Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
KVM: s390: Fix gmap_link()
The slow path of the fault handler ultimately called gmap_link(), which
assumed the fault was a major fault, and blindly called dat_link().
In case of minor faults, things were not always handled properly; in
particular the prefix and vsie marker bits were ignored.
Move dat_link() into gmap.c, renaming it accordingly. Once moved, the
new _gmap_link() function will be able to correctly honour the prefix
and vsie markers.
This will cause spurious unshadows in some uncommon cases.
Fixes: 94fd9b16cc67 ("KVM: s390: KVM page table management functions: lifecycle management")
Fixes: a2c17f9270cc ("KVM: s390: New gmap code")
Reviewed-by: Steffen Eiden <seiden@linux.ibm.com>
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
+52
-54
-48
arch/s390/kvm/dat.c
-48
arch/s390/kvm/dat.c
···
997
997
return _dat_walk_gfn_range(start, end, asce, &test_age_ops, 0, NULL) > 0;
998
998
}
999
999
1000
-
int dat_link(struct kvm_s390_mmu_cache *mc, union asce asce, int level,
1001
-
bool uses_skeys, struct guest_fault *f)
1002
-
{
1003
-
union crste oldval, newval;
1004
-
union pte newpte, oldpte;
1005
-
union pgste pgste;
1006
-
int rc = 0;
1007
-
1008
-
rc = dat_entry_walk(mc, f->gfn, asce, DAT_WALK_ALLOC_CONTINUE, level, &f->crstep, &f->ptep);
1009
-
if (rc == -EINVAL || rc == -ENOMEM)
1010
-
return rc;
1011
-
if (rc)
1012
-
return -EAGAIN;
1013
-
1014
-
if (WARN_ON_ONCE(unlikely(get_level(f->crstep, f->ptep) > level)))
1015
-
return -EINVAL;
1016
-
1017
-
if (f->ptep) {
1018
-
pgste = pgste_get_lock(f->ptep);
1019
-
oldpte = *f->ptep;
1020
-
newpte = _pte(f->pfn, f->writable, f->write_attempt | oldpte.s.d, !f->page);
1021
-
newpte.s.sd = oldpte.s.sd;
1022
-
oldpte.s.sd = 0;
1023
-
if (oldpte.val == _PTE_EMPTY.val || oldpte.h.pfra == f->pfn) {
1024
-
pgste = __dat_ptep_xchg(f->ptep, pgste, newpte, f->gfn, asce, uses_skeys);
1025
-
if (f->callback)
1026
-
f->callback(f);
1027
-
} else {
1028
-
rc = -EAGAIN;
1029
-
}
1030
-
pgste_set_unlock(f->ptep, pgste);
1031
-
} else {
1032
-
oldval = READ_ONCE(*f->crstep);
1033
-
newval = _crste_fc1(f->pfn, oldval.h.tt, f->writable,
1034
-
f->write_attempt | oldval.s.fc1.d);
1035
-
newval.s.fc1.sd = oldval.s.fc1.sd;
1036
-
if (oldval.val != _CRSTE_EMPTY(oldval.h.tt).val &&
1037
-
crste_origin_large(oldval) != crste_origin_large(newval))
1038
-
return -EAGAIN;
1039
-
if (!dat_crstep_xchg_atomic(f->crstep, oldval, newval, f->gfn, asce))
1040
-
return -EAGAIN;
1041
-
if (f->callback)
1042
-
f->callback(f);
1043
-
}
1044
-
1045
-
return rc;
1046
-
}
1047
-
1048
1000
static long dat_set_pn_crste(union crste *crstep, gfn_t gfn, gfn_t next, struct dat_walk *walk)
1049
1001
{
1050
1002
union crste newcrste, oldcrste;
-2
arch/s390/kvm/dat.h
-2
arch/s390/kvm/dat.h
···
540
540
u16 type, u16 param);
