169 		unsigned long pfra : 52; /* Page-Frame Real Address */
170 		unsigned long z  : 1; /* Zero Bit */
171 		unsigned long i  : 1; /* Page-Invalid Bit */
172 		unsigned long p  : 1; /* DAT-Protection Bit */
173 		unsigned long iep: 1; /* Instruction-Execution-Protection */
174 		unsigned long	 : 8;
175 	};
176 };
177 
178 /*
179  * vaddress union in order to easily decode a virtual address into its
180  * region first index, region second index etc. parts.
181  */
182 union vaddress {
183 	unsigned long addr;
184 	struct {
185 		unsigned long rfx : 11;
186 		unsigned long rsx : 11;
187 		unsigned long rtx : 11;
188 		unsigned long sx  : 11;
189 		unsigned long px  : 8;
190 		unsigned long bx  : 12;
191 	};
192 	struct {
193 		unsigned long rfx01 : 2;
194 		unsigned long	    : 9;
195 		unsigned long rsx01 : 2;
196 		unsigned long	    : 9;
197 		unsigned long rtx01 : 2;
198 		unsigned long	    : 9;
199 		unsigned long sx01  : 2;
200 		unsigned long	    : 29;
201 	};
202 };
203 
204 /*
205  * raddress union which will contain the result (real or absolute address)
206  * after a page table walk. The rfaa, sfaa and pfra members are used to
207  * simply assign them the value of a region, segment or page table entry.
208  */
209 union raddress {
210 	unsigned long addr;
211 	unsigned long rfaa : 33; /* Region-Frame Absolute Address */
212 	unsigned long sfaa : 44; /* Segment-Frame Absolute Address */
213 	unsigned long pfra : 52; /* Page-Frame Real Address */
214 };
215 
216 union alet {
217 	u32 val;
218 	struct {
219 		u32 reserved : 7;
220 		u32 p        : 1;
221 		u32 alesn    : 8;
222 		u32 alen     : 16;
223 	};
224 };
225 
226 union ald {
227 	u32 val;
228 	struct {
229 		u32     : 1;
230 		u32 alo : 24;
231 		u32 all : 7;
232 	};
233 };
234 
235 struct ale {
236 	unsigned long i      : 1; /* ALEN-Invalid Bit */
237 	unsigned long        : 5;
238 	unsigned long fo     : 1; /* Fetch-Only Bit */
239 	unsigned long p      : 1; /* Private Bit */
240 	unsigned long alesn  : 8; /* Access-List-Entry Sequence Number */
241 	unsigned long aleax  : 16; /* Access-List-Entry Authorization Index */
242 	unsigned long        : 32;
243 	unsigned long        : 1;
244 	unsigned long asteo  : 25; /* ASN-Second-Table-Entry Origin */
245 	unsigned long        : 6;
246 	unsigned long astesn : 32; /* ASTE Sequence Number */
247 };
248 
249 struct aste {
250 	unsigned long i      : 1; /* ASX-Invalid Bit */
251 	unsigned long ato    : 29; /* Authority-Table Origin */
252 	unsigned long        : 1;
253 	unsigned long b      : 1; /* Base-Space Bit */
254 	unsigned long ax     : 16; /* Authorization Index */
255 	unsigned long atl    : 12; /* Authority-Table Length */
256 	unsigned long        : 2;
257 	unsigned long ca     : 1; /* Controlled-ASN Bit */
258 	unsigned long ra     : 1; /* Reusable-ASN Bit */
259 	unsigned long asce   : 64; /* Address-Space-Control Element */
260 	unsigned long ald    : 32;
261 	unsigned long astesn : 32;
262 	/* .. more fields there */
263 };
264 
265 int ipte_lock_held(struct kvm *kvm)
266 {
267 	if (sclp.has_siif) {
268 		int rc;
269 
270 		read_lock(&kvm->arch.sca_lock);
271 		rc = kvm_s390_get_ipte_control(kvm)->kh != 0;
272 		read_unlock(&kvm->arch.sca_lock);
273 		return rc;
274 	}
275 	return kvm->arch.ipte_lock_count != 0;
276 }
277 
278 static void ipte_lock_simple(struct kvm *kvm)
279 {
280 	union ipte_control old, new, *ic;
281 
282 	mutex_lock(&kvm->arch.ipte_mutex);
283 	kvm->arch.ipte_lock_count++;
284 	if (kvm->arch.ipte_lock_count > 1)
285 		goto out;
286 retry:
287 	read_lock(&kvm->arch.sca_lock);
288 	ic = kvm_s390_get_ipte_control(kvm);
289 	do {
290 		old = READ_ONCE(*ic);
291 		if (old.k) {
292 			read_unlock(&kvm->arch.sca_lock);
293 			cond_resched();
294 			goto retry;
295 		}
296 		new = old;
297 		new.k = 1;
298 	} while (cmpxchg(&ic->val, old.val, new.val) != old.val);
299 	read_unlock(&kvm->arch.sca_lock);
300 out:
301 	mutex_unlock(&kvm->arch.ipte_mutex);
302 }
303 
304 static void ipte_unlock_simple(struct kvm *kvm)
305 {
306 	union ipte_control old, new, *ic;
307 
308 	mutex_lock(&kvm->arch.ipte_mutex);
309 	kvm->arch.ipte_lock_count--;
310 	if (kvm->arch.ipte_lock_count)
311 		goto out;
312 	read_lock(&kvm->arch.sca_lock);
313 	ic = kvm_s390_get_ipte_control(kvm);
314 	do {
315 		old = READ_ONCE(*ic);
316 		new = old;
317 		new.k = 0;
318 	} while (cmpxchg(&ic->val, old.val, new.val) != old.val);
319 	read_unlock(&kvm->arch.sca_lock);
320 	wake_up(&kvm->arch.ipte_wq);
321 out:
322 	mutex_unlock(&kvm->arch.ipte_mutex);
323 }
324 
325 static void ipte_lock_siif(struct kvm *kvm)
326 {
327 	union ipte_control old, new, *ic;
328 
329 retry:
330 	read_lock(&kvm->arch.sca_lock);
331 	ic = kvm_s390_get_ipte_control(kvm);
332 	do {
333 		old = READ_ONCE(*ic);
334 		if (old.kg) {
335 			read_unlock(&kvm->arch.sca_lock);
336 			cond_resched();
337 			goto retry;
338 		}
339 		new = old;
340 		new.k = 1;
341 		new.kh++;
342 	} while (cmpxchg(&ic->val, old.val, new.val) != old.val);
343 	read_unlock(&kvm->arch.sca_lock);
344 }
345 
346 static void ipte_unlock_siif(struct kvm *kvm)
347 {
