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
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
ipte_lock_held(struct kvm * kvm)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
ipte_lock_simple(struct kvm * kvm)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
ipte_unlock_simple(struct kvm * kvm)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
ipte_lock_siif(struct kvm * kvm)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
ipte_unlock_siif(struct kvm * kvm)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
ipte_lock(struct kvm * kvm)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
ipte_unlock(struct kvm * kvm)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
ar_translation(struct kvm_vcpu * vcpu,union asce * asce,u8 ar,enum gacc_mode mode)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 struct trans_exc_code_bits {
470 unsigned long addr : 52; /* Translation-exception Address */
471 unsigned long fsi : 2; /* Access Exception Fetch/Store Indication */
472 unsigned long : 2;
473 unsigned long b56 : 1;
474 unsigned long : 3;
475 unsigned long b60 : 1;
476 unsigned long b61 : 1;
477 unsigned long as : 2; /* ASCE Identifier */
478 };
479
480 enum {
481 FSI_UNKNOWN = 0, /* Unknown whether fetch or store */
482 FSI_STORE = 1, /* Exception was due to store operation */
483 FSI_FETCH = 2 /* Exception was due to fetch operation */
484 };
485
486 enum prot_type {
487 PROT_TYPE_LA = 0,
488 PROT_TYPE_KEYC = 1,
489 PROT_TYPE_ALC = 2,
490 PROT_TYPE_DAT = 3,
491 PROT_TYPE_IEP = 4,
492 /* Dummy value for passing an initialized value when code != PGM_PROTECTION */
493 PROT_NONE,
494 };
495
trans_exc_ending(struct kvm_vcpu * vcpu,int code,unsigned long gva,u8 ar,enum gacc_mode mode,enum prot_type prot,bool terminate)496 static int trans_exc_ending(struct kvm_vcpu *vcpu, int code, unsigned long gva, u8 ar,
497 enum gacc_mode mode, enum prot_type prot, bool terminate)
498 {
499 struct kvm_s390_pgm_info *pgm = &vcpu->arch.pgm;
500 struct trans_exc_code_bits *tec;
501
502 memset(pgm, 0, sizeof(*pgm));
503 pgm->code = code;
504 tec = (struct trans_exc_code_bits *)&pgm->trans_exc_code;
505
506 switch (code) {
507 case PGM_PROTECTION:
508 switch (prot) {
509 case PROT_NONE:
510 /* We should never get here, acts like termination */
511 WARN_ON_ONCE(1);
512 break;
513 case PROT_TYPE_IEP:
514 tec->b61 = 1;
515 fallthrough;
516 case PROT_TYPE_LA:
517 tec->b56 = 1;
518 break;
519 case PROT_TYPE_KEYC:
520 tec->b60 = 1;
521 break;
522 case PROT_TYPE_ALC:
523 tec->b60 = 1;
524 fallthrough;
525 case PROT_TYPE_DAT:
526 tec->b61 = 1;
527 break;
528 }
529 if (terminate) {
530 tec->b56 = 0;
531 tec->b60 = 0;
532 tec->b61 = 0;
533 }
534 fallthrough;
535 case PGM_ASCE_TYPE:
536 case PGM_PAGE_TRANSLATION:
537 case PGM_REGION_FIRST_TRANS:
538 case PGM_REGION_SECOND_TRANS:
539 case PGM_REGION_THIRD_TRANS:
540 case PGM_SEGMENT_TRANSLATION:
541 /*
542 * op_access_id only applies to MOVE_PAGE -> set bit 61
543 * exc_access_id has to be set to 0 for some instructions. Both
544 * cases have to be handled by the caller.
545 */
546 tec->addr = gva >> PAGE_SHIFT;
547 tec->fsi = mode == GACC_STORE ? FSI_STORE : FSI_FETCH;
548 tec->as = psw_bits(vcpu->arch.sie_block->gpsw).as;
549 fallthrough;
550 case PGM_ALEN_TRANSLATION:
551 case PGM_ALE_SEQUENCE:
552 case PGM_ASTE_VALIDITY:
553 case PGM_ASTE_SEQUENCE:
554 case PGM_EXTENDED_AUTHORITY:
555 /*
556 * We can always store exc_access_id, as it is
557 * undefined for non-ar cases. It is undefined for
558 * most DAT protection exceptions.
559 */
560 pgm->exc_access_id = ar;
561 break;
562 }
563 return code;
564 }
565
trans_exc(struct kvm_vcpu * vcpu,int code,unsigned long gva,u8 ar,enum gacc_mode mode,enum prot_type prot)566 static int trans_exc(struct kvm_vcpu *vcpu, int code, unsigned long gva, u8 ar,
567 enum gacc_mode mode, enum prot_type prot)
568 {
569 return trans_exc_ending(vcpu, code, gva, ar, mode, prot, false);
570 }
571
get_vcpu_asce(struct kvm_vcpu * vcpu,union asce * asce,unsigned long ga,u8 ar,enum gacc_mode mode)572 static int get_vcpu_asce(struct kvm_vcpu *vcpu, union asce *asce,
573 unsigned long ga, u8 ar, enum gacc_mode mode)
574 {
575 int rc;
576 struct psw_bits psw = psw_bits(vcpu->arch.sie_block->gpsw);
577
578 if (!psw.dat) {
579 asce->val = 0;
580 asce->r = 1;
581 return 0;
582 }
583
584 if ((mode == GACC_IFETCH) && (psw.as != PSW_BITS_AS_HOME))
585 psw.as = PSW_BITS_AS_PRIMARY;
586
587 switch (psw.as) {
588 case PSW_BITS_AS_PRIMARY:
589 asce->val = vcpu->arch.sie_block->gcr[1];
590 return 0;
591 case PSW_BITS_AS_SECONDARY:
592 asce->val = vcpu->arch.sie_block->gcr[7];
593 return 0;
594 case PSW_BITS_AS_HOME:
595 asce->val = vcpu->arch.sie_block->gcr[13];
596 return 0;
597 case PSW_BITS_AS_ACCREG:
598 rc = ar_translation(vcpu, asce, ar, mode);
599 if (rc > 0)
600 return trans_exc(vcpu, rc, ga, ar, mode, PROT_TYPE_ALC);
601 return rc;
602 }
603 return 0;
604 }
605
deref_table(struct kvm * kvm,unsigned long gpa,unsigned long * val)606 static int deref_table(struct kvm *kvm, unsigned long gpa, unsigned long *val)
607 {
608 return kvm_read_guest(kvm, gpa, val, sizeof(*val));
609 }
610
611 /**
612 * guest_translate - translate a guest virtual into a guest absolute address
613 * @vcpu: virtual cpu
614 * @gva: guest virtual address
615 * @gpa: points to where guest physical (absolute) address should be stored
616 * @asce: effective asce
617 * @mode: indicates the access mode to be used
618 * @prot: returns the type for protection exceptions
619 *
620 * Translate a guest virtual address into a guest absolute address by means
621 * of dynamic address translation as specified by the architecture.
