]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - arch/s390/kvm/gaccess.c
s390/mm: push rte protection down to shadow pte
[mirror_ubuntu-zesty-kernel.git] / arch / s390 / kvm / gaccess.c
CommitLineData
22938978
HC
1/*
2 * guest access functions
3 *
4 * Copyright IBM Corp. 2014
5 *
6 */
7
8#include <linux/vmalloc.h>
9#include <linux/err.h>
10#include <asm/pgtable.h>
aa17aa57 11#include <asm/gmap.h>
22938978
HC
12#include "kvm-s390.h"
13#include "gaccess.h"
664b4973 14#include <asm/switch_to.h>
22938978
HC
15
16union asce {
17 unsigned long val;
18 struct {
19 unsigned long origin : 52; /* Region- or Segment-Table Origin */
20 unsigned long : 2;
21 unsigned long g : 1; /* Subspace Group Control */
22 unsigned long p : 1; /* Private Space Control */
23 unsigned long s : 1; /* Storage-Alteration-Event Control */
24 unsigned long x : 1; /* Space-Switch-Event Control */
25 unsigned long r : 1; /* Real-Space Control */
26 unsigned long : 1;
27 unsigned long dt : 2; /* Designation-Type Control */
28 unsigned long tl : 2; /* Region- or Segment-Table Length */
29 };
30};
31
32enum {
33 ASCE_TYPE_SEGMENT = 0,
34 ASCE_TYPE_REGION3 = 1,
35 ASCE_TYPE_REGION2 = 2,
36 ASCE_TYPE_REGION1 = 3
37};
38
39union region1_table_entry {
40 unsigned long val;
41 struct {
42 unsigned long rto: 52;/* Region-Table Origin */
43 unsigned long : 2;
44 unsigned long p : 1; /* DAT-Protection Bit */
45 unsigned long : 1;
46 unsigned long tf : 2; /* Region-Second-Table Offset */
47 unsigned long i : 1; /* Region-Invalid Bit */
48 unsigned long : 1;
49 unsigned long tt : 2; /* Table-Type Bits */
50 unsigned long tl : 2; /* Region-Second-Table Length */
51 };
52};
53
54union region2_table_entry {
55 unsigned long val;
56 struct {
57 unsigned long rto: 52;/* Region-Table Origin */
58 unsigned long : 2;
59 unsigned long p : 1; /* DAT-Protection Bit */
60 unsigned long : 1;
61 unsigned long tf : 2; /* Region-Third-Table Offset */
62 unsigned long i : 1; /* Region-Invalid Bit */
63 unsigned long : 1;
64 unsigned long tt : 2; /* Table-Type Bits */
65 unsigned long tl : 2; /* Region-Third-Table Length */
66 };
67};
68
69struct region3_table_entry_fc0 {
70 unsigned long sto: 52;/* Segment-Table Origin */
71 unsigned long : 1;
72 unsigned long fc : 1; /* Format-Control */
73 unsigned long p : 1; /* DAT-Protection Bit */
74 unsigned long : 1;
75 unsigned long tf : 2; /* Segment-Table Offset */
76 unsigned long i : 1; /* Region-Invalid Bit */
77 unsigned long cr : 1; /* Common-Region Bit */
78 unsigned long tt : 2; /* Table-Type Bits */
79 unsigned long tl : 2; /* Segment-Table Length */
80};
81
82struct region3_table_entry_fc1 {
83 unsigned long rfaa : 33; /* Region-Frame Absolute Address */
84 unsigned long : 14;
85 unsigned long av : 1; /* ACCF-Validity Control */
86 unsigned long acc: 4; /* Access-Control Bits */
87 unsigned long f : 1; /* Fetch-Protection Bit */
88 unsigned long fc : 1; /* Format-Control */
89 unsigned long p : 1; /* DAT-Protection Bit */
90 unsigned long co : 1; /* Change-Recording Override */
91 unsigned long : 2;
92 unsigned long i : 1; /* Region-Invalid Bit */
93 unsigned long cr : 1; /* Common-Region Bit */
94 unsigned long tt : 2; /* Table-Type Bits */
95 unsigned long : 2;
96};
97
98union region3_table_entry {
99 unsigned long val;
100 struct region3_table_entry_fc0 fc0;
101 struct region3_table_entry_fc1 fc1;
102 struct {
103 unsigned long : 53;
104 unsigned long fc : 1; /* Format-Control */
105 unsigned long : 4;
106 unsigned long i : 1; /* Region-Invalid Bit */
107 unsigned long cr : 1; /* Common-Region Bit */
108 unsigned long tt : 2; /* Table-Type Bits */
109 unsigned long : 2;
110 };
111};
112
113struct segment_entry_fc0 {
114 unsigned long pto: 53;/* Page-Table Origin */
115 unsigned long fc : 1; /* Format-Control */
116 unsigned long p : 1; /* DAT-Protection Bit */
117 unsigned long : 3;
118 unsigned long i : 1; /* Segment-Invalid Bit */
119 unsigned long cs : 1; /* Common-Segment Bit */
120 unsigned long tt : 2; /* Table-Type Bits */
121 unsigned long : 2;
122};
123
124struct segment_entry_fc1 {
125 unsigned long sfaa : 44; /* Segment-Frame Absolute Address */
126 unsigned long : 3;
127 unsigned long av : 1; /* ACCF-Validity Control */
128 unsigned long acc: 4; /* Access-Control Bits */
129 unsigned long f : 1; /* Fetch-Protection Bit */
130 unsigned long fc : 1; /* Format-Control */
131 unsigned long p : 1; /* DAT-Protection Bit */
132 unsigned long co : 1; /* Change-Recording Override */
133 unsigned long : 2;
134 unsigned long i : 1; /* Segment-Invalid Bit */
135 unsigned long cs : 1; /* Common-Segment Bit */
136 unsigned long tt : 2; /* Table-Type Bits */
137 unsigned long : 2;
138};
139
140union segment_table_entry {
141 unsigned long val;
142 struct segment_entry_fc0 fc0;
143 struct segment_entry_fc1 fc1;
144 struct {
145 unsigned long : 53;
146 unsigned long fc : 1; /* Format-Control */
147 unsigned long : 4;
148 unsigned long i : 1; /* Segment-Invalid Bit */
149 unsigned long cs : 1; /* Common-Segment Bit */
150 unsigned long tt : 2; /* Table-Type Bits */
151 unsigned long : 2;
152 };
153};
154
155enum {
156 TABLE_TYPE_SEGMENT = 0,
157 TABLE_TYPE_REGION3 = 1,
158 TABLE_TYPE_REGION2 = 2,
159 TABLE_TYPE_REGION1 = 3
160};
161
162union page_table_entry {
163 unsigned long val;
164 struct {
165 unsigned long pfra : 52; /* Page-Frame Real Address */
166 unsigned long z : 1; /* Zero Bit */
167 unsigned long i : 1; /* Page-Invalid Bit */
168 unsigned long p : 1; /* DAT-Protection Bit */
169 unsigned long co : 1; /* Change-Recording Override */
170 unsigned long : 8;
171 };
172};
173
174/*
175 * vaddress union in order to easily decode a virtual address into its
176 * region first index, region second index etc. parts.
