blob: ae0be3de408e02ebe0db7eea9d25f92c1779e568 [file] [log] [blame]
/*
* mm/low-mem-notify.c
*
* Sends low-memory notifications to processes via /dev/low-mem.
*
* Copyright (C) 2012 The Chromium OS Authors
* This program is free software, released under the GPL.
* Based on a proposal by Minchan Kim
*
* A process that polls /dev/low-mem is notified of a low-memory situation.
* The intent is to allow the process to free some memory before the OOM killer
* is invoked.
*
* A low-memory condition is estimated by subtracting anonymous memory
* (i.e. process data segments), kernel memory, and a fixed amount of
* file-backed memory from total memory. This is just a heuristic, as in
* general we don't know how much memory can be reclaimed before we try to
* reclaim it, and that's too expensive or too late.
*
* This is tailored to Chromium OS, where a single program (the browser)
* controls most of the memory, and (currently) no swap space is used.
*/
#include <linux/low-mem-notify.h>
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/wait.h>
#include <linux/poll.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/ctype.h>
#define MB (1 << 20)
static DECLARE_WAIT_QUEUE_HEAD(low_mem_wait);
static atomic_t low_mem_state = ATOMIC_INIT(0);
/* This is a list of thresholds in pages and should be in ascending order. */
unsigned long low_mem_thresholds[LOW_MEM_THRESHOLD_MAX] = {
50 * MB / PAGE_SIZE };
unsigned int low_mem_threshold_count = 1;
/* last observed threshold */
unsigned int low_mem_threshold_last = UINT_MAX;
bool low_mem_margin_enabled = true;
unsigned int low_mem_ram_vs_swap_weight = 4;
/* Limit logging low memory to once per second. */
DEFINE_RATELIMIT_STATE(low_mem_logging_ratelimit, 1 * HZ, 1);
unsigned long low_mem_lowest_seen_anon_mem;
const unsigned long low_mem_anon_mem_delta = 10 * MB / PAGE_SIZE;
static struct kernfs_node *low_mem_available_dirent;
struct low_mem_notify_file_info {
unsigned long unused;
};
void low_mem_notify(void)
{
atomic_set(&low_mem_state, true);
wake_up(&low_mem_wait);
}
static int low_mem_notify_open(struct inode *inode, struct file *file)
{
struct low_mem_notify_file_info *info;
int err = 0;
info = kmalloc(sizeof(*info), GFP_KERNEL);
if (!info) {
err = -ENOMEM;
goto out;
}
file->private_data = info;
out:
return err;
}
static int low_mem_notify_release(struct inode *inode, struct file *file)
{
kfree(file->private_data);
return 0;
}
static unsigned int low_mem_notify_poll(struct file *file, poll_table *wait)
{
unsigned int ret = 0;
/* Update state to reflect any recent freeing. */
atomic_set(&low_mem_state, low_mem_check());
poll_wait(file, &low_mem_wait, wait);
if (low_mem_margin_enabled && atomic_read(&low_mem_state) != 0)
ret = POLLIN;
return ret;
}
const struct file_operations low_mem_notify_fops = {
.open = low_mem_notify_open,
.release = low_mem_notify_release,
.poll = low_mem_notify_poll,
};
EXPORT_SYMBOL(low_mem_notify_fops);
#ifdef CONFIG_SYSFS
#define LOW_MEM_ATTR(_name) \
static struct kobj_attribute low_mem_##_name##_attr = \
__ATTR(_name, 0644, low_mem_##_name##_show, \
low_mem_##_name##_store)
static ssize_t low_mem_margin_show(struct kobject *kobj,
struct kobj_attribute *attr, char *buf)
{
if (low_mem_margin_enabled && low_mem_threshold_count) {
int i;
ssize_t written = 0;
for (i = 0; i < low_mem_threshold_count; i++)
written += sprintf(buf + written, "%lu ",
low_mem_thresholds[i] * PAGE_SIZE / MB);
written += sprintf(buf + written, "\n");
return written;
} else
return sprintf(buf, "off\n");
}
static ssize_t low_mem_margin_store(struct kobject *kobj,
struct kobj_attribute *attr,
const char *buf, size_t count)
{
int i = 0, consumed = 0;
const char *start = buf;
char *endp;
unsigned long thresholds[LOW_MEM_THRESHOLD_MAX];
memset(thresholds, 0, sizeof(thresholds));
/*
* Even though the API does not say anything about this, the string in
* buf is zero-terminated (as long as count < PAGE_SIZE) because buf is
* a newly allocated zero-filled page. Most other sysfs handlers rely
* on this too.
