Re: [GIT PULL 1/6] KVM: s390: fix ESSA addressing exception injection in migration mode

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [Low] KVM: s390: commit message states -PGM_ADDRESSING instead of
  PGM_ADDRESSING
It also noticed 1 related pre-existing issue this patch does not
introduce:
- [High] KVM: s390: missing CMMA init and page table allocation in
  handle_essa() [1]
--

commit ee014980cbe371ebff00ee828f1f5961b34040ff
Author: Gautam Gala <ggala@linux.ibm.com>

KVM: s390: fix ESSA addressing exception injection in migration mode

This commit updates dat_perform_essa() to pass DAT_WALK_ANY to
dat_entry_walk() and return PGM_ADDRESSING on walk failure, so __do_essa()
injects an addressing exception instead of silently dropping it.

>     Return -PGM_ADDRESSING on a walk failure and inject the program
>     interrupt in __do_essa() for the -PGM_ADDRESSING return value.

[Severity: Low]
This isn't a bug, but the commit message says -PGM_ADDRESSING whereas the
code returns and checks positive PGM_ADDRESSING.

> diff --git a/arch/s390/kvm/s390/priv.c b/arch/s390/kvm/s390/priv.c
> index 856939acf012f..b252c812dcbe7 100644
> --- a/arch/s390/kvm/s390/priv.c
> +++ b/arch/s390/kvm/s390/priv.c
> @@ -1169,6 +1169,8 @@ static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
>  	entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
>  
>  	nappended = dat_perform_essa(vcpu->arch.gmap->asce, gfn, orc, &state, &dirtied);
> +	if (nappended == PGM_ADDRESSING)
> +		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
>  	vcpu->run->s.regs.gprs[r1] = state.val;
>  	if (nappended < 0)
>  		return 0;

[Severity: High]
This problem wasn't introduced by this patch, but can an initial ESSA
intercept arriving after migration has already started reach __do_essa()
without GMAP_FLAG_USES_CMM set or guest page tables allocated [1]?

In handle_essa(), setting GMAP_FLAG_USES_CMM and calling
gmap_allocate_all_page_tables_best_effort() only happen in the
!migration_mode branch:

arch/s390/kvm/s390/priv.c:handle_essa() {
	...
	if (!READ_ONCE(vcpu->kvm->arch.migration_mode)) {
		...
		if (!uses_cmm(vcpu->arch.gmap)) {
			set_bit(GMAP_FLAG_USES_CMM, &vcpu->arch.gmap->flags);
			gmap_allocate_all_page_tables_best_effort(vcpu->arch.gmap);
		}
		...
	} else {
		scoped_guard(mutex, &vcpu->kvm->slots_arch_lock)
			scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
				i = __do_essa(vcpu, orc);
	...
}

If the guest executes its first ESSA instruction while migration_mode is
already active, dat_perform_essa() walks the gmap without allocating missing
page tables, returning an exception state and dropping the PGSTE update for
any unallocated segment.  In addition, because GMAP_FLAG_USES_CMM remains
unset on kvm->arch.gmap, kvm_s390_get_cmma_bits() sees
!uses_cmm(kvm->arch.gmap) and returns zeroed data, losing dirtied guest CMMA
attributes during live migration.

[1] https://sashiko.dev/#/bug/linux-20a07467-1a8a-41bb-8117-0f046fed9112

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106164958.431978-2-borntraeger@linux.ibm.com?part=1