[PATCH 2/3 v6] cpufreq: stats: Add 'load_table' debugfs file to showaccumulated data of CPUs

From: Chanwoo Choi
Date: Thu Jul 18 2013 - 07:18:23 EST


This patch add new 'load_table' debugfs file to show previous accumulated data
of CPUs load as following path and add CPUFREQ_LOADCHECK notification to
CPUFREQ_TRANSITION_NOTIFIER notifier chain.
- /sys/kernel/debug/cpufreq/cpuX/load_table

When governor calculates CPUs load on dbs_check_cpu(), governor send
CPUFREQ_LOADCHECK notification with CPUs load, so that cpufreq_stats
accumulates calculated CPUs load on 'load_table' storage.

This debugfs file is used to judge the correct system state or determine
suitable system resource according to current CPUs load on user-space.

This debugfs file include following data:
- Measurement point of time
- CPU frequency
- Per-CPU load

Signed-off-by: Chanwoo Choi <cw00.choi@xxxxxxxxxxx>
Signed-off-by: Kyungmin Park <kyungmin.park@xxxxxxxxxxx>
Signed-off-by: Myungjoo Ham <myungjoo.ham@xxxxxxxxxxx>
---
Changes since v5:
- Determine index value of policy->cpu_debugfs[] according to
cpumask_weight(policy->cpus) value
- Bug fix, store 'policy->cpu' to 'freq->cpu' before notify
CPUFREQ_LOADCHECK notification

Changes since v4:
- Reset the data of CPUs load when cpufreq governor is changed
- Move code about creating debugfs directory to below first patch
: [PATCH 1/2] cpufreq: Add debugfs directory for cpufreq

Changes since v3:
- Extend a range of accumulated data (10 ~ 1000)
- Add unit information of time/freq and align 'Time' field as left for readability
- Use CONFIG_CPU_FREQ_STAT depdendency instead of CONFIG_CPU_FREQ_STAT_DETATILS
- Initialize load of offline CPUx as zero(0)
- Create/remove debugfs root directory on cpufreq_stats_init/exit() because
debugfs root is used on all CPUs.

Changes since v2:
- Code clean according to Viresh Kumar's comment
- Show both old frequency and new frequency on 'load_table' debugfs file
- Change debufs file patch as below
old: /sys/kernel/debugfs/cpufreq/load_table
new: /sys/kernel/debugfs/cpufreq/cpuX/load_table

Changes since v1:
- Set maximum storage size to save CPUs load on Kconfig
- Use spinlock to synchronize read/write operation for CPUs load
- Use local variable instead of global variable(struct cpufreq_freqs *freqs)
- Use pointer of data structure to get correct size of data structure
in sizeof() macro instead of structure name
: sizeof(struct cpufreq_freqs) -> sizeof(*stat->load_table)
- Change time unit from nanosecond to microsecond
- Remove unnecessary memory copy

drivers/cpufreq/Kconfig | 6 +
drivers/cpufreq/cpufreq.c | 4 +
drivers/cpufreq/cpufreq_governor.c | 14 ++
drivers/cpufreq/cpufreq_stats.c | 258 +++++++++++++++++++++++++++++++++----
include/linux/cpufreq.h | 6 +
5 files changed, 264 insertions(+), 24 deletions(-)

diff --git a/drivers/cpufreq/Kconfig b/drivers/cpufreq/Kconfig
index 534fcb8..5c3f406 100644
--- a/drivers/cpufreq/Kconfig
+++ b/drivers/cpufreq/Kconfig
@@ -36,6 +36,12 @@ config CPU_FREQ_STAT

If in doubt, say N.

