| From: Li Zhe <lizhe.67@bytedance.com> |
| Subject: hugetlb: increase number of reserving hugepages via cmdline |
| Date: Fri, 19 Sep 2025 17:23:53 +0800 |
| |
| Commit 79359d6d24df ("hugetlb: perform vmemmap optimization on a list of |
| pages") batches the submission of HugeTLB vmemmap optimization (HVO) |
| during hugepage reservation. With HVO enabled, hugepages obtained from |
| the buddy allocator are not submitted for optimization and their |
| struct-page memory is therefore not released—until the entire |
| reservation request has been satisfied. As a result, any struct-page |
| memory freed in the course of the allocation cannot be reused for the |
| ongoing reservation, artificially limiting the number of huge pages that |
| can ultimately be provided. |
| |
| As commit b1222550fbf7 ("mm/hugetlb: do pre-HVO for bootmem allocated |
| pages") already applies early HVO to bootmem-allocated huge pages, this |
| patch extends the same benefit to non-bootmem pages by incrementally |
| submitting them for HVO as they are allocated, thereby returning |
| struct-page memory to the buddy allocator in real time. The change raises |
| the maximum 2 MiB hugepage reservation from just under 376 GB to more than |
| 381 GB on a 384 GB x86 VM. |
| |
| Link: https://lkml.kernel.org/r/20250919092353.41671-1-lizhe.67@bytedance.com |
| Signed-off-by: Li Zhe <lizhe.67@bytedance.com> |
| Cc: David Hildenbrand <david@redhat.com> |
| Cc: Muchun Song <muchun.song@linux.dev> |
| Cc: Oscar Salvador <osalvador@suse.de> |
| Signed-off-by: Andrew Morton <akpm@linux-foundation.org> |
| --- |
| |
| mm/hugetlb.c | 9 ++++++++- |
| 1 file changed, 8 insertions(+), 1 deletion(-) |
| |
| --- a/mm/hugetlb.c~hugetlb-increase-number-of-reserving-hugepages-via-cmdline |
| +++ a/mm/hugetlb.c |
| @@ -3538,7 +3538,14 @@ static void __init hugetlb_pages_alloc_b |
| nodes_clear(node_alloc_noretry); |
| |
| for (i = 0; i < num; ++i) { |
| - struct folio *folio = alloc_pool_huge_folio(h, &node_states[N_MEMORY], |
| + struct folio *folio; |
| + |
| + if (hugetlb_vmemmap_optimizable_size(h) && |
| + (si_mem_available() == 0) && !list_empty(&folio_list)) { |
| + prep_and_add_allocated_folios(h, &folio_list); |
| + INIT_LIST_HEAD(&folio_list); |
| + } |
| + folio = alloc_pool_huge_folio(h, &node_states[N_MEMORY], |
| &node_alloc_noretry, &next_node); |
| if (!folio) |
| break; |
| _ |
