On 4/25/26 14:00, Zi Yan wrote: > On 17 Apr 2026, at 22:44, Zi Yan wrote: > >> Replace it with a check on the max folio order of the file's address space >> mapping, making sure PMD THP is supported. Also remove the read-only fd >> check, since collapse_file() now makes sure all to-be-collapsed folios are >> clean and the created PMD file THP can be handled by FSes properly. >> >> Signed-off-by: Zi Yan <[email protected]> >> --- >> mm/huge_memory.c | 8 ++++---- >> 1 file changed, 4 insertions(+), 4 deletions(-) >> >> diff --git a/mm/huge_memory.c b/mm/huge_memory.c >> index 970e077019b7..7e9cf8c0985f 100644 >> --- a/mm/huge_memory.c >> +++ b/mm/huge_memory.c >> @@ -86,9 +86,6 @@ static inline bool file_thp_enabled(struct vm_area_struct >> *vma) >> { >> struct inode *inode; >> >> - if (!IS_ENABLED(CONFIG_READ_ONLY_THP_FOR_FS)) >> - return false; >> - >> if (!vma->vm_file) >> return false; >> >> @@ -97,7 +94,10 @@ static inline bool file_thp_enabled(struct vm_area_struct >> *vma) >> if (IS_ANON_FILE(inode)) >> return false; >> >> - return !inode_is_open_for_write(inode) && S_ISREG(inode->i_mode); > > Hi Matthew, David, and Lorenzo, > > After some discussions on irc, I feel that we probably should not allow > read-write fd for PMD THP collapse at the moment. Combining with the > filemap_flush() under folio_dirty() check from collapse_file(), khugepaged > would become a kwritebackd that scans pagecache folios and writes them back. > If we limit it to read-only fds, at least khugepaged would only write > back once for the pagecache folios from these fds.
Makes sense. The comment above the filemap_flush() is valuable :) > > I am planning to restore inode_is_open_for_write() check in the next version. > Let me know your thoughts. That would be better. Do we plan on handling races with concurrent writable-opening? Or could we simply gate the filemap_flush() by a inode_is_open_for_write(), to not have it turn into a kwritebackd()? -- Cheers, David