+config NR_CPU_LOAD_STORAGE
+ int "Maximum storage size to save CPU load (10-1000)"
+ range 10 1000
+ depends on CPU_FREQ_STAT
+ default "10"
+
config CPU_FREQ_STAT_DETAILS
bool "CPU frequency translation statistics details"
depends on CPU_FREQ_STAT
diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index 924d654..cb68873 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -293,6 +293,10 @@ void __cpufreq_notify_transition(struct cpufreq_policy *policy,
if (likely(policy) && likely(policy->cpu == freqs->cpu))
policy->cur = freqs->new;
break;
+ case CPUFREQ_LOADCHECK:
+ srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
+ CPUFREQ_LOADCHECK, freqs);
+ break;
}
}
/**
diff --git a/drivers/cpufreq/cpufreq_governor.c b/drivers/cpufreq/cpufreq_governor.c
index dc9b72e..decabcb 100644
--- a/drivers/cpufreq/cpufreq_governor.c
+++ b/drivers/cpufreq/cpufreq_governor.c
@@ -87,6 +87,9 @@ void dbs_check_cpu(struct dbs_data *dbs_data, int cpu)
struct od_dbs_tuners *od_tuners = dbs_data->tuners;
struct cs_dbs_tuners *cs_tuners = dbs_data->tuners;
struct cpufreq_policy *policy;
+#ifdef CONFIG_CPU_FREQ_STAT
+ struct cpufreq_freqs freq = {0};
+#endif
unsigned int max_load = 0;
unsigned int ignore_nice;
unsigned int j;
@@ -148,6 +151,9 @@ void dbs_check_cpu(struct dbs_data *dbs_data, int cpu)
continue;

load = 100 * (wall_time - idle_time) / wall_time;
+#ifdef CONFIG_CPU_FREQ_STAT
+ freq.load[j] = load;
+#endif

if (dbs_data->cdata->governor == GOV_ONDEMAND) {
int freq_avg = __cpufreq_driver_getavg(policy, j);
@@ -161,6 +167,14 @@ void dbs_check_cpu(struct dbs_data *dbs_data, int cpu)
max_load = load;
}

+#ifdef CONFIG_CPU_FREQ_STAT
+ freq.time = ktime_to_ms(ktime_get());
+ freq.old = policy->cur;
+ freq.cpu = policy->cpu;
+
+ cpufreq_notify_transition(policy, &freq, CPUFREQ_LOADCHECK);
+#endif
+
dbs_data->cdata->gov_check_cpu(cpu, max_load);
}
EXPORT_SYMBOL_GPL(dbs_check_cpu);
diff --git a/drivers/cpufreq/cpufreq_stats.c b/drivers/cpufreq/cpufreq_stats.c
index fb65dec..fd171ba 100644
--- a/drivers/cpufreq/cpufreq_stats.c
+++ b/drivers/cpufreq/cpufreq_stats.c
@@ -12,6 +12,7 @@
#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/cpu.h>
+#include <linux/debugfs.h>
#include <linux/sysfs.h>
#include <linux/cpufreq.h>
#include <linux/module.h>
@@ -36,6 +37,12 @@ struct cpufreq_stats {
#ifdef CONFIG_CPU_FREQ_STAT_DETAILS
unsigned int *trans_table;
#endif
+
+ /* Debugfs file for load_table */
+ struct cpufreq_freqs *load_table;
+ unsigned int load_last_index;
+ unsigned int load_max_index;
+ struct dentry *debugfs_load_table;
};

static DEFINE_PER_CPU(struct cpufreq_stats *, cpufreq_stats_table);
@@ -149,6 +156,181 @@ static struct attribute_group stats_attr_group = {
.name = "stats"
};

+#define MAX_LINE_SIZE 255
+static ssize_t load_table_read(struct file *file, char __user *user_buf,
+ size_t count, loff_t *ppos)
+{
+ struct cpufreq_policy *policy = file->private_data;
+ struct cpufreq_stats *stat = per_cpu(cpufreq_stats_table, policy->cpu);
+ struct cpufreq_freqs *load_table = stat->load_table;
+ ssize_t len = 0;
+ char *buf;
+ int i, cpu, ret;
+
+ buf = kzalloc(MAX_LINE_SIZE * stat->load_max_index, GFP_KERNEL);
+ if (!buf)
+ return 0;
+
+ spin_lock(&cpufreq_stats_lock);
+ len += sprