重构MADV_COLD/PAGEOUT页表遍历 · v2

把 MADV_COLD/PAGEOUT 共享的 PMD 回调由 PTE 实现名改为 madvise_lru_pmd_entry()，为后续拆分 PMD/PTE 路径做准备，无功能变化。

The huge-PMD and PTE paths duplicate folio filtering, reference clearing, deactivation and pageout isolation. Keeping both copies synchronized obscures the page-table-specific control flow. Factor th

A non-present huge PMD must contain a software leaf type supported at PMD level. The open-coded check names the currently supported migration and device-private entries instead of expressing that inva

Both the huge-PMD and PTE paths of madvise_lru_pmd_entry() split a folio they hold locked and referenced, then drop the lock and the reference, with the same three calls open-coded at each site. Move

The huge-PMD branch of madvise_lru_pmd_entry() decides inline whether the PMD maps a folio worth advising: it checks that the PMD is present, looks up the folio and applies the shared filters, all bet

madvise_lru_pmd_entry() still handles a huge PMD inline, and jumps into its PTE walk with a goto when the huge-PMD folio was split. The PMD lock, the split and the reclaim of the advised folio all sh

Each pass of the PTE loop validates the entry, batches the PTEs that map one large folio, decides whether a partly mapped large folio must be split, and ages or isolates the folio. The split path als

The PTE loop’s scan is broken up by the two places that drop the PTE lock: the periodic reschedule and the large-folio split. Move the scan into madvise_lru_pte_range_locked(), which runs entirely un

madvise_lru_pmd_entry() maps and unmaps the PTE table through a restart label and an out label, so finding where the PTE lock is held means following every goto. Move the PTE path into madvise_lru_pt

madvise_cold() and madvise_pageout() each open-code the same mmu_gather and page-walk sequence, differing only in the pageout flag they pass to the walk. Move that sequence into madvise_lru_vma_range