177 */
178union vaddress {
179 unsigned long addr;
180 struct {
181 unsigned long rfx : 11;
182 unsigned long rsx : 11;
183 unsigned long rtx : 11;
184 unsigned long sx : 11;
185 unsigned long px : 8;
186 unsigned long bx : 12;
187 };
188 struct {
189 unsigned long rfx01 : 2;
190 unsigned long : 9;
191 unsigned long rsx01 : 2;
192 unsigned long : 9;
193 unsigned long rtx01 : 2;
194 unsigned long : 9;
195 unsigned long sx01 : 2;
196 unsigned long : 29;
197 };
198};
199
200/*
201 * raddress union which will contain the result (real or absolute address)
202 * after a page table walk. The rfaa, sfaa and pfra members are used to
203 * simply assign them the value of a region, segment or page table entry.
204 */
205union raddress {
206 unsigned long addr;
207 unsigned long rfaa : 33; /* Region-Frame Absolute Address */
208 unsigned long sfaa : 44; /* Segment-Frame Absolute Address */
209 unsigned long pfra : 52; /* Page-Frame Real Address */
210};
211
664b4973
AY
212union alet {
213 u32 val;
214 struct {
215 u32 reserved : 7;
216 u32 p : 1;
217 u32 alesn : 8;
218 u32 alen : 16;
219 };
220};
221
222union ald {
223 u32 val;
224 struct {
225 u32 : 1;
226 u32 alo : 24;
227 u32 all : 7;
228 };
229};
230
231struct ale {
232 unsigned long i : 1; /* ALEN-Invalid Bit */
233 unsigned long : 5;
234 unsigned long fo : 1; /* Fetch-Only Bit */
235 unsigned long p : 1; /* Private Bit */
236 unsigned long alesn : 8; /* Access-List-Entry Sequence Number */
237 unsigned long aleax : 16; /* Access-List-Entry Authorization Index */
238 unsigned long : 32;
239 unsigned long : 1;
240 unsigned long asteo : 25; /* ASN-Second-Table-Entry Origin */
241 unsigned long : 6;
242 unsigned long astesn : 32; /* ASTE Sequence Number */
243} __packed;
244
245struct aste {
246 unsigned long i : 1; /* ASX-Invalid Bit */
247 unsigned long ato : 29; /* Authority-Table Origin */
248 unsigned long : 1;
249 unsigned long b : 1; /* Base-Space Bit */
250 unsigned long ax : 16; /* Authorization Index */
251 unsigned long atl : 12; /* Authority-Table Length */
252 unsigned long : 2;
253 unsigned long ca : 1; /* Controlled-ASN Bit */
254 unsigned long ra : 1; /* Reusable-ASN Bit */
255 unsigned long asce : 64; /* Address-Space-Control Element */
256 unsigned long ald : 32;
257 unsigned long astesn : 32;
258 /* .. more fields there */
259} __packed;
8a242234
HC
260
261int ipte_lock_held(struct kvm_vcpu *vcpu)
262{
5e044315
ED
263 if (vcpu->arch.sie_block->eca & 1) {
264 int rc;
265
266 read_lock(&vcpu->kvm->arch.sca_lock);
267 rc = kvm_s390_get_ipte_control(vcpu->kvm)->kh != 0;
268 read_unlock(&vcpu->kvm->arch.sca_lock);
269 return rc;
270 }
a6b7e459 271 return vcpu->kvm->arch.ipte_lock_count != 0;
8a242234
HC
272}
273
274static void ipte_lock_simple(struct kvm_vcpu *vcpu)
275{
276 union ipte_control old, new, *ic;
277
a6b7e459
TH
278 mutex_lock(&vcpu->kvm->arch.ipte_mutex);
279 vcpu->kvm->arch.ipte_lock_count++;
280 if (vcpu->kvm->arch.ipte_lock_count > 1)
8a242234 281 goto out;
5e044315
ED
282retry:
283 read_lock(&vcpu->kvm->arch.sca_lock);
60514510 284 ic = kvm_s390_get_ipte_control(vcpu->kvm);
8a242234 285 do {
5de72a22 286 old = READ_ONCE(*ic);
5e044315
ED
287 if (old.k) {
288 read_unlock(&vcpu->kvm->arch.sca_lock);
8a242234 289 cond_resched();
5e044315 290 goto retry;
8a242234
HC
291 }
292 new = old;
293 new.k = 1;
294 } while (cmpxchg(&ic->val, old.val, new.val) != old.val);
5e044315 295 read_unlock(&vcpu->kvm->arch.sca_lock);
8a242234 296out:
a6b7e459 297 mutex_unlock(&vcpu->kvm->arch.ipte_mutex);
8a242234
HC
298}
299
300static void ipte_unlock_simple(struct kvm_vcpu *vcpu)
301{
302 union ipte_control old, new, *ic;
303
a6b7e459
TH
304 mutex_lock(&vcpu->kvm->arch.ipte_mutex);
305 vcpu->kvm->arch.ipte_lock_count--;
306 if (vcpu->kvm->arch.ipte_lock_count)
8a242234 307 goto out;
5e044315 308 read_lock(&vcpu->kvm->arch.sca_lock);
60514510 309 ic = kvm_s390_get_ipte_control(vcpu->kvm);
8a242234 310 do {
5de72a22 311 old = READ_ONCE(*ic);
1365039d 312 new = old;
8a242234
HC
313 new.k = 0;
314 } while (cmpxchg(&ic->val, old.val, new.val) != old.val);
5e044315 315 read_unlock(&vcpu->kvm->arch.sca_lock);
6b331952 316 wake_up(&vcpu->kvm->arch.ipte_wq);
8a242234 317out:
a6b7e459 318 mutex_unlock(&vcpu->kvm->arch.ipte_mutex);
8a242234
HC
319}
320
321static void ipte_lock_siif(struct kvm_vcpu *vcpu)
322{
323 union ipte_control old, new, *ic;
324
5e044315
ED
325retry:
326 read_lock(&vcpu->kvm->arch.sca_lock);
60514510 327 ic = kvm_s390_get_ipte_control(vcpu->kvm);
8a242234 328 do {
5de72a22 329 old = READ_ONCE(*ic);
5e044315
ED
330 if (old.kg) {
331 read_unlock(&vcpu->kvm->arch.sca_lock);
8a242234 332 cond_resched();
5e044315 333 goto retry;
8a242234
HC
334 }
335 new = old;
336 new.k = 1;
337 new.kh++;
338 } while (cmpxchg(&ic->val, old.val, new.val) != old.val);
5e044315 339 read_unlock(&vcpu->kvm->arch.sca_lock);
8a242234
HC
340}
341
342static void ipte_unlock_siif(struct kvm_vcpu *vcpu)
343{
344 union ipte_control old, new, *ic;
345
5e044315 346 read_lock(&vcpu->kvm->arch.sca_lock);
60514510 347 ic = kvm_s390_get_ipte_control(vcpu->kvm);
8a242234 348 do {
5de72a22 349 old = READ_ONCE(*ic);
1365039d 350 new = old;
8a242234
HC
351 new.kh--;
352 if (!new.kh)
353 new.k = 0;
354 } while (cmpxchg(&ic->val, old.val, new.val) != old.val);
5e044315 355 read_unlock(&vcpu->kvm->arch.sca_lock);
8a242234
HC
356 if (!new.kh)
357 wake_up(&vcpu->kvm->arch.ipte_wq);
358}
359
a0465f9a 360void ipte_lock(struct kvm_vcpu *vcpu)
8a242234
HC
361{
362 if (vcpu->arch.sie_block->eca & 1)
363 ipte_lock_siif(vcpu);
364 else
365 ipte_lock_simple(vcpu);
366}
367
a0465f9a 368void ipte_unlock(struct kvm_vcpu *vcpu)
8a242234
HC
369{
370 if (vcpu->arch.sie_block->eca & 1)
371 ipte_unlock_siif(vcpu);
372 else
373 ipte_unlock_simple(vcpu);
374}
375
664b4973 376static int ar_translation(struct kvm_vcpu *vcpu, union asce *asce, ar_t ar,
92c96321 377 enum gacc_mode mode)
664b4973
AY
378{
379 union alet alet;
380 struct ale ale;
381 struct aste aste;
382 unsigned long ald_addr, authority_table_addr;
383 union ald ald;
384 int eax, rc;
385 u8 authority_table;
386
387 if (ar >= NUM_ACRS)
388 return -EINVAL;
389
390 save_access_regs(vcpu->run->s.regs.acrs);
391 alet.val = vcpu->run->s.regs.acrs[ar];
392
393 if (ar == 0 || alet.val == 0) {
394 asce->val = vcpu->arch.sie_block->gcr[1];
395 return 0;
396 } else if (alet.val == 1) {
397 asce->val = vcpu->arch.sie_block->gcr[7];
398 return 0;
399 }
400
401 if (alet.reserved)
402 return PGM_ALET_SPECIFICATION;
403
404 if (alet.p)
405 ald_addr = vcpu->arch.sie_block->gcr[5];
406 else
407 ald_addr = vcpu->arch.sie_block->gcr[2];
408 ald_addr &= 0x7fffffc0;
409
410 rc = read_guest_real(vcpu, ald_addr + 16, &ald.val, sizeof(union ald));
411 if (rc)
412 return rc;
413
414 if (alet.alen / 8 > ald.all)
415 return PGM_ALEN_TRANSLATION;
416
417 if (0x7fffffff - ald.alo * 128 < alet.alen * 16)
418 return PGM_ADDRESSING;
419
420 rc = read_guest_real(vcpu, ald.alo * 128 + alet.alen * 16, &ale,
421 sizeof(struct ale));
422 if (rc)
423 return rc;
424
425 if (ale.i == 1)
426 return PGM_ALEN_TRANSLATION;
427 if (ale.alesn != alet.alesn)
428 return PGM_ALE_SEQUENCE;
429
430 rc = read_guest_real(vcpu, ale.asteo * 64, &aste, sizeof(struct aste));
431 if (rc)
432 return rc;
433
434 if (aste.i)
435 return PGM_ASTE_VALIDITY;
436 if (aste.astesn != ale.astesn)
437 return PGM_ASTE_SEQUENCE;
438
439 if (ale.p == 1) {
440 eax = (vcpu->arch.sie_block->gcr[8] >> 16) & 0xffff;
441 if (ale.aleax != eax) {
442 if (eax / 16 > aste.atl)
443 return PGM_EXTENDED_AUTHORITY;
444
445 authority_table_addr = aste.ato * 4 + eax / 4;
446
447 rc = read_guest_real(vcpu, authority_table_addr,
448 &authority_table,
449 sizeof(u8));
450 if (rc)
451 return rc;
452
453 if ((authority_table & (0x40 >> ((eax & 3) * 2))) == 0)
454 return PGM_EXTENDED_AUTHORITY;
455 }
456 }
457
92c96321 458 if (ale.fo == 1 && mode == GACC_STORE)
664b4973
AY
459 return PGM_PROTECTION;
460
461 asce->val = aste.asce;
462 return 0;
463}
464
465struct trans_exc_code_bits {
466 unsigned long addr : 52; /* Translation-exception Address */
467 unsigned long fsi : 2; /* Access Exception Fetch/Store Indication */
468 unsigned long : 6;
469 unsigned long b60 : 1;
470 unsigned long b61 : 1;
471 unsigned long as : 2; /* ASCE Identifier */
472};
473
474enum {
475 FSI_UNKNOWN = 0, /* Unknown wether fetch or store */
476 FSI_STORE = 1, /* Exception was due to store operation */
477 FSI_FETCH = 2 /* Exception was due to fetch operation */
478};
479
d03193de
DH
480enum prot_type {
481 PROT_TYPE_LA = 0,
482 PROT_TYPE_KEYC = 1,
483 PROT_TYPE_ALC = 2,
484 PROT_TYPE_DAT = 3,
485};
486
487static int trans_exc(struct kvm_vcpu *vcpu, int code, unsigned long gva,
488 ar_t ar, enum gacc_mode mode, enum prot_type prot)
489{
490 struct kvm_s390_pgm_info *pgm = &vcpu->arch.pgm;
491 struct trans_exc_code_bits *tec;
492
493 memset(pgm, 0, sizeof(*pgm));
494 pgm->code = code;
495 tec = (struct trans_exc_code_bits *)&pgm->trans_exc_code;
496
497 switch (code) {
498 case PGM_ASCE_TYPE:
499 case PGM_PAGE_TRANSLATION:
500 case PGM_REGION_FIRST_TRANS:
501 case PGM_REGION_SECOND_TRANS:
502 case PGM_REGION_THIRD_TRANS:
503 case PGM_SEGMENT_TRANSLATION:
504 /*
505 * op_access_id only applies to MOVE_PAGE -> set bit 61
506 * exc_access_id has to be set to 0 for some instructions. Both
507 * cases have to be handled by the caller. We can always store
508 * exc_access_id, as it is undefined for non-ar cases.