348 	union ipte_control old, new, *ic;
349 
350 	read_lock(&kvm->arch.sca_lock);
351 	ic = kvm_s390_get_ipte_control(kvm);
352 	do {
353 		old = READ_ONCE(*ic);
354 		new = old;
355 		new.kh--;
356 		if (!new.kh)
357 			new.k = 0;
358 	} while (cmpxchg(&ic->val, old.val, new.val) != old.val);
359 	read_unlock(&kvm->arch.sca_lock);
360 	if (!new.kh)
361 		wake_up(&kvm->arch.ipte_wq);
362 }
363 
364 void ipte_lock(struct kvm *kvm)
365 {
366 	if (sclp.has_siif)
367 		ipte_lock_siif(kvm);
368 	else
369 		ipte_lock_simple(kvm);
370 }
371 
372 void ipte_unlock(struct kvm *kvm)
373 {
374 	if (sclp.has_siif)
375 		ipte_unlock_siif(kvm);
376 	else
377 		ipte_unlock_simple(kvm);
378 }
379 
380 static int ar_translation(struct kvm_vcpu *vcpu, union asce *asce, u8 ar,
381 			  enum gacc_mode mode)
382 {
383 	union alet alet;
384 	struct ale ale;
385 	struct aste aste;
386 	unsigned long ald_addr, authority_table_addr;
387 	union ald ald;
388 	int eax, rc;
389 	u8 authority_table;
390 
391 	if (ar >= NUM_ACRS)
392 		return -EINVAL;
393 
394 	save_access_regs(vcpu->run->s.regs.acrs);
395 	alet.val = vcpu->run->s.regs.acrs[ar];
396 
397 	if (ar == 0 || alet.val == 0) {
398 		asce->val = vcpu->arch.sie_block->gcr[1];
399 		return 0;
400 	} else if (alet.val == 1) {
401 		asce->val = vcpu->arch.sie_block->gcr[7];
402 		return 0;
403 	}
404 
405 	if (alet.reserved)
406 		return PGM_ALET_SPECIFICATION;
407 
408 	if (alet.p)
409 		ald_addr = vcpu->arch.sie_block->gcr[5];
410 	else
411 		ald_addr = vcpu->arch.sie_block->gcr[2];
412 	ald_addr &= 0x7fffffc0;
413 
414 	rc = read_guest_real(vcpu, ald_addr + 16, &ald.val, sizeof(union ald));
415 	if (rc)
416 		return rc;
417 
418 	if (alet.alen / 8 > ald.all)
419 		return PGM_ALEN_TRANSLATION;
420 
421 	if (0x7fffffff - ald.alo * 128 < alet.alen * 16)
422 		return PGM_ADDRESSING;
423 
424 	rc = read_guest_real(vcpu, ald.alo * 128 + alet.alen * 16, &ale,
425 			     sizeof(struct ale));
426 	if (rc)
427 		return rc;
428 
429 	if (ale.i == 1)
430 		return PGM_ALEN_TRANSLATION;
431 	if (ale.alesn != alet.alesn)
432 		return PGM_ALE_SEQUENCE;
433 
434 	rc = read_guest_real(vcpu, ale.asteo * 64, &aste, sizeof(struct aste));
435 	if (rc)
436 		return rc;
437 
438 	if (aste.i)
439 		return PGM_ASTE_VALIDITY;
440 	if (aste.astesn != ale.astesn)
441 		return PGM_ASTE_SEQUENCE;
442 
443 	if (ale.p == 1) {
444 		eax = (vcpu->arch.sie_block->gcr[8] >> 16) & 0xffff;
445 		if (ale.aleax != eax) {
446 			if (eax / 16 > aste.atl)
447 				return PGM_EXTENDED_AUTHORITY;
448 
449 			authority_table_addr = aste.ato * 4 + eax / 4;
450 
451 			rc = read_guest_real(vcpu, authority_table_addr,
452 					     &authority_table,
453 					     sizeof(u8));
454 			if (rc)
455 				return rc;
456 
457 			if ((authority_table & (0x40 >> ((eax & 3) * 2))) == 0)
458 				return PGM_EXTENDED_AUTHORITY;
459 		}
460 	}
461 
462 	if (ale.fo == 1 && mode == GACC_STORE)
463 		return PGM_PROTECTION;
464 
465 	asce->val = aste.asce;
466 	return 0;
467 }
468 
469 enum prot_type {
470 	PROT_TYPE_LA   = 0,
471 	PROT_TYPE_KEYC = 1,
472 	PROT_TYPE_ALC  = 2,
473 	PROT_TYPE_DAT  = 3,
474 	PROT_TYPE_IEP  = 4,
475 	/* Dummy value for passing an initialized value when code != PGM_PROTECTION */
476 	PROT_NONE,
477 };
478 
479 static int trans_exc_ending(struct kvm_vcpu *vcpu, int code, unsigned long gva, u8 ar,
480 			    enum gacc_mode mode, enum prot_type prot, bool terminate)
481 {
482 	struct kvm_s390_pgm_info *pgm = &vcpu->arch.pgm;
483 	union teid *teid;
484 
485 	memset(pgm, 0, sizeof(*pgm));
486 	pgm->code = code;
487 	teid = (union teid *)&pgm->trans_exc_code;
488 
489 	switch (code) {
490 	case PGM_PROTECTION:
491 		switch (prot) {
492 		case PROT_NONE:
493 			/* We should never get here, acts like termination */
494 			WARN_ON_ONCE(1);
495 			break;
496 		case PROT_TYPE_IEP:
497 			teid->b61 = 1;
498 			fallthrough;
499 		case PROT_TYPE_LA:
500 			teid->b56 = 1;
501 			break;
502 		case PROT_TYPE_KEYC:
503 			teid->b60 = 1;
504 			break;
505 		case PROT_TYPE_ALC:
506 			teid->b60 = 1;
507 			fallthrough;
508 		case PROT_TYPE_DAT:
509 			teid->b61 = 1;
510 			break;
511 		}
512 		if (terminate) {
513 			teid->b56 = 0;
514 			teid->b60 = 0;
515 			teid->b61 = 0;
516 		}
517 		fallthrough;
518 	case PGM_ASCE_TYPE:
519 	case PGM_PAGE_TRANSLATION:
520 	case PGM_REGION_FIRST_TRANS:
521 	case PGM_REGION_SECOND_TRANS:
522 	case PGM_REGION_THIRD_TRANS:
523 	case PGM_SEGMENT_TRANSLATION:
524 		/*
525 		 * op_access_id only applies to MOVE_PAGE -> set bit 61
526 		 * exc_access_id has to be set to 0 for some instructions. Both
527 		 * cases have to be handled by the caller.
528 		 */
529 		teid->addr = gva >> PAGE_SHIFT;
530 		teid->fsi = mode == GACC_STORE ? TEID_FSI_STORE : TEID_FSI_FETCH;
531 		teid->as = psw_bits(vcpu->arch.sie_block->gpsw).as;
532 		fallthrough;
533 	case PGM_ALEN_TRANSLATION:
534 	case PGM_ALE_SEQUENCE:
535 	case PGM_ASTE_VALIDITY:
536 	case PGM_ASTE_SEQUENCE:
537 	case PGM_EXTENDED_AUTHORITY:
538 		/*
539 		 * We can always store exc_access_id, as it is
540 		 * undefined for non-ar cases. It is undefined for
541 		 * most DAT protection exceptions.
