Messages in this thread | | | Date | Wed, 14 Jan 2015 15:34:49 -0800 | From | Andrew Morton <> | Subject | Re: [PATCH -mm] vmscan: move reclaim_state handling to shrink_slab |
| |
On Wed, 14 Jan 2015 16:55:36 +0300 Vladimir Davydov <vdavydov@parallels.com> wrote:
> current->reclaim_state is only used to count the number of slab pages > reclaimed by shrink_slab(). So instead of initializing it before we are > going to call try_to_free_pages() or shrink_zone(), let's set in > directly in shrink_slab(). > > This patch also makes shrink_slab() return the number of reclaimed slab > pages (obtained from reclaim_state) instead of the number of reclaimed > objects, because the latter is not of much use - it was only checked by > drop_slab() to decide whether it should continue reclaim or abort. The > number of reclaimed pages is more appropriate, because it also can be > used by shrink_zone() to accumulate scan_control->nr_reclaimed.
Not sure that this is a good change. If shrink_slab() managed to free some objects but didn't free any pages then that's a good sign that additional calls to shrink_slab() *will* free some pages. With this change, drop_slab_node() can give up too early.
The general philosophy throughout here is: "pass it nr_to_scan, it returns nr_scanned/nr_freed". Switching the return value to nr_pages_freed kinda breaks that paradigm.
> Note that after this patch try_to_free_mem_cgroup_pages() will count not > only reclaimed user pages, but also slab pages, which is expected, > because it can reclaim kmem from kmem-active sub cgroups. > > mm/page_alloc.c | 4 --- > mm/vmscan.c | 73 ++++++++++++++++++++----------------------------------- > 2 files changed, 27 insertions(+), 50 deletions(-)
That's nice though.
| |