622 * If the resulting absolute address is not available in the configuration
623 * an addressing exception is indicated and @gpa will not be changed.
624 *
625 * Returns: - zero on success; @gpa contains the resulting absolute address
626 * - a negative value if guest access failed due to e.g. broken
627 * guest mapping
628 * - a positive value if an access exception happened. In this case
629 * the returned value is the program interruption code as defined
630 * by the architecture
631 */
guest_translate(struct kvm_vcpu * vcpu,unsigned long gva,unsigned long * gpa,const union asce asce,enum gacc_mode mode,enum prot_type * prot)632 static unsigned long guest_translate(struct kvm_vcpu *vcpu, unsigned long gva,
633 unsigned long *gpa, const union asce asce,
634 enum gacc_mode mode, enum prot_type *prot)
635 {
636 union vaddress vaddr = {.addr = gva};
637 union raddress raddr = {.addr = gva};
638 union page_table_entry pte;
639 int dat_protection = 0;
640 int iep_protection = 0;
641 union ctlreg0 ctlreg0;
642 unsigned long ptr;
643 int edat1, edat2, iep;
644
645 ctlreg0.val = vcpu->arch.sie_block->gcr[0];
646 edat1 = ctlreg0.edat && test_kvm_facility(vcpu->kvm, 8);
647 edat2 = edat1 && test_kvm_facility(vcpu->kvm, 78);
648 iep = ctlreg0.iep && test_kvm_facility(vcpu->kvm, 130);
649 if (asce.r)
650 goto real_address;
651 ptr = asce.origin * PAGE_SIZE;
652 switch (asce.dt) {
653 case ASCE_TYPE_REGION1:
654 if (vaddr.rfx01 > asce.tl)
655 return PGM_REGION_FIRST_TRANS;
656 ptr += vaddr.rfx * 8;
657 break;
658 case ASCE_TYPE_REGION2:
659 if (vaddr.rfx)
660 return PGM_ASCE_TYPE;
661 if (vaddr.rsx01 > asce.tl)
662 return PGM_REGION_SECOND_TRANS;
663 ptr += vaddr.rsx * 8;
664 break;
665 case ASCE_TYPE_REGION3:
666 if (vaddr.rfx || vaddr.rsx)
667 return PGM_ASCE_TYPE;
668 if (vaddr.rtx01 > asce.tl)
669 return PGM_REGION_THIRD_TRANS;
670 ptr += vaddr.rtx * 8;
671 break;
672 case ASCE_TYPE_SEGMENT:
673 if (vaddr.rfx || vaddr.rsx || vaddr.rtx)
674 return PGM_ASCE_TYPE;
675 if (vaddr.sx01 > asce.tl)
676 return PGM_SEGMENT_TRANSLATION;
677 ptr += vaddr.sx * 8;
678 break;
679 }
680 switch (asce.dt) {
681 case ASCE_TYPE_REGION1: {
682 union region1_table_entry rfte;
683
684 if (kvm_is_error_gpa(vcpu->kvm, ptr))
685 return PGM_ADDRESSING;
686 if (deref_table(vcpu->kvm, ptr, &rfte.val))
687 return -EFAULT;
688 if (rfte.i)
689 return PGM_REGION_FIRST_TRANS;
690 if (rfte.tt != TABLE_TYPE_REGION1)
691 return PGM_TRANSLATION_SPEC;
692 if (vaddr.rsx01 < rfte.tf || vaddr.rsx01 > rfte.tl)
693 return PGM_REGION_SECOND_TRANS;
694 if (edat1)
695 dat_protection |= rfte.p;
696 ptr = rfte.rto * PAGE_SIZE + vaddr.rsx * 8;
697 }
698 fallthrough;
699 case ASCE_TYPE_REGION2: {
700 union region2_table_entry rste;
701
702 if (kvm_is_error_gpa(vcpu->kvm, ptr))
703 return PGM_ADDRESSING;
704 if (deref_table(vcpu->kvm, ptr, &rste.val))
705 return -EFAULT;
706 if (rste.i)
707 return PGM_REGION_SECOND_TRANS;
708 if (rste.tt != TABLE_TYPE_REGION2)
709 return PGM_TRANSLATION_SPEC;
710 if (vaddr.rtx01 < rste.tf || vaddr.rtx01 > rste.tl)
711 return PGM_REGION_THIRD_TRANS;
712 if (edat1)
713 dat_protection |= rste.p;
714 ptr = rste.rto * PAGE_SIZE + vaddr.rtx * 8;
715 }
716 fallthrough;
717 case ASCE_TYPE_REGION3: {
718 union region3_table_entry rtte;
719
720 if (kvm_is_error_gpa(vcpu->kvm, ptr))
721 return PGM_ADDRESSING;
722 if (deref_table(vcpu->kvm, ptr, &rtte.val))
723 return -EFAULT;
724 if (rtte.i)
725 return PGM_REGION_THIRD_TRANS;
726 if (rtte.tt != TABLE_TYPE_REGION3)
727 return PGM_TRANSLATION_SPEC;
728 if (rtte.cr && asce.p && edat2)
729 return PGM_TRANSLATION_SPEC;
730 if (rtte.fc && edat2) {
731 dat_protection |= rtte.fc1.p;
732 iep_protection = rtte.fc1.iep;
733 raddr.rfaa = rtte.fc1.rfaa;
734 goto absolute_address;
735 }
736 if (vaddr.sx01 < rtte.fc0.tf)
737 return PGM_SEGMENT_TRANSLATION;
738 if (vaddr.sx01 > rtte.fc0.tl)
739 return PGM_SEGMENT_TRANSLATION;
740 if (edat1)
741 dat_protection |= rtte.fc0.p;
742 ptr = rtte.fc0.sto * PAGE_SIZE + vaddr.sx * 8;
743 }
744 fallthrough;
745 case ASCE_TYPE_SEGMENT: {
746 union segment_table_entry ste;
747
748 if (kvm_is_error_gpa(vcpu->kvm, ptr))
749 return PGM_ADDRESSING;
750 if (deref_table(vcpu->kvm, ptr, &ste.val))
751 return -EFAULT;
752 if (ste.i)
753 return PGM_SEGMENT_TRANSLATION;
754 if (ste.tt != TABLE_TYPE_SEGMENT)
755 return PGM_TRANSLATION_SPEC;
756 if (ste.cs && asce.p)
757 return PGM_TRANSLATION_SPEC;
758 if (ste.fc && edat1) {
759 dat_protection |= ste.fc1.p;
760 iep_protection = ste.fc1.iep;
761 raddr.sfaa = ste.fc1.sfaa;
762 goto absolute_address;
763 }
764 dat_protection |= ste.fc0.p;
765 ptr = ste.fc0.pto * (PAGE_SIZE / 2) + vaddr.px * 8;
766 }
767 }
768 if (kvm_is_error_gpa(vcpu->kvm, ptr))
769 return PGM_ADDRESSING;
770 if (deref_table(vcpu->kvm, ptr, &pte.val))
771 return -EFAULT;
772 if (pte.i)
773 return PGM_PAGE_TRANSLATION;
774 if (pte.z)
775 return PGM_TRANSLATION_SPEC;
776 dat_protection |= pte.p;
777 iep_protection = pte.iep;
778 raddr.pfra = pte.pfra;
779 real_address:
780 raddr.addr = kvm_s390_real_to_abs(vcpu, raddr.addr);
781 absolute_address:
782 if (mode == GACC_STORE && dat_protection) {
783 *prot = PROT_TYPE_DAT;
784 return PGM_PROTECTION;
785 }
786 if (mode == GACC_IFETCH && iep_protection && iep) {
787 *prot = PROT_TYPE_IEP;
788 return PGM_PROTECTION;
789 }
790 if (kvm_is_error_gpa(vcpu->kvm, raddr.addr))
791 return PGM_ADDRESSING;
792 *gpa = raddr.addr;
793 return 0;
794 }
795
is_low_address(unsigned long ga)796 static inline int is_low_address(unsigned long ga)
797 {
798 /* Check for address ranges 0..511 and 4096..4607 */
799 return (ga & ~0x11fful) == 0;
800 }
801
low_address_protection_enabled(struct kvm_vcpu * vcpu,const union asce asce)802 static int low_address_protection_enabled(struct kvm_vcpu *vcpu,
803 const union asce asce)
804 {
805 union ctlreg0 ctlreg0 = {.val = vcpu->arch.sie_block->gcr[0]};
806 psw_t *psw = &vcpu->arch.sie_block->gpsw;
807
808 if (!ctlreg0.lap)
809 return 0;
810 if (psw_bits(*psw).dat && asce.p)
811 return 0;
812 return 1;
813 }
814
vm_check_access_key(struct kvm * kvm,u8 access_key,enum gacc_mode mode,gpa_t gpa)815 static int vm_check_access_key(struct kvm *kvm, u8 access_key,
816 enum gacc_mode mode, gpa_t gpa)
817 {
818 u8 storage_key, access_control;
819 bool fetch_protected;
820 unsigned long hva;
821 int r;
822
823 if (access_key == 0)
824 return 0;
825
826 hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
827 if (kvm_is_error_hva(hva))
828 return PGM_ADDRESSING;
829
830 mmap_read_lock(current->mm);
831 r = get_guest_storage_key(current->mm, hva, &storage_key);
832 mmap_read_unlock(current->mm);
833 if (r)
834 return r;
835 access_control = FIELD_GET(_PAGE_ACC_BITS, storage_key);
836 if (access_control == access_key)
837 return 0;
838 fetch_protected = storage_key & _PAGE_FP_BIT;
839 if ((mode == GACC_FETCH || mode == GACC_IFETCH) && !fetch_protected)
840 return 0;
841 return PGM_PROTECTION;
842 }
843
fetch_prot_override_applicable(struct kvm_vcpu * vcpu,enum gacc_mode mode,union asce asce)844 static bool fetch_prot_override_applicable(struct kvm_vcpu *vcpu, enum gacc_mode mode,
845 union asce asce)
846 {
847 psw_t *psw = &vcpu->arch.sie_block->gpsw;
848 unsigned long override;
849
850 if (mode == GACC_FETCH || mode == GACC_IFETCH) {
851 /* check if fetch protection override enabled */
852 override = vcpu->arch.sie_block->gcr[0];
853 override &= CR0_FETCH_PROTECTION_OVERRIDE;
854 /* not applicable if subject to DAT && private space */
855 override = override && !(psw_bits(*psw).dat && asce.p);
856 return override;
857 }
858 return false;
859 }
860
fetch_prot_override_applies(unsigned long ga,unsigned int len)861 static bool fetch_prot_override_applies(unsigned long ga, unsigned int len)
862 {
863 return ga < 2048 && ga + len <= 2048;
864 }
865
storage_prot_override_applicable(struct kvm_vcpu * vcpu)866 static bool storage_prot_override_applicable(struct kvm_vcpu *vcpu)
867 {
868 /* check if storage protection override enabled */
869 return vcpu->arch.sie_block->gcr[0] & CR0_STORAGE_PROTECTION_OVERRIDE;
870 }
871
storage_prot_override_applies(u8 access_control)872 static bool storage_prot_override_applies(u8 access_control)
873 {
874 /* matches special storage protection override key (9) -> allow */
875 return access_control == PAGE_SPO_ACC;
876 }
877
vcpu_check_access_key(struct kvm_vcpu * vcpu,u8 access_key,enum gacc_mode mode,union asce asce,gpa_t gpa,unsigned long ga,unsigned int len)878 static int vcpu_check_access_key(struct kvm_vcpu *vcpu, u8 access_key,
879 enum gacc_mode mode, union asce asce, gpa_t gpa,
880 unsigned long ga, unsigned int len)
881 {
882 u8 storage_key, access_control;
883 unsigned long hva;
884 int r;
885
886 /* access key 0 matches any storage key -> allow */
887 if (access_key == 0)
888 return 0;
889 /*
890 * caller needs to ensure that gfn is accessible, so we can
891 * assume that this cannot fail
892 */
893 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gpa));
894 mmap_read_lock(current->mm);
895 r = get_guest_storage_key(current->mm, hva, &storage_key);
896 mmap_read_unlock(current->mm);
897 if (r)
898 return r;
899 access_control = FIELD_GET(_PAGE_ACC_BITS, storage_key);
900 /* access key matches storage key -> allow */
901 if (access_control == access_key)
902 return 0;
903 if (mode == GACC_FETCH || mode == GACC_IFETCH) {
904 /* it is a fetch and fetch protection is off -> allow */
905 if (!(storage_key & _PAGE_FP_BIT))
906 return 0;
907 if (fetch_prot_override_applicable(vcpu, mode, asce) &&
908 fetch_prot_override_applies(ga, len))
909 return 0;
910 }
911 if (storage_prot_override_applicable(vcpu) &&
912 storage_prot_override_applies(access_control))
913 return 0;
914 return PGM_PROTECTION;
915 }
916
917 /**
918 * guest_range_to_gpas() - Calculate guest physical addresses of page fragments
919 * covering a logical range
920 * @vcpu: virtual cpu
921 * @ga: guest address, start of range
922 * @ar: access register
923 * @gpas: output argument, may be NULL
924 * @len: length of range in bytes
925 * @asce: address-space-control element to use for translation
926 * @mode: access mode
927 * @access_key: access key to mach the range's storage keys against
928 *
929 * Translate a logical range to a series of guest absolute addresses,
930 * such that the concatenation of page fragments starting at each gpa make up
931 * the whole range.
932 * The translation is performed as if done by the cpu for the given @asce, @ar,
933 * @mode and state of the @vcpu.
934 * If the translation causes an exception, its program interruption code is
935 * returned and the &struct kvm_s390_pgm_info pgm member of @vcpu is modified
936 * such that a subsequent call to kvm_s390_inject_prog_vcpu() will inject
937 * a correct exception into the guest.
938 * The resulting gpas are stored into @gpas, unless it is NULL.
939 *
940 * Note: All fragments except the first one start at the beginning of a page.
941 * When deriving the boundaries of a fragment from a gpa, all but the last
942 * fragment end at the end of the page.
943 *
944 * Return:
945 * * 0 - success
946 * * <0 - translation could not be performed, for example if guest
947 * memory could not be accessed
948 * * >0 - an access exception occurred. In this case the returned value
949 * is the program interruption code and the contents of pgm may
950 * be used to inject an exception into the guest.
951 */
guest_range_to_gpas(struct kvm_vcpu * vcpu,unsigned long ga,u8 ar,unsigned long * gpas,unsigned long len,const union asce asce,enum gacc_mode mode,u8 access_key)952 static int guest_range_to_gpas(struct kvm_vcpu *vcpu, unsigned long ga, u8 ar,
953 unsigned long *gpas, unsigned long len,
954 const union asce asce, enum gacc_mode mode,
955 u8 access_key)
956 {
957 psw_t *psw = &vcpu->arch.sie_block->gpsw;
958 unsigned int offset = offset_in_page(ga);
959 unsigned int fragment_len;
960 int lap_enabled, rc = 0;
961 enum prot_type prot;
962 unsigned long gpa;
963
964 lap_enabled = low_address_protection_enabled(vcpu, asce);
965 while (min(PAGE_SIZE - offset, len) > 0) {
966 fragment_len = min(PAGE_SIZE - offset, len);
967 ga = kvm_s390_logical_to_effective(vcpu, ga);
968 if (mode == GACC_STORE && lap_enabled && is_low_address(ga))
969 return trans_exc(vcpu, PGM_PROTECTION, ga, ar, mode,
970 PROT_TYPE_LA);
971 if (psw_bits(*psw).dat) {
972 rc = guest_translate(vcpu, ga, &gpa, asce, mode, &prot);
973 if (rc < 0)
974 return rc;
975 } else {
976 gpa = kvm_s390_real_to_abs(vcpu, ga);
977 if (kvm_is_error_gpa(vcpu->kvm, gpa)) {
978 rc = PGM_ADDRESSING;
979 prot = PROT_NONE;
980 }
981 }
982 if (rc)
983 return trans_exc(vcpu, rc, ga, ar, mode, prot);
984 rc = vcpu_check_access_key(vcpu, access_key, mode, asce, gpa, ga,
985 fragment_len);
986 if (rc)
987 return trans_exc(vcpu, rc, ga, ar, mode, PROT_TYPE_KEYC);
988 if (gpas)
989 *gpas++ = gpa;
990 offset = 0;
991 ga += fragment_len;
992 len -= fragment_len;
993 }
994 return 0;
995 }
996
access_guest_page(struct kvm * kvm,enum gacc_mode mode,gpa_t gpa,void * data,unsigned int len)997 static int access_guest_page(struct kvm *kvm, enum gacc_mode mode, gpa_t gpa,
998 void *data, unsigned int len)
999 {
1000 const unsigned int offset = offset_in_page(gpa);
1001 const gfn_t gfn = gpa_to_gfn(gpa);
1002 int rc;
1003
1004 if (!gfn_to_memslot(kvm, gfn))
1005 return PGM_ADDRESSING;
1006 if (mode == GACC_STORE)
1007 rc = kvm_write_guest_page(kvm, gfn, data, offset, len);
1008 else
1009 rc = kvm_read_guest_page(kvm, gfn, data, offset, len);
1010 return rc;
1011 }
1012
1013 static int
access_guest_page_with_key(struct kvm * kvm,enum gacc_mode mode,gpa_t gpa,void * data,unsigned int len,u8 access_key)1014 access_guest_page_with_key(struct kvm *kvm, enum gacc_mode mode, gpa_t gpa,
1015 void *data, unsigned int len, u8 access_key)
1016 {
1017 struct kvm_memory_slot *slot;
1018 bool writable;
1019 gfn_t gfn;
1020 hva_t hva;
1021 int rc;
1022
1023 gfn = gpa >> PAGE_SHIFT;
1024 slot = gfn_to_memslot(kvm, gfn);
1025 hva = gfn_to_hva_memslot_prot(slot, gfn, &writable);
1026
1027 if (kvm_is_error_hva(hva))
1028 return PGM_ADDRESSING;
1029 /*
1030 * Check if it's a ro memslot, even tho that can't occur (they're unsupported).
1031 * Don't try to actually handle that case.