509 */
510 tec->addr = gva >> PAGE_SHIFT;
511 tec->fsi = mode == GACC_STORE ? FSI_STORE : FSI_FETCH;
512 tec->as = psw_bits(vcpu->arch.sie_block->gpsw).as;
513 /* FALL THROUGH */
514 case PGM_ALEN_TRANSLATION:
515 case PGM_ALE_SEQUENCE:
516 case PGM_ASTE_VALIDITY:
517 case PGM_ASTE_SEQUENCE:
518 case PGM_EXTENDED_AUTHORITY:
519 pgm->exc_access_id = ar;
520 break;
521 case PGM_PROTECTION:
522 switch (prot) {
523 case PROT_TYPE_ALC:
524 tec->b60 = 1;
525 /* FALL THROUGH */
526 case PROT_TYPE_DAT:
527 tec->b61 = 1;
528 tec->addr = gva >> PAGE_SHIFT;
529 tec->fsi = mode == GACC_STORE ? FSI_STORE : FSI_FETCH;
530 tec->as = psw_bits(vcpu->arch.sie_block->gpsw).as;
531 /* exc_access_id is undefined for most cases */
532 pgm->exc_access_id = ar;
533 break;
534 default: /* LA and KEYC set b61 to 0, other params undefined */
535 break;
536 }
537 break;
538 }
539 return code;
540}
541
664b4973 542static int get_vcpu_asce(struct kvm_vcpu *vcpu, union asce *asce,
6167375b 543 unsigned long ga, ar_t ar, enum gacc_mode mode)
22938978 544{
664b4973 545 int rc;
34346b9a 546 struct psw_bits psw = psw_bits(vcpu->arch.sie_block->gpsw);
664b4973 547
34346b9a 548 if (!psw.t) {
664b4973
AY
549 asce->val = 0;
550 asce->r = 1;
551 return 0;
552 }
553
34346b9a
DH
554 if (mode == GACC_IFETCH)
555 psw.as = psw.as == PSW_AS_HOME ? PSW_AS_HOME : PSW_AS_PRIMARY;
556
557 switch (psw.as) {
22938978 558 case PSW_AS_PRIMARY:
664b4973
AY
559 asce->val = vcpu->arch.sie_block->gcr[1];
560 return 0;
22938978 561 case PSW_AS_SECONDARY:
664b4973
AY
562 asce->val = vcpu->arch.sie_block->gcr[7];
563 return 0;
22938978 564 case PSW_AS_HOME:
664b4973
AY
565 asce->val = vcpu->arch.sie_block->gcr[13];
566 return 0;
567 case PSW_AS_ACCREG:
92c96321 568 rc = ar_translation(vcpu, asce, ar, mode);
664b4973 569 if (rc > 0)
bcfa01d7 570 return trans_exc(vcpu, rc, ga, ar, mode, PROT_TYPE_ALC);
664b4973 571 return rc;
22938978
HC
572 }
573 return 0;
574}
575
576static int deref_table(struct kvm *kvm, unsigned long gpa, unsigned long *val)
577{
578 return kvm_read_guest(kvm, gpa, val, sizeof(*val));
579}
580
581/**
582 * guest_translate - translate a guest virtual into a guest absolute address
583 * @vcpu: virtual cpu
584 * @gva: guest virtual address
585 * @gpa: points to where guest physical (absolute) address should be stored
75a18122 586 * @asce: effective asce
92c96321 587 * @mode: indicates the access mode to be used
22938978
HC
588 *
589 * Translate a guest virtual address into a guest absolute address by means
16b0fc13 590 * of dynamic address translation as specified by the architecture.
22938978
HC
591 * If the resulting absolute address is not available in the configuration
592 * an addressing exception is indicated and @gpa will not be changed.
