Files
kernel_xiaomi_sm8250/drivers/devfreq/governor_memlat.c
Rama Aparna Mallavarapu eccbcb112c PM / devfreq: memlat: Look for min stall% in addition to ratio criteria
Some workloads doing memory access might appear memory latency bound even
though they might not actually be memory latency bound.

This error can happen when the core that's running the workload is very
parallelized or can do out of order executions, etc so not all memory
accesses would actually stall the core.

This can also happen when the the memory access monitoring capabilities
aren't ideal and end up counting more kinds of memory accesses than what
would be ideal. In this case, the IPM ratio can be lower than what it would
be if we had ideal monitoring capabilities.

To account for these errors, if the core has a stall cycle counting
capabilities, check for a minimum stall% before the workload is considered
memory latency bound. This would help reduce the inaccuracies, but is not a
replacement for IPM ratio scheme because the stall% method doesn't allow us
to detect which level of memory the workload is latency bound on, but the
IPM ratio does (based on which memory accesses we use for calculating the
ratio).

Change-Id: I4363d7848584e5562f6683b5ad6b0f99017ec71b
Signed-off-by: Saravana Kannan <skannan@codeaurora.org>
Signed-off-by: Rama Aparna Mallavarapu <aparnam@codeaurora.org>
2019-01-17 14:37:16 -08:00

