forked from torvalds/linux
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
PM: hibernate: don't store zero pages in the image file.
On ChromeOS we've observed a considerable number of in-use pages filled with zeros. Today with hibernate it's entirely possible that saveable pages are just zero filled. Since we're already copying pages word-by-word in do_copy_page it becomes almost free to determine if a page was completely filled with zeros. This change introduces a new bitmap which will track these zero pages. If a page is zero it will not be included in the saved image, instead to track these zero pages in the image file we will introduce a new flag which we will set on the packed PFN list. When reading back in the image file we will detect these zero page PFNs and rebuild the zero page bitmap. When the image is being loaded through calls to snapshot_write_next if we encounter a zero page we will silently memset it to 0 and then continue on to the next page. Given the implementation in snapshot_read_next and snapshot_write_next this change will be transparent to non-compressed, compressed, and swsusp modes of operation. To provide some concrete numbers from simple ad-hoc testing, on a device which was lightly in use we saw that: PM: hibernation: Image created (964408 pages copied, 548304 zero pages) Of the approximately 6.2GB of saveable pages 2.2GB (36%) were just zero filled and could be tracked entirely within the packed PFN list. The savings would obviously be much lower for lzo compressed images, but even in the case of compression not copying pages across to the compression threads will still speed things up. It's also possible that we would see better overall compression ratios as larger regions of "real data" would improve the compressibility. Finally, such an approach could dramatically improve swsusp performance as each one of those zero pages requires a write syscall to reload, by handling it as part of the packed PFN list we're able to fully avoid that. Signed-off-by: Brian Geffon <bgeffon@google.com>
- Loading branch information
1 parent
1b929c0
commit 658520c
Showing
1 changed file
with
97 additions
and
30 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters