KVM: s390: Fix minor bugs in STFLE shadowing
v4 cover KVM: s390: Fix minor bugs in STFLE shadowing
v3 -> v4:
* pick up tags (thanks {David, Janosch, Heiko})
* changes to commit messages
* flip lines and add comment (Janosch)
v2 -> v3:
* pick up tags (thanks Claudio)
* reverse Christmas tree
v1 -> v2:
* pick up tags (thanks {Claudio, David})
* drop Fixes tag on cleanup patch, change message (thanks David)
* drop Fixes tag on second patch since the length of the facility list copied wasn't initially specified and only clarified in later revisions
* use READ/WRITE_ONCE (thanks {David, Heiko})
Improve the STFLE vsie implementation.
Firstly, fix a bug concerning the identification if the guest is intending to use interpretive execution for STFLE for its guest. Secondly, decrease the amount of guest memory accessed to the minimum.
Also do some (optional) cleanups.
Nina Schoetterl-Glausch (4):
KVM: s390: vsie: Fix STFLE interpretive execution identification
KVM: s390: vsie: Fix length of facility list shadowed
KVM: s390: cpu model: Use proper define for facility mask size
KVM: s390: Minor refactor of base/ext facility lists
arch/s390/include/asm/facility.h | 6 +++++ arch/s390/include/asm/kvm_host.h | 2 +- arch/s390/kernel/Makefile | 2 +- arch/s390/kernel/facility.c | 21 +++++++++++++++ arch/s390/kvm/kvm-s390.c | 44 ++++++++++++++------------------ arch/s390/kvm/vsie.c | 19 ++++++++++++-- 6 files changed, 65 insertions(+), 29 deletions(-) create mode 100644 arch/s390/kernel/facility.c
Range-diff against v3:
1: de77a2c36786 ! 1: 69599bb38487 KVM: s390: vsie: Fix STFLE interpretive execution identification
@@ arch/s390/kvm/vsie.c: static void retry_vsie_icpt(struct vsie_page *vsie_page)
+ __u32 fac = READ_ONCE(vsie_page->scb_o->fac);
if (fac && test_kvm_facility(vcpu->kvm, 7)) {
-+ fac = fac & 0x7ffffff8U;
retry_vsie_icpt(vsie_page);
++ /*
++ * The facility list origin (FLO) is in bits 1 - 28 of the FLD
++ * so we need to mask here before reading.
++ */
++ fac = fac & 0x7ffffff8U;
if (read_guest_real(vcpu, fac, &vsie_page->fac,
sizeof(vsie_page->fac)))
+ return set_validity_icpt(scb_s, 0x1090U);2: e4b44c4d2400 ! 2: cba3c32a8db7 KVM: s390: vsie: Fix length of facility list shadowed
@@ Commit message
The length of the facility list accessed when interpretively executing STFLE is the same as the hosts facility list (in case of format-0)
- When shadowing, copy only those bytes.
- The memory following the facility list need not be accessible, in which
- case we'd wrongly inject a validity intercept.
+ The memory following the facility list doesn't need to be accessible.
+ The current VSIE implementation accesses a fixed length that exceeds the
+ guest/host facility list length and can therefore wrongly inject a
+ validity intercept.
+ Instead, find out the host facility list length by running STFLE and
+ copy only as much as necessary when shadowing.
+ Acked-by: Heiko Carstens <[email protected]>
## arch/s390/include/asm/facility.h ##
@@ arch/s390/include/asm/facility.h: static inline void stfle(u64 *stfle_fac_list,
#endif /* __ASM_FACILITY_H */
## arch/s390/kernel/Makefile ##
-@@ arch/s390/kernel/Makefile: obj-y += sysinfo.o lgr.o os_info.o
+@@ arch/s390/kernel/Makefile: obj-y += sysinfo.o lgr.o os_info.o ctlreg.o
obj-y += runtime_instr.o cache.o fpu.o dumpstack.o guarded_storage.o sthyi.o
obj-y += entry.o reipl.o kdebugfs.o alternative.o
obj-y += nospec-branch.o ipl_vmparm.o machine_kexec_reloc.o unwind_bc.o
@@ arch/s390/kvm/vsie.c: static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie
+ * -> format-0 flcb
+ */
if (fac && test_kvm_facility(vcpu->kvm, 7)) {
- fac = fac & 0x7ffffff8U;
retry_vsie_icpt(vsie_page);
+ /*
+@@ arch/s390/kvm/vsie.c: static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
+ * so we need to mask here before reading.
