When readahead pulls in all the remaining pages for a file, setting the readahead bit is counter productive. The async readahead it would trigger would almost certainly be a no-op. Additionally, for mmap'd file IO, the readahead bit limits the fault around [0], causing an extra minor fault when the page is accessed. This was discovered when looking at /sys/kernel/tracing/events/readahead traces for a simple program. With the patch applied, fewer page_cache_ra_unbounded calls are observed. [1] do_fault_around calls filemap_map_pages, which finds eligible pages by calling next_uptodate_folio [2]. next_uptodate_folio skips pages with PG_readahead set [3]. Link: https://github.com/torvalds/linux/blob/v7.0/mm/filemap.c#L3921-L3939 [2] Link: https://github.com/torvalds/linux/blob/v7.0/mm/filemap.c#L3721-L3722 [3] Cc: Kalesh Singh Cc: Suren Baghdasaryan Signed-off-by: Frederick Mayle --- mm/readahead.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) diff --git a/mm/readahead.c b/mm/readahead.c index 8c12b63ccd4a..784a3bde82be 100644 --- a/mm/readahead.c +++ b/mm/readahead.c @@ -329,8 +329,11 @@ static void do_page_cache_ra(struct readahead_control *ractl, if (index > end_index) return; /* Don't read past the page containing the last byte of the file */ - if (nr_to_read > end_index - index) + if (nr_to_read > end_index - index) { nr_to_read = end_index - index + 1; + /* We've reached the end, so don't set a readahead marker. */ + lookahead_size = 0; + } filemap_invalidate_lock_shared(mapping); page_cache_ra_unbounded(ractl, nr_to_read, lookahead_size); @@ -474,7 +477,7 @@ void page_cache_ra_order(struct readahead_control *ractl, pgoff_t index = start; unsigned int min_order = mapping_min_folio_order(mapping); pgoff_t limit; - pgoff_t mark = index + ra->size - ra->async_size; + pgoff_t mark; unsigned int nofs; int err = 0; gfp_t gfp = readahead_gfp_mask(mapping); @@ -488,7 +491,13 @@ void page_cache_ra_order(struct readahead_control *ractl, limit = (i_size_read(mapping->host) - 1) >> PAGE_SHIFT; limit = min(limit, ractl->_max_index); - limit = min(limit, index + ra->size - 1); + if (limit > index + ra->size - 1) { + limit = index + ra->size - 1; + mark = index + ra->size - ra->async_size; + } else { + /* We've reached the end, so don't set a readahead marker. */ + mark = ULONG_MAX; + } new_order = min(mapping_max_folio_order(mapping), new_order); new_order = min_t(unsigned int, new_order, ilog2(ra->size)); base-commit: 2d565cbaafd43b10b75da56e43e5db9852a56afd -- 2.54.0.563.g4f69b47b94-goog