[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20200323234147.558EBA81@viggo.jf.intel.com>
Date: Mon, 23 Mar 2020 16:41:47 -0700
From: Dave Hansen <dave.hansen@...ux.intel.com>
To: linux-kernel@...r.kernel.org
Cc: Dave Hansen <dave.hansen@...ux.intel.com>, mhocko@...e.com,
jannh@...gle.com, vbabka@...e.cz, minchan@...nel.org,
dancol@...gle.com, joel@...lfernandes.org,
akpm@...ux-foundation.org
Subject: [PATCH 0/2] mm/madvise: teach MADV_PAGEOUT about swap cache
MADV_PAGEOUT ignores swap cache pages which are not mapped
into the calling process. That's nasty because it means that
some of the most easily reclaimable pages are inaccessible to
a mechanism designed to reclaim pages.
This all turned out to be a bit nastier and more complicated
than I would have liked. This has been lightly tested, but I
did pass a normal barrage of compile tests.
I rigged up a little test program to try to create these
situations. Basically, if the parent "reader" RSS changes
in response to MADV_PAGEOUT actions in the child, there is
a problem.
https://www.sr71.net/~dave/intel/madv-pageout.c
I'd add this to selftests, but it *requires* swap to work
and its parsing of /proc/self/maps is quite icky.
P.S. mincore() really doesn't work at all in this situation,
probably something we need to look at too.
Cc: Michal Hocko <mhocko@...e.com>
Cc: Jann Horn <jannh@...gle.com>
Cc: Vlastimil Babka <vbabka@...e.cz>
Cc: Minchan Kim <minchan@...nel.org>
Cc: Daniel Colascione <dancol@...gle.com>
Cc: "Joel Fernandes (Google)" <joel@...lfernandes.org>
Cc: Andrew Morton <akpm@...ux-foundation.org>
Powered by blists - more mailing lists