+ */
+ fac = fac & 0x7ffffff8U;
+ /*
+ * format-0 -> size of nested guest's facility list == guest's size
+ * guest's size == host's size, since STFLE is interpretatively executed
3: 8b02ac33defb ! 3: 4b52e432d736 KVM: s390: cpu model: Use proper define for facility mask size
@@ Commit message
Note that both values are the same, there is no functional change.
+ Reviewed-by: David Hildenbrand <[email protected]>
+ Reviewed-by: Janosch Frank <[email protected]>
## arch/s390/include/asm/kvm_host.h ##
4: a592be823576 = 4: 9e551ba53b14 KVM: s390: Minor refactor of base/ext facility lists
Signature · 1 line
2.40.1
PATCH v4 1/4 KVM: s390: vsie: Fix STFLE interpretive execution identification
STFLE can be interpretively executed.
This occurs when the facility list designation is unequal to zero. Perform the check before applying the address mask instead of after.
--- arch/s390/kvm/vsie.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
arch/s390/kvm/vsie.c +6 -1
@@ -984,10 +984,15 @@ static void retry_vsie_icpt(struct vsie_page *vsie_page)static int handle_stfle (struct kvm_vcpu *vcpu,struct vsie_page *vsie_page) {struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; -__u32 fac =READ_ONCE (vsie_page->scb_o->fac) &0x7ffffff8U ; +__u32 fac =READ_ONCE (vsie_page->scb_o->fac);if (fac &&test_kvm_facility (vcpu->kvm,7 )) {retry_vsie_icpt (vsie_page); +/* +* The facility list origin (FLO) is in bits 1 - 28 of the FLD +* so we need to mask here before reading. +*/ + fac = fac &0x7ffffff8U ;if (read_guest_real (vcpu, fac, &vsie_page->fac,sizeof (vsie_page->fac)))return set_validity_icpt (scb_s,0x1090U );
Signature · 1 line
2.40.1
Re: [PATCH v4 1/4] KVM: s390: vsie: Fix STFLE interpretive execution identification
Am 19.12.23 um 15:08 schrieb Nina Schoetterl-Glausch:
Quoted 8 lines
STFLE can be interpretively executed.
This occurs when the facility list designation is unequal to zero. Perform the check before applying the address mask instead of after.Fixes 66b630d5b7f2 ("KVM: s390: vsie: support STFLE interpretation")Reviewed-by Claudio Imbrenda <[email protected]>Signed-off-by Nina Schoetterl-Glausch <[email protected]>
this should not matter in reality but maybe some weird guests puts this at address 0. Do we want a unit test for that case?
Quoted 25 lines
--- arch/s390/kvm/vsie.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)arch/s390/kvm/vsie.c +6
-1@@ -984,10 +984,15 @@ static void retry_vsie_icpt(struct vsie_page *vsie_page)static int handle_stfle (struct kvm_vcpu *vcpu,struct vsie_page *vsie_page) {struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; -__u32 fac =READ_ONCE (vsie_page->scb_o->fac) &0x7ffffff8U ; +__u32 fac =READ_ONCE (vsie_page->scb_o->fac);if (fac &&test_kvm_facility (vcpu->kvm,7 )) {retry_vsie_icpt (vsie_page); +/* +* The facility list origin (FLO) is in bits 1 - 28 of the FLD +* so we need to mask here before reading. +*/ + fac = fac &0x7ffffff8U ;if (read_guest_real (vcpu, fac, &vsie_page->fac,sizeof (vsie_page->fac)))return set_validity_icpt (scb_s,0x1090U );
Re: [PATCH v4 1/4] KVM: s390: vsie: Fix STFLE interpretive execution identification
In reply to Nina Schoetterl-GlauschQuoted 8 lines · Nina Schoetterl-Glausch
STFLE can be interpretively executed.
This occurs when the facility list designation is unequal to zero. Perform the check before applying the address mask instead of after.Fixes 66b630d5b7f2 ("KVM: s390: vsie: support STFLE interpretation")Reviewed-by Claudio Imbrenda <[email protected]>Signed-off-by Nina Schoetterl-Glausch <[email protected]>
Quoted 25 lines
--- arch/s390/kvm/vsie.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)arch/s390/kvm/vsie.c +6
-1@@ -984,10 +984,15 @@ static void retry_vsie_icpt(struct vsie_page *vsie_page)static int handle_stfle (struct kvm_vcpu *vcpu,struct vsie_page *vsie_page) {struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; -__u32 fac =READ_ONCE (vsie_page->scb_o->fac) &0x7ffffff8U ; +__u32 fac =READ_ONCE (vsie_page->scb_o->fac);if (fac &&test_kvm_facility (vcpu->kvm,7 )) {retry_vsie_icpt (vsie_page); +/* +* The facility list origin (FLO) is in bits 1 - 28 of the FLD +* so we need to mask here before reading. +*/ + fac = fac &0x7ffffff8U ;if (read_guest_real (vcpu, fac, &vsie_page->fac,sizeof (vsie_page->fac)))return set_validity_icpt (scb_s,0x1090U );
PATCH v4 2/4 KVM: s390: vsie: Fix length of facility list shadowed
In reply to Nina Schoetterl-GlauschThe length of the facility list accessed when interpretively executing STFLE is the same as the hosts facility list (in case of format-0) The memory following the facility list doesn't need to be accessible. The current VSIE implementation accesses a fixed length that exceeds the guest/host facility list length and can therefore wrongly inject a validity intercept.
Instead, find out the host facility list length by running STFLE and copy only as much as necessary when shadowing.
--- arch/s390/include/asm/facility.h | 6 ++++++ arch/s390/kernel/Makefile | 2 +- arch/s390/kernel/facility.c | 21 +++++++++++++++++++++ arch/s390/kvm/vsie.c | 12 +++++++++++- 4 files changed, 39 insertions(+), 2 deletions(-) create mode 100644 arch/s390/kernel/facility.c
arch/s390/include/asm/facility.h +6 -0
@@ -111,4 +111,10 @@ static inline void stfle(u64 *stfle_fac_list, int size) preempt_enable(); } +/** +* stfle_size - Actual size of the facility list as specified by stfle +* (number of double words) +*/ +unsigned int stfle_size (void ); +#endif /* __ASM_FACILITY_H */
arch/s390/kernel/Makefile +1 -1
@@ -41,7 +41,7 @@ obj-y += sysinfo.o lgr.o os_info.o ctlreg.o obj-y += runtime_instr.o cache.o fpu.o dumpstack.o guarded_storage.o sthyi.o obj-y += entry.o reipl.o kdebugfs.o alternative.o obj-y += nospec-branch.o ipl_vmparm.o machine_kexec_reloc.o unwind_bc.o -obj-y += smp.o text_amode31.o stacktrace.o abs_lowcore.o +obj-y += smp.o text_amode31.o stacktrace.o abs_lowcore.o facility.o extra-y += vmlinux.lds
arch/s390/kernel/facility.c new file +21 -0
@@ -0,0 +1,21 @@ +// SPDX-License-Identifier: GPL-2.0 +/* +* Copyright IBM Corp. 2023 +*/ + +#include <asm/facility.h> + +unsigned int stfle_size (void ) +{ +static unsigned int size; +unsigned int r; +u64 dummy; + + r =READ_ONCE (size); +if (!r) { + r =__stfle_asm (&dummy,1 ) +1 ; +WRITE_ONCE (size, r); + } +return r; +} +EXPORT_SYMBOL (stfle_size);
