| Message ID | 20180110060439.21822-1-david@gibson.dropbear.id.au (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Jan 10, 2018 at 05:04:39PM +1100, David Gibson wrote: > The KVM_PPC_ALLOCATE_HTAB ioctl(), implemented by kvmppc_alloc_reset_hpt() > is supposed to completely clear and reset a guest's Hashed Page Table (HPT) > allocating or re-allocating it if necessary. > > In the case where an HPT of the right size already exists and it just > zeroes it, it forces a TLB flush on all guest CPUs, to remove any stale TLB > entries loaded from the old HPT. > > However, that situation can arise when the HPT is resizing as well - or > even when switching from an RPT to HPT - so those cases need a TLB flush as > well. > > So, move the TLB flush to trigger in all cases except for errors. > > Signed-off-by: David Gibson <david@gibson.dropbear.id.au> > --- > arch/powerpc/kvm/book3s_64_mmu_hv.c | 6 ++++-- > 1 file changed, 4 insertions(+), 2 deletions(-) > > Paul, this is based on Paolo's KVM tree, but it should apply without > modification to pretty much any vaguely current tree. It's a pretty > nasty bug - the case we've found hitting it in the wild is a bit > esoteric, but it could in theory affect other situations as well. > > Please apply ASAP, and should probably be queued for the stable > branches as well. Thanks, applied to my kvm-ppc-fixes branch, and I added cc: stable@vger.kernel.org. Paul.
diff --git a/arch/powerpc/kvm/book3s_64_mmu_hv.c b/arch/powerpc/kvm/book3s_64_mmu_hv.c index 966097232d21..51a275cc8a4d 100644 --- a/arch/powerpc/kvm/book3s_64_mmu_hv.c +++ b/arch/powerpc/kvm/book3s_64_mmu_hv.c @@ -159,8 +159,6 @@ long kvmppc_alloc_reset_hpt(struct kvm *kvm, int order) * Reset all the reverse-mapping chains for all memslots */ kvmppc_rmap_reset(kvm); - /* Ensure that each vcpu will flush its TLB on next entry. */ - cpumask_setall(&kvm->arch.need_tlb_flush); err = 0; goto out; } @@ -176,6 +174,10 @@ long kvmppc_alloc_reset_hpt(struct kvm *kvm, int order) kvmppc_set_hpt(kvm, &info); out: + if (err == 0) + /* Ensure that each vcpu will flush its TLB on next entry. */ + cpumask_setall(&kvm->arch.need_tlb_flush); + mutex_unlock(&kvm->lock); return err; }
The KVM_PPC_ALLOCATE_HTAB ioctl(), implemented by kvmppc_alloc_reset_hpt() is supposed to completely clear and reset a guest's Hashed Page Table (HPT) allocating or re-allocating it if necessary. In the case where an HPT of the right size already exists and it just zeroes it, it forces a TLB flush on all guest CPUs, to remove any stale TLB entries loaded from the old HPT. However, that situation can arise when the HPT is resizing as well - or even when switching from an RPT to HPT - so those cases need a TLB flush as well. So, move the TLB flush to trigger in all cases except for errors. Signed-off-by: David Gibson <david@gibson.dropbear.id.au> --- arch/powerpc/kvm/book3s_64_mmu_hv.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) Paul, this is based on Paolo's KVM tree, but it should apply without modification to pretty much any vaguely current tree. It's a pretty nasty bug - the case we've found hitting it in the wild is a bit esoteric, but it could in theory affect other situations as well. Please apply ASAP, and should probably be queued for the stable branches as well.