542 		 */
543 		pgm->exc_access_id = ar;
544 		break;
545 	}
546 	return code;
547 }
548 
549 static int trans_exc(struct kvm_vcpu *vcpu, int code, unsigned long gva, u8 ar,
550 		     enum gacc_mode mode, enum prot_type prot)
551 {
552 	return trans_exc_ending(vcpu, code, gva, ar, mode, prot, false);
553 }
554 
555 static int get_vcpu_asce(struct kvm_vcpu *vcpu, union asce *asce,
556 			 unsigned long ga, u8 ar, enum gacc_mode mode)
557 {
558 	int rc;
559 	struct psw_bits psw = psw_bits(vcpu->arch.sie_block->gpsw);
560 
561 	if (!psw.dat) {
562 		asce->val = 0;
563 		asce->r = 1;
564 		return 0;
565 	}
566 
567 	if ((mode == GACC_IFETCH) && (psw.as != PSW_BITS_AS_HOME))
568 		psw.as = PSW_BITS_AS_PRIMARY;
569 
570 	switch (psw.as) {
571 	case PSW_BITS_AS_PRIMARY:
572 		asce->val = vcpu->arch.sie_block->gcr[1];
573 		return 0;
574 	case PSW_BITS_AS_SECONDARY:
575 		asce->val = vcpu->arch.sie_block->gcr[7];
576 		return 0;
577 	case PSW_BITS_AS_HOME:
578 		asce->val = vcpu->arch.sie_block->gcr[13];
579 		return 0;
580 	case PSW_BITS_AS_ACCREG:
581 		rc = ar_translation(vcpu, asce, ar, mode);
582 		if (rc > 0)
583 			return trans_exc(vcpu, rc, ga, ar, mode, PROT_TYPE_ALC);
584 		return rc;
585 	}
586 	return 0;
587 }
588 
589 static int deref_table(struct kvm *kvm, unsigned long gpa, unsigned long *val)
590 {
591 	return kvm_read_guest(kvm, gpa, val, sizeof(*val));
592 }
593 
594 /**
595  * guest_translate - translate a guest virtual into a guest absolute address
596  * @vcpu: virtual cpu
597  * @gva: guest virtual address
598  * @gpa: points to where guest physical (absolute) address should be stored
599  * @asce: effective asce
600  * @mode: indicates the access mode to be used
601  * @prot: returns the type for protection exceptions
602  *
603  * Translate a guest virtual address into a guest absolute address by means
604  * of dynamic address translation as specified by the architecture.
605  * If the resulting absolute address is not available in the configuration
606  * an addressing exception is indicated and @gpa will not be changed.
607  *
608  * Returns: - zero on success; @gpa contains the resulting absolute address
609  *	    - a negative value if guest access failed due to e.g. broken
610  *	      guest mapping
611  *	    - a positive value if an access exception happened. In this case
612  *	      the returned value is the program interruption code as defined
613  *	      by the architecture
614  */
615 static unsigned long guest_translate(struct kvm_vcpu *vcpu, unsigned long gva,
616 				     unsigned long *gpa, const union asce asce,
617 				     enum gacc_mode mode, enum prot_type *prot)
618 {
619 	union vaddress vaddr = {.addr = gva};
620 	union raddress raddr = {.addr = gva};
621 	union page_table_entry pte;
622 	int dat_protection = 0;
623 	int iep_protection = 0;
624 	union ctlreg0 ctlreg0;
625 	unsigned long ptr;
626 	int edat1, edat2, iep;
627 
628 	ctlreg0.val = vcpu->arch.sie_block->gcr[0];
629 	edat1 = ctlreg0.edat && test_kvm_facility(vcpu->kvm, 8);
630 	edat2 = edat1 && test_kvm_facility(vcpu->kvm, 78);
631 	iep = ctlreg0.iep && test_kvm_facility(vcpu->kvm, 130);
632 	if (asce.r)
633 		goto real_address;
634 	ptr = asce.origin * PAGE_SIZE;
635 	switch (asce.dt) {
636 	case ASCE_TYPE_REGION1:
637 		if (vaddr.rfx01 > asce.tl)
638 			return PGM_REGION_FIRST_TRANS;
639 		ptr += vaddr.rfx * 8;
640 		break;
641 	case ASCE_TYPE_REGION2:
642 		if (vaddr.rfx)
643 			return PGM_ASCE_TYPE;
644 		if (vaddr.rsx01 > asce.tl)
645 			return PGM_REGION_SECOND_TRANS;
646 		ptr += vaddr.rsx * 8;
647 		break;
648 	case ASCE_TYPE_REGION3:
649 		if (vaddr.rfx || vaddr.rsx)
650 			return PGM_ASCE_TYPE;
651 		if (vaddr.rtx01 > asce.tl)
652 			return PGM_REGION_THIRD_TRANS;
653 		ptr += vaddr.rtx * 8;
654 		break;
655 	case ASCE_TYPE_SEGMENT:
656 		if (vaddr.rfx || vaddr.rsx || vaddr.rtx)
657 			return PGM_ASCE_TYPE;
658 		if (vaddr.sx01 > asce.tl)
659 			return PGM_SEGMENT_TRANSLATION;
660 		ptr += vaddr.sx * 8;
661 		break;
662 	}
663 	switch (asce.dt) {
664 	case ASCE_TYPE_REGION1:	{
665 		union region1_table_entry rfte;
666 
667 		if (kvm_is_error_gpa(vcpu->kvm, ptr))
668 			return PGM_ADDRESSING;
669 		if (deref_table(vcpu->kvm, ptr, &rfte.val))
670 			return -EFAULT;
671 		if (rfte.i)
672 			return PGM_REGION_FIRST_TRANS;
673 		if (rfte.tt != TABLE_TYPE_REGION1)
674 			return PGM_TRANSLATION_SPEC;
675 		if (vaddr.rsx01 < rfte.tf || vaddr.rsx01 > rfte.tl)
676 			return PGM_REGION_SECOND_TRANS;
677 		if (edat1)
678 			dat_protection |= rfte.p;
679 		ptr = rfte.rto * PAGE_SIZE + vaddr.rsx * 8;
680 	}
681 		fallthrough;
682 	case ASCE_TYPE_REGION2: {
683 		union region2_table_entry rste;
684 
685 		if (kvm_is_error_gpa(vcpu->kvm, ptr))
686 			return PGM_ADDRESSING;
687 		if (deref_table(vcpu->kvm, ptr, &rste.val))
688 			return -EFAULT;
689 		if (rste.i)
690 			return PGM_REGION_SECOND_TRANS;
691 		if (rste.tt != TABLE_TYPE_REGION2)
692 			return PGM_TRANSLATION_SPEC;
693 		if (vaddr.rtx01 < rste.tf || vaddr.rtx01 > rste.tl)
694 			return PGM_REGION_THIRD_TRANS;
695 		if (edat1)
696 			dat_protection |= rste.p;
697 		ptr = rste.rto * PAGE_SIZE + vaddr.rtx * 8;
698 	}
699 		fallthrough;
700 	case ASCE_TYPE_REGION3: {
701 		union region3_table_entry rtte;
702 
703 		if (kvm_is_error_gpa(vcpu->kvm, ptr))
704 			return PGM_ADDRESSING;
705 		if (deref_table(vcpu->kvm, ptr, &rtte.val))
706 			return -EFAULT;
707 		if (rtte.i)
708 			return PGM_REGION_THIRD_TRANS;
709 		if (rtte.tt != TABLE_TYPE_REGION3)
710 			return PGM_TRANSLATION_SPEC;
711 		if (rtte.cr && asce.p && edat2)
712 			return PGM_TRANSLATION_SPEC;
713 		if (rtte.fc && edat2) {
714 			dat_protection |= rtte.fc1.p;
715 			iep_protection = rtte.fc1.iep;
716 			raddr.rfaa = rtte.fc1.rfaa;
717 			goto absolute_address;
718 		}
719 		if (vaddr.sx01 < rtte.fc0.tf)
720 			return PGM_SEGMENT_TRANSLATION;
721 		if (vaddr.sx01 > rtte.fc0.tl)
722 			return PGM_SEGMENT_TRANSLATION;
723 		if (edat1)
724 			dat_protection |= rtte.fc0.p;
725 		ptr = rtte.fc0.sto * PAGE_SIZE + vaddr.sx * 8;
726 	}
727 		fallthrough;
728 	case ASCE_TYPE_SEGMENT: {
729 		union segment_table_entry ste;
730 
731 		if (kvm_is_error_gpa(vcpu->kvm, ptr))
732 			return PGM_ADDRESSING;
733 		if (deref_table(vcpu->kvm, ptr, &ste.val))
734 			return -EFAULT;
735 		if (ste.i)
736 			return PGM_SEGMENT_TRANSLATION;
737 		if (ste.tt != TABLE_TYPE_SEGMENT)
738 			return PGM_TRANSLATION_SPEC;
739 		if (ste.cs && asce.p)
740 			return PGM_TRANSLATION_SPEC;
741 		if (ste.fc && edat1) {
742 			dat_protection |= ste.fc1.p;
743 			iep_protection = ste.fc1.iep;
744 			raddr.sfaa = ste.fc1.sfaa;
745 			goto absolute_address;
746 		}
747 		dat_protection |= ste.fc0.p;
748 		ptr = ste.fc0.pto * (PAGE_SIZE / 2) + vaddr.px * 8;
749 	}
750 	}
751 	if (kvm_is_error_gpa(vcpu->kvm, ptr))
752 		return PGM_ADDRESSING;
753 	if (deref_table(vcpu->kvm, ptr, &pte.val))
754 		return -EFAULT;
755 	if (pte.i)
756 		return PGM_PAGE_TRANSLATION;
757 	if (pte.z)
758 		return PGM_TRANSLATION_SPEC;
759 	dat_protection |= pte.p;
760 	iep_protection = pte.iep;
761 	raddr.pfra = pte.pfra;
762 real_address:
763 	raddr.addr = kvm_s390_real_to_abs(vcpu, raddr.addr);
764 absolute_address:
765 	if (mode == GACC_STORE && dat_protection) {
766 		*prot = PROT_TYPE_DAT;
767 		return PGM_PROTECTION;
768 	}
769 	if (mode == GACC_IFETCH && iep_protection && iep) {
770 		*prot = PROT_TYPE_IEP;
771 		return PGM_PROTECTION;
772 	}
773 	if (kvm_is_error_gpa(vcpu->kvm, raddr.addr))
774 		return PGM_ADDRESSING;
775 	*gpa = raddr.addr;
776 	return 0;
777 }
778 
779 static inline int is_low_address(unsigned long ga)
780 {
781 	/* Check for address ranges 0..511 and 4096..4607 */
782 	return (ga & ~0x11fful) == 0;
783 }
784 
785 static int low_address_protection_enabled(struct kvm_vcpu *vcpu,
786 					  const union asce asce)
787 {
788 	union ctlreg0 ctlreg0 = {.val = vcpu->arch.sie_block->gcr[0]};
789 	psw_t *psw = &vcpu->arch.sie_block->gpsw;
790 
791 	if (!ctlreg0.lap)
792 		return 0;
793 	if (psw_bits(*psw).dat && asce.p)
794 		return 0;
795 	return 1;
796 }
797 
798 static int vm_check_access_key(struct kvm *kvm, u8 access_key,
799 			       enum gacc_mode mode, gpa_t gpa)
800 {
801 	u8 storage_key, access_control;
802 	bool fetch_protected;
803 	unsigned long hva;
804 	int r;
805 
806 	if (access_key == 0)
807 		return 0;
808 
809 	hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
810 	if (kvm_is_error_hva(hva))
811 		return PGM_ADDRESSING;
812 
813 	mmap_read_lock(current->mm);
814 	r = get_guest_storage_key(current->mm, hva, &storage_key);
815 	mmap_read_unlock(current->mm);
816 	if (r)
817 		return r;
818 	access_control = FIELD_GET(_PAGE_ACC_BITS, storage_key);
819 	if (access_control == access_key)
820 		return 0;
821 	fetch_protected = storage_key & _PAGE_FP_BIT;
822 	if ((mode == GACC_FETCH || mode == GACC_IFETCH) && !fetch_protected)
823 		return 0;
824 	return PGM_PROTECTION;
825 }
826 
827 static bool fetch_prot_override_applicable(struct kvm_vcpu *vcpu, enum gacc_mode mode,
828 					   union asce asce)
829 {
830 	psw_t *psw = &vcpu->arch.sie_block->gpsw;
831 	unsigned long override;
832 
833 	if (mode == GACC_FETCH || mode == GACC_IFETCH) {
834 		/* check if fetch protection override enabled */
835 		override = vcpu->arch.sie_block->gcr[0];
836 		override &= CR0_FETCH_PROTECTION_OVERRIDE;
837 		/* not applicable if subject to DAT && private space */
838 		override = override && !(psw_bits(*psw).dat && asce.p);
839 		return override;
840 	}
841 	return false;
842 }
843 
844 static bool fetch_prot_override_applies(unsigned long ga, unsigned int len)
845 {
846 	return ga < 2048 && ga + len <= 2048;
847 }
848 
849 static bool storage_prot_override_applicable(struct kvm_vcpu *vcpu)
850 {
851 	/* check if storage protection override enabled */
852 	return vcpu->arch.sie_block->gcr[0] & CR0_STORAGE_PROTECTION_OVERRIDE;
853 }
854 
855 static bool storage_prot_override_applies(u8 access_control)
856 {
857 	/* matches special storage protection override key (9) -> allow */
858 	return access_control == PAGE_SPO_ACC;
859 }
860 
861 static int vcpu_check_access_key(struct kvm_vcpu *vcpu, u8 access_key,
862 				 enum gacc_mode mode, union asce asce, gpa_t gpa,
863 				 unsigned long ga, unsigned int len)
864 {
865 	u8 storage_key, access_control;
866 	unsigned long hva;
867 	int r;
868 
869 	/* access key 0 matches any storage key -> allow */
870 	if (access_key == 0)
871 		return 0;
872 	/*
873 	 * caller needs to ensure that gfn is accessible, so we can
874 	 * assume that this cannot fail
875 	 */
876 	hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gpa));
877 	mmap_read_lock(current->mm);
878 	r = get_guest_storage_key(current->mm, hva, &storage_key);
879 	mmap_read_unlock(current->mm);
880 	if (r)
881 		return r;
882 	access_control = FIELD_GET(_PAGE_ACC_BITS, storage_key);
883 	/* access key matches storage key -> allow */
884 	if (access_control == access_key)
885 		return 0;
886 	if (mode == GACC_FETCH || mode == GACC_IFETCH) {
887 		/* it is a fetch and fetch protection is off -> allow */
888 		if (!(storage_key & _PAGE_FP_BIT))
889 			return 0;
890 		if (fetch_prot_override_applicable(vcpu, mode, asce) &&
891 		    fetch_prot_override_applies(ga, len))
892 			return 0;
893 	}
894 	if (storage_prot_override_applicable(vcpu) &&
895 	    storage_prot_override_applies(access_control))
896 		return 0;
897 	return PGM_PROTECTION;
898 }
899 
900 /**
901  * guest_range_to_gpas() - Calculate guest physical addresses of page fragments
902  * covering a logical range
903  * @vcpu: virtual cpu
904  * @ga: guest address, start of range
905  * @ar: access register
906  * @gpas: output argument, may be NULL
907  * @len: length of range in bytes
908  * @asce: address-space-control element to use for translation
909  * @mode: access mode
910  * @access_key: access key to mach the range's storage keys against
911  *
912  * Translate a logical range to a series of guest absolute addresses,
913  * such that the concatenation of page fragments starting at each gpa make up
914  * the whole range.
915  * The translation is performed as if done by the cpu for the given @asce, @ar,
916  * @mode and state of the @vcpu.
917  * If the translation causes an exception, its program interruption code is
918  * returned and the &struct kvm_s390_pgm_info pgm member of @vcpu is modified
919  * such that a subsequent call to kvm_s390_inject_prog_vcpu() will inject
920  * a correct exception into the guest.
921  * The resulting gpas are stored into @gpas, unless it is NULL.
922  *
923  * Note: All fragments except the first one start at the beginning of a page.