420 lines
9.0 KiB
C

// SPDX-License-Identifier: GPL-2.0-only
/*
* Copyright (c) 2015-2017, 2019, The Linux Foundation. All rights reserved.
*/
#define pr_fmt(fmt) "mem_lat: " fmt
#include <linux/kernel.h>
#include <linux/sizes.h>
#include <linux/module.h>
#include <linux/init.h>
#include <linux/io.h>
#include <linux/delay.h>
#include <linux/ktime.h>
#include <linux/time.h>
#include <linux/err.h>
#include <linux/errno.h>
#include <linux/mutex.h>
#include <linux/interrupt.h>
#include <linux/platform_device.h>
#include <linux/of.h>
#include <linux/devfreq.h>
#include "governor.h"
#include "governor_memlat.h"
#include <trace/events/power.h>
struct memlat_node {
unsigned int ratio_ceil;
unsigned int stall_floor;
bool mon_started;
bool already_zero;
struct list_head list;
void *orig_data;
struct memlat_hwmon *hw;
struct devfreq_governor *gov;
struct attribute_group *attr_grp;
};
static LIST_HEAD(memlat_list);
static DEFINE_MUTEX(list_lock);
static int use_cnt;
static DEFINE_MUTEX(state_lock);
#define show_attr(name) \
static ssize_t show_##name(struct device *dev, \
struct device_attribute *attr, char *buf) \
{ \
struct devfreq *df = to_devfreq(dev); \
struct memlat_node *hw = df->data; \
return scnprintf(buf, PAGE_SIZE, "%u\n", hw->name); \
}
#define store_attr(name, _min, _max) \
static ssize_t store_##name(struct device *dev, \
struct device_attribute *attr, const char *buf, \
size_t count) \
{ \
struct devfreq *df = to_devfreq(dev); \
struct memlat_node *hw = df->data; \
int ret; \
unsigned int val; \
ret = kstrtouint(buf, 10, &val); \
if (ret) \
return ret; \
val = max(val, _min); \
val = min(val, _max); \
hw->name = val; \
return count; \
}
#define gov_attr(__attr, min, max) \
show_attr(__attr) \
store_attr(__attr, min, max) \
static DEVICE_ATTR(__attr, 0644, show_##__attr, store_##__attr)
static ssize_t freq_map_show(struct device *dev, struct device_attribute *attr,
char *buf)
{
struct devfreq *df = to_devfreq(dev);
struct memlat_node *n = df->data;
struct core_dev_map *map = n->hw->freq_map;
unsigned int cnt = 0;
cnt += scnprintf(buf, PAGE_SIZE, "Core freq (MHz)\tDevice BW\n");
while (map->core_mhz && cnt < PAGE_SIZE) {
cnt += scnprintf(buf + cnt, PAGE_SIZE - cnt, "%15u\t%9u\n",
map->core_mhz, map->target_freq);
map++;
}
if (cnt < PAGE_SIZE)
cnt += scnprintf(buf + cnt, PAGE_SIZE - cnt, "\n");
return cnt;
}
static DEVICE_ATTR_RO(freq_map);
static unsigned long core_to_dev_freq(struct memlat_node *node,
unsigned long coref)
{
struct memlat_hwmon *hw = node->hw;
struct core_dev_map *map = hw->freq_map;
unsigned long freq = 0;
if (!map)
goto out;
while (map->core_mhz && map->core_mhz < coref)
map++;
if (!map->core_mhz)
map--;
freq = map->target_freq;
out:
pr_debug("freq: %lu -> dev: %lu\n", coref, freq);
return freq;
}
static struct memlat_node *find_memlat_node(struct devfreq *df)
{
struct memlat_node *node, *found = NULL;
mutex_lock(&list_lock);
list_for_each_entry(node, &memlat_list, list)
if (node->hw->dev == df->dev.parent ||
node->hw->of_node == df->dev.parent->of_node) {
found = node;
break;
}
mutex_unlock(&list_lock);
return found;
}
static int start_monitor(struct devfreq *df)
{
struct memlat_node *node = df->data;
struct memlat_hwmon *hw = node->hw;
struct device *dev = df->dev.parent;
int ret;
ret = hw->start_hwmon(hw);
if (ret) {
dev_err(dev, "Unable to start HW monitor! (%d)\n", ret);
return ret;
}
devfreq_monitor_start(df);
node->mon_started = true;
return 0;
}
static void stop_monitor(struct devfreq *df)
{
struct memlat_node *node = df->data;
struct memlat_hwmon *hw = node->hw;
node->mon_started = false;
devfreq_monitor_stop(df);
hw->stop_hwmon(hw);
}
static int gov_start(struct devfreq *df)
{
int ret = 0;
struct device *dev = df->dev.parent;
struct memlat_node *node;
struct memlat_hwmon *hw;
node = find_memlat_node(df);
if (!node) {
dev_err(dev, "Unable to find HW monitor!\n");
return -ENODEV;
}
hw = node->hw;
hw->df = df;
node->orig_data = df->data;
df->data = node;
if (start_monitor(df))
goto err_start;
ret = sysfs_create_group(&df->dev.kobj, node->attr_grp);
if (ret)
goto err_sysfs;
return 0;
err_sysfs:
stop_monitor(df);
err_start:
df->data = node->orig_data;
node->orig_data = NULL;
hw->df = NULL;
return ret;
}
static void gov_stop(struct devfreq *df)
{
struct memlat_node *node = df->data;
struct memlat_hwmon *hw = node->hw;
