1 // SPDX-License-Identifier: GPL-2.0
3 * guest access functions
5 * Copyright IBM Corp. 2014
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>
17 #include <asm/dat-bits.h>
22 * vaddress union in order to easily decode a virtual address into its
23 * region first index, region second index etc. parts.
28 unsigned long rfx : 11;
29 unsigned long rsx : 11;
30 unsigned long rtx : 11;
31 unsigned long sx : 11;
33 unsigned long bx : 12;
36 unsigned long rfx01 : 2;
38 unsigned long rsx01 : 2;
40 unsigned long rtx01 : 2;
42 unsigned long sx01 : 2;
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.
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 */
79 unsigned long i : 1; /* ALEN-Invalid Bit */
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 */
87 unsigned long asteo : 25; /* ASN-Second-Table-Entry Origin */
89 unsigned long astesn : 32; /* ASTE Sequence Number */
93 unsigned long i : 1; /* ASX-Invalid Bit */
94 unsigned long ato : 29; /* Authority-Table Origin */
96 unsigned long b : 1; /* Base-Space Bit */
97 unsigned long ax : 16; /* Authorization Index */
98 unsigned long atl : 12; /* Authority-Table Length */
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 */
108 int ipte_lock_held(struct kvm *kvm)
113 read_lock(&kvm->arch.sca_lock);
114 rc = kvm_s390_get_ipte_control(kvm)->kh != 0;
115 read_unlock(&kvm->arch.sca_lock);
118 return kvm->arch.ipte_lock_count != 0;
121 static void ipte_lock_simple(struct kvm *kvm)
123 union ipte_control old, new, *ic;
125 mutex_lock(&kvm->arch.ipte_mutex);
126 kvm->arch.ipte_lock_count++;
127 if (kvm->arch.ipte_lock_count > 1)
130 read_lock(&kvm->arch.sca_lock);
131 ic = kvm_s390_get_ipte_control(kvm);
132 old = READ_ONCE(*ic);
135 read_unlock(&kvm->arch.sca_lock);
141 } while (!try_cmpxchg(&ic->val, &old.val, new.val));
142 read_unlock(&kvm->arch.sca_lock);
144 mutex_unlock(&kvm->arch.ipte_mutex);
147 static void ipte_unlock_simple(struct kvm *kvm)
149 union ipte_control old, new, *ic;
151 mutex_lock(&kvm->arch.ipte_mutex);
152 kvm->arch.ipte_lock_count--;
153 if (kvm->arch.ipte_lock_count)
155 read_lock(&kvm->arch.sca_lock);
156 ic = kvm_s390_get_ipte_control(kvm);
157 old = READ_ONCE(*ic);
161 } while (!try_cmpxchg(&ic->val, &old.val, new.val));
162 read_unlock(&kvm->arch.sca_lock);
163 wake_up(&kvm->arch.ipte_wq);
165 mutex_unlock(&kvm->arch.ipte_mutex);
168 static void ipte_lock_siif(struct kvm *kvm)
170 union ipte_control old, new, *ic;
173 read_lock(&kvm->arch.sca_lock);
174 ic = kvm_s390_get_ipte_control(kvm);
175 old = READ_ONCE(*ic);
178 read_unlock(&kvm->arch.sca_lock);
185 } while (!try_cmpxchg(&ic->val, &old.val, new.val));
186 read_unlock(&kvm->arch.sca_lock);
189 static void ipte_unlock_siif(struct kvm *kvm)
191 union ipte_control old, new, *ic;
193 read_lock(&kvm->arch.sca_lock);
194 ic = kvm_s390_get_ipte_control(kvm);
195 old = READ_ONCE(*ic);
201 } while (!try_cmpxchg(&ic->val, &old.val, new.val));
202 read_unlock(&kvm->arch.sca_lock);
204 wake_up(&kvm->arch.ipte_wq);
207 void ipte_lock(struct kvm *kvm)
212 ipte_lock_simple(kvm);
215 void ipte_unlock(struct kvm *kvm)
218 ipte_unlock_siif(kvm);
220 ipte_unlock_simple(kvm);
223 static int ar_translation(struct kvm_vcpu *vcpu, union asce *asce, u8 ar,
229 unsigned long ald_addr, authority_table_addr;
237 if (vcpu->arch.acrs_loaded)
238 save_access_regs(vcpu->run->s.regs.acrs);
239 alet.val = vcpu->run->s.regs.acrs[ar];
241 if (ar == 0 || alet.val == 0) {
242 asce->val = vcpu->arch.sie_block->gcr[1];
244 } else if (alet.val == 1) {
245 asce->val = vcpu->arch.sie_block->gcr[7];
250 return PGM_ALET_SPECIFICATION;
253 ald_addr = vcpu->arch.sie_block->gcr[5];
255 ald_addr = vcpu->arch.sie_block->gcr[2];
256 ald_addr &= 0x7fffffc0;
258 rc = read_guest_real(vcpu, ald_addr + 16, &ald.val, sizeof(union ald));
262 if (alet.alen / 8 > ald.all)
263 return PGM_ALEN_TRANSLATION;
265 if (0x7fffffff - ald.alo * 128 < alet.alen * 16)
266 return PGM_ADDRESSING;
268 rc = read_guest_real(vcpu, ald.alo * 128 + alet.alen * 16, &ale,
274 return PGM_ALEN_TRANSLATION;
275 if (ale.alesn != alet.alesn)
276 return PGM_ALE_SEQUENCE;
278 rc = read_guest_real(vcpu, ale.asteo * 64, &aste, sizeof(struct aste));
283 return PGM_ASTE_VALIDITY;
284 if (aste.astesn != ale.astesn)
285 return PGM_ASTE_SEQUENCE;
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;
293 authority_table_addr = aste.ato * 4 + eax / 4;
295 rc = read_guest_real(vcpu, authority_table_addr,
301 if ((authority_table & (0x40 >> ((eax & 3) * 2))) == 0)
302 return PGM_EXTENDED_AUTHORITY;
306 if (ale.fo == 1 && mode == GACC_STORE)
307 return PGM_PROTECTION;
309 asce->val = aste.asce;
319 /* Dummy value for passing an initialized value when code != PGM_PROTECTION */
323 static 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)
326 struct kvm_s390_pgm_info *pgm = &vcpu->arch.pgm;
329 memset(pgm, 0, sizeof(*pgm));
331 teid = (union teid *)&pgm->trans_exc_code;
337 /* We should never get here, acts like termination */
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:
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.
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;
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:
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.
387 pgm->exc_access_id = ar;
393 static int trans_exc(struct kvm_vcpu *vcpu, int code, unsigned long gva, u8 ar,
394 enum gacc_mode mode, enum prot_type prot)
396 return trans_exc_ending(vcpu, code, gva, ar, mode, prot, false);
399 static int get_vcpu_asce(struct kvm_vcpu *vcpu, union asce *asce,
400 unsigned long ga, u8 ar, enum gacc_mode mode)
403 struct psw_bits psw = psw_bits(vcpu->arch.sie_block->gpsw);
411 if ((mode == GACC_IFETCH) && (psw.as != PSW_BITS_AS_HOME))
412 psw.as = PSW_BITS_AS_PRIMARY;
415 case PSW_BITS_AS_PRIMARY:
416 asce->val = vcpu->arch.sie_block->gcr[1];
418 case PSW_BITS_AS_SECONDARY:
419 asce->val = vcpu->arch.sie_block->gcr[7];
421 case PSW_BITS_AS_HOME:
422 asce->val = vcpu->arch.sie_block->gcr[13];
424 case PSW_BITS_AS_ACCREG:
425 rc = ar_translation(vcpu, asce, ar, mode);
427 return trans_exc(vcpu, rc, ga, ar, mode, PROT_TYPE_ALC);
433 static int deref_table(struct kvm *kvm, unsigned long gpa, unsigned long *val)
435 return kvm_read_guest(kvm, gpa, val, sizeof(*val));
439 * guest_translate - translate a guest virtual into a guest absolute address
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
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.
452 * Returns: - zero on success; @gpa contains the resulting absolute address
453 * - a negative value if guest access failed due to e.g. broken
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
459 static 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)
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;
470 int edat1, edat2, iep;
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);
478 ptr = asce.rsto * PAGE_SIZE;
480 case ASCE_TYPE_REGION1:
481 if (vaddr.rfx01 > asce.tl)
482 return PGM_REGION_FIRST_TRANS;
483 ptr += vaddr.rfx * 8;
485 case ASCE_TYPE_REGION2:
487 return PGM_ASCE_TYPE;
488 if (vaddr.rsx01 > asce.tl)
489 return PGM_REGION_SECOND_TRANS;
490 ptr += vaddr.rsx * 8;
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;
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;
508 case ASCE_TYPE_REGION1: {
509 union region1_table_entry rfte;
511 if (!kvm_is_gpa_in_memslot(vcpu->kvm, ptr))
512 return PGM_ADDRESSING;
513 if (deref_table(vcpu->kvm, ptr, &rfte.val))
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;
522 dat_protection |= rfte.p;
523 ptr = rfte.rto * PAGE_SIZE + vaddr.rsx * 8;
526 case ASCE_TYPE_REGION2: {
527 union region2_table_entry rste;
529 if (!kvm_is_gpa_in_memslot(vcpu->kvm, ptr))
530 return PGM_ADDRESSING;
531 if (deref_table(vcpu->kvm, ptr, &rste.val))
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;
540 dat_protection |= rste.p;
541 ptr = rste.rto * PAGE_SIZE + vaddr.rtx * 8;
544 case ASCE_TYPE_REGION3: {
545 union region3_table_entry rtte;
547 if (!kvm_is_gpa_in_memslot(vcpu->kvm, ptr))
548 return PGM_ADDRESSING;
549 if (deref_table(vcpu->kvm, ptr, &rtte.val))
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;
563 if (vaddr.sx01 < rtte.fc0.tf)
564 return PGM_SEGMENT_TRANSLATION;
565 if (vaddr.sx01 > rtte.fc0.tl)
566 return PGM_SEGMENT_TRANSLATION;
568 dat_protection |= rtte.fc0.p;
569 ptr = rtte.fc0.sto * PAGE_SIZE + vaddr.sx * 8;
572 case ASCE_TYPE_SEGMENT: {
573 union segment_table_entry ste;
575 if (!kvm_is_gpa_in_memslot(vcpu->kvm, ptr))
576 return PGM_ADDRESSING;
577 if (deref_table(vcpu->kvm, ptr, &ste.val))
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;
591 dat_protection |= ste.fc0.p;
592 ptr = ste.fc0.pto * (PAGE_SIZE / 2) + vaddr.px * 8;
595 if (!kvm_is_gpa_in_memslot(vcpu->kvm, ptr))
596 return PGM_ADDRESSING;
597 if (deref_table(vcpu->kvm, ptr, &pte.val))
600 return PGM_PAGE_TRANSLATION;
602 return PGM_TRANSLATION_SPEC;
603 dat_protection |= pte.p;
604 iep_protection = pte.iep;
605 raddr.pfra = pte.pfra;
607 raddr.addr = kvm_s390_real_to_abs(vcpu, raddr.addr);
609 if (mode == GACC_STORE && dat_protection) {
610 *prot = PROT_TYPE_DAT;
611 return PGM_PROTECTION;
613 if (mode == GACC_IFETCH && iep_protection && iep) {
614 *prot = PROT_TYPE_IEP;
615 return PGM_PROTECTION;
617 if (!kvm_is_gpa_in_memslot(vcpu->kvm, raddr.addr))
618 return PGM_ADDRESSING;
623 static inline int is_low_address(unsigned long ga)
