This will make sure we get invoked in time,
and combined with a kernel change to increase
the window size (and thus rate-limit the
notifications) based on the amount of memory
on the device, this should provide a nice
balance between receiving too many notifications
vs getting notified too late.
Change-Id: I2f5e92390b94e582aaa06d88763a232e39673496
Opening /proc/zoneinfo may return -ENOMEM under memory pressure,
assume something needs to be killed if that happens.
Bug: 16236289
Change-Id: I25aab09f4c3f2b9e176eaaeca94fd6bd790dad4e
The task selected to die may be small, add its approximate size
to other_free and other_file and keep killing until all thresholds
are met.
Bug: 16236289
Change-Id: Iceeca4c63fec98cae2bf53e258f7707cea408b07
If kswapd runs more than lmkd it may throw away all page cache pages
before lmkd gets a chance to kill a process. Make lmkd SCHED_FIFO
so it can react quickly to kernel low memory notifications.
Bug: 16236289
Change-Id: I46767b4ec81ef8638d0c9e47d168a68b62e3cac5
Use mlockall(MCL_FUTURE) to lock all lmkd pages in memory. This
avoids lmkd thrashing when the system is low on memory.
As a side effect, it will also keep the .text sections of liblog, libm,
libc, and libprocessgroup in memory at all times.
Bug: 16236289
Change-Id: Idd70557efa4b1e14bc86f14220672a30f6c956e3
After sending sigkill to the main process, also call into
libprocessgroup to kill any processes that were forked.
Bug: 15313911
Change-Id: I05577c6f5b70800ce0a01f480c8870c2c601afda
Use oom_score_adj for setting OOM score adjustment values instead
of the older, deprecated oom_adj interface.
Change-Id: I5ba7801d4b8971be092b31ae0fd8af8c3001d613
Move kernel low memory killer logic to new daemon lmkd. ActivityManager
communicates with this daemon over a named socket.
Change-Id: Ie957da7e9b1f7150e23c689fdadf23ff260b47be
Martijn Coenen