sysfs_remove_group(&df->dev.kobj, node->attr_grp);
stop_monitor(df);
df->data = node->orig_data;
node->orig_data = NULL;
hw->df = NULL;
}
static int devfreq_memlat_get_freq(struct devfreq *df,
unsigned long *freq)
{
int i, lat_dev = 0;
struct memlat_node *node = df->data;
struct memlat_hwmon *hw = node->hw;
unsigned long max_freq = 0;
unsigned int ratio;
hw->get_cnt(hw);
for (i = 0; i < hw->num_cores; i++) {
ratio = hw->core_stats[i].inst_count;
if (hw->core_stats[i].mem_count)
ratio /= hw->core_stats[i].mem_count;
if (!hw->core_stats[i].inst_count
|| !hw->core_stats[i].freq)
continue;
trace_memlat_dev_meas(dev_name(df->dev.parent),
hw->core_stats[i].id,
hw->core_stats[i].inst_count,
hw->core_stats[i].mem_count,
hw->core_stats[i].freq,
hw->core_stats[i].stall_pct, ratio);
if (ratio <= node->ratio_ceil
&& hw->core_stats[i].stall_pct >= node->stall_floor
&& hw->core_stats[i].freq > max_freq) {
lat_dev = i;
max_freq = hw->core_stats[i].freq;
}
}
if (max_freq)
max_freq = core_to_dev_freq(node, max_freq);
if (max_freq || !node->already_zero) {
trace_memlat_dev_update(dev_name(df->dev.parent),
hw->core_stats[lat_dev].id,
hw->core_stats[lat_dev].inst_count,
hw->core_stats[lat_dev].mem_count,
hw->core_stats[lat_dev].freq,
max_freq);
}
node->already_zero = !max_freq;
*freq = max_freq;
return 0;
}
gov_attr(ratio_ceil, 1U, 10000U);
gov_attr(stall_floor, 0U, 100U);
static struct attribute *dev_attr[] = {
&dev_attr_ratio_ceil.attr,
&dev_attr_stall_floor.attr,
&dev_attr_freq_map.attr,
NULL,
};
static struct attribute_group dev_attr_group = {
.name = "mem_latency",
.attrs = dev_attr,
};
#define MIN_MS 10U
#define MAX_MS 500U
static int devfreq_memlat_ev_handler(struct devfreq *df,
unsigned int event, void *data)
{
int ret;
unsigned int sample_ms;
switch (event) {
case DEVFREQ_GOV_START:
sample_ms = df->profile->polling_ms;
sample_ms = max(MIN_MS, sample_ms);
sample_ms = min(MAX_MS, sample_ms);
df->profile->polling_ms = sample_ms;
ret = gov_start(df);
if (ret)
return ret;
dev_dbg(df->dev.parent,
"Enabled Memory Latency governor\n");
break;
case DEVFREQ_GOV_STOP:
gov_stop(df);
dev_dbg(df->dev.parent,
"Disabled Memory Latency governor\n");
break;
case DEVFREQ_GOV_INTERVAL:
sample_ms = *(unsigned int *)data;
sample_ms = max(MIN_MS, sample_ms);
sample_ms = min(MAX_MS, sample_ms);
devfreq_interval_update(df, &sample_ms);
break;
}
return 0;
}
static struct devfreq_governor devfreq_gov_memlat = {
.name = "mem_latency",
.get_target_freq = devfreq_memlat_get_freq,
.event_handler = devfreq_memlat_ev_handler,
};
#define NUM_COLS 2
static struct core_dev_map *init_core_dev_map(struct device *dev,
char *prop_name)
{
int len, nf, i, j;
u32 data;
struct core_dev_map *tbl;
int ret;
if (!of_find_property(dev->of_node, prop_name, &len))
return NULL;
len /= sizeof(data);
if (len % NUM_COLS || len == 0)
return NULL;
nf = len / NUM_COLS;
tbl = devm_kzalloc(dev, (nf + 1) * sizeof(struct core_dev_map),
GFP_KERNEL);
if (!tbl)
return NULL;
for (i = 0, j = 0; i < nf; i++, j += 2) {
ret = of_property_read_u32_index(dev->of_node, prop_name, j,
&data);
if (ret)
return NULL;
tbl[i].core_mhz = data / 1000;
ret = of_property_read_u32_index(dev->of_node, prop_name, j + 1,
&data);
if (ret)
return NULL;
tbl[i].target_freq = data;
pr_debug("Entry%d CPU:%u, Dev:%u\n", i, tbl[i].core_mhz,
tbl[i].target_freq);
}
tbl[i].core_mhz = 0;
return tbl;
}
int register_memlat(struct device *dev, struct memlat_hwmon *hw)
{
int ret = 0;
struct memlat_node *node;
if (!hw->dev && !hw->of_node)
return -EINVAL;
node = devm_kzalloc(dev, sizeof(*node), GFP_KERNEL);
if (!node)
return -ENOMEM;
node->gov = &devfreq_gov_memlat;
node->attr_grp = &dev_attr_group;
node->ratio_ceil = 10;
node->hw = hw;
hw->freq_map = init_core_dev_map(dev, "qcom,core-dev-table");
if (!hw->freq_map) {
dev_err(dev, "Couldn't find the core-dev freq table!\n");
return -EINVAL;
}
mutex_lock(&list_lock);
list_add_tail(&node->list, &memlat_list);
mutex_unlock(&list_lock);
mutex_lock(&state_lock);
if (!use_cnt)
ret = devfreq_add_governor(&devfreq_gov_memlat);
if (!ret)
use_cnt++;
mutex_unlock(&state_lock);
if (!ret)
dev_info(dev, "Memory Latency governor registered.\n");
else
dev_err(dev, "Memory Latency governor registration failed!\n");
return ret;
}
MODULE_DESCRIPTION("HW monitor based dev DDR bandwidth voting driver");
MODULE_LICENSE("GPL v2");