924  *	 When deriving the boundaries of a fragment from a gpa, all but the last
925  *	 fragment end at the end of the page.
926  *
927  * Return:
928  * * 0		- success
929  * * <0		- translation could not be performed, for example if  guest
930  *		  memory could not be accessed
931  * * >0		- an access exception occurred. In this case the returned value
932  *		  is the program interruption code and the contents of pgm may
933  *		  be used to inject an exception into the guest.
934  */
935 static int guest_range_to_gpas(struct kvm_vcpu *vcpu, unsigned long ga, u8 ar,
936 			       unsigned long *gpas, unsigned long len,
937 			       const union asce asce, enum gacc_mode mode,
938 			       u8 access_key)
939 {
940 	psw_t *psw = &vcpu->arch.sie_block->gpsw;
941 	unsigned int offset = offset_in_page(ga);
942 	unsigned int fragment_len;
943 	int lap_enabled, rc = 0;
944 	enum prot_type prot;
945 	unsigned long gpa;
946 
947 	lap_enabled = low_address_protection_enabled(vcpu, asce);
948 	while (min(PAGE_SIZE - offset, len) > 0) {
949 		fragment_len = min(PAGE_SIZE - offset, len);
950 		ga = kvm_s390_logical_to_effective(vcpu, ga);
951 		if (mode == GACC_STORE && lap_enabled && is_low_address(ga))
952 			return trans_exc(vcpu, PGM_PROTECTION, ga, ar, mode,
953 					 PROT_TYPE_LA);
954 		if (psw_bits(*psw).dat) {
955 			rc = guest_translate(vcpu, ga, &gpa, asce, mode, &prot);
956 			if (rc < 0)
957 				return rc;
958 		} else {
959 			gpa = kvm_s390_real_to_abs(vcpu, ga);
960 			if (kvm_is_error_gpa(vcpu->kvm, gpa)) {
961 				rc = PGM_ADDRESSING;
962 				prot = PROT_NONE;
963 			}
964 		}
965 		if (rc)
966 			return trans_exc(vcpu, rc, ga, ar, mode, prot);
967 		rc = vcpu_check_access_key(vcpu, access_key, mode, asce, gpa, ga,
968 					   fragment_len);
969 		if (rc)
970 			return trans_exc(vcpu, rc, ga, ar, mode, PROT_TYPE_KEYC);
971 		if (gpas)
972 			*gpas++ = gpa;
973 		offset = 0;
974 		ga += fragment_len;
975 		len -= fragment_len;
976 	}
977 	return 0;
978 }
979 
980 static int access_guest_page(struct kvm *kvm, enum gacc_mode mode, gpa_t gpa,
981 			     void *data, unsigned int len)
982 {
983 	const unsigned int offset = offset_in_page(gpa);
984 	const gfn_t gfn = gpa_to_gfn(gpa);
985 	int rc;
986 
987 	if (mode == GACC_STORE)
988 		rc = kvm_write_guest_page(kvm, gfn, data, offset, len);
989 	else
990 		rc = kvm_read_guest_page(kvm, gfn, data, offset, len);
991 	return rc;
992 }
993 
994 static int
995 access_guest_page_with_key(struct kvm *kvm, enum gacc_mode mode, gpa_t gpa,
996 			   void *data, unsigned int len, u8 access_key)
997 {
998 	struct kvm_memory_slot *slot;
999 	bool writable;
1000 	gfn_t gfn;
1001 	hva_t hva;
1002 	int rc;
1003 
1004 	gfn = gpa >> PAGE_SHIFT;
1005 	slot = gfn_to_memslot(kvm, gfn);
1006 	hva = gfn_to_hva_memslot_prot(slot, gfn, &writable);
1007 
1008 	if (kvm_is_error_hva(hva))
1009 		return PGM_ADDRESSING;
1010 	/*
1011 	 * Check if it's a ro memslot, even tho that can't occur (they're unsupported).
1012 	 * Don't try to actually handle that case.
1013 	 */
1014 	if (!writable && mode == GACC_STORE)
1015 		return -EOPNOTSUPP;
1016 	hva += offset_in_page(gpa);
1017 	if (mode == GACC_STORE)
1018 		rc = copy_to_user_key((void __user *)hva, data, len, access_key);
1019 	else
1020 		rc = copy_from_user_key(data, (void __user *)hva, len, access_key);
1021 	if (rc)
1022 		return PGM_PROTECTION;
1023 	if (mode == GACC_STORE)
1024 		mark_page_dirty_in_slot(kvm, slot, gfn);
1025 	return 0;
1026 }
1027 
1028 int access_guest_abs_with_key(struct kvm *kvm, gpa_t gpa, void *data,
1029 			      unsigned long len, enum gacc_mode mode, u8 access_key)
1030 {
1031 	int offset = offset_in_page(gpa);
1032 	int fragment_len;
1033 	int rc;
1034 
1035 	while (min(PAGE_SIZE - offset, len) > 0) {
1036 		fragment_len = min(PAGE_SIZE - offset, len);
1037 		rc = access_guest_page_with_key(kvm, mode, gpa, data, fragment_len, access_key);
1038 		if (rc)
1039 			return rc;
1040 		offset = 0;
1041 		len -= fragment_len;
1042 		data += fragment_len;
1043 		gpa += fragment_len;
1044 	}
1045 	return 0;
1046 }
1047 
1048 int access_guest_with_key(struct kvm_vcpu *vcpu, unsigned long ga, u8 ar,
1049 			  void *data, unsigned long len, enum gacc_mode mode,
1050 			  u8 access_key)
1051 {
1052 	psw_t *psw = &vcpu->arch.sie_block->gpsw;
1053 	unsigned long nr_pages, idx;
1054 	unsigned long gpa_array[2];
1055 	unsigned int fragment_len;
1056 	unsigned long *gpas;
1057 	enum prot_type prot;
1058 	int need_ipte_lock;
1059 	union asce asce;
1060 	bool try_storage_prot_override;
1061 	bool try_fetch_prot_override;
1062 	int rc;
1063 
1064 	if (!len)
1065 		return 0;
1066 	ga = kvm_s390_logical_to_effective(vcpu, ga);
1067 	rc = get_vcpu_asce(vcpu, &asce, ga, ar, mode);
1068 	if (rc)
1069 		return rc;
1070 	nr_pages = (((ga & ~PAGE_MASK) + len - 1) >> PAGE_SHIFT) + 1;
1071 	gpas = gpa_array;
1072 	if (nr_pages > ARRAY_SIZE(gpa_array))
1073 		gpas = vmalloc(array_size(nr_pages, sizeof(unsigned long)));
1074 	if (!gpas)
1075 		return -ENOMEM;
1076 	try_fetch_prot_override = fetch_prot_override_applicable(vcpu, mode, asce);
1077 	try_storage_prot_override = storage_prot_override_applicable(vcpu);
1078 	need_ipte_lock = psw_bits(*psw).dat && !asce.r;
1079 	if (need_ipte_lock)
1080 		ipte_lock(vcpu->kvm);
1081 	/*
1082 	 * Since we do the access further down ultimately via a move instruction
1083 	 * that does key checking and returns an error in case of a protection
1084 	 * violation, we don't need to do the check during address translation.
1085 	 * Skip it by passing access key 0, which matches any storage key,
1086 	 * obviating the need for any further checks. As a result the check is
1087 	 * handled entirely in hardware on access, we only need to take care to
1088 	 * forego key protection checking if fetch protection override applies or
1089 	 * retry with the special key 9 in case of storage protection override.
