* [PATCH v1 0/2] KVM: s390: fix migration mode teardown in memslot handling @ 2026-11-20 7:53 Gautam Gala 2026-11-20 7:53 ` [PATCH v1 1/2] KVM: s390: reserve mmu cache before stopping migration mode Gautam Gala 2026-11-20 7:53 ` [PATCH v1 2/2] KVM: s390: make kvm_s390_vm_stop_migration() void Gautam Gala 0 siblings, 2 replies; 6+ messages in thread From: Gautam Gala @ 2026-11-20 7:53 UTC (permalink / raw) To: kvm Cc: linux-s390, linux-kernel, Christian Borntraeger, Janosch Frank, Claudio Imbrenda, David Hildenbrand, Steffen Eiden, Andreas Grapentin, Marc Zyngier, Heiko Carstens, Vasily Gorbik, Alexander Gordeev, Sven Schnelle Two small fixes for migration mode handling in s390_kvm_mmu_prepare_memory_region(). Patch 1 fixes an error-path bug: if the mmu cache reservation fails after migration mode was already stopped, the ioctl fails but the CMMA state cleared by kvm_s390_vm_stop_migration() cannot be restored, so an in-flight migration aborts over an unrelated allocation failure. Reserving the cache first keeps both memslots and migration state untouched on failure. Patch 2 is a cleanup on top: kvm_s390_vm_stop_migration() cannot fail, so it returns void and the always-false WARN() at its call site is dropped. Gautam Gala (2): KVM: s390: reserve mmu cache before stopping migration mode KVM: s390: make kvm_s390_vm_stop_migration() void arch/s390/kvm/arm64/arm.h | 3 +-- arch/s390/kvm/gmap/kvm_mmu.c | 18 ++++++++++-------- arch/s390/kvm/s390/s390.c | 10 +++++----- arch/s390/kvm/s390/s390.h | 2 +- 4 files changed, 17 insertions(+), 16 deletions(-) -- 2.56.0 ^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH v1 1/2] KVM: s390: reserve mmu cache before stopping migration mode 2026-11-20 7:53 [PATCH v1 0/2] KVM: s390: fix migration mode teardown in memslot handling Gautam Gala @ 2026-11-20 7:53 ` Gautam Gala 2026-11-20 7:57 ` sashiko-bot 2026-11-20 7:53 ` [PATCH v1 2/2] KVM: s390: make kvm_s390_vm_stop_migration() void Gautam Gala 1 sibling, 1 reply; 6+ messages in thread From: Gautam Gala @ 2026-11-20 7:53 UTC (permalink / raw) To: kvm Cc: linux-s390, linux-kernel, Christian Borntraeger, Janosch Frank, Claudio Imbrenda, David Hildenbrand, Steffen Eiden, Andreas Grapentin, Marc Zyngier, Heiko Carstens, Vasily Gorbik, Alexander Gordeev, Sven Schnelle Creating or moving a memslot without KVM_MEM_LOG_DIRTY_PAGES stops migration mode in s390_kvm_mmu_prepare_memory_region(). If the mmu cache reservation fails afterwards, the ioctl returns -ENOMEM with the memslots unchanged, but migration mode stays stopped: the CMMA state cleared by kvm_s390_vm_stop_migration() cannot be restored, so an in-flight migration aborts over an unrelated allocation failure. Reserve the mmu cache before tearing down migration mode so the error path leaves both memslots and migration state untouched. Fixes: 19192a404327 ("KVM: s390: Move all code into s390_kvm_mmu_prepare_memory_region()") Reported-by: Christian Borntraeger <borntraeger@linux.ibm.com> Signed-off-by: Gautam Gala <ggala@linux.ibm.com> Reviewed-by: Christian Borntraeger <borntraeger@linux.ibm.com> Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com> --- arch/s390/kvm/gmap/kvm_mmu.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-) diff --git a/arch/s390/kvm/gmap/kvm_mmu.c b/arch/s390/kvm/gmap/kvm_mmu.c index 4e2da60b9859..120980cf6ac6 100644 --- a/arch/s390/kvm/gmap/kvm_mmu.c +++ b/arch/s390/kvm/gmap/kvm_mmu.c @@ -78,6 +78,14 @@ int s390_kvm_mmu_prepare_memory_region(struct kvm *kvm, return -EINVAL; if (!asce_contains_gfn(kvm->arch.gmap->asce, new->base_gfn + new->npages - 1)) return -EINVAL; + + /* + * Must be allocated before migration mode is stopped below: + * kvm_s390_vm_stop_migration() is not rolled back on failure. + */ + mc = kvm_s390_new_mmu_cache(); + if (!mc) + return -ENOMEM; } if (kvm_s390_is_migration_mode(kvm)) { @@ -97,12 +105,7 @@ int s390_kvm_mmu_prepare_memory_region(struct kvm *kvm, if (change == KVM_MR_FLAGS_ONLY) return 0; - if (change != KVM_MR_DELETE) { - /* Enough capacity to add a new memslot */ - mc = kvm_s390_new_mmu_cache(); - if (!mc) - return -ENOMEM; - } + scoped_guard(write_lock, &kvm->mmu_lock) { kvm_s390_update_cmma_dirty(kvm, old); if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) -- 2.56.0 ^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH v1 1/2] KVM: s390: reserve mmu cache before stopping migration mode 2026-11-20 7:53 ` [PATCH v1 1/2] KVM: s390: reserve mmu cache before stopping migration mode Gautam Gala @ 2026-11-20 7:57 ` sashiko-bot 0 siblings, 0 replies; 6+ messages in thread From: sashiko-bot @ 2026-11-20 7:57 UTC (permalink / raw) To: Gautam Gala Cc: linux-s390, Heiko Carstens, Vasily Gorbik, Alexander Gordeev, Christian Borntraeger > Creating or moving a memslot without KVM_MEM_LOG_DIRTY_PAGES stops > migration mode in s390_kvm_mmu_prepare_memory_region(). If the mmu > cache reservation fails afterwards, the ioctl returns -ENOMEM with > the memslots unchanged, but migration mode stays stopped: the CMMA > state cleared by kvm_s390_vm_stop_migration() cannot be restored, > so an in-flight migration aborts over an unrelated allocation > failure. > > Reserve the mmu cache before tearing down migration mode so the > error path leaves both memslots and migration state untouched. > > Fixes: 19192a404327 ("KVM: s390: Move all code into s390_kvm_mmu_prepare_memory_region()") > Reported-by: Christian Borntraeger <borntraeger@linux.ibm.com> > Signed-off-by: Gautam Gala <ggala@linux.ibm.com> Sashiko has reviewed this patch and found no issues. -- Sashiko AI review · https://sashiko.dev/#/patchset/20261120075345.116521-1-ggala@linux.ibm.com?part=1 ^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH v1 2/2] KVM: s390: make kvm_s390_vm_stop_migration() void 2026-11-20 7:53 [PATCH v1 0/2] KVM: s390: fix migration mode teardown in memslot handling Gautam Gala 2026-11-20 7:53 ` [PATCH v1 1/2] KVM: s390: reserve mmu cache before stopping migration mode Gautam Gala @ 2026-11-20 7:53 ` Gautam Gala 2026-11-20 7:58 ` sashiko-bot 2026-11-20 10:30 ` Andreas Grapentin 1 sibling, 2 replies; 6+ messages in thread From: Gautam Gala @ 2026-11-20 7:53 UTC (permalink / raw) To: kvm Cc: linux-s390, linux-kernel, Christian Borntraeger, Janosch Frank, Claudio Imbrenda, David Hildenbrand, Steffen Eiden, Andreas Grapentin, Marc Zyngier, Heiko Carstens, Vasily Gorbik, Alexander Gordeev, Sven Schnelle kvm_s390_vm_stop_migration() returns 0 on all paths; it cannot fail. Its caller in s390_kvm_mmu_prepare_memory_region() wraps it in WARN(), producing a condition that is always false. The arm64 stub returns -EINVAL but is unreachable because kvm_s390_is_migration_mode() returns false there. Change the return type to void and drop the dead WARN(). The KVM_S390_VM_MIGRATION_STOP attribute handler sets its result to 0 explicitly; userspace behavior is unchanged since stopping migration never failed. Signed-off-by: Gautam Gala <ggala@linux.ibm.com> Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com> --- arch/s390/kvm/arm64/arm.h | 3 +-- arch/s390/kvm/gmap/kvm_mmu.c | 3 +-- arch/s390/kvm/s390/s390.c | 10 +++++----- arch/s390/kvm/s390/s390.h | 2 +- 4 files changed, 8 insertions(+), 10 deletions(-) diff --git a/arch/s390/kvm/arm64/arm.h