arch/s390/kvm/vsie.c +11 -1
@@ -19,6 +19,7 @@#include <asm/nmi.h> #include <asm/dis.h> #include <asm/fpu/api.h> +#include <asm/facility.h> #include "kvm-s390.h" #include "gaccess.h"
@@ -986,6 +987,10 @@ static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;__u32 fac =READ_ONCE (vsie_page->scb_o->fac); +/* +* Alternate-STFLE-Interpretive-Execution facilities are not supported +* -> format-0 flcb +*/ if (fac &&test_kvm_facility (vcpu->kvm,7 )) {retry_vsie_icpt (vsie_page);/*
@@ -993,8 +998,13 @@ static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)* so we need to mask here before reading. */ fac = fac &0x7ffffff8U ; +/* +* format-0 -> size of nested guest's facility list == guest's size +* guest's size == host's size, since STFLE is interpretatively executed +* using a format-0 for the guest, too. +*/ if (read_guest_real (vcpu, fac, &vsie_page->fac, -sizeof (vsie_page->fac))) +stfle_size () *sizeof (u64)))return set_validity_icpt (scb_s,0x1090U ); scb_s->fac = (__u32)(__u64) &vsie_page->fac; }
Signature · 1 line
2.40.1
Re: [PATCH v4 2/4] KVM: s390: vsie: Fix length of facility list shadowed
Quoted 9 lines · Nina Schoetterl-Glausch
The length of the facility list accessed when interpretively executing STFLE is the same as the hosts facility list (in case of format-0) The memory following the facility list doesn't need to be accessible. The current VSIE implementation accesses a fixed length that exceeds the guest/host facility list length and can therefore wrongly inject a validity intercept.
Instead, find out the host facility list length by running STFLE and copy only as much as necessary when shadowing.
Quoted 4 lines
Reviewed-by Claudio Imbrenda <[email protected]>Acked-by Heiko Carstens <[email protected]>Signed-off-by Nina Schoetterl-Glausch <[email protected]>
PATCH v4 3/4 KVM: s390: cpu model: Use proper define for facility mask size
In reply to Nina Schoetterl-GlauschUse the previously unused S390_ARCH_FAC_MASK_SIZE_U64 instead of S390_ARCH_FAC_LIST_SIZE_U64 for defining the fac_mask array. Note that both values are the same, there is no functional change.
--- arch/s390/include/asm/kvm_host.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/s390/include/asm/kvm_host.h +1 -1
@@ -818,7 +818,7 @@ struct s390_io_adapter {struct kvm_s390_cpu_model {/* facility mask supported by kvm & hosting machine */ -__u64 fac_mask[S390_ARCH_FAC_LIST_SIZE_U64 ]; +__u64 fac_mask[S390_ARCH_FAC_MASK_SIZE_U64 ];struct kvm_s390_vm_cpu_subfunc subfuncs;/* facility list requested by guest (in dma page) */ __u64 *fac_list;
Signature · 1 line
2.40.1
PATCH v4 4/4 KVM: s390: Minor refactor of base/ext facility lists
In reply to Nina Schoetterl-GlauschDirectly use the size of the arrays instead of going through the indirection of kvm_s390_fac_size().
Don't use magic number for the number of entries in the non hypervisor managed facility bit mask list.
Make the constraint of that number on kvm_s390_fac_base obvious. Get rid of implicit double anding of stfle_fac_list.
---
Notes:
I think it's nicer this way but it might be needless churn.