1090 	 */
1091 	rc = guest_range_to_gpas(vcpu, ga, ar, gpas, len, asce, mode, 0);
1092 	if (rc)
1093 		goto out_unlock;
1094 	for (idx = 0; idx < nr_pages; idx++) {
1095 		fragment_len = min(PAGE_SIZE - offset_in_page(gpas[idx]), len);
1096 		if (try_fetch_prot_override && fetch_prot_override_applies(ga, fragment_len)) {
1097 			rc = access_guest_page(vcpu->kvm, mode, gpas[idx],
1098 					       data, fragment_len);
1099 		} else {
1100 			rc = access_guest_page_with_key(vcpu->kvm, mode, gpas[idx],
1101 							data, fragment_len, access_key);
1102 		}
1103 		if (rc == PGM_PROTECTION && try_storage_prot_override)
1104 			rc = access_guest_page_with_key(vcpu->kvm, mode, gpas[idx],
1105 							data, fragment_len, PAGE_SPO_ACC);
1106 		if (rc)
1107 			break;
1108 		len -= fragment_len;
1109 		data += fragment_len;
1110 		ga = kvm_s390_logical_to_effective(vcpu, ga + fragment_len);
1111 	}
1112 	if (rc > 0) {
1113 		bool terminate = (mode == GACC_STORE) && (idx > 0);
1114 
1115 		if (rc == PGM_PROTECTION)
1116 			prot = PROT_TYPE_KEYC;
1117 		else
1118 			prot = PROT_NONE;
1119 		rc = trans_exc_ending(vcpu, rc, ga, ar, mode, prot, terminate);
1120 	}
1121 out_unlock:
1122 	if (need_ipte_lock)
1123 		ipte_unlock(vcpu->kvm);
1124 	if (nr_pages > ARRAY_SIZE(gpa_array))
1125 		vfree(gpas);
1126 	return rc;
1127 }
1128 
1129 int access_guest_real(struct kvm_vcpu *vcpu, unsigned long gra,
1130 		      void *data, unsigned long len, enum gacc_mode mode)
1131 {
1132 	unsigned int fragment_len;
1133 	unsigned long gpa;
1134 	int rc = 0;
1135 
1136 	while (len && !rc) {
1137 		gpa = kvm_s390_real_to_abs(vcpu, gra);
1138 		fragment_len = min(PAGE_SIZE - offset_in_page(gpa), len);
1139 		rc = access_guest_page(vcpu->kvm, mode, gpa, data, fragment_len);
1140 		len -= fragment_len;
1141 		gra += fragment_len;
1142 		data += fragment_len;
1143 	}
1144 	return rc;
1145 }
1146 
1147 /**
1148  * cmpxchg_guest_abs_with_key() - Perform cmpxchg on guest absolute address.
1149  * @kvm: Virtual machine instance.
1150  * @gpa: Absolute guest address of the location to be changed.
1151  * @len: Operand length of the cmpxchg, required: 1 <= len <= 16. Providing a
1152  *       non power of two will result in failure.
1153  * @old_addr: Pointer to old value. If the location at @gpa contains this value,
1154  *            the exchange will succeed. After calling cmpxchg_guest_abs_with_key()
1155  *            *@old_addr contains the value at @gpa before the attempt to
1156  *            exchange the value.
1157  * @new: The value to place at @gpa.
1158  * @access_key: The access key to use for the guest access.
1159  * @success: output value indicating if an exchange occurred.
1160  *
1161  * Atomically exchange the value at @gpa by @new, if it contains *@old.
1162  * Honors storage keys.
1163  *
1164  * Return: * 0: successful exchange
1165  *         * >0: a program interruption code indicating the reason cmpxchg could
1166  *               not be attempted
1167  *         * -EINVAL: address misaligned or len not power of two
1168  *         * -EAGAIN: transient failure (len 1 or 2)
1169  *         * -EOPNOTSUPP: read-only memslot (should never occur)
1170  */
1171 int cmpxchg_guest_abs_with_key(struct kvm *kvm, gpa_t gpa, int len,
1172 			       __uint128_t *old_addr, __uint128_t new,
1173 			       u8 access_key, bool *success)
1174 {
1175 	gfn_t gfn = gpa_to_gfn(gpa);
1176 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1177 	bool writable;
1178 	hva_t hva;
1179 	int ret;
1180 
1181 	if (!IS_ALIGNED(gpa, len))
1182 		return -EINVAL;
1183 
1184 	hva = gfn_to_hva_memslot_prot(slot, gfn, &writable);
1185 	if (kvm_is_error_hva(hva))
1186 		return PGM_ADDRESSING;
1187 	/*
1188 	 * Check if it's a read-only memslot, even though that cannot occur
1189 	 * since those are unsupported.
1190 	 * Don't try to actually handle that case.
1191 	 */
1192 	if (!writable)
1193 		return -EOPNOTSUPP;
1194 
1195 	hva += offset_in_page(gpa);
1196 	/*
1197 	 * The cmpxchg_user_key macro depends on the type of "old", so we need
1198 	 * a case for each valid length and get some code duplication as long
1199 	 * as we don't introduce a new macro.
1200 	 */
1201 	switch (len) {
1202 	case 1: {
1203 		u8 old;
1204 
1205 		ret = cmpxchg_user_key((u8 __user *)hva, &old, *old_addr, new, access_key);
1206 		*success = !ret && old == *old_addr;
1207 		*old_addr = old;
1208 		break;
1209 	}
1210 	case 2: {
1211 		u16 old;
1212 
1213 		ret = cmpxchg_user_key((u16 __user *)hva, &old, *old_addr, new, access_key);
1214 		*success = !ret && old == *old_addr;
1215 		*old_addr = old;
1216 		break;
1217 	}
1218 	case 4: {
1219 		u32 old;
1220 
1221 		ret = cmpxchg_user_key((u32 __user *)hva, &old, *old_addr, new, access_key);
1222 		*success = !ret && old == *old_addr;
1223 		*old_addr = old;
1224 		break;
1225 	}
1226 	case 8: {
1227 		u64 old;
1228 
1229 		ret = cmpxchg_user_key((u64 __user *)hva, &old, *old_addr, new, access_key);
1230 		*success = !ret && old == *old_addr;
1231 		*old_addr = old;
1232 		break;
1233 	}
1234 	case 16: {
1235 		__uint128_t old;
1236 
1237 		ret = cmpxchg_user_key((__uint128_t __user *)hva, &old, *old_addr, new, access_key);
1238 		*success = !ret && old == *old_addr;
1239 		*old_addr = old;
1240 		break;
1241 	}
1242 	default:
1243 		return -EINVAL;
1244 	}
1245 	if (*success)
1246 		mark_page_dirty_in_slot(kvm, slot, gfn);
1247 	/*
1248 	 * Assume that the fault is caused by protection, either key protection
1249 	 * or user page write protection.
1250 	 */
1251 	if (ret == -EFAULT)
1252 		ret = PGM_PROTECTION;
1253 	return ret;
1254 }
1255 
1256 /**
1257  * guest_translate_address_with_key - translate guest logical into guest absolute address
1258  * @vcpu: virtual cpu
1259  * @gva: Guest virtual address
1260  * @ar: Access register
1261  * @gpa: Guest physical address
1262  * @mode: Translation access mode
1263  * @access_key: access key to mach the storage key with
1264  *
1265  * Parameter semantics are the same as the ones from guest_translate.
1266  * The memory contents at the guest address are not changed.
1267  *
1268  * Note: The IPTE lock is not taken during this function, so the caller
1269  * has to take care of this.