541
541
int dat_set_prefix_notif_bit(union asce asce, gfn_t gfn);
542
542
bool dat_test_age_gfn(union asce asce, gfn_t start, gfn_t end);
543
-
int dat_link(struct kvm_s390_mmu_cache *mc, union asce asce, int level,
544
-
bool uses_skeys, struct guest_fault *f);
545
543
546
544
int dat_perform_essa(union asce asce, gfn_t gfn, int orc, union essa_state *state, bool *dirty);
547
545
long dat_reset_cmma(union asce asce, gfn_t start_gfn);
+52
-4
arch/s390/kvm/gmap.c
+52
-4
arch/s390/kvm/gmap.c
···
631
631
return test_bit(GMAP_FLAG_ALLOW_HPAGE_1M, &gmap->flags);
632
632
}
633
633
634
+
static int _gmap_link(struct kvm_s390_mmu_cache *mc, struct gmap *gmap, int level,
635
+
struct guest_fault *f)
636
+
{
637
+
union crste oldval, newval;
638
+
union pte newpte, oldpte;
639
+
union pgste pgste;
640
+
int rc = 0;
641
+
642
+
rc = dat_entry_walk(mc, f->gfn, gmap->asce, DAT_WALK_ALLOC_CONTINUE, level,
643
+
&f->crstep, &f->ptep);
644
+
if (rc == -ENOMEM)
645
+
return rc;
646
+
if (KVM_BUG_ON(rc == -EINVAL, gmap->kvm))
647
+
return rc;
648
+
if (rc)
649
+
return -EAGAIN;
650
+
if (KVM_BUG_ON(get_level(f->crstep, f->ptep) > level, gmap->kvm))
651
+
return -EINVAL;
652
+
653
+
if (f->ptep) {
654
+
pgste = pgste_get_lock(f->ptep);
655
+
oldpte = *f->ptep;
656
+
newpte = _pte(f->pfn, f->writable, f->write_attempt | oldpte.s.d, !f->page);
657
+
newpte.s.sd = oldpte.s.sd;
658
+
oldpte.s.sd = 0;
659
+
if (oldpte.val == _PTE_EMPTY.val || oldpte.h.pfra == f->pfn) {
660
+
pgste = gmap_ptep_xchg(gmap, f->ptep, newpte, pgste, f->gfn);
661
+
if (f->callback)
662
+
f->callback(f);
663
+
} else {
664
+
rc = -EAGAIN;
665
+
}
666
+
pgste_set_unlock(f->ptep, pgste);
667
+
} else {
668
+
do {
669
+
oldval = READ_ONCE(*f->crstep);
670
+
newval = _crste_fc1(f->pfn, oldval.h.tt, f->writable,
671
+
f->write_attempt | oldval.s.fc1.d);
672
+
newval.s.fc1.sd = oldval.s.fc1.sd;
673
+
if (oldval.val != _CRSTE_EMPTY(oldval.h.tt).val &&
674
+
crste_origin_large(oldval) != crste_origin_large(newval))
675
+
return -EAGAIN;
676
+
} while (!gmap_crstep_xchg_atomic(gmap, f->crstep, oldval, newval, f->gfn));
677
+
if (f->callback)
678
+
f->callback(f);
679
+
}
680
+
681
+
return rc;
682
+
}
683
+
634
684
int gmap_link(struct kvm_s390_mmu_cache *mc, struct gmap *gmap, struct guest_fault *f)
635
685
{
636
686
unsigned int order;
637
-
int rc, level;
687
+
int level;
638
688
639
689
lockdep_assert_held(&gmap->kvm->mmu_lock);
640
690
···
696
646
else if (order >= get_order(_SEGMENT_SIZE) && gmap_1m_allowed(gmap, f->gfn))
697
647
level = TABLE_TYPE_SEGMENT;
698
648
}
699
-
rc = dat_link(mc, gmap->asce, level, uses_skeys(gmap), f);
700
-
KVM_BUG_ON(rc == -EINVAL, gmap->kvm);
701
-
return rc;
649
+
return _gmap_link(mc, gmap, level, f);
702
650
}
703
651
704
652
static int gmap_ucas_map_one(struct kvm_s390_mmu_cache *mc, struct gmap *gmap,