*/
if (strncmp("off", buf, 3) == 0) {
pr_info("low_mem: disabling notifier\n");
low_mem_margin_enabled = false;
return count;
}
if (strncmp("on", buf, 2) == 0) {
pr_info("low_mem: enabling notifier\n");
low_mem_margin_enabled = true;
return count;
}
/*
* This takes a space separated list of thresholds in ascending order,
* and a trailing newline is optional.
*/
while (consumed < count) {
if (i >= LOW_MEM_THRESHOLD_MAX) {
pr_warn("low-mem: too many thresholds");
return -EINVAL;
}
/* special case for trailing newline */
if (*start == '\n')
break;
thresholds[i] = simple_strtoul(start, &endp, 0);
if ((endp == start) && *endp != '\n')
return -EINVAL;
/* make sure each is larger than the last one */
if (i && thresholds[i] <= thresholds[i - 1]) {
pr_warn("low-mem: thresholds not in increasing order: %lu then %lu\n",
thresholds[i - 1], thresholds[i]);
return -EINVAL;
}
if (thresholds[i] * (MB / PAGE_SIZE) > totalram_pages) {
pr_warn("low-mem: threshold too high\n");
return -EINVAL;
}
consumed += endp - start + 1;
start = endp + 1;
i++;
}
low_mem_threshold_count = i;
low_mem_margin_enabled = !!low_mem_threshold_count;
/* Convert to pages outside the allocator fast path. */
for (i = 0; i < low_mem_threshold_count; i++) {
low_mem_thresholds[i] =
thresholds[i] * (MB / PAGE_SIZE);
pr_info("low_mem: threshold[%d] %lu MB\n", i,
low_mem_thresholds[i] * PAGE_SIZE / MB);
}
return count;
}
LOW_MEM_ATTR(margin);
static ssize_t low_mem_ram_vs_swap_weight_show(struct kobject *kobj,
struct kobj_attribute *attr,
char *buf)
{
return sprintf(buf, "%u\n", low_mem_ram_vs_swap_weight);
}
static ssize_t low_mem_ram_vs_swap_weight_store(struct kobject *kobj,
struct kobj_attribute *attr,
const char *buf, size_t count)
{
unsigned long weight;
int err;
err = kstrtoul(buf, 10, &weight);
if (err)
return -EINVAL;
/* The special value 0 represents infinity. */
low_mem_ram_vs_swap_weight = weight == 0 ?
-1U : (unsigned int) weight;
pr_info("low_mem: setting ram weight to %u\n",
low_mem_ram_vs_swap_weight);
return count;
}
LOW_MEM_ATTR(ram_vs_swap_weight);
static ssize_t low_mem_available_show(struct kobject *kobj,
struct kobj_attribute *attr,
char *buf)
{
unsigned long available_mem = get_available_mem_adj();
return sprintf(buf, "%lu\n",
available_mem / (MB / PAGE_SIZE));
}
static ssize_t low_mem_available_store(struct kobject *kobj,
struct kobj_attribute *attr,
const char *buf, size_t count)
{
return -EINVAL;
}
LOW_MEM_ATTR(available);
static struct attribute *low_mem_attrs[] = {
&low_mem_margin_attr.attr,
&low_mem_ram_vs_swap_weight_attr.attr,
&low_mem_available_attr.attr,
NULL,
};
static struct attribute_group low_mem_attr_group = {
.attrs = low_mem_attrs,
.name = "chromeos-low_mem",
};
void low_mem_threshold_notify(void)
{
if (low_mem_available_dirent)
sysfs_notify_dirent(low_mem_available_dirent);
}
static int __init low_mem_init(void)
{
struct kernfs_node *low_mem_node;
int err = sysfs_create_group(mm_kobj, &low_mem_attr_group);
if (err)
pr_err("low_mem: register sysfs failed\n");
low_mem_node = sysfs_get_dirent(mm_kobj->sd, "chromeos-low_mem");
if (low_mem_node) {
low_mem_available_dirent =
sysfs_get_dirent(low_mem_node, "available");
sysfs_put(low_mem_node);
}
if (!low_mem_available_dirent)
pr_warn("unable to find dirent for \"available\" attribute\n");
low_mem_lowest_seen_anon_mem = totalram_pages;
return err;
}
module_init(low_mem_init)
#endif