On Sat, Jul 26, 2014 at 11:08:06PM +0900, Jungseok Lee wrote: > On Jul 23, 2014, at 9:40 PM, Richard W.M. Jones wrote: > > FWIW I have reproduced this problem with the Fedora Rawhide kernel, > > which is very close to upstream (it still has some hardware enablement > > patches). > > > > 3.16.0-0.rc6.git1.1.efirtcfix1.fc22.aarch64 > > > > Page size host & guest is 4K. > > > > The loss is around 2.2 MB per guest. > > > > Rich. > > Hi, Richard > > Could you try [arm/arm64: KVM: Fix and refactor unmap_range] patch staged > on queue branch of [1]? Since Mark Salter points out unmap_range recently, > it can help you figure out the issue. > > [1]: git://git.kernel.org/pub/scm/linux/kernel/git/kvmarm/kvmarm.git Indeed that patch *does* fix the leak. I was able to start 1000 VMs and afterwards free memory remains pretty much unchanged. Be great to have that patch upstream, since although Mark called it a "small" leak it quite badly affects libguestfs where we routinely run 1000s of small VMs. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com virt-top is 'top' for virtual machines. Tiny program with many powerful monitoring features, net stats, disk stats, logging, etc. http://people.redhat.com/~rjones/virt-top _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
