On Tue 01-08-23 14:58:29, Aneesh Kumar K V wrote: > On 8/1/23 2:28 PM, Michal Hocko wrote: > > On Tue 01-08-23 10:11:16, Aneesh Kumar K.V wrote: > >> Allow updating memmap_on_memory mode after the kernel boot. Memory > >> hotplug done after the mode update will use the new mmemap_on_memory > >> value. > > > > Well, this is a user space kABI extension and as such you should spend > > more words about the usecase. Why we could live with this static and now > > need dynamic? > > > > This enables easy testing of memmap_on_memory feature without a kernel reboot.
Testing alone is rather weak argument to be honest. > I also expect people wanting to use that when they find dax kmem memory online > failing because of struct page allocation failures[1]. User could reboot back > with > memmap_on_memory=y kernel parameter. But being able to enable it via sysfs > makes > the feature much more useful. Sure it can be useful but that holds for any feature, right. The main question is whether this is worth maintaing. The current implementation seems rather trivial which is an argument to have it but are there any risks long term? Have you evaluated a potential long term maintenance cost? There is no easy way to go back and disable it later on without breaking some userspace. All that should be in the changelog! -- Michal Hocko SUSE Labs