625 /* Check for address ranges 0..511 and 4096..4607 */
626 return (ga & ~0x11fful) == 0;
629 static int low_address_protection_enabled(struct kvm_vcpu *vcpu,
630 const union asce asce)
632 union ctlreg0 ctlreg0 = {.val = vcpu->arch.sie_block->gcr[0]};
633 psw_t *psw = &vcpu->arch.sie_block->gpsw;
637 if (psw_bits(*psw).dat && asce.p)
642 static int vm_check_access_key(struct kvm *kvm, u8 access_key,
643 enum gacc_mode mode, gpa_t gpa)
645 u8 storage_key, access_control;
646 bool fetch_protected;
653 hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
654 if (kvm_is_error_hva(hva))
655 return PGM_ADDRESSING;
657 mmap_read_lock(current->mm);
658 r = get_guest_storage_key(current->mm, hva, &storage_key);
659 mmap_read_unlock(current->mm);
662 access_control = FIELD_GET(_PAGE_ACC_BITS, storage_key);
663 if (access_control == access_key)
665 fetch_protected = storage_key & _PAGE_FP_BIT;
666 if ((mode == GACC_FETCH || mode == GACC_IFETCH) && !fetch_protected)
668 return PGM_PROTECTION;
671 static bool fetch_prot_override_applicable(struct kvm_vcpu *vcpu, enum gacc_mode mode,
674 psw_t *psw = &vcpu->arch.sie_block->gpsw;
675 unsigned long override;
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);
688 static bool fetch_prot_override_applies(unsigned long ga, unsigned int len)
690 return ga < 2048 && ga + len <= 2048;
693 static bool storage_prot_override_applicable(struct kvm_vcpu *vcpu)
695 /* check if storage protection override enabled */
696 return vcpu->arch.sie_block->gcr[0] & CR0_STORAGE_PROTECTION_OVERRIDE;
699 static bool storage_prot_override_applies(u8 access_control)
701 /* matches special storage protection override key (9) -> allow */
702 return access_control == PAGE_SPO_ACC;
705 static 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)
709 u8 storage_key, access_control;
713 /* access key 0 matches any storage key -> allow */
717 * caller needs to ensure that gfn is accessible, so we can
718 * assume that this cannot fail
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);
726 access_control = FIELD_GET(_PAGE_ACC_BITS, storage_key);
727 /* access key matches storage key -> allow */
728 if (access_control == access_key)
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))
734 if (fetch_prot_override_applicable(vcpu, mode, asce) &&
735 fetch_prot_override_applies(ga, len))
738 if (storage_prot_override_applicable(vcpu) &&
739 storage_prot_override_applies(access_control))
741 return PGM_PROTECTION;
745 * guest_range_to_gpas() - Calculate guest physical addresses of page fragments
746 * covering a logical range
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
754 * @access_key: access key to mach the range's storage keys against
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
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.
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.
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.
