On Mon, Jun 15, 2026 at 03:35:39PM +0200, Pratyush Yadav wrote: > On Sun, Jun 14 2026, Mike Rapoport wrote: > > On Fri, 05 Jun 2026 20:34:49 +0200, Pratyush Yadav <[email protected]> > > wrote: > > > >> First, it does not use mirrored memory for hugetlb. Mirrored memory is a > >> limited resource that is best saved for kernel data structures, not user > >> memory. > >> > >> Second, if the memory found overlaps with KHO scratch areas, it discards > >> the memory and retries. > > > > This sentence is somewhat hard to parse. > > Okay, let me retry: > > Second, if the free memory area found by memblock_find_in_range_node() > is a part of a KHO scratch area, the free area is not used. Allocation > is retried starting after the free area to ensure no hugepages come from > KHO scratch. > > Any better?
Yep :) > > > >> > >> > >> diff --git a/mm/memblock.c b/mm/memblock.c > >> index 6349c48154f4..131e54dd5d8d 100644 > >> --- a/mm/memblock.c > >> +++ b/mm/memblock.c > >> @@ -1756,6 +1761,69 @@ void * __init memblock_alloc_try_nid_raw( > >> [ ... skip 51 lines ... ] > >> + if (memblock_bottom_up()) > >> + start = addr + size; > >> + else > >> + start = addr - size; > >> + > >> + goto retry; > > > > Hmm, two goto retry don't seem nice :/ > > Although I can't see how to imporove it really. > > Dunno, looked easy enough to understand to me. > > > > Maybe add a helper for going the node fallback? > > There is a small downside. There will then be no way to know the > fallback was tried already, so if a retry is done because of scratch > overlap, the fallback needs to be done again. > > I don't think it should be too bad, so if you still prefer this then I > can do it. Let's keep it for now, maybe will find something smarter later. > -- > Regards, > Pratyush Yadav -- Sincerely yours, Mike.