1032 */
1033 if (!writable && mode == GACC_STORE)
1034 return -EOPNOTSUPP;
1035 hva += offset_in_page(gpa);
1036 if (mode == GACC_STORE)
1037 rc = copy_to_user_key((void __user *)hva, data, len, access_key);
1038 else
1039 rc = copy_from_user_key(data, (void __user *)hva, len, access_key);
1040 if (rc)
1041 return PGM_PROTECTION;
1042 if (mode == GACC_STORE)
1043 mark_page_dirty_in_slot(kvm, slot, gfn);
1044 return 0;
1045 }
1046
access_guest_abs_with_key(struct kvm * kvm,gpa_t gpa,void * data,unsigned long len,enum gacc_mode mode,u8 access_key)1047 int access_guest_abs_with_key(struct kvm *kvm, gpa_t gpa, void *data,
1048 unsigned long len, enum gacc_mode mode, u8 access_key)
1049 {
1050 int offset = offset_in_page(gpa);
1051 int fragment_len;
1052 int rc;
1053
1054 while (min(PAGE_SIZE - offset, len) > 0) {
1055 fragment_len = min(PAGE_SIZE - offset, len);
1056 rc = access_guest_page_with_key(kvm, mode, gpa, data, fragment_len, access_key);
1057 if (rc)
1058 return rc;
1059 offset = 0;
1060 len -= fragment_len;
1061 data += fragment_len;
1062 gpa += fragment_len;
1063 }
1064 return 0;
1065 }
1066
access_guest_with_key(struct kvm_vcpu * vcpu,unsigned long ga,u8 ar,void * data,unsigned long len,enum gacc_mode mode,u8 access_key)1067 int access_guest_with_key(struct kvm_vcpu *vcpu, unsigned long ga, u8 ar,
1068 void *data, unsigned long len, enum gacc_mode mode,
1069 u8 access_key)
1070 {
1071 psw_t *psw = &vcpu->arch.sie_block->gpsw;
1072 unsigned long nr_pages, idx;
1073 unsigned long gpa_array[2];
1074 unsigned int fragment_len;
1075 unsigned long *gpas;
1076 enum prot_type prot;
1077 int need_ipte_lock;
1078 union asce asce;
1079 bool try_storage_prot_override;
1080 bool try_fetch_prot_override;
1081 int rc;
1082
1083 if (!len)
1084 return 0;
1085 ga = kvm_s390_logical_to_effective(vcpu, ga);
1086 rc = get_vcpu_asce(vcpu, &asce, ga, ar, mode);
1087 if (rc)
1088 return rc;
1089 nr_pages = (((ga & ~PAGE_MASK) + len - 1) >> PAGE_SHIFT) + 1;
1090 gpas = gpa_array;
1091 if (nr_pages > ARRAY_SIZE(gpa_array))
1092 gpas = vmalloc(array_size(nr_pages, sizeof(unsigned long)));
1093 if (!gpas)
1094 return -ENOMEM;
1095 try_fetch_prot_override = fetch_prot_override_applicable(vcpu, mode, asce);
1096 try_storage_prot_override = storage_prot_override_applicable(vcpu);
1097 need_ipte_lock = psw_bits(*psw).dat && !asce.r;
1098 if (need_ipte_lock)
1099 ipte_lock(vcpu->kvm);
1100 /*
1101 * Since we do the access further down ultimately via a move instruction
1102 * that does key checking and returns an error in case of a protection
1103 * violation, we don't need to do the check during address translation.
1104 * Skip it by passing access key 0, which matches any storage key,
1105 * obviating the need for any further checks. As a result the check is
1106 * handled entirely in hardware on access, we only need to take care to
1107 * forego key protection checking if fetch protection override applies or
1108 * retry with the special key 9 in case of storage protection override.
1109 */
1110 rc = guest_range_to_gpas(vcpu, ga, ar, gpas, len, asce, mode, 0);
1111 if (rc)
1112 goto out_unlock;
1113 for (idx = 0; idx < nr_pages; idx++) {
1114 fragment_len = min(PAGE_SIZE - offset_in_page(gpas[idx]), len);
1115 if (try_fetch_prot_override && fetch_prot_override_applies(ga, fragment_len)) {
1116 rc = access_guest_page(vcpu->kvm, mode, gpas[idx],
1117 data, fragment_len);
1118 } else {
1119 rc = access_guest_page_with_key(vcpu->kvm, mode, gpas[idx],
1120 data, fragment_len, access_key);
1121 }
1122 if (rc == PGM_PROTECTION && try_storage_prot_override)
1123 rc = access_guest_page_with_key(vcpu->kvm, mode, gpas[idx],
1124 data, fragment_len, PAGE_SPO_ACC);
1125 if (rc)
1126 break;
1127 len -= fragment_len;
1128 data += fragment_len;
1129 ga = kvm_s390_logical_to_effective(vcpu, ga + fragment_len);
1130 }
1131 if (rc > 0) {
1132 bool terminate = (mode == GACC_STORE) && (idx > 0);
1133
1134 if (rc == PGM_PROTECTION)
1135 prot = PROT_TYPE_KEYC;
1136 else
1137 prot = PROT_NONE;
1138 rc = trans_exc_ending(vcpu, rc, ga, ar, mode, prot, terminate);
1139 }
1140 out_unlock:
1141 if (need_ipte_lock)
1142 ipte_unlock(vcpu->kvm);
1143 if (nr_pages > ARRAY_SIZE(gpa_array))
1144 vfree(gpas);
1145 return rc;
1146 }
1147
access_guest_real(struct kvm_vcpu * vcpu,unsigned long gra,void * data,unsigned long len,enum gacc_mode mode)1148 int access_guest_real(struct kvm_vcpu *vcpu, unsigned long gra,
1149 void *data, unsigned long len, enum gacc_mode mode)
1150 {
1151 unsigned int fragment_len;
1152 unsigned long gpa;
1153 int rc = 0;
1154
1155 while (len && !rc) {
1156 gpa = kvm_s390_real_to_abs(vcpu, gra);
1157 fragment_len = min(PAGE_SIZE - offset_in_page(gpa), len);
1158 rc = access_guest_page(vcpu->kvm, mode, gpa, data, fragment_len);
1159 len -= fragment_len;
1160 gra += fragment_len;
1161 data += fragment_len;
1162 }
1163 if (rc > 0)
1164 vcpu->arch.pgm.code = rc;
1165 return rc;
1166 }
1167
1168 /**
1169 * cmpxchg_guest_abs_with_key() - Perform cmpxchg on guest absolute address.
1170 * @kvm: Virtual machine instance.
1171 * @gpa: Absolute guest address of the location to be changed.
1172 * @len: Operand length of the cmpxchg, required: 1 <= len <= 16. Providing a
1173 * non power of two will result in failure.
1174 * @old_addr: Pointer to old value. If the location at @gpa contains this value,
1175 * the exchange will succeed. After calling cmpxchg_guest_abs_with_key()
1176 * *@old_addr contains the value at @gpa before the attempt to
1177 * exchange the value.
1178 * @new: The value to place at @gpa.
1179 * @access_key: The access key to use for the guest access.
1180 * @success: output value indicating if an exchange occurred.
1181 *
1182 * Atomically exchange the value at @gpa by @new, if it contains *@old.
