I have an SOC I could try to reproduce this on (maybe the same one?). Care to send me any scripts to help reproduce? -Joel On 07/21/2014 11:37 AM, Richard W.M. Jones wrote: > On Mon, Jul 21, 2014 at 01:45:15PM +0100, Richard W.M. Jones wrote: >> The kernel uses 64 KB pages. > I retested this with 4 KB pages (guest & host), and it leaks about > 3MB per guest in that configuration. > > So this *should* be perfectly reproducible for anyone using an > upstream kernel. You just have to start of the order of 1000 guests > to see significant amounts of memory disappearing (say 3+ GB). > > (Negative reproductions are also interesting, since it would indicate > a problem with one of the hardware enablement patches we are using) > > Rich. > _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
