On 12.05.22 15:59, Muchun Song wrote: > On Thu, May 12, 2022 at 03:04:57PM +0200, David Hildenbrand wrote: >> On 12.05.22 14:50, Muchun Song wrote: >>> On Thu, May 12, 2022 at 09:36:15AM +0200, David Hildenbrand wrote: >>>> On 09.05.22 08:27, Muchun Song wrote: >>>>> Optimizing HugeTLB vmemmap pages is not compatible with allocating memmap on >>>>> hot added memory. If "hugetlb_free_vmemmap=on" and >>>>> memory_hotplug.memmap_on_memory" are both passed on the kernel command line, >>>>> optimizing hugetlb pages takes precedence. >>>> >>>> Why? >>>> >>> >>> Because both two features are not compatible since hugetlb_free_vmemmap cannot >>> optimize the vmemmap pages allocated from alternative allocator (when >>> memory_hotplug.memmap_on_memory=1). So when the feature of hugetlb_free_vmemmap >>> is introduced, I made hugetlb_free_vmemmap take precedence. BTW, I have a plan >>> to remove this restriction, I'll post it out ASAP. >> >> I was asking why vmemmap optimization should take precedence. >> memmap_on_memory makes it more likely to succeed memory hotplug in >> close-to-OOM situations -- which is IMHO more important than a vmemmap >> optimization. >> > > I thought the users who enable hugetlb_free_vmemmap value memory > savings more, so I made a decision in commit 4bab4964a59f. Seems > I made a bad decision from your description. Depends on the perspective I guess. :) > >> But anyhow, the proper approach should most probably be to simply not >> mess with the vmemmap if we stumble over a vmemmap that's special due to >> memmap_on_memory. I assume that's what you're talking about sending out. >> > > I mean I want to have hugetlb_vmemmap.c do the check whether the section > which the HugeTLB pages belong to can be optimized instead of making > hugetlb_free_vmemmap take precedence. E.g. If the section's vmemmap pages > are allocated from the added memory block itself, hugetlb_free_vmemmap will > refuse to optimize the vmemmap, otherwise, do the optimization. Then > both kernel parameters are compatible. I have done those patches, but > haven't send them out. Yeah, that's exactly what I thought. How complicated are they? If they are easy, can we just avoid this patch here and do it "properly"? :) -- Thanks, David / dhildenb