1270  */
1271 int guest_translate_address_with_key(struct kvm_vcpu *vcpu, unsigned long gva, u8 ar,
1272 				     unsigned long *gpa, enum gacc_mode mode,
1273 				     u8 access_key)
1274 {
1275 	union asce asce;
1276 	int rc;
1277 
1278 	gva = kvm_s390_logical_to_effective(vcpu, gva);
1279 	rc = get_vcpu_asce(vcpu, &asce, gva, ar, mode);
1280 	if (rc)
1281 		return rc;
1282 	return guest_range_to_gpas(vcpu, gva, ar, gpa, 1, asce, mode,
1283 				   access_key);
1284 }
1285 
1286 /**
1287  * check_gva_range - test a range of guest virtual addresses for accessibility
1288  * @vcpu: virtual cpu
1289  * @gva: Guest virtual address
1290  * @ar: Access register
1291  * @length: Length of test range
1292  * @mode: Translation access mode
1293  * @access_key: access key to mach the storage keys with
1294  */
1295 int check_gva_range(struct kvm_vcpu *vcpu, unsigned long gva, u8 ar,
1296 		    unsigned long length, enum gacc_mode mode, u8 access_key)
1297 {
1298 	union asce asce;
1299 	int rc = 0;
1300 
1301 	rc = get_vcpu_asce(vcpu, &asce, gva, ar, mode);
1302 	if (rc)
1303 		return rc;
1304 	ipte_lock(vcpu->kvm);
1305 	rc = guest_range_to_gpas(vcpu, gva, ar, NULL, length, asce, mode,
1306 				 access_key);
1307 	ipte_unlock(vcpu->kvm);
1308 
1309 	return rc;
1310 }
1311 
1312 /**
1313  * check_gpa_range - test a range of guest physical addresses for accessibility
1314  * @kvm: virtual machine instance
1315  * @gpa: guest physical address
1316  * @length: length of test range
1317  * @mode: access mode to test, relevant for storage keys
1318  * @access_key: access key to mach the storage keys with
1319  */
1320 int check_gpa_range(struct kvm *kvm, unsigned long gpa, unsigned long length,
1321 		    enum gacc_mode mode, u8 access_key)
1322 {
1323 	unsigned int fragment_len;
1324 	int rc = 0;
1325 
1326 	while (length && !rc) {
1327 		fragment_len = min(PAGE_SIZE - offset_in_page(gpa), length);
1328 		rc = vm_check_access_key(kvm, access_key, mode, gpa);
1329 		length -= fragment_len;
1330 		gpa += fragment_len;
1331 	}
1332 	return rc;
1333 }
1334 
1335 /**
1336  * kvm_s390_check_low_addr_prot_real - check for low-address protection
1337  * @vcpu: virtual cpu
1338  * @gra: Guest real address
1339  *
1340  * Checks whether an address is subject to low-address protection and set
1341  * up vcpu->arch.pgm accordingly if necessary.
1342  *
1343  * Return: 0 if no protection exception, or PGM_PROTECTION if protected.
1344  */
1345 int kvm_s390_check_low_addr_prot_real(struct kvm_vcpu *vcpu, unsigned long gra)
1346 {
1347 	union ctlreg0 ctlreg0 = {.val = vcpu->arch.sie_block->gcr[0]};
1348 
1349 	if (!ctlreg0.lap || !is_low_address(gra))
1350 		return 0;
1351 	return trans_exc(vcpu, PGM_PROTECTION, gra, 0, GACC_STORE, PROT_TYPE_LA);
1352 }
1353 
1354 /**
1355  * kvm_s390_shadow_tables - walk the guest page table and create shadow tables
1356  * @sg: pointer to the shadow guest address space structure
1357  * @saddr: faulting address in the shadow gmap
1358  * @pgt: pointer to the beginning of the page table for the given address if
1359  *	 successful (return value 0), or to the first invalid DAT entry in
1360  *	 case of exceptions (return value > 0)
1361  * @dat_protection: referenced memory is write protected
1362  * @fake: pgt references contiguous guest memory block, not a pgtable
1363  */
1364 static int kvm_s390_shadow_tables(struct gmap *sg, unsigned long saddr,
1365 				  unsigned long *pgt, int *dat_protection,
1366 				  int *fake)
1367 {
1368 	struct kvm *kvm;
1369 	struct gmap *parent;
1370 	union asce asce;
1371 	union vaddress vaddr;
1372 	unsigned long ptr;
1373 	int rc;
1374 
1375 	*fake = 0;
1376 	*dat_protection = 0;
1377 	kvm = sg->private;
1378 	parent = sg->parent;
1379 	vaddr.addr = saddr;
1380 	asce.val = sg->orig_asce;
1381 	ptr = asce.origin * PAGE_SIZE;
1382 	if (asce.r) {
1383 		*fake = 1;
1384 		ptr = 0;
1385 		asce.dt = ASCE_TYPE_REGION1;
1386 	}
1387 	switch (asce.dt) {
1388 	case ASCE_TYPE_REGION1:
1389 		if (vaddr.rfx01 > asce.tl && !*fake)
1390 			return PGM_REGION_FIRST_TRANS;
1391 		break;
1392 	case ASCE_TYPE_REGION2:
1393 		if (vaddr.rfx)
1394 			return PGM_ASCE_TYPE;
1395 		if (vaddr.rsx01 > asce.tl)
1396 			return PGM_REGION_SECOND_TRANS;
1397 		break;
1398 	case ASCE_TYPE_REGION3:
1399 		if (vaddr.rfx || vaddr.rsx)
1400 			return PGM_ASCE_TYPE;
1401 		if (vaddr.rtx01 > asce.tl)
1402 			return PGM_REGION_THIRD_TRANS;
1403 		break;
1404 	case ASCE_TYPE_SEGMENT:
1405 		if (vaddr.rfx || vaddr.rsx || vaddr.rtx)
1406 			return PGM_ASCE_TYPE;
1407 		if (vaddr.sx01 > asce.tl)
1408 			return PGM_SEGMENT_TRANSLATION;
1409 		break;
1410 	}
1411 
1412 	switch (asce.dt) {
1413 	case ASCE_TYPE_REGION1: {
1414 		union region1_table_entry rfte;
1415 
1416 		if (*fake) {
1417 			ptr += vaddr.rfx * _REGION1_SIZE;
1418 			rfte.val = ptr;
1419 			goto shadow_r2t;
1420 		}
1421 		*pgt = ptr + vaddr.rfx * 8;
1422 		rc = gmap_read_table(parent, ptr + vaddr.rfx * 8, &rfte.val);
1423 		if (rc)
1424 			return rc;
1425 		if (rfte.i)
1426 			return PGM_REGION_FIRST_TRANS;
1427 		if (rfte.tt != TABLE_TYPE_REGION1)
1428 			return PGM_TRANSLATION_SPEC;
1429 		if (vaddr.rsx01 < rfte.tf || vaddr.rsx01 > rfte.tl)
1430 			return PGM_REGION_SECOND_TRANS;
1431 		if (sg->edat_level >= 1)
1432 			*dat_protection |= rfte.p;
1433 		ptr = rfte.rto * PAGE_SIZE;
1434 shadow_r2t:
1435 		rc = gmap_shadow_r2t(sg, saddr, rfte.val, *fake);
1436 		if (rc)
1437 			return rc;
1438 		kvm->stat.gmap_shadow_r1_entry++;
