staging, android: remove lowmemory killer from the tree
authorMichal Hocko <mhocko@suse.com>
Wed, 22 Feb 2017 12:01:21 +0000 (13:01 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 9 Mar 2017 12:47:03 +0000 (13:47 +0100)
Lowmemory killer is sitting in the staging tree since 2008 without any
serious interest for fixing issues brought up by the MM folks. The main
objection is that the implementation is basically broken by design:
- it hooks into slab shrinker API which is not suitable for this
  purpose. lowmem_count implementation just shows this nicely.
  There is no scaling based on the memory pressure and no
  feedback to the generic shrinker infrastructure.
  Moreover lowmem_scan is called way too often for the heavy
  work it performs.
- it is not reclaim context aware - no NUMA and/or memcg
  awareness.

As the code stands right now it just adds a maintenance overhead when
core MM changes have to update lowmemorykiller.c as well. It also seems
that the alternative LMK implementation will be solely in the userspace
so this code has no perspective it seems. The staging tree is supposed
to be for a code which needs to be put in shape before it can be merged
which is not the case here obviously.

Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/android/Kconfig
drivers/staging/android/Makefile
drivers/staging/android/lowmemorykiller.c [deleted file]
include/linux/sched.h

index 6c00d6f765c669027e4bf2ef5c5ed801342da074..71a50b99caff160b9f5603ac4229f87c84cacb09 100644 (file)
@@ -14,16 +14,6 @@ config ASHMEM
          It is, in theory, a good memory allocator for low-memory devices,
          because it can discard shared memory units when under memory pressure.
 
-config ANDROID_LOW_MEMORY_KILLER
-       bool "Android Low Memory Killer"
-       ---help---
-         Registers processes to be killed when low memory conditions, this is useful
-         as there is no particular swap space on android.
-
-         The registered process will kill according to the priorities in android init
-         scripts (/init.rc), and it defines priority values with minimum free memory size
-         for each priority.
-
 source "drivers/staging/android/ion/Kconfig"
 
 endif # if ANDROID
index 7ed1be7989097d3f45be6cb63cb741839649cbba..7cf1564a49a5eec86acf2c6f09a9196940beef25 100644 (file)
@@ -3,4 +3,3 @@ ccflags-y += -I$(src)                   # needed for trace events
 obj-y                                  += ion/
 
 obj-$(CONFIG_ASHMEM)                   += ashmem.o
-obj-$(CONFIG_ANDROID_LOW_MEMORY_KILLER)        += lowmemorykiller.o
diff --git a/drivers/staging/android/lowmemorykiller.c b/drivers/staging/android/lowmemorykiller.c
deleted file mode 100644 (file)
index 0546600..0000000
+++ /dev/null
@@ -1,212 +0,0 @@
-/* drivers/misc/lowmemorykiller.c
- *
- * The lowmemorykiller driver lets user-space specify a set of memory thresholds
- * where processes with a range of oom_score_adj values will get killed. Specify
- * the minimum oom_score_adj values in
- * /sys/module/lowmemorykiller/parameters/adj and the number of free pages in
- * /sys/module/lowmemorykiller/parameters/minfree. Both files take a comma
- * separated list of numbers in ascending order.
- *
- * For example, write "0,8" to /sys/module/lowmemorykiller/parameters/adj and
- * "1024,4096" to /sys/module/lowmemorykiller/parameters/minfree to kill
- * processes with a oom_score_adj value of 8 or higher when the free memory
- * drops below 4096 pages and kill processes with a oom_score_adj value of 0 or
- * higher when the free memory drops below 1024 pages.
- *
- * The driver considers memory used for caches to be free, but if a large
- * percentage of the cached memory is locked this can be very inaccurate
- * and processes may not get killed until the normal oom killer is triggered.
- *
- * Copyright (C) 2007-2008 Google, Inc.
- *
- * This software is licensed under the terms of the GNU General Public
- * License version 2, as published by the Free Software Foundation, and
- * may be copied, distributed, and modified under those terms.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- */
-
-#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
-
-#include <linux/init.h>
-#include <linux/moduleparam.h>
-#include <linux/kernel.h>
-#include <linux/mm.h>
-#include <linux/oom.h>
-#include <linux/sched/signal.h>
-#include <linux/swap.h>
-#include <linux/rcupdate.h>
-#include <linux/profile.h>
-#include <linux/notifier.h>
-
-static u32 lowmem_debug_level = 1;
-static short lowmem_adj[6] = {
-       0,
-       1,
-       6,
-       12,
-};
-
-static int lowmem_adj_size = 4;
-static int lowmem_minfree[6] = {
-       3 * 512,        /* 6MB */
-       2 * 1024,       /* 8MB */
-       4 * 1024,       /* 16MB */
-       16 * 1024,      /* 64MB */
-};
-
-static int lowmem_minfree_size = 4;
-
-static unsigned long lowmem_deathpending_timeout;
-
-#define lowmem_print(level, x...)                      \
-       do {                                            \
-               if (lowmem_debug_level >= (level))      \
-                       pr_info(x);                     \
-       } while (0)
-
-static unsigned long lowmem_count(struct shrinker *s,
-                                 struct shrink_control *sc)
-{
-       return global_node_page_state(NR_ACTIVE_ANON) +
-               global_node_page_state(NR_ACTIVE_FILE) +
-               global_node_page_state(NR_INACTIVE_ANON) +
-               global_node_page_state(NR_INACTIVE_FILE);
-}
-
-static unsigned long lowmem_scan(struct shrinker *s, struct shrink_control *sc)
-{
-       struct task_struct *tsk;
-       struct task_struct *selected = NULL;
-       unsigned long rem = 0;
-       int tasksize;
-       int i;
-       short min_score_adj = OOM_SCORE_ADJ_MAX + 1;
-       int minfree = 0;
-       int selected_tasksize = 0;
-       short selected_oom_score_adj;
-       int array_size = ARRAY_SIZE(lowmem_adj);
-       int other_free = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
-       int other_file = global_node_page_state(NR_FILE_PAGES) -
-                               global_node_page_state(NR_SHMEM) -
-                               total_swapcache_pages();
-
-       if (lowmem_adj_size < array_size)
-               array_size = lowmem_adj_size;
-       if (lowmem_minfree_size < array_size)
-               array_size = lowmem_minfree_size;
-       for (i = 0; i < array_size; i++) {
-               minfree = lowmem_minfree[i];
-               if (other_free < minfree && other_file < minfree) {
-                       min_score_adj = lowmem_adj[i];
-                       break;
-               }
-       }
-
-       lowmem_print(3, "lowmem_scan %lu, %x, ofree %d %d, ma %hd\n",
-                    sc->nr_to_scan, sc->gfp_mask, other_free,
-                    other_file, min_score_adj);
-
-       if (min_score_adj == OOM_SCORE_ADJ_MAX + 1) {
-               lowmem_print(5, "lowmem_scan %lu, %x, return 0\n",
-                            sc->nr_to_scan, sc->gfp_mask);
-               return 0;
