On 10/15/2019 04:01 PM, Michal Hocko wrote: > On Tue 15-10-19 15:27:46, Anshuman Khandual wrote: >> >> >> On 10/14/2019 10:22 PM, Mike Kravetz wrote: >>> On 10/14/19 6:08 AM, Anshuman Khandual wrote: >>>> On 10/14/2019 06:30 PM, Michal Hocko wrote: >>>>> >>>>> OK, thanks for the clarification. This all means that this patch is not >>>>> the right approach. If you need a more generic alloc_contig_range then >>>>> add it to page_alloc.c and make it completely independent on the hugetlb >>>>> config and the code. Hugetlb allocator can reuse that helper. >>> >>> Should we revisit this previous attempt at such an interface? >>> >>> https://lkml.org/lkml/2018/4/16/1072 >>> >>> This looks like another use case. >>> >> >> The current proposal [v6] does not go far enough to unify all callers >> of alloc_contig_range() looking for contiguous pages of certain size, >> but instead it just tries not to duplicate HugeTLB gigantic allocation >> code in the test case for it's purpose. > > Please go and extract that functionality to a commonly usable helper. > Dependency on hugetlb is just ugly as hell. I did that in V6 and posted some time back. Could you please review that and let me know if anything needs to be improved. - Anshuman
