On Fri, Apr 28, 2023 at 9:14 AM Suren Baghdasaryan <surenb@xxxxxxxxxx> wrote: > > On Fri, Apr 28, 2023 at 9:08 AM Linus Torvalds > <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: > > > > On Fri, Apr 28, 2023 at 9:03 AM Suren Baghdasaryan <surenb@xxxxxxxxxx> wrote: > > > > > > I wanted these stats enabled by default to be able to identify > > > possible pathological cases and to also let users disable them if they > > > can't tolerate even a small overhead in the pagefault path. Should I > > > document this reasoning for the config option? > > > > You should document what the stats actually count (at a high enough > > level for a user to understand), and why anybody would want to keep > > them on. > > > > Honestly, 99% of the time, these are things that *developers* think > > they might want, but that nobody else will ever ever use. > > > > Really, ask yourself if a normal user would ever look at them? > > > > Now, ask yourself whether this might be something that a cloud > > provider would want to look at to gather statistics. > > > > And if it's the latter case, then it should be "default n", because > > the default should be for the people who DO NOT KNOW, AND DO NOT CARE. > > > > The cloud provider will be using a custom config anyway. The default > > is irrelevant for that use. The use that *matters* is literally the > > clueless end user who I bet will never look at these numbers, and will > > never be asked for them. > > Ok, sounds like this should be 'default n'. I'll prepare a patch. Thanks! Should I send a replacement patch for "mm: introduce per-VMA lock statistics" or a followup patch fixing it? > > > > > Linus