593 *
594 * Returns: - zero on success; @gpa contains the resulting absolute address
595 * - a negative value if guest access failed due to e.g. broken
596 * guest mapping
597 * - a positve value if an access exception happened. In this case
598 * the returned value is the program interruption code as defined
599 * by the architecture
600 */
601static unsigned long guest_translate(struct kvm_vcpu *vcpu, unsigned long gva,
75a18122 602 unsigned long *gpa, const union asce asce,
92c96321 603 enum gacc_mode mode)
22938978
HC
604{
605 union vaddress vaddr = {.addr = gva};
606 union raddress raddr = {.addr = gva};
607 union page_table_entry pte;
608 int dat_protection = 0;
609 union ctlreg0 ctlreg0;
610 unsigned long ptr;
611 int edat1, edat2;
22938978
HC
612
613 ctlreg0.val = vcpu->arch.sie_block->gcr[0];
9d8d5786
MM
614 edat1 = ctlreg0.edat && test_kvm_facility(vcpu->kvm, 8);
615 edat2 = edat1 && test_kvm_facility(vcpu->kvm, 78);
22938978
HC
616 if (asce.r)
617 goto real_address;
618 ptr = asce.origin * 4096;
619 switch (asce.dt) {
620 case ASCE_TYPE_REGION1:
621 if (vaddr.rfx01 > asce.tl)
622 return PGM_REGION_FIRST_TRANS;
623 ptr += vaddr.rfx * 8;
624 break;
625 case ASCE_TYPE_REGION2:
626 if (vaddr.rfx)
627 return PGM_ASCE_TYPE;
628 if (vaddr.rsx01 > asce.tl)
629 return PGM_REGION_SECOND_TRANS;
630 ptr += vaddr.rsx * 8;
631 break;
632 case ASCE_TYPE_REGION3:
633 if (vaddr.rfx || vaddr.rsx)
634 return PGM_ASCE_TYPE;
635 if (vaddr.rtx01 > asce.tl)
636 return PGM_REGION_THIRD_TRANS;
637 ptr += vaddr.rtx * 8;
638 break;
639 case ASCE_TYPE_SEGMENT:
640 if (vaddr.rfx || vaddr.rsx || vaddr.rtx)
641 return PGM_ASCE_TYPE;
642 if (vaddr.sx01 > asce.tl)
643 return PGM_SEGMENT_TRANSLATION;
644 ptr += vaddr.sx * 8;
645 break;
646 }
647 switch (asce.dt) {
648 case ASCE_TYPE_REGION1: {
649 union region1_table_entry rfte;
650
651 if (kvm_is_error_gpa(vcpu->kvm, ptr))
652 return PGM_ADDRESSING;
653 if (deref_table(vcpu->kvm, ptr, &rfte.val))
654 return -EFAULT;
655 if (rfte.i)
656 return PGM_REGION_FIRST_TRANS;
657 if (rfte.tt != TABLE_TYPE_REGION1)
658 return PGM_TRANSLATION_SPEC;
659 if (vaddr.rsx01 < rfte.tf || vaddr.rsx01 > rfte.tl)
660 return PGM_REGION_SECOND_TRANS;
661 if (edat1)
662 dat_protection |= rfte.p;
663 ptr = rfte.rto * 4096 + vaddr.rsx * 8;
664 }
665 /* fallthrough */
666 case ASCE_TYPE_REGION2: {
667 union region2_table_entry rste;
668
669 if (kvm_is_error_gpa(vcpu->kvm, ptr))
670 return PGM_ADDRESSING;
671 if (deref_table(vcpu->kvm, ptr, &rste.val))
672 return -EFAULT;
673 if (rste.i)
674 return PGM_REGION_SECOND_TRANS;
675 if (rste.tt != TABLE_TYPE_REGION2)
676 return PGM_TRANSLATION_SPEC;
677 if (vaddr.rtx01 < rste.tf || vaddr.rtx01 > rste.tl)
678 return PGM_REGION_THIRD_TRANS;
679 if (edat1)
680 dat_protection |= rste.p;
681 ptr = rste.rto * 4096 + vaddr.rtx * 8;
682 }
683 /* fallthrough */
684 case ASCE_TYPE_REGION3: {
685 union region3_table_entry rtte;
686
687 if (kvm_is_error_gpa(vcpu->kvm, ptr))
688 return PGM_ADDRESSING;
689 if (deref_table(vcpu->kvm, ptr, &rtte.val))
690 return -EFAULT;
691 if (rtte.i)
692 return PGM_REGION_THIRD_TRANS;
693 if (rtte.tt != TABLE_TYPE_REGION3)
694 return PGM_TRANSLATION_SPEC;
695 if (rtte.cr && asce.p && edat2)
696 return PGM_TRANSLATION_SPEC;
697 if (rtte.fc && edat2) {
698 dat_protection |= rtte.fc1.p;
699 raddr.rfaa = rtte.fc1.rfaa;
700 goto absolute_address;
701 }
702 if (vaddr.sx01 < rtte.fc0.tf)
703 return PGM_SEGMENT_TRANSLATION;
704 if (vaddr.sx01 > rtte.fc0.tl)
705 return PGM_SEGMENT_TRANSLATION;
706 if (edat1)
707 dat_protection |= rtte.fc0.p;
708 ptr = rtte.fc0.sto * 4096 + vaddr.sx * 8;
709 }
710 /* fallthrough */
711 case ASCE_TYPE_SEGMENT: {
712 union segment_table_entry ste;
713
714 if (kvm_is_error_gpa(vcpu->kvm, ptr))
715 return PGM_ADDRESSING;
716 if (deref_table(vcpu->kvm, ptr, &ste.val))
717 return -EFAULT;
718 if (ste.i)
719 return PGM_SEGMENT_TRANSLATION;
720 if (ste.tt != TABLE_TYPE_SEGMENT)
721 return PGM_TRANSLATION_SPEC;
722 if (ste.cs && asce.p)
723 return PGM_TRANSLATION_SPEC;
724 if (ste.fc && edat1) {
725 dat_protection |= ste.fc1.p;
726 raddr.sfaa = ste.fc1.sfaa;
727 goto absolute_address;
728 }
729 dat_protection |= ste.fc0.p;
730 ptr = ste.fc0.pto * 2048 + vaddr.px * 8;
731 }
732 }
733 if (kvm_is_error_gpa(vcpu->kvm, ptr))
734 return PGM_ADDRESSING;
735 if (deref_table(vcpu->kvm, ptr, &pte.val))
736 return -EFAULT;
737 if (pte.i)
738 return PGM_PAGE_TRANSLATION;
739 if (pte.z)
740 return PGM_TRANSLATION_SPEC;
741 if (pte.co && !edat1)