-       }
-
-       selected_oom_score_adj = min_score_adj;
-
-       rcu_read_lock();
-       for_each_process(tsk) {
-               struct task_struct *p;
-               short oom_score_adj;
-
-               if (tsk->flags & PF_KTHREAD)
-                       continue;
-
-               p = find_lock_task_mm(tsk);
-               if (!p)
-                       continue;
-
-               if (task_lmk_waiting(p) &&
-                   time_before_eq(jiffies, lowmem_deathpending_timeout)) {
-                       task_unlock(p);
-                       rcu_read_unlock();
-                       return 0;
-               }
-               oom_score_adj = p->signal->oom_score_adj;
-               if (oom_score_adj < min_score_adj) {
-                       task_unlock(p);
-                       continue;
-               }
-               tasksize = get_mm_rss(p->mm);
-               task_unlock(p);
-               if (tasksize <= 0)
-                       continue;
-               if (selected) {
-                       if (oom_score_adj < selected_oom_score_adj)
-                               continue;
-                       if (oom_score_adj == selected_oom_score_adj &&
-                           tasksize <= selected_tasksize)
-                               continue;
-               }
-               selected = p;
-               selected_tasksize = tasksize;
-               selected_oom_score_adj = oom_score_adj;
-               lowmem_print(2, "select '%s' (%d), adj %hd, size %d, to kill\n",
-                            p->comm, p->pid, oom_score_adj, tasksize);
-       }
-       if (selected) {
-               task_lock(selected);
-               send_sig(SIGKILL, selected, 0);
-               if (selected->mm)
-                       task_set_lmk_waiting(selected);
-               task_unlock(selected);
-               lowmem_print(1, "Killing '%s' (%d), adj %hd,\n"
-                                "   to free %ldkB on behalf of '%s' (%d) because\n"
-                                "   cache %ldkB is below limit %ldkB for oom_score_adj %hd\n"
-                                "   Free memory is %ldkB above reserved\n",
-                            selected->comm, selected->pid,
-                            selected_oom_score_adj,
-                            selected_tasksize * (long)(PAGE_SIZE / 1024),
-                            current->comm, current->pid,
-                            other_file * (long)(PAGE_SIZE / 1024),
-                            minfree * (long)(PAGE_SIZE / 1024),
-                            min_score_adj,
-                            other_free * (long)(PAGE_SIZE / 1024));
-               lowmem_deathpending_timeout = jiffies + HZ;
-               rem += selected_tasksize;
-       }
-
-       lowmem_print(4, "lowmem_scan %lu, %x, return %lu\n",
-                    sc->nr_to_scan, sc->gfp_mask, rem);
-       rcu_read_unlock();
-       return rem;
-}
-
-static struct shrinker lowmem_shrinker = {
-       .scan_objects = lowmem_scan,
-       .count_objects = lowmem_count,
-       .seeks = DEFAULT_SEEKS * 16
-};
-
-static int __init lowmem_init(void)
-{
-       register_shrinker(&lowmem_shrinker);
-       return 0;
-}
-device_initcall(lowmem_init);
-
-/*
- * not really modular, but the easiest way to keep compat with existing
- * bootargs behaviour is to continue using module_param here.
- */
-module_param_named(cost, lowmem_shrinker.seeks, int, 0644);
-module_param_array_named(adj, lowmem_adj, short, &lowmem_adj_size, 0644);
-module_param_array_named(minfree, lowmem_minfree, uint, &lowmem_minfree_size,
-                        0644);
-module_param_named(debug_level, lowmem_debug_level, uint, 0644);
-
index d67eee84fd430f3c44b77d4ba007ec5d2dcabb2b..942c2250301b46a7f601dc07d6ddeabdfd7d7e13 100644 (file)
@@ -1256,7 +1256,6 @@ extern struct pid *cad_pid;
 #define PFA_NO_NEW_PRIVS               0       /* May not gain new privileges. */
 #define PFA_SPREAD_PAGE                        1       /* Spread page cache over cpuset */
 #define PFA_SPREAD_SLAB                        2       /* Spread some slab caches over cpuset */
-#define PFA_LMK_WAITING                        3       /* Lowmemorykiller is waiting */
 
 
 #define TASK_PFA_TEST(name, func)                                      \
@@ -1282,9 +1281,6 @@ TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
 
-TASK_PFA_TEST(LMK_WAITING, lmk_waiting)
-TASK_PFA_SET(LMK_WAITING, lmk_waiting)
-
 static inline void
 tsk_restore_flags(struct task_struct *task, unsigned long orig_flags, unsigned long flags)
 {