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