b/arch/s390/kvm/arm64/arm.h index 760230685ca1..e60801e4a467 100644 --- a/arch/s390/kvm/arm64/arm.h +++ b/arch/s390/kvm/arm64/arm.h @@ -50,9 +50,8 @@ static __always_inline void kvm_s390_update_cmma_dirty(struct kvm *kvm, const st } /* should never be called */ -static __always_inline int kvm_s390_vm_stop_migration(struct kvm *kvm) +static __always_inline void kvm_s390_vm_stop_migration(struct kvm *kvm) { - return -EINVAL; } #endif /* ARCH_S390_KVM_ARM64_H */ diff --git a/arch/s390/kvm/gmap/kvm_mmu.c b/arch/s390/kvm/gmap/kvm_mmu.c index 120980cf6ac6..f1d1374c92a1 100644 --- a/arch/s390/kvm/gmap/kvm_mmu.c +++ b/arch/s390/kvm/gmap/kvm_mmu.c @@ -99,8 +99,7 @@ int s390_kvm_mmu_prepare_memory_region(struct kvm *kvm, */ if (change != KVM_MR_DELETE && !(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) - WARN(kvm_s390_vm_stop_migration(kvm), - "Failed to stop migration mode"); + kvm_s390_vm_stop_migration(kvm); } if (change == KVM_MR_FLAGS_ONLY) diff --git a/arch/s390/kvm/s390/s390.c b/arch/s390/kvm/s390/s390.c index 18fb200f854a..04e4fa7dfd1d 100644 --- a/arch/s390/kvm/s390/s390.c +++ b/arch/s390/kvm/s390/s390.c @@ -1242,11 +1242,11 @@ static int kvm_s390_vm_start_migration(struct kvm *kvm) * Must be called with kvm->slots_arch_lock to avoid races with ourselves, * kvm_s390_vm_start_migration() and kvm_s390_get_cmma_bits(). */ -int kvm_s390_vm_stop_migration(struct kvm *kvm) +void kvm_s390_vm_stop_migration(struct kvm *kvm) { /* migration mode already disabled */ if (!kvm->arch.migration_mode) - return 0; + return; /* * Unset the flag and propagate to all vCPUs. From now on the cmma_d * bit will not be touched on any PGSTE. @@ -1255,7 +1255,7 @@ int kvm_s390_vm_stop_migration(struct kvm *kvm) */ WRITE_ONCE(kvm->arch.migration_mode, 0); if (!kvm->arch.use_cmma) - return 0; + return; kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION); /* Clear cmma_d on all existing PGSTEs and set cmma_dirty_pages to 0. */ @@ -1266,7 +1266,6 @@ int kvm_s390_vm_stop_migration(struct kvm *kvm) * cmma_dirty_pages is 0, and all existing PGSTEs have their cmma_d bit * set to 0. */ - return 0; } static int kvm_s390_vm_set_migration(struct kvm *kvm, @@ -1282,7 +1281,8 @@ static int kvm_s390_vm_set_migration(struct kvm *kvm, res = kvm_s390_vm_start_migration(kvm); break; case KVM_S390_VM_MIGRATION_STOP: - res = kvm_s390_vm_stop_migration(kvm); + kvm_s390_vm_stop_migration(kvm); + res = 0; break; default: break; diff --git a/arch/s390/kvm/s390/s390.h b/arch/s390/kvm/s390/s390.h index 9ed1af5583e3..45f397663d5f 100644 --- a/arch/s390/kvm/s390/s390.h +++ b/arch/s390/kvm/s390/s390.h @@ -474,7 +474,7 @@ int __kvm_s390_mprotect_many(struct gmap *gmap, gpa_t gpa, u8 npages, unsigned i bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu); void kvm_s390_update_cmma_dirty(struct kvm *kvm, const struct kvm_memory_slot *old); -int kvm_s390_vm_stop_migration(struct kvm *kvm); +void kvm_s390_vm_stop_migration(struct kvm *kvm); /* implemented in diag.c */ -- 2.56.0 ^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH v1 2/2] KVM: s390: make kvm_s390_vm_stop_migration() void 2026-11-20 7:53 ` [PATCH v1 2/2] KVM: s390: make kvm_s390_vm_stop_migration() void Gautam Gala @ 2026-11-20 7:58 ` sashiko-bot 2026-11-20 10:30 ` Andreas Grapentin 1 sibling, 0 replies; 6+ messages in thread From: sashiko-bot @ 2026-11-20 7:58 UTC (permalink / raw) To: Gautam Gala Cc: linux-s390, Heiko Carstens, Vasily Gorbik, Alexander Gordeev, Christian Borntraeger > kvm_s390_vm_stop_migration() returns 0 on all paths; it cannot fail. > Its caller in s390_kvm_mmu_prepare_memory_region() wraps