drivers:xen-selfballoon:reset 'frontswap_inertia_counter' after frontswap_shrink
When I looked at this issue https://lkml.org/lkml/2013/11/21/158, I found that frontswap_selfshrink() doesn't work as expected sometimes. Pages are continuously added to frontswap and gotten back soon. It's a waste of cpu time and increases the memory pressue of Guest OS. Take an example. First time in frontswap_selfshrink(): 1. last_frontswap_pages = cur_frontswap_pages = 0 2. cur_frontswap_pages = frontswap_curr_pages() = 100 When 'frontswap_inertia_counter' decreased to 0: 1. last_frontswap_pages = cur_frontswap_pages = 100 2. cur_frontswap_pages = frontswap_curr_pages() = 100 3. call frontswap_shrink() and let's assumption that 10 pages are gotten back from frontswap. 4. now frontswap_curr_pages() is 90. If then memory is not enough in Guest OS and 9 more pages(smaller than gotten back) added to frontswap. Now frontswap_curr_pages() is 99 and we don't expect to get back more pages from frontswap because geust os is under memory pressure. But next time in frontswap_selfshrink(): 1. last_frontswap_pages is set to the old value of cur_frontswap_pages(still 100) 2. cur_frontswap_pages(99) is still smaller than last_frontswap_pages. 3. call frontswap_shrink() and continue to get back pages from frontswap!! Signed-off-by: Bob Liu <bob.liu@oracle.com> Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
This commit is contained in:
parent
6cb606f102
commit
d4c7abdff7
1 changed files with 1 additions and 0 deletions
|
@ -170,6 +170,7 @@ static void frontswap_selfshrink(void)
|
||||||
tgt_frontswap_pages = cur_frontswap_pages -
|
tgt_frontswap_pages = cur_frontswap_pages -
|
||||||
(cur_frontswap_pages / frontswap_hysteresis);
|
(cur_frontswap_pages / frontswap_hysteresis);
|
||||||
frontswap_shrink(tgt_frontswap_pages);
|
frontswap_shrink(tgt_frontswap_pages);
|
||||||
|
frontswap_inertia_counter = frontswap_inertia;
|
||||||
}
|
}
|
||||||
|
|
||||||
#endif /* CONFIG_FRONTSWAP */
|
#endif /* CONFIG_FRONTSWAP */
|
||||||
|
|
Loading…
Reference in a new issue