On Fri, May 11, 2018 at 03:20:11PM +0100, Andre Przywara wrote: > Jan recently reported lockdep complaints regarding various locks in our > VGIC emulation [1][2]. > This boiled down to two separate issues: > - When promoting the vgic_irq->irq_lock to require IRQs being disabled, > we forgot to amend some instances of this lock on the way. Also this > needs to be applied to dependent locks as well. The first two patches > fix that. The patch split is designed to simplify backporting. > Those patches have been posted before, I am resending them as part > of this series. > - Calling kvm_read_guest() requires us to be inside an SRCU critical > section. On some architectures we are always in it when handling VCPU > exits, but on ARM we need to lock it individually. Patches 3 and 4 > fix that, the split is again made to ease backporting. > Each of the hunks fix an indiviual commit, but I refrained from > splitting this down into eight patches just to put proper Fixes: tags > on it. Eventually those commits are part of one out of two series, I put > the respective kernel release version as a tag to the Cc: stable line. > > I couldn't reproduce the full lockdep splat on my setup, but at least > could show one instance and prove that these patches fixes that. > > For the series: Acked-by: Christoffer Dall <christoffer.dall@xxxxxxx> Thanks for fixing this, -Christoffer _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