1183 * Honors storage keys.
1184 *
1185 * Return: * 0: successful exchange
1186 * * >0: a program interruption code indicating the reason cmpxchg could
1187 * not be attempted
1188 * * -EINVAL: address misaligned or len not power of two
1189 * * -EAGAIN: transient failure (len 1 or 2)
1190 * * -EOPNOTSUPP: read-only memslot (should never occur)
1191 */
cmpxchg_guest_abs_with_key(struct kvm * kvm,gpa_t gpa,int len,__uint128_t * old_addr,__uint128_t new,u8 access_key,bool * success)1192 int cmpxchg_guest_abs_with_key(struct kvm *kvm, gpa_t gpa, int len,
1193 __uint128_t *old_addr, __uint128_t new,
1194 u8 access_key, bool *success)
1195 {
1196 gfn_t gfn = gpa_to_gfn(gpa);
1197 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1198 bool writable;
1199 hva_t hva;
1200 int ret;
1201
1202 if (!IS_ALIGNED(gpa, len))
1203 return -EINVAL;
1204
1205 hva = gfn_to_hva_memslot_prot(slot, gfn, &writable);
1206 if (kvm_is_error_hva(hva))
1207 return PGM_ADDRESSING;
1208 /*
1209 * Check if it's a read-only memslot, even though that cannot occur
1210 * since those are unsupported.
1211 * Don't try to actually handle that case.
1212 */
1213 if (!writable)
1214 return -EOPNOTSUPP;
1215
1216 hva += offset_in_page(gpa);
1217 /*
1218 * The cmpxchg_user_key macro depends on the type of "old", so we need
1219 * a case for each valid length and get some code duplication as long
1220 * as we don't introduce a new macro.
1221 */
1222 switch (len) {
1223 case 1: {
1224 u8 old;
1225
1226 ret = cmpxchg_user_key((u8 __user *)hva, &old, *old_addr, new, access_key);
1227 *success = !ret && old == *old_addr;
1228 *old_addr = old;
1229 break;
1230 }
1231 case 2: {
1232 u16 old;
1233
1234 ret = cmpxchg_user_key((u16 __user *)hva, &old, *old_addr, new, access_key);
1235 *success = !ret && old == *old_addr;
1236 *old_addr = old;
1237 break;
1238 }
1239 case 4: {
1240 u32 old;
1241
1242 ret = cmpxchg_user_key((u32 __user *)hva, &old, *old_addr, new, access_key);
1243 *success = !ret && old == *old_addr;
1244 *old_addr = old;
1245 break;
1246 }
1247 case 8: {
1248 u64 old;
1249
1250 ret = cmpxchg_user_key((u64 __user *)hva, &old, *old_addr, new, access_key);
1251 *success = !ret && old == *old_addr;
1252 *old_addr = old;
1253 break;
1254 }
1255 case 16: {
1256 __uint128_t old;
1257
1258 ret = cmpxchg_user_key((__uint128_t __user *)hva, &old, *old_addr, new, access_key);
1259 *success = !ret && old == *old_addr;
1260 *old_addr = old;
1261 break;
1262 }
1263 default:
1264 return -EINVAL;
1265 }
1266 if (*success)
1267 mark_page_dirty_in_slot(kvm, slot, gfn);
1268 /*
1269 * Assume that the fault is caused by protection, either key protection
1270 * or user page write protection.
1271 */
1272 if (ret == -EFAULT)
1273 ret = PGM_PROTECTION;
1274 return ret;
1275 }
1276
1277 /**
1278 * guest_translate_address_with_key - translate guest logical into guest absolute address
1279 * @vcpu: virtual cpu
1280 * @gva: Guest virtual address
1281 * @ar: Access register
1282 * @gpa: Guest physical address
1283 * @mode: Translation access mode
1284 * @access_key: access key to mach the storage key with
1285 *
1286 * Parameter semantics are the same as the ones from guest_translate.
1287 * The memory contents at the guest address are not changed.
1288 *
1289 * Note: The IPTE lock is not taken during this function, so the caller
1290 * has to take care of this.
1291 */
guest_translate_address_with_key(struct kvm_vcpu * vcpu,unsigned long gva,u8 ar,unsigned long * gpa,enum gacc_mode mode,u8 access_key)1292 int guest_translate_address_with_key(struct kvm_vcpu *vcpu, unsigned long gva, u8 ar,
1293 unsigned long *gpa, enum gacc_mode mode,
1294 u8 access_key)
1295 {
1296 union asce asce;
1297 int rc;
1298
1299 gva = kvm_s390_logical_to_effective(vcpu, gva);
1300 rc = get_vcpu_asce(vcpu, &asce, gva, ar, mode);
1301 if (rc)
1302 return rc;
1303 return guest_range_to_gpas(vcpu, gva, ar, gpa, 1, asce, mode,
1304 access_key);
1305 }
1306
1307 /**
1308 * check_gva_range - test a range of guest virtual addresses for accessibility
1309 * @vcpu: virtual cpu
1310 * @gva: Guest virtual address
1311 * @ar: Access register
1312 * @length: Length of test range
1313 * @mode: Translation access mode
1314 * @access_key: access key to mach the storage keys with
1315 */
check_gva_range(struct kvm_vcpu * vcpu,unsigned long gva,u8 ar,unsigned long length,enum gacc_mode mode,u8 access_key)1316 int check_gva_range(struct kvm_vcpu *vcpu, unsigned long gva, u8 ar,
1317 unsigned long length, enum gacc_mode mode, u8 access_key)
1318 {
1319 union asce asce;
1320 int rc = 0;
1321
1322 rc = get_vcpu_asce(vcpu, &asce, gva, ar, mode);
1323 if (rc)
1324 return rc;
1325 ipte_lock(vcpu->kvm);
1326 rc = guest_range_to_gpas(vcpu, gva, ar, NULL, length, asce, mode,
1327 access_key);
1328 ipte_unlock(vcpu->kvm);
1329
1330 return rc;
1331 }
1332
1333 /**
1334 * check_gpa_range - test a range of guest physical addresses for accessibility
1335 * @kvm: virtual machine instance
1336 * @gpa: guest physical address
1337 * @length: length of test range
1338 * @mode: access mode to test, relevant for storage keys
1339 * @access_key: access key to mach the storage keys with
1340 */
check_gpa_range(struct kvm * kvm,unsigned long gpa,unsigned long length,enum gacc_mode mode,u8 access_key)1341 int check_gpa_range(struct kvm *kvm, unsigned long gpa, unsigned long length,
1342 enum gacc_mode mode, u8 access_key)
1343 {
1344 unsigned int fragment_len;
1345 int rc = 0;
1346
1347 while (length && !rc) {
1348 fragment_len = min(PAGE_SIZE - offset_in_page(gpa), length);
1349 rc = vm_check_access_key(kvm, access_key, mode, gpa);
1350 length -= fragment_len;
1351 gpa += fragment_len;
1352 }
1353 return rc;
1354 }
1355
1356 /**
1357 * kvm_s390_check_low_addr_prot_real - check for low-address protection
1358 * @vcpu: virtual cpu
1359 * @gra: Guest real address
1360 *
1361 * Checks whether an address is subject to low-address protection and set
1362 * up vcpu->arch.pgm accordingly if necessary.
1363 *
1364 * Return: 0 if no protection exception, or PGM_PROTECTION if protected.
1365 */
kvm_s390_check_low_addr_prot_real(struct kvm_vcpu * vcpu,unsigned long gra)1366 int kvm_s390_check_low_addr_prot_real(struct kvm_vcpu *vcpu, unsigned long gra)
1367 {
1368 union ctlreg0 ctlreg0 = {.val = vcpu->arch.sie_block->gcr[0]};
1369
1370 if (!ctlreg0.lap || !is_low_address(gra))
1371 return 0;
1372 return trans_exc(vcpu, PGM_PROTECTION, gra, 0, GACC_STORE, PROT_TYPE_LA);
1373 }
1374
1375 /**
1376 * kvm_s390_shadow_tables - walk the guest page table and create shadow tables
1377 * @sg: pointer to the shadow guest address space structure
1378 * @saddr: faulting address in the shadow gmap
1379 * @pgt: pointer to the beginning of the page table for the given address if
1380 * successful (return value 0), or to the first invalid DAT entry in
1381 * case of exceptions (return value > 0)
1382 * @dat_protection: referenced memory is write protected
1383 * @fake: pgt references contiguous guest memory block, not a pgtable
1384 */
kvm_s390_shadow_tables(struct gmap * sg,unsigned long saddr,unsigned long * pgt,int * dat_protection,int * fake)1385 static int kvm_s390_shadow_tables(struct gmap *sg, unsigned long saddr,
1386 unsigned long *pgt, int *dat_protection,
1387 int *fake)
1388 {
1389 struct kvm *kvm;
1390 struct gmap *parent;
1391 union asce asce;
1392 union vaddress vaddr;
1393 unsigned long ptr;
1394 int rc;
1395
1396 *fake = 0;
1397 *dat_protection = 0;
1398 kvm = sg->private;
1399 parent = sg->parent;
1400 vaddr.addr = saddr;
1401 asce.val = sg->orig_asce;
1402 ptr = asce.origin * PAGE_SIZE;
1403 if (asce.r) {
1404 *fake = 1;
1405 ptr = 0;
1406 asce.dt = ASCE_TYPE_REGION1;
1407 }
1408 switch (asce.dt) {
1409 case ASCE_TYPE_REGION1:
1410 if (vaddr.rfx01 > asce.tl && !*fake)
1411 return PGM_REGION_FIRST_TRANS;
1412 break;
1413 case ASCE_TYPE_REGION2:
1414 if (vaddr.rfx)
1415 return PGM_ASCE_TYPE;
1416 if (vaddr.rsx01 > asce.tl)
1417 return PGM_REGION_SECOND_TRANS;
1418 break;
1419 case ASCE_TYPE_REGION3:
1420 if (vaddr.rfx || vaddr.rsx)
1421 return PGM_ASCE_TYPE;
1422 if (vaddr.rtx01 > asce.tl)
1423 return PGM_REGION_THIRD_TRANS;
1424 break;
1425 case ASCE_TYPE_SEGMENT:
1426 if (vaddr.rfx || vaddr.rsx || vaddr.rtx)
1427 return PGM_ASCE_TYPE;
1428 if (vaddr.sx01 > asce.tl)
1429 return PGM_SEGMENT_TRANSLATION;
1430 break;
1431 }
1432
1433 switch (asce.dt) {
1434 case ASCE_TYPE_REGION1: {
1435 union region1_table_entry rfte;
1436
1437 if (*fake) {
1438 ptr += vaddr.rfx * _REGION1_SIZE;
1439 rfte.val = ptr;
1440 goto shadow_r2t;
1441 }
1442 *pgt = ptr + vaddr.rfx * 8;
1443 rc = gmap_read_table(parent, ptr + vaddr.rfx * 8, &rfte.val);
1444 if (rc)
1445 return rc;
1446 if (rfte.i)
1447 return PGM_REGION_FIRST_TRANS;
1448 if (rfte.tt != TABLE_TYPE_REGION1)
1449 return PGM_TRANSLATION_SPEC;
1450 if (vaddr.rsx01 < rfte.tf || vaddr.rsx01 > rfte.tl)
1451 return PGM_REGION_SECOND_TRANS;
1452 if (sg->edat_level >= 1)
1453 *dat_protection |= rfte.p;
1454 ptr = rfte.rto * PAGE_SIZE;
1455 shadow_r2t:
1456 rc = gmap_shadow_r2t(sg, saddr, rfte.val, *fake);
1457 if (rc)
1458 return rc;
1459 kvm->stat.gmap_shadow_r1_entry++;
1460 }
1461 fallthrough;
1462 case ASCE_TYPE_REGION2: {
1463 union region2_table_entry rste;
1464
1465 if (*fake) {
1466 ptr += vaddr.rsx * _REGION2_SIZE;
1467 rste.val = ptr;
1468 goto shadow_r3t;
1469 }
1470 *pgt = ptr + vaddr.rsx * 8;
1471 rc = gmap_read_table(parent, ptr + vaddr.rsx * 8, &rste.val);
1472 if (rc)
1473 return rc;
1474 if (rste.i)
1475 return PGM_REGION_SECOND_TRANS;
1476 if (rste.tt != TABLE_TYPE_REGION2)
1477 return PGM_TRANSLATION_SPEC;
1478 if (vaddr.rtx01 < rste.tf || vaddr.rtx01 > rste.tl)
1479 return PGM_REGION_THIRD_TRANS;
1480 if (sg->edat_level >= 1)
1481 *dat_protection |= rste.p;
1482 ptr = rste.rto * PAGE_SIZE;
1483 shadow_r3t:
1484 rste.p |= *dat_protection;
1485 rc = gmap_shadow_r3t(sg, saddr, rste.val, *fake);
1486 if (rc)
1487 return rc;
1488 kvm->stat.gmap_shadow_r2_entry++;
1489 }
1490 fallthrough;
1491 case ASCE_TYPE_REGION3: {
1492 union region3_table_entry rtte;
1493
1494 if (*fake) {
1495 ptr += vaddr.rtx * _REGION3_SIZE;
1496 rtte.val = ptr;
1497 goto shadow_sgt;
1498 }