779 static 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,
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;
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,
798 if (psw_bits(*psw).dat) {
799 rc = guest_translate(vcpu, ga, &gpa, asce, mode, &prot);
803 gpa = kvm_s390_real_to_abs(vcpu, ga);
804 if (!kvm_is_gpa_in_memslot(vcpu->kvm, gpa)) {
810 return trans_exc(vcpu, rc, ga, ar, mode, prot);
811 rc = vcpu_check_access_key(vcpu, access_key, mode, asce, gpa, ga,
814 return trans_exc(vcpu, rc, ga, ar, mode, PROT_TYPE_KEYC);
824 static int access_guest_page(struct kvm *kvm, enum gacc_mode mode, gpa_t gpa,
825 void *data, unsigned int len)
827 const unsigned int offset = offset_in_page(gpa);
828 const gfn_t gfn = gpa_to_gfn(gpa);
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);
836 rc = kvm_read_guest_page(kvm, gfn, data, offset, len);
841 access_guest_page_with_key(struct kvm *kvm, enum gacc_mode mode, gpa_t gpa,
842 void *data, unsigned int len, u8 access_key)
844 struct kvm_memory_slot *slot;
850 gfn = gpa >> PAGE_SHIFT;
851 slot = gfn_to_memslot(kvm, gfn);
852 hva = gfn_to_hva_memslot_prot(slot, gfn, &writable);
854 if (kvm_is_error_hva(hva))
855 return PGM_ADDRESSING;
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.
860 if (!writable && mode == GACC_STORE)
862 hva += offset_in_page(gpa);
863 if (mode == GACC_STORE)
864 rc = copy_to_user_key((void __user *)hva, data, len, access_key);
866 rc = copy_from_user_key(data, (void __user *)hva, len, access_key);
868 return PGM_PROTECTION;
869 if (mode == GACC_STORE)
870 mark_page_dirty_in_slot(kvm, slot, gfn);
874 int access_guest_abs_with_key(struct kvm *kvm, gpa_t gpa, void *data,
875 unsigned long len, enum gacc_mode mode, u8 access_key)
877 int offset = offset_in_page(gpa);
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);
888 data += fragment_len;
894 int access_guest_with_key(struct kvm_vcpu *vcpu, unsigned long ga, u8 ar,
895 void *data, unsigned long len, enum gacc_mode mode,
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;
906 bool try_storage_prot_override;
907 bool try_fetch_prot_override;
912 ga = kvm_s390_logical_to_effective(vcpu, ga);
913 rc = get_vcpu_asce(vcpu, &asce, ga, ar, mode);
916 nr_pages = (((ga & ~PAGE_MASK) + len - 1) >> PAGE_SHIFT) + 1;
918 if (nr_pages > ARRAY_SIZE(gpa_array))
919 gpas = vmalloc(array_size(nr_pages, sizeof(unsigned long)));
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;
926 ipte_lock(vcpu->kvm);
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.
937 rc = guest_range_to_gpas(vcpu, ga, ar, gpas, len, asce, mode, 0);
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],
946 rc = access_guest_page_with_key(vcpu->kvm, mode, gpas[idx],
947 data, fragment_len, access_key);
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);
955 data += fragment_len;
956 ga = kvm_s390_logical_to_effective(vcpu, ga + fragment_len);
959 bool terminate = (mode == GACC_STORE) && (idx > 0);
961 if (rc == PGM_PROTECTION)
962 prot = PROT_TYPE_KEYC;
965 rc = trans_exc_ending(vcpu, rc, ga, ar, mode, prot, terminate);
969 ipte_unlock(vcpu->kvm);
970 if (nr_pages > ARRAY_SIZE(gpa_array))
975 int access_guest_real(struct kvm_vcpu *vcpu, unsigned long gra,
976 void *data, unsigned long len, enum gacc_mode mode)
978 unsigned int fragment_len;
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);
988 data += fragment_len;
991 vcpu->arch.pgm.code = rc;
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.
1009 * Atomically exchange the value at @gpa by @new, if it contains *@old.
1010 * Honors storage keys.
1012 * Return: * 0: successful exchange
1013 * * >0: a program interruption code indicating the reason cmpxchg could
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)
1019 int 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)
1023 gfn_t gfn = gpa_to_gfn(gpa);
1024 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1029 if (!IS_ALIGNED(gpa, len))
1032 hva = gfn_to_hva_memslot_prot(slot, gfn, &writable);
1033 if (kvm_is_error_hva(hva))
1034 return PGM_ADDRESSING;
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.
1043 hva += offset_in_page(gpa);
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.
