On Mon, 4 Dec 2023 20:57:33 +0100 Dmytro Maluka <dmaluka@xxxxxxxxxxxx> wrote: > On Mon, Dec 04, 2023 at 11:13:01AM -0800, Andrew Morton wrote: > > On Mon, 4 Dec 2023 17:32:54 +0100 Dmytro Maluka <dmaluka@xxxxxxxxxxxx> wrote: > > > > > Add an option to disable transparent hugepages by default, in line with > > > the existing transparent_hugepage=never command line setting. > > > > > > Rationale: khugepaged has its own non-negligible memory cost even if it > > > is not used by any applications, since it bumps up vm.min_free_kbytes to > > > its own required minimum in set_recommended_min_free_kbytes(). For > > > example, on a machine with 4GB RAM, with 3 mm zones and pageblock_order > > > == MAX_ORDER, starting khugepaged causes vm.min_free_kbytes increase > > > from 8MB to 132MB. > > > > > > So if we use THP on machines with e.g. >=8GB of memory for better > > > performance, but avoid using it on lower-memory machines to avoid its > > > memory overhead, then for the same reason we also want to avoid even > > > starting khugepaged on those <8GB machines. So with > > > CONFIG_TRANSPARENT_HUGEPAGE_NEVER we can use the same kernel image on > > > both >=8GB and <8GB machines, with THP support enabled but khugepaged > > > not started by default. The userspace can then decide to enable THP > > > (i.e. start khugepaged) via sysfs if needed, based on the total amount > > > of memory. > > > > > > This could also be achieved with the existing transparent_hugepage=never > > > setting in the kernel command line instead. But it seems cleaner to > > > avoid tweaking the command line for such a basic setting. > > > > > > P.S. I see that CONFIG_TRANSPARENT_HUGEPAGE_NEVER was already proposed > > > in the past [1] but without an explanation of the purpose. > > > > > > ... > > > > > > --- a/mm/Kconfig > > > +++ b/mm/Kconfig > > > @@ -859,6 +859,12 @@ choice > > > madvise(MADV_HUGEPAGE) but it won't risk to increase the > > > memory footprint of applications without a guaranteed > > > benefit. > > > + > > > + config TRANSPARENT_HUGEPAGE_NEVER > > > + bool "never" > > > + help > > > + Disabling Transparent Hugepage by default. It can still be > > > > s/Disabling/Disable/ > > It is in line with the descriptions of TRANSPARENT_HUGEPAGE_ALWAYS and > TRANSPARENT_HUGEPAGE_MADVISE: "Enabling Transparent Hugepage ..." Those are incorrect also. > > > + enabled at runtime via sysfs. > > > endchoice > > > > The patch adds the config option but doesn't use it? > > I should have been more precise: it is not a new option but a new choice > for CONFIG_TRANSPARENT_HUGEPAGE, in addition to the existing ALWAYS and > MADVISE choices. In mm/huge_memory.c in the declaration of the > transparent_hugepage_flags variable, if either ALWAYS or MADVISE is > chosen, transparent_hugepage_flags is initialized with such a value > that makes khugepaged being started by default during bootup. This patch > allows enabling CONFIG_TRANSPARENT_HUGEPAGE without setting either > ALWAYS or MADVISE, so that transparent_hugepage_flags is initialized > with such a value that khugepaged is not started by default. OK, thanks.
