The HugeTLB address ranges are linearly scanned during fork, unmap and remap operations, and the linear scan can skip to the end of range mapped by the page table page if hitting a non-present entry, which can help to speed linear scanning of the HugeTLB address ranges. So hugetlb_mask_last_hp() is introduced to help to update the address in the loop of HugeTLB linear scanning with getting the last huge page mapped by the associated page table page[1], when a non-present entry is encountered. Considering ARM64 specific cont-pte/pmd size HugeTLB, this patch implemented an ARM64 specific hugetlb_mask_last_hp() to help this case. [1] https://lore.kernel.org/linux-mm/20220527225849.284839-1-mike.kravetz@xxxxxxxxxx/ Signed-off-by: Baolin Wang <baolin.wang@xxxxxxxxxxxxxxxxx> --- Note: this patch is based on the series: "hugetlb: speed up linear address scanning" from Mike. Mike, please fold it into your series. Thanks. --- arch/arm64/mm/hugetlbpage.c | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+) diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c index e2a5ec9..958935c 100644 --- a/arch/arm64/mm/hugetlbpage.c +++ b/arch/arm64/mm/hugetlbpage.c @@ -368,6 +368,26 @@ pte_t *huge_pte_offset(struct mm_struct *mm, return NULL; } +unsigned long hugetlb_mask_last_hp(struct hstate *h) +{ + unsigned long hp_size = huge_page_size(h); + + switch (hp_size) { + case PUD_SIZE: + return PGDIR_SIZE - PUD_SIZE; + case CONT_PMD_SIZE: + return PUD_SIZE - CONT_PMD_SIZE; + case PMD_SIZE: + return PUD_SIZE - PMD_SIZE; + case CONT_PTE_SIZE: + return PMD_SIZE - CONT_PTE_SIZE; + default: + break; + } + + return ~0UL; +} + pte_t arch_make_huge_pte(pte_t entry, unsigned int shift, vm_flags_t flags) { size_t pagesize = 1UL << shift; -- 1.8.3.1