1053 ret = cmpxchg_user_key((u8 __user *)hva, &old, *old_addr, new, access_key);
1054 *success = !ret && old == *old_addr;
1061 ret = cmpxchg_user_key((u16 __user *)hva, &old, *old_addr, new, access_key);
1062 *success = !ret && old == *old_addr;
1069 ret = cmpxchg_user_key((u32 __user *)hva, &old, *old_addr, new, access_key);
1070 *success = !ret && old == *old_addr;
1077 ret = cmpxchg_user_key((u64 __user *)hva, &old, *old_addr, new, access_key);
1078 *success = !ret && old == *old_addr;
1085 ret = cmpxchg_user_key((__uint128_t __user *)hva, &old, *old_addr, new, access_key);
1086 *success = !ret && old == *old_addr;
1094 mark_page_dirty_in_slot(kvm, slot, gfn);
1096 * Assume that the fault is caused by protection, either key protection
1097 * or user page write protection.
1100 ret = PGM_PROTECTION;
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
1113 * Parameter semantics are the same as the ones from guest_translate.
1114 * The memory contents at the guest address are not changed.
1116 * Note: The IPTE lock is not taken during this function, so the caller
1117 * has to take care of this.
1119 int guest_translate_address_with_key(struct kvm_vcpu *vcpu, unsigned long gva, u8 ar,
1120 unsigned long *gpa, enum gacc_mode mode,
1126 gva = kvm_s390_logical_to_effective(vcpu, gva);
1127 rc = get_vcpu_asce(vcpu, &asce, gva, ar, mode);
1130 return guest_range_to_gpas(vcpu, gva, ar, gpa, 1, asce, mode,
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
1143 int check_gva_range(struct kvm_vcpu *vcpu, unsigned long gva, u8 ar,
1144 unsigned long length, enum gacc_mode mode, u8 access_key)
1149 rc = get_vcpu_asce(vcpu, &asce, gva, ar, mode);
1152 ipte_lock(vcpu->kvm);
1153 rc = guest_range_to_gpas(vcpu, gva, ar, NULL, length, asce, mode,
1155 ipte_unlock(vcpu->kvm);
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
1168 int check_gpa_range(struct kvm *kvm, unsigned long gpa, unsigned long length,
1169 enum gacc_mode mode, u8 access_key)
1171 unsigned int fragment_len;
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;
1184 * kvm_s390_check_low_addr_prot_real - check for low-address protection
1185 * @vcpu: virtual cpu
1186 * @gra: Guest real address
1188 * Checks whether an address is subject to low-address protection and set
1189 * up vcpu->arch.pgm accordingly if necessary.
1191 * Return: 0 if no protection exception, or PGM_PROTECTION if protected.
1193 int kvm_s390_check_low_addr_prot_real(struct kvm_vcpu *vcpu, unsigned long gra)
1195 union ctlreg0 ctlreg0 = {.val = vcpu->arch.sie_block->gcr[0]};
1197 if (!ctlreg0.lap || !is_low_address(gra))
1199 return trans_exc(vcpu, PGM_PROTECTION, gra, 0, GACC_STORE, PROT_TYPE_LA);
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
1212 static int kvm_s390_shadow_tables(struct gmap *sg, unsigned long saddr,
1213 unsigned long *pgt, int *dat_protection,
1217 struct gmap *parent;
1219 union vaddress vaddr;
1224 *dat_protection = 0;
1226 parent = sg->parent;
1228 asce.val = sg->orig_asce;
1229 ptr = asce.rsto * PAGE_SIZE;
1233 asce.dt = ASCE_TYPE_REGION1;
1236 case ASCE_TYPE_REGION1:
1237 if (vaddr.rfx01 > asce.tl && !*fake)
1238 return PGM_REGION_FIRST_TRANS;
1240 case ASCE_TYPE_REGION2:
1242 return PGM_ASCE_TYPE;
1243 if (vaddr.rsx01 > asce.tl)
1244 return PGM_REGION_SECOND_TRANS;
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;
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;
1261 case ASCE_TYPE_REGION1: {
1262 union region1_table_entry rfte;
1265 ptr += vaddr.rfx * _REGION1_SIZE;
1269 *pgt = ptr + vaddr.rfx * 8;
1270 rc = gmap_read_table(parent, ptr + vaddr.rfx * 8, &rfte.val);
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;
1283 rc = gmap_shadow_r2t(sg, saddr, rfte.val, *fake);
1286 kvm->stat.gmap_shadow_r1_entry++;
1289 case ASCE_TYPE_REGION2: {
1290 union region2_table_entry rste;
1293 ptr += vaddr.rsx * _REGION2_SIZE;
1297 *pgt = ptr + vaddr.rsx * 8;
1298 rc = gmap_read_table(parent, ptr + vaddr.rsx * 8, &rste.val);
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;
1311 rste.p |= *dat_protection;
1312 rc = gmap_shadow_r3t(sg, saddr, rste.val, *fake);
1315 kvm->stat.gmap_shadow_r2_entry++;
1318 case ASCE_TYPE_REGION3: {
1319 union region3_table_entry rtte;
1322 ptr += vaddr.rtx * _REGION3_SIZE;
1326 *pgt = ptr + vaddr.rtx * 8;
1327 rc = gmap_read_table(parent, ptr + vaddr.rtx * 8, &rtte.val);
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;
1339 ptr = rtte.fc1.rfaa * _REGION3_SIZE;
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;
1349 rtte.fc0.p |= *dat_protection;
1350 rc = gmap_shadow_sgt(sg, saddr, rtte.val, *fake);
1353 kvm->stat.gmap_shadow_r3_entry++;
1356 case ASCE_TYPE_SEGMENT: {
1357 union segment_table_entry ste;
1360 ptr += vaddr.sx * _SEGMENT_SIZE;
1364 *pgt = ptr + vaddr.sx * 8;
1365 rc = gmap_read_table(parent, ptr + vaddr.sx * 8, &ste.val);
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) {
1377 ptr = ste.fc1.sfaa * _SEGMENT_SIZE;
1381 ptr = ste.fc0.pto * (PAGE_SIZE / 2);
1383 ste.fc0.p |= *dat_protection;
1384 rc = gmap_shadow_pgt(sg, saddr, ste.val, *fake);
1387 kvm->stat.gmap_shadow_sg_entry++;
1390 /* Return the parent address of the page table */
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
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
1409 int kvm_s390_shadow_fault(struct kvm_vcpu *vcpu, struct gmap *sg,
1410 unsigned long saddr, unsigned long *datptr)
1412 union vaddress vaddr;
1413 union page_table_entry pte;
1414 unsigned long pgt = 0;
1415 int dat_protection, fake;
1418 mmap_read_lock(sg->mm);
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.
1424 ipte_lock(vcpu->kvm);
1426 rc = gmap_shadow_pgt_lookup(sg, saddr, &pgt, &dat_protection, &fake);
1428 rc = kvm_s390_shadow_tables(sg, saddr, &pgt, &dat_protection,
1433 pte.val = pgt + vaddr.px * PAGE_SIZE;
1438 case PGM_SEGMENT_TRANSLATION:
1439 case PGM_REGION_THIRD_TRANS:
1440 case PGM_REGION_SECOND_TRANS:
1441 case PGM_REGION_FIRST_TRANS:
1445 pgt += vaddr.px * 8;
1446 rc = gmap_read_table(sg->parent, pgt, &pte.val);
1449 *datptr = pgt | dat_protection * PEI_DAT_PROT;
1451 rc = PGM_PAGE_TRANSLATION;
1453 rc = PGM_TRANSLATION_SPEC;
1455 pte.p |= dat_protection;
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);