mm: readahead: change ra size for random read
Subsystems (2)
RFC 1/1 mm: readahead: change ra size for random read
Set a smaller ra size for random read than contiguous one. It is deemed random if the lower-level device is unable to meet the read pattern even with its IO capability doubled.
Signed-off-by Hillf Danton <[email protected]>
---
mm/readahead.c +29 -2
@@ -388,6 +388,7 @@ ondemand_readahead(struct address_spaceunsigned long max_pages = ra->ra_pages;unsigned long add_pages;pgoff_t prev_offset; +bool random = true;/* * If the request exceeds the readahead window, allow the read to
@@ -399,8 +400,34 @@ ondemand_readahead(struct address_space/* * start of file */ -if (!offset) - goto initial_readahead; +if (!offset) { +fill_ra: + ra->start = offset; + ra->size = random ? +min (bdi->io_pages, bdi->ra_pages) : +max (bdi->io_pages, bdi->ra_pages); + + ra->async_size = ra->size > req_size ? + ra->size - req_size : ra->size; + +return ra_submit (ra, mapping, filp); + }else { +unsigned long leap; + +if (offset > ra->start) + leap = offset - ra->start; +else + leap = ra->start - offset; + +/* +* anything other than page cache cannot help if it is +* too great a leap for the lower-level device to back +* up so feel free to put ra into fire +*/ + random = leap >max (bdi->io_pages, bdi->ra_pages) *2 ; + + goto fill_ra; + }/* * It's the expected callback offset, assume sequential access.