1439 	}
1440 		fallthrough;
1441 	case ASCE_TYPE_REGION2: {
1442 		union region2_table_entry rste;
1443 
1444 		if (*fake) {
1445 			ptr += vaddr.rsx * _REGION2_SIZE;
1446 			rste.val = ptr;
1447 			goto shadow_r3t;
1448 		}
1449 		*pgt = ptr + vaddr.rsx * 8;
1450 		rc = gmap_read_table(parent, ptr + vaddr.rsx * 8, &rste.val);
1451 		if (rc)
1452 			return rc;
1453 		if (rste.i)
1454 			return PGM_REGION_SECOND_TRANS;
1455 		if (rste.tt != TABLE_TYPE_REGION2)
1456 			return PGM_TRANSLATION_SPEC;
1457 		if (vaddr.rtx01 < rste.tf || vaddr.rtx01 > rste.tl)
1458 			return PGM_REGION_THIRD_TRANS;
1459 		if (sg->edat_level >= 1)
1460 			*dat_protection |= rste.p;
1461 		ptr = rste.rto * PAGE_SIZE;
1462 shadow_r3t:
1463 		rste.p |= *dat_protection;
1464 		rc = gmap_shadow_r3t(sg, saddr, rste.val, *fake);
1465 		if (rc)
1466 			return rc;
1467 		kvm->stat.gmap_shadow_r2_entry++;
1468 	}
1469 		fallthrough;
1470 	case ASCE_TYPE_REGION3: {
1471 		union region3_table_entry rtte;
1472 
1473 		if (*fake) {
1474 			ptr += vaddr.rtx * _REGION3_SIZE;
1475 			rtte.val = ptr;
1476 			goto shadow_sgt;
1477 		}
1478 		*pgt = ptr + vaddr.rtx * 8;
1479 		rc = gmap_read_table(parent, ptr + vaddr.rtx * 8, &rtte.val);
1480 		if (rc)
1481 			return rc;
1482 		if (rtte.i)
1483 			return PGM_REGION_THIRD_TRANS;
1484 		if (rtte.tt != TABLE_TYPE_REGION3)
1485 			return PGM_TRANSLATION_SPEC;
1486 		if (rtte.cr && asce.p && sg->edat_level >= 2)
1487 			return PGM_TRANSLATION_SPEC;
1488 		if (rtte.fc && sg->edat_level >= 2) {
1489 			*dat_protection |= rtte.fc0.p;
1490 			*fake = 1;
1491 			ptr = rtte.fc1.rfaa * _REGION3_SIZE;
1492 			rtte.val = ptr;
1493 			goto shadow_sgt;
1494 		}
1495 		if (vaddr.sx01 < rtte.fc0.tf || vaddr.sx01 > rtte.fc0.tl)
1496 			return PGM_SEGMENT_TRANSLATION;
1497 		if (sg->edat_level >= 1)
1498 			*dat_protection |= rtte.fc0.p;
1499 		ptr = rtte.fc0.sto * PAGE_SIZE;
1500 shadow_sgt:
1501 		rtte.fc0.p |= *dat_protection;
1502 		rc = gmap_shadow_sgt(sg, saddr, rtte.val, *fake);
1503 		if (rc)
1504 			return rc;
1505 		kvm->stat.gmap_shadow_r3_entry++;
1506 	}
1507 		fallthrough;
1508 	case ASCE_TYPE_SEGMENT: {
1509 		union segment_table_entry ste;
1510 
1511 		if (*fake) {
1512 			ptr += vaddr.sx * _SEGMENT_SIZE;
1513 			ste.val = ptr;
1514 			goto shadow_pgt;
1515 		}
1516 		*pgt = ptr + vaddr.sx * 8;
1517 		rc = gmap_read_table(parent, ptr + vaddr.sx * 8, &ste.val);
1518 		if (rc)
1519 			return rc;
1520 		if (ste.i)
1521 			return PGM_SEGMENT_TRANSLATION;
1522 		if (ste.tt != TABLE_TYPE_SEGMENT)
1523 			return PGM_TRANSLATION_SPEC;
1524 		if (ste.cs && asce.p)
1525 			return PGM_TRANSLATION_SPEC;
1526 		*dat_protection |= ste.fc0.p;
1527 		if (ste.fc && sg->edat_level >= 1) {
1528 			*fake = 1;
1529 			ptr = ste.fc1.sfaa * _SEGMENT_SIZE;
1530 			ste.val = ptr;
1531 			goto shadow_pgt;
1532 		}
1533 		ptr = ste.fc0.pto * (PAGE_SIZE / 2);
1534 shadow_pgt:
1535 		ste.fc0.p |= *dat_protection;
1536 		rc = gmap_shadow_pgt(sg, saddr, ste.val, *fake);
1537 		if (rc)
1538 			return rc;
1539 		kvm->stat.gmap_shadow_sg_entry++;
1540 	}
1541 	}
1542 	/* Return the parent address of the page table */
1543 	*pgt = ptr;
1544 	return 0;
1545 }
1546 
1547 /**
1548  * kvm_s390_shadow_fault - handle fault on a shadow page table
1549  * @vcpu: virtual cpu
1550  * @sg: pointer to the shadow guest address space structure
1551  * @saddr: faulting address in the shadow gmap
1552  * @datptr: will contain the address of the faulting DAT table entry, or of
1553  *	    the valid leaf, plus some flags
1554  *
1555  * Returns: - 0 if the shadow fault was successfully resolved
1556  *	    - > 0 (pgm exception code) on exceptions while faulting
1557  *	    - -EAGAIN if the caller can retry immediately
1558  *	    - -EFAULT when accessing invalid guest addresses
1559  *	    - -ENOMEM if out of memory
1560  */
1561 int kvm_s390_shadow_fault(struct kvm_vcpu *vcpu, struct gmap *sg,
1562 			  unsigned long saddr, unsigned long *datptr)
1563 {
1564 	union vaddress vaddr;
1565 	union page_table_entry pte;
1566 	unsigned long pgt = 0;
1567 	int dat_protection, fake;
1568 	int rc;
1569 
1570 	mmap_read_lock(sg->mm);
1571 	/*
1572 	 * We don't want any guest-2 tables to change - so the parent
1573 	 * tables/pointers we read stay valid - unshadowing is however
1574 	 * always possible - only guest_table_lock protects us.
1575 	 */
1576 	ipte_lock(vcpu->kvm);
1577 
1578 	rc = gmap_shadow_pgt_lookup(sg, saddr, &pgt, &dat_protection, &fake);
1579 	if (rc)
1580 		rc = kvm_s390_shadow_tables(sg, saddr, &pgt, &dat_protection,
1581 					    &fake);
1582 
1583 	vaddr.addr = saddr;
1584 	if (fake) {
1585 		pte.val = pgt + vaddr.px * PAGE_SIZE;
1586 		goto shadow_page;
1587 	}
1588 
1589 	switch (rc) {
1590 	case PGM_SEGMENT_TRANSLATION:
1591 	case PGM_REGION_THIRD_TRANS:
1592 	case PGM_REGION_SECOND_TRANS:
1593 	case PGM_REGION_FIRST_TRANS:
1594 		pgt |= PEI_NOT_PTE;
1595 		break;
1596 	case 0:
1597 		pgt += vaddr.px * 8;
1598 		rc = gmap_read_table(sg->parent, pgt, &pte.val);
1599 	}
1600 	if (datptr)
1601 		*datptr = pgt | dat_protection * PEI_DAT_PROT;
1602 	if (!rc && pte.i)
1603 		rc = PGM_PAGE_TRANSLATION;
1604 	if (!rc && pte.z)
1605 		rc = PGM_TRANSLATION_SPEC;
1606 shadow_page:
1607 	pte.p |= dat_protection;
1608 	if (!rc)
1609 		rc = gmap_shadow_page(sg, saddr, __pte(pte.val));
1610 	vcpu->kvm->stat.gmap_shadow_pg_entry++;
1611 	ipte_unlock(vcpu->kvm);
1612 	mmap_read_unlock(sg->mm);
1613 	return rc;
1614 }
1615