On Tue, 14 Jan 2014 16:25:10 -0800 (PST) David Rientjes <[email protected]> wrote:
> On Tue, 14 Jan 2014, Andrew Morton wrote: > > > This is all a bit nasty, isn't it? THP goes and alters min_free_kbytes > > to improve its own reliability, but min_free_kbytes is also > > user-modifiable. And over many years we have trained a *lot* of users > > to alter min_free_kbytes. Often to prevent nasty page allocation > > failure warnings from net drivers. > > > > I can vouch for kernel logs that are spammed with tons of net page > allocation failure warnings, in fact we're going through and adding > __GFP_NOWARN to most of these. > > > So there are probably quite a lot of people out there who are manually > > rubbing out THP's efforts. And there may also be people who are > > setting min_free_kbytes to a value which is unnecessarily high for more > > recent kernels. > > > > Indeed, we have initscripts that modified min_free_kbytes before thp was > introduced but luckily they were comparing their newly computed value to > the existing value and only writing if the new value is greater. > Hopefully most users are doing the same thing. I've been waiting 10+ years for us to decide to delete that warning due to the false positives. Hasn't happened yet, and the warning does find bugs/issues/misconfigurations/etc. But I do worry this has led to users unnecessarily increasing min_free_kbytes just to shut the warnings up. Net result: they have less memory available for cache, etc. > Would it be overkill to save the kernel default both with and without thp > and then doing a WARN_ON_ONCE() if a user-written value is ever less? Well, min_free_kbytes is a userspace thing, not a kernel thing - maybe THP shouldn't be dinking with it. What effect is THP trying to achieve and can we achieve it by other/better means? -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