it in > WARN(), producing a condition that is always false. The arm64 stub > returns -EINVAL but is unreachable because > kvm_s390_is_migration_mode() returns false there. > > Change the return type to void and drop the dead WARN(). The > KVM_S390_VM_MIGRATION_STOP attribute handler sets its result to 0 > explicitly; userspace behavior is unchanged since stopping > migration never failed. > > Signed-off-by: Gautam Gala <ggala@linux.ibm.com> Sashiko has reviewed this patch and found no issues. -- Sashiko AI review · https://sashiko.dev/#/patchset/20261120075345.116521-1-ggala@linux.ibm.com?part=2 ^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH v1 2/2] KVM: s390: make kvm_s390_vm_stop_migration() void 2026-11-20 7:53 ` [PATCH v1 2/2] KVM: s390: make kvm_s390_vm_stop_migration() void Gautam Gala 2026-11-20 7:58 ` sashiko-bot @ 2026-11-20 10:30 ` Andreas Grapentin 1 sibling, 0 replies; 6+ messages in thread From: Andreas Grapentin @ 2026-11-20 10:30 UTC (permalink / raw) To: Gautam Gala Cc: kvm, linux-s390, linux-kernel, Christian Borntraeger, Janosch Frank, Claudio Imbrenda, David Hildenbrand, Steffen Eiden, Marc Zyngier, Heiko Carstens, Vasily Gorbik, Alexander Gordeev, Sven Schnelle On Nov 20 26, Gautam Gala wrote: > kvm_s390_vm_stop_migration() returns 0 on all paths; it cannot fail. > Its caller in s390_kvm_mmu_prepare_memory_region() wraps it in > WARN(), producing a condition that is always false. The arm64 stub > returns -EINVAL but is unreachable because > kvm_s390_is_migration_mode() returns false there. > > Change the return type to void and drop the dead WARN(). The > KVM_S390_VM_MIGRATION_STOP attribute handler sets its result to 0 > explicitly; userspace behavior is unchanged since stopping > migration never failed. > > Signed-off-by: Gautam Gala <ggala@linux.ibm.com> > Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com> > --- > arch/s390/kvm/arm64/arm.h | 3 +-- > arch/s390/kvm/gmap/kvm_mmu.c | 3 +-- > arch/s390/kvm/s390/s390.c | 10 +++++----- > arch/s390/kvm/s390/s390.h | 2 +- > 4 files changed, 8 insertions(+), 10 deletions(-) > > diff --git a/arch/s390/kvm/arm64/arm.h b/arch/s390/kvm/arm64/arm.h > index 760230685ca1..e60801e4a467 100644 > --- a/arch/s390/kvm/arm64/arm.h > +++ b/arch/s390/kvm/arm64/arm.h > @@ -50,9 +50,8 @@ static __always_inline void kvm_s390_update_cmma_dirty(struct kvm *kvm, const st > } > > /* should never be called */ > -static __always_inline int kvm_s390_vm_stop_migration(struct kvm *kvm) > +static __always_inline void kvm_s390_vm_stop_migration(struct kvm *kvm) > { > - return -EINVAL; should we maybe add something like BUILD_BUG_ON(1) here, so it doesn't fail silently when the underlying assumption (will never be called because kvm_s390_is_migration_mode() always return false) stops holding for some reason? BUILD_BUG_ON(1) won't fire if the compiler can prove the path is never taken. Beyond that, this change makes sense to me on the arm64 side. Reviewed-by: Andreas Grapentin <gra@linux.ibm.com> ^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2026-11-20 10:30 UTC | newest] Thread overview: 6+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2026-11-20 7:53 [PATCH v1 0/2] KVM: s390: fix migration mode teardown in memslot handling Gautam Gala 2026-11-20 7:53 ` [PATCH v1 1/2] KVM: s390: reserve mmu cache before stopping migration mode Gautam Gala 2026-11-20 7:57 ` sashiko-bot 2026-11-20 7:53 ` [PATCH v1 2/2] KVM: s390: make kvm_s390_vm_stop_migration() void Gautam Gala 2026-11-20 7:58 ` sashiko-bot 2026-11-20 10:30 ` Andreas Grapentin
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.