On 14.10.20 16:15, Michal Privoznik wrote: > On 10/14/20 4:07 PM, David Hildenbrand wrote: > >> >> >> Note: prealloc=yes is a bad choice in this environment. It >> contradicts memory overcommit - what we want to optimize with >> free page reporting. You allocate all VM memory to throw it away >> once the guest is up. Is there an option to turn this of with >> hugetlbfs? I hope so. > > Is this something that should be addressed on qemu side? Since I'm > writing a follow up patches for libvirt I can forbid this combination in > libvirt. You mean, bailing out in QEMU if "prealloc=yes" is used along free page reporting? Hmm, not sure if that's the right approach ... we would also have to check for any DIMMs we hotplug and things like that. It's definitely a setup issue to me (where we could try to warn the user, but QEMU warnings tend to get ignored by everybody ...). It's also going to be "not what you want" once we have virtio-mem support in libvirt. (https://virtio-mem.gitlab.io/) But there, it's easier to check+warn. You only want "prealloc=no" for the memory device assigned to a virtio-mem device, not for all system RAM. Thanks! -- Thanks, David / dhildenb