742 return PGM_TRANSLATION_SPEC;
743 dat_protection |= pte.p;
744 raddr.pfra = pte.pfra;
745real_address:
746 raddr.addr = kvm_s390_real_to_abs(vcpu, raddr.addr);
747absolute_address:
92c96321 748 if (mode == GACC_STORE && dat_protection)
22938978
HC
749 return PGM_PROTECTION;
750 if (kvm_is_error_gpa(vcpu->kvm, raddr.addr))
751 return PGM_ADDRESSING;
752 *gpa = raddr.addr;
753 return 0;
754}
755
756static inline int is_low_address(unsigned long ga)
757{
758 /* Check for address ranges 0..511 and 4096..4607 */
759 return (ga & ~0x11fful) == 0;
760}
761
75a18122
AY
762static int low_address_protection_enabled(struct kvm_vcpu *vcpu,
763 const union asce asce)
22938978
HC
764{
765 union ctlreg0 ctlreg0 = {.val = vcpu->arch.sie_block->gcr[0]};
766 psw_t *psw = &vcpu->arch.sie_block->gpsw;
22938978
HC
767
768 if (!ctlreg0.lap)
769 return 0;
22938978
HC
770 if (psw_bits(*psw).t && asce.p)
771 return 0;
772 return 1;
773}
774
cde0dcfb 775static int guest_page_range(struct kvm_vcpu *vcpu, unsigned long ga, ar_t ar,
22938978 776 unsigned long *pages, unsigned long nr_pages,
92c96321 777 const union asce asce, enum gacc_mode mode)
22938978 778{
22938978 779 psw_t *psw = &vcpu->arch.sie_block->gpsw;
cde0dcfb 780 int lap_enabled, rc = 0;
22938978 781
75a18122 782 lap_enabled = low_address_protection_enabled(vcpu, asce);
22938978
HC
783 while (nr_pages) {
784 ga = kvm_s390_logical_to_effective(vcpu, ga);
cde0dcfb
DH
785 if (mode == GACC_STORE && lap_enabled && is_low_address(ga))
786 return trans_exc(vcpu, PGM_PROTECTION, ga, ar, mode,
787 PROT_TYPE_LA);
22938978
HC
788 ga &= PAGE_MASK;
789 if (psw_bits(*psw).t) {
92c96321 790 rc = guest_translate(vcpu, ga, pages, asce, mode);
22938978
HC
791 if (rc < 0)
792 return rc;
22938978
HC
793 } else {
794 *pages = kvm_s390_real_to_abs(vcpu, ga);
795 if (kvm_is_error_gpa(vcpu->kvm, *pages))
cde0dcfb 796 rc = PGM_ADDRESSING;
22938978 797 }
cde0dcfb
DH
798 if (rc)
799 return trans_exc(vcpu, rc, ga, ar, mode, PROT_TYPE_DAT);
22938978
HC
800 ga += PAGE_SIZE;
801 pages++;
802 nr_pages--;
803 }
804 return 0;
805}
806
8ae04b8f 807int access_guest(struct kvm_vcpu *vcpu, unsigned long ga, ar_t ar, void *data,
92c96321 808 unsigned long len, enum gacc_mode mode)
22938978
HC
809{
810 psw_t *psw = &vcpu->arch.sie_block->gpsw;
811 unsigned long _len, nr_pages, gpa, idx;
812 unsigned long pages_array[2];
813 unsigned long *pages;
8a242234
HC
814 int need_ipte_lock;
815 union asce asce;
22938978
HC
816 int rc;
817
818 if (!len)
819 return 0;
6167375b
DH
820 ga = kvm_s390_logical_to_effective(vcpu, ga);
821 rc = get_vcpu_asce(vcpu, &asce, ga, ar, mode);
664b4973
AY
822 if (rc)
823 return rc;
22938978
HC
824 nr_pages = (((ga & ~PAGE_MASK) + len - 1) >> PAGE_SHIFT) + 1;
825 pages = pages_array;
826 if (nr_pages > ARRAY_SIZE(pages_array))
827 pages = vmalloc(nr_pages * sizeof(unsigned long));
828 if (!pages)
829 return -ENOMEM;
8a242234
HC
830 need_ipte_lock = psw_bits(*psw).t && !asce.r;
831 if (need_ipte_lock)
832 ipte_lock(vcpu);
cde0dcfb 833 rc = guest_page_range(vcpu, ga, ar, pages, nr_pages, asce, mode);
22938978
HC
834 for (idx = 0; idx < nr_pages && !rc; idx++) {
835 gpa = *(pages + idx) + (ga & ~PAGE_MASK);
836 _len = min(PAGE_SIZE - (gpa & ~PAGE_MASK), len);
92c96321 837 if (mode == GACC_STORE)
22938978
HC
838 rc = kvm_write_guest(vcpu->kvm, gpa, data, _len);
839 else
840 rc = kvm_read_guest(vcpu->kvm, gpa, data, _len);
841 len -= _len;
842 ga += _len;
843 data += _len;
844 }
8a242234
HC
845 if (need_ipte_lock)
846 ipte_unlock(vcpu);
22938978
HC
847 if (nr_pages > ARRAY_SIZE(pages_array))
848 vfree(pages);
849 return rc;
850}
851
852int access_guest_real(struct kvm_vcpu *vcpu, unsigned long gra,
92c96321 853 void *data, unsigned long len, enum gacc_mode mode)
22938978
HC
854{
855 unsigned long _len, gpa;
856 int rc = 0;
857
858 while (len && !rc) {
859 gpa = kvm_s390_real_to_abs(vcpu, gra);
860 _len = min(PAGE_SIZE - (gpa & ~PAGE_MASK), len);
92c96321 861 if (mode)
22938978
HC
862 rc = write_guest_abs(vcpu, gpa, data, _len);
863 else
864 rc = read_guest_abs(vcpu, gpa, data, _len);
865 len -= _len;
866 gra += _len;
867 data += _len;
868 }
869 return rc;
870}
f8232c8c 871
9fbc0276
TH
872/**
873 * guest_translate_address - translate guest logical into guest absolute address
874 *
875 * Parameter semantics are the same as the ones from guest_translate.
876 * The memory contents at the guest address are not changed.
877 *
878 * Note: The IPTE lock is not taken during this function, so the caller
879 * has to take care of this.
880 */
8ae04b8f 881int guest_translate_address(struct kvm_vcpu *vcpu, unsigned long gva, ar_t ar,
92c96321 882 unsigned long *gpa, enum gacc_mode mode)
9fbc0276 883{
9fbc0276 884 psw_t *psw = &vcpu->arch.sie_block->gpsw;
9fbc0276
TH
885 union asce asce;
886 int rc;
887
9fbc0276 888 gva = kvm_s390_logical_to_effective(vcpu, gva);
6167375b 889 rc = get_vcpu_asce(vcpu, &asce, gva, ar, mode);
664b4973
AY
890 if (rc)
891 return rc;
75a18122 892 if (is_low_address(gva) && low_address_protection_enabled(vcpu, asce)) {
fbcb7d51
DH
893 if (mode == GACC_STORE)
894 return trans_exc(vcpu, PGM_PROTECTION, gva, 0,
895 mode, PROT_TYPE_LA);
9fbc0276
TH
896 }
897
9fbc0276 898 if (psw_bits(*psw).t && !asce.r) { /* Use DAT? */
92c96321 899 rc = guest_translate(vcpu, gva, gpa, asce, mode);
fbcb7d51
DH
900 if (rc > 0)
901 return trans_exc(vcpu, rc, gva, 0, mode, PROT_TYPE_DAT);
9fbc0276 902 } else {
9fbc0276
TH
903 *gpa = kvm_s390_real_to_abs(vcpu, gva);
904 if (kvm_is_error_gpa(vcpu->kvm, *gpa))
fbcb7d51 905 return trans_exc(vcpu, rc, gva, PGM_ADDRESSING, mode, 0);
9fbc0276
TH
906 }
907
908 return rc;
909}
910
41408c28
TH
911/**
912 * check_gva_range - test a range of guest virtual addresses for accessibility
913 */
914int check_gva_range(struct kvm_vcpu *vcpu, unsigned long gva, ar_t ar,
92c96321 915 unsigned long length, enum gacc_mode mode)
41408c28
TH
916{
917 unsigned long gpa;
918 unsigned long currlen;
919 int rc = 0;
920
921 ipte_lock(vcpu);
922 while (length > 0 && !rc) {
923 currlen = min(length, PAGE_SIZE - (gva % PAGE_SIZE));
92c96321 924 rc = guest_translate_address(vcpu, gva, ar, &gpa, mode);
41408c28
TH
925 gva += currlen;
926 length -= currlen;
927 }
928 ipte_unlock(vcpu);
929
930 return rc;
931}
932
f8232c8c 933/**
dd9e5b7b
AY
934 * kvm_s390_check_low_addr_prot_real - check for low-address protection
935 * @gra: Guest real address
f8232c8c
TH
936 *
937 * Checks whether an address is subject to low-address protection and set
938 * up vcpu->arch.pgm accordingly if necessary.
939 *
940 * Return: 0 if no protection exception, or PGM_PROTECTION if protected.
941 */
dd9e5b7b 942int kvm_s390_check_low_addr_prot_real(struct kvm_vcpu *vcpu, unsigned long gra)
f8232c8c 943{
dd9e5b7b 944 union ctlreg0 ctlreg0 = {.val = vcpu->arch.sie_block->gcr[0]};
f8232c8c 945
dd9e5b7b 946 if (!ctlreg0.lap || !is_low_address(gra))
f8232c8c 947 return 0;
3e3c67f6 948 return trans_exc(vcpu, PGM_PROTECTION, gra, 0, GACC_STORE, PROT_TYPE_LA);
f8232c8c 949}
aa17aa57
MS
950
951/**
952 * kvm_s390_shadow_tables - walk the guest page table and create shadow tables
953 * @sg: pointer to the shadow guest address space structure
954 * @saddr: faulting address in the shadow gmap
955 * @pgt: pointer to the page table address result
fd8d4e3a 956 * @fake: pgt references contiguous guest memory block, not a pgtable
aa17aa57
MS
957 */
958static int kvm_s390_shadow_tables(struct gmap *sg, unsigned long saddr,
fd8d4e3a
DH
959 unsigned long *pgt, int *dat_protection,
960 int *fake)
aa17aa57
MS
961{
962 struct gmap *parent;
963 union asce asce;
964 union vaddress vaddr;
965 unsigned long ptr;
966 int rc;
967
fd8d4e3a 968 *fake = 0;
1c65781b 969 *dat_protection = 0;
aa17aa57
MS
970 parent = sg->parent;
971 vaddr.addr = saddr;
972 asce.val = sg->orig_asce;
973 ptr = asce.origin * 4096;
974 switch (asce.dt) {
975 case ASCE_TYPE_REGION1:
976 if (vaddr.rfx01 > asce.tl)
977 return PGM_REGION_FIRST_TRANS;
978 break;
979 case ASCE_TYPE_REGION2:
980 if (vaddr.rfx)
981 return PGM_ASCE_TYPE;
982 if (vaddr.rsx01 > asce.tl)
983 return PGM_REGION_SECOND_TRANS;
984 break;
985 case ASCE_TYPE_REGION3:
986 if (vaddr.rfx || vaddr.rsx)
987 return PGM_ASCE_TYPE;
988 if (vaddr.rtx01 > asce.tl)
989 return PGM_REGION_THIRD_TRANS;
990 break;
991 case ASCE_TYPE_SEGMENT:
992 if (vaddr.rfx || vaddr.rsx || vaddr.rtx)
993 return PGM_ASCE_TYPE;
994 if (vaddr.sx01 > asce.tl)
995 return PGM_SEGMENT_TRANSLATION;
996 break;
997 }
998
999 switch (asce.dt) {
1000 case ASCE_TYPE_REGION1: {
1001 union region1_table_entry rfte;
1002
1003 rc = gmap_read_table(parent, ptr + vaddr.rfx * 8, &rfte.val);
1004 if (rc)
1005 return rc;
1006 if (rfte.i)
1007 return PGM_REGION_FIRST_TRANS;
1008 if (rfte.tt != TABLE_TYPE_REGION1)
1009 return PGM_TRANSLATION_SPEC;
1010 if (vaddr.rsx01 < rfte.tf || vaddr.rsx01 > rfte.tl)
1011 return PGM_REGION_SECOND_TRANS;
1c65781b
DH
1012 if (sg->edat_level >= 1)
1013 *dat_protection |= rfte.p;
aa17aa57
MS
1014 rc = gmap_shadow_r2t(sg, saddr, rfte.val);
1015 if (rc)
1016 return rc;
1017 ptr = rfte.rto * 4096;
1018 /* fallthrough */
1019 }
1020 case ASCE_TYPE_REGION2: {
1021 union region2_table_entry rste;
1022
1023 rc = gmap_read_table(parent, ptr + vaddr.rsx * 8, &rste.val);