arch/s390/kvm/kvm-s390.c | 44 +++++++++++++++++----------------------- 1 file changed, 19 insertions(+), 25 deletions(-)
arch/s390/kvm/kvm-s390.c +19 -25
@@ -224,33 +224,25 @@ static int async_destroy = 1;module_param (async_destroy, int,0444 );MODULE_PARM_DESC (async_destroy,"Asynchronous destroy for protected guests" ); -/* -* For now we handle at most 16 double words as this is what the s390 base -* kernel handles and stores in the prefix page. If we ever need to go beyond -* this, this requires changes to code, but the external uapi can stay. -*/ -#define SIZE_INTERNAL 16 - +#define HMFAI_DWORDS 16/* * Base feature mask that defines default mask for facilities. Consists of the * defines in FACILITIES_KVM and the non-hypervisor managed bits. */ -static unsigned long kvm_s390_fac_base[SIZE_INTERNAL ] = {FACILITIES_KVM }; +static unsigned long kvm_s390_fac_base[HMFAI_DWORDS ] = {FACILITIES_KVM }; +static_assert (ARRAY_SIZE (((long []){FACILITIES_KVM })) <=HMFAI_DWORDS ); +static_assert (ARRAY_SIZE (kvm_s390_fac_base) <=S390_ARCH_FAC_MASK_SIZE_U64 ); +static_assert (ARRAY_SIZE (kvm_s390_fac_base) <=S390_ARCH_FAC_LIST_SIZE_U64 ); +static_assert (ARRAY_SIZE (kvm_s390_fac_base) <=ARRAY_SIZE (stfle_fac_list)); +/* * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL * and defines the facilities that can be enabled via a cpu model. */ -static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL ] = {FACILITIES_KVM_CPUMODEL }; - -static unsigned long kvm_s390_fac_size (void ) -{ -BUILD_BUG_ON (SIZE_INTERNAL >S390_ARCH_FAC_MASK_SIZE_U64 ); -BUILD_BUG_ON (SIZE_INTERNAL >S390_ARCH_FAC_LIST_SIZE_U64 ); -BUILD_BUG_ON (SIZE_INTERNAL *sizeof (unsigned long ) > -sizeof (stfle_fac_list)); - -return SIZE_INTERNAL ; -} +static const unsigned long kvm_s390_fac_ext[] = {FACILITIES_KVM_CPUMODEL }; +static_assert (ARRAY_SIZE (kvm_s390_fac_ext) <=S390_ARCH_FAC_MASK_SIZE_U64 ); +static_assert (ARRAY_SIZE (kvm_s390_fac_ext) <=S390_ARCH_FAC_LIST_SIZE_U64 ); +static_assert (ARRAY_SIZE (kvm_s390_fac_ext) <=ARRAY_SIZE (stfle_fac_list));/* available cpu features supported by kvm */ static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
@@ -3348,13 +3340,16 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) kvm->arch.sie_page2->kvm = kvm; kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list; -for (i =0 ; i <kvm_s390_fac_size (); i++) { +for (i =0 ; i <ARRAY_SIZE (kvm_s390_fac_base); i++) { kvm->arch.model.fac_mask[i] = stfle_fac_list[i] & - (kvm_s390_fac_base[i] | - kvm_s390_fac_ext[i]); + kvm_s390_fac_base[i]; kvm->arch.model.fac_list[i] = stfle_fac_list[i] & kvm_s390_fac_base[i]; } +for (i =0 ; i <ARRAY_SIZE (kvm_s390_fac_ext); i++) { + kvm->arch.model.fac_mask[i] |= stfle_fac_list[i] & + kvm_s390_fac_ext[i]; + } kvm->arch.model.subfuncs = kvm_s390_available_subfunc;/* we are always in czam mode - even on pre z14 machines */
@@ -5868,9 +5863,8 @@ static int __init kvm_s390_init(void)return -EINVAL ; } -for (i =0 ; i <16 ; i++) - kvm_s390_fac_base[i] |= - stfle_fac_list[i] &nonhyp_mask (i); +for (i =0 ; i <HMFAI_DWORDS ; i++) + kvm_s390_fac_base[i] |=nonhyp_mask (i); r =__kvm_s390_init ();if (r)
Signature · 1 line
2.40.1
Janosch Frank
maintainerRe: [PATCH v4 4/4] KVM: s390: Minor refactor of base/ext facility lists
Quoted 9 lines · Nina Schoetterl-Glausch
Directly use the size of the arrays instead of going through the indirection of
kvm_s390_fac_size().
Don't use magic number for the number of entries in the non hypervisor managed facility bit mask list.
Make the constraint of that number onkvm_s390_fac_baseobvious. Get rid of implicit double anding ofstfle_fac_list.Reviewed-by Claudio Imbrenda <[email protected]>Signed-off-by Nina Schoetterl-Glausch <[email protected]>
@Nina: I'm currently still recovering from a cold and hence I'm not fully able to grasp this patch.
May I drop it and we re-visit it next year for 6.9?