1499 *pgt = ptr + vaddr.rtx * 8;
1500 rc = gmap_read_table(parent, ptr + vaddr.rtx * 8, &rtte.val);
1501 if (rc)
1502 return rc;
1503 if (rtte.i)
1504 return PGM_REGION_THIRD_TRANS;
1505 if (rtte.tt != TABLE_TYPE_REGION3)
1506 return PGM_TRANSLATION_SPEC;
1507 if (rtte.cr && asce.p && sg->edat_level >= 2)
1508 return PGM_TRANSLATION_SPEC;
1509 if (rtte.fc && sg->edat_level >= 2) {
1510 *dat_protection |= rtte.fc0.p;
1511 *fake = 1;
1512 ptr = rtte.fc1.rfaa * _REGION3_SIZE;
1513 rtte.val = ptr;
1514 goto shadow_sgt;
1515 }
1516 if (vaddr.sx01 < rtte.fc0.tf || vaddr.sx01 > rtte.fc0.tl)
1517 return PGM_SEGMENT_TRANSLATION;
1518 if (sg->edat_level >= 1)
1519 *dat_protection |= rtte.fc0.p;
1520 ptr = rtte.fc0.sto * PAGE_SIZE;
1521 shadow_sgt:
1522 rtte.fc0.p |= *dat_protection;
1523 rc = gmap_shadow_sgt(sg, saddr, rtte.val, *fake);
1524 if (rc)
1525 return rc;
1526 kvm->stat.gmap_shadow_r3_entry++;
1527 }
1528 fallthrough;
1529 case ASCE_TYPE_SEGMENT: {
1530 union segment_table_entry ste;
1531
1532 if (*fake) {
1533 ptr += vaddr.sx * _SEGMENT_SIZE;
1534 ste.val = ptr;
1535 goto shadow_pgt;
1536 }
1537 *pgt = ptr + vaddr.sx * 8;
1538 rc = gmap_read_table(parent, ptr + vaddr.sx * 8, &ste.val);
1539 if (rc)
1540 return rc;
1541 if (ste.i)
1542 return PGM_SEGMENT_TRANSLATION;
1543 if (ste.tt != TABLE_TYPE_SEGMENT)
1544 return PGM_TRANSLATION_SPEC;
1545 if (ste.cs && asce.p)
1546 return PGM_TRANSLATION_SPEC;
1547 *dat_protection |= ste.fc0.p;
1548 if (ste.fc && sg->edat_level >= 1) {
1549 *fake = 1;
1550 ptr = ste.fc1.sfaa * _SEGMENT_SIZE;
1551 ste.val = ptr;
1552 goto shadow_pgt;
1553 }
1554 ptr = ste.fc0.pto * (PAGE_SIZE / 2);
1555 shadow_pgt:
1556 ste.fc0.p |= *dat_protection;
1557 rc = gmap_shadow_pgt(sg, saddr, ste.val, *fake);
1558 if (rc)
1559 return rc;
1560 kvm->stat.gmap_shadow_sg_entry++;
1561 }
1562 }
1563 /* Return the parent address of the page table */
1564 *pgt = ptr;
1565 return 0;
1566 }
1567
1568 /**
1569 * kvm_s390_shadow_fault - handle fault on a shadow page table
1570 * @vcpu: virtual cpu
1571 * @sg: pointer to the shadow guest address space structure
1572 * @saddr: faulting address in the shadow gmap
1573 * @datptr: will contain the address of the faulting DAT table entry, or of
1574 * the valid leaf, plus some flags
1575 *
1576 * Returns: - 0 if the shadow fault was successfully resolved
1577 * - > 0 (pgm exception code) on exceptions while faulting
1578 * - -EAGAIN if the caller can retry immediately
1579 * - -EFAULT when accessing invalid guest addresses
1580 * - -ENOMEM if out of memory
1581 */
kvm_s390_shadow_fault(struct kvm_vcpu * vcpu,struct gmap * sg,unsigned long saddr,unsigned long * datptr)1582 int kvm_s390_shadow_fault(struct kvm_vcpu *vcpu, struct gmap *sg,
1583 unsigned long saddr, unsigned long *datptr)
1584 {
1585 union vaddress vaddr;
1586 union page_table_entry pte;
1587 unsigned long pgt = 0;
1588 int dat_protection, fake;
1589 int rc;
1590
1591 mmap_read_lock(sg->mm);
1592 /*
1593 * We don't want any guest-2 tables to change - so the parent
1594 * tables/pointers we read stay valid - unshadowing is however
1595 * always possible - only guest_table_lock protects us.
1596 */
1597 ipte_lock(vcpu->kvm);
1598
1599 rc = gmap_shadow_pgt_lookup(sg, saddr, &pgt, &dat_protection, &fake);
1600 if (rc)
1601 rc = kvm_s390_shadow_tables(sg, saddr, &pgt, &dat_protection,
1602 &fake);
1603
1604 vaddr.addr = saddr;
1605 if (fake) {
1606 pte.val = pgt + vaddr.px * PAGE_SIZE;
1607 goto shadow_page;
1608 }
1609
1610 switch (rc) {
1611 case PGM_SEGMENT_TRANSLATION:
1612 case PGM_REGION_THIRD_TRANS:
1613 case PGM_REGION_SECOND_TRANS:
1614 case PGM_REGION_FIRST_TRANS:
1615 pgt |= PEI_NOT_PTE;
1616 break;
1617 case 0:
1618 pgt += vaddr.px * 8;
1619 rc = gmap_read_table(sg->parent, pgt, &pte.val);
1620 }
1621 if (datptr)
1622 *datptr = pgt | dat_protection * PEI_DAT_PROT;
1623 if (!rc && pte.i)
1624 rc = PGM_PAGE_TRANSLATION;
1625 if (!rc && pte.z)
1626 rc = PGM_TRANSLATION_SPEC;
1627 shadow_page:
1628 pte.p |= dat_protection;
1629 if (!rc)
1630 rc = gmap_shadow_page(sg, saddr, __pte(pte.val));
1631 vcpu->kvm->stat.gmap_shadow_pg_entry++;
1632 ipte_unlock(vcpu->kvm);
1633 mmap_read_unlock(sg->mm);
1634 return rc;
1635 }
1636