1024 if (rc)
1025 return rc;
1026 if (rste.i)
1027 return PGM_REGION_SECOND_TRANS;
1028 if (rste.tt != TABLE_TYPE_REGION2)
1029 return PGM_TRANSLATION_SPEC;
1030 if (vaddr.rtx01 < rste.tf || vaddr.rtx01 > rste.tl)
1031 return PGM_REGION_THIRD_TRANS;
1c65781b
DH
1032 if (sg->edat_level >= 1)
1033 *dat_protection |= rste.p;
1034 rste.p |= *dat_protection;
aa17aa57
MS
1035 rc = gmap_shadow_r3t(sg, saddr, rste.val);
1036 if (rc)
1037 return rc;
1038 ptr = rste.rto * 4096;
1039 /* fallthrough */
1040 }
1041 case ASCE_TYPE_REGION3: {
1042 union region3_table_entry rtte;
1043
1044 rc = gmap_read_table(parent, ptr + vaddr.rtx * 8, &rtte.val);
1045 if (rc)
1046 return rc;
1047 if (rtte.i)
1048 return PGM_REGION_THIRD_TRANS;
1049 if (rtte.tt != TABLE_TYPE_REGION3)
1050 return PGM_TRANSLATION_SPEC;
18b89809
DH
1051 if (rtte.cr && asce.p && sg->edat_level >= 2)
1052 return PGM_TRANSLATION_SPEC;
1053 if (rtte.fc && sg->edat_level >= 2) {
1c65781b 1054 *dat_protection |= rtte.fc0.p;
18b89809
DH
1055 *fake = 1;
1056 ptr = rtte.fc1.rfaa << 31UL;
1057 rtte.val = ptr;
18b89809
DH
1058 goto shadow_sgt;
1059 }
aa17aa57
MS
1060 if (vaddr.sx01 < rtte.fc0.tf || vaddr.sx01 > rtte.fc0.tl)
1061 return PGM_SEGMENT_TRANSLATION;
1c65781b
DH
1062 if (sg->edat_level >= 1)
1063 *dat_protection |= rtte.fc0.p;
18b89809
DH
1064 ptr = rtte.fc0.sto << 12UL;
1065shadow_sgt:
1c65781b 1066 rtte.fc0.p |= *dat_protection;
18b89809 1067 rc = gmap_shadow_sgt(sg, saddr, rtte.val, *fake);
aa17aa57
MS
1068 if (rc)
1069 return rc;
aa17aa57
MS
1070 /* fallthrough */
1071 }
1072 case ASCE_TYPE_SEGMENT: {
1073 union segment_table_entry ste;
1074
18b89809
DH
1075 if (*fake) {
1076 /* offset in 2G guest memory block */
1077 ptr = ptr + ((unsigned long) vaddr.sx << 20UL);
1078 ste.val = ptr;
1079 goto shadow_pgt;
1080 }
aa17aa57
MS
1081 rc = gmap_read_table(parent, ptr + vaddr.sx * 8, &ste.val);
1082 if (rc)
1083 return rc;
1084 if (ste.i)
1085 return PGM_SEGMENT_TRANSLATION;
1086 if (ste.tt != TABLE_TYPE_SEGMENT)
1087 return PGM_TRANSLATION_SPEC;
1088 if (ste.cs && asce.p)
1089 return PGM_TRANSLATION_SPEC;
1c65781b 1090 *dat_protection |= ste.fc0.p;
fd8d4e3a 1091 if (ste.fc && sg->edat_level >= 1) {
fd8d4e3a
DH
1092 *fake = 1;
1093 ptr = ste.fc1.sfaa << 20UL;
1094 ste.val = ptr;
fd8d4e3a
DH
1095 goto shadow_pgt;
1096 }
1097 ptr = ste.fc0.pto << 11UL;
1098shadow_pgt:
1c65781b 1099 ste.fc0.p |= *dat_protection;
fd8d4e3a 1100 rc = gmap_shadow_pgt(sg, saddr, ste.val, *fake);
aa17aa57
MS
1101 if (rc)
1102 return rc;
aa17aa57
MS
1103 }
1104 }
1105 /* Return the parent address of the page table */
1106 *pgt = ptr;
1107 return 0;
1108}
1109
1110/**
1111 * kvm_s390_shadow_fault - handle fault on a shadow page table
f4debb40 1112 * @vcpu: virtual cpu
aa17aa57
MS
1113 * @sg: pointer to the shadow guest address space structure
1114 * @saddr: faulting address in the shadow gmap
aa17aa57
MS
1115 *
1116 * Returns: - 0 if the shadow fault was successfully resolved
1117 * - > 0 (pgm exception code) on exceptions while faulting
1118 * - -EAGAIN if the caller can retry immediately
1119 * - -EFAULT when accessing invalid guest addresses
1120 * - -ENOMEM if out of memory
1121 */
f4debb40
DH
1122int kvm_s390_shadow_fault(struct kvm_vcpu *vcpu, struct gmap *sg,
1123 unsigned long saddr)
aa17aa57
MS
1124{
1125 union vaddress vaddr;
1126 union page_table_entry pte;
1127 unsigned long pgt;
fd8d4e3a 1128 int dat_protection, fake;
aa17aa57
MS
1129 int rc;
1130
e52f8b61 1131 down_read(&sg->mm->mmap_sem);
f4debb40
DH
1132 /*
1133 * We don't want any guest-2 tables to change - so the parent
1134 * tables/pointers we read stay valid - unshadowing is however
1135 * always possible - only guest_table_lock protects us.
1136 */
1137 ipte_lock(vcpu);
e52f8b61 1138
fd8d4e3a 1139 rc = gmap_shadow_pgt_lookup(sg, saddr, &pgt, &dat_protection, &fake);
e52f8b61 1140 if (rc)
fd8d4e3a
DH
1141 rc = kvm_s390_shadow_tables(sg, saddr, &pgt, &dat_protection,
1142 &fake);
aa17aa57
MS
1143
1144 vaddr.addr = saddr;
fd8d4e3a
DH
1145 if (fake) {
1146 /* offset in 1MB guest memory block */
1147 pte.val = pgt + ((unsigned long) vaddr.px << 12UL);
1148 goto shadow_page;
1149 }
e52f8b61
DH
1150 if (!rc)
1151 rc = gmap_read_table(sg->parent, pgt + vaddr.px * 8, &pte.val);
1152 if (!rc && pte.i)
1153 rc = PGM_PAGE_TRANSLATION;
fd8d4e3a 1154 if (!rc && (pte.z || (pte.co && sg->edat_level < 1)))
e52f8b61 1155 rc = PGM_TRANSLATION_SPEC;
fd8d4e3a 1156shadow_page:
00fc062d 1157 pte.p |= dat_protection;
e52f8b61
DH
1158 if (!rc)
1159 rc = gmap_shadow_page(sg, saddr, __pte(pte.val));
f4debb40 1160 ipte_unlock(vcpu);
e52f8b61
DH
1161 up_read(&sg->mm->mmap_sem);
1162 return rc;
aa17aa57 1163}