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