Hi Drew, On Mon, Feb 27, 2017 at 06:54:59PM +0100, Andrew Jones wrote: > This series fixes four races. Two are easy to produce with a > kvm-unit-test test[1], but the other two would be quite hard. I > didn't even try to produce those. The two hard to produce races are > addressed by changing vcpu->arch.pause and vcpu->arch.power_off to > vcpu requests. The two easy to produce races are addressed in two > different ways: the first takes advantage of power_off having been > changed to a vcpu request, the second caches vcpu MPIDRs in order > to avoid extracting them from sys_regs. When introducing the MPIDR > cache we also introduce a new feature (userspace settable MPIDRs). > > Support for userspace settable MPIDRs was already posted once[2], > but rejected due to not having a use case. We have one now, which > is to satisfy QEMU's need for the MPDIR information very early, > before vcpu-init has even run. While the original posting author > wasn't me, I've taken authorship now, as I've changed the patch > substantially. If anybody disagrees with that, then feel free to > suggest alternatives. The QEMU counterpart has been posted[3]. > > This series is based on Radim's recent posting[4] that improves > the vcpu-request framework. I've tested the series on a couple > AArch64 platforms and compile-tested the arm bits. So as we discussed during lunch today, it would be awesome if you could split this up into a series that fixes the identified races and a separate series for the introduction of supporting setting the MPIDR from user space. Thanks a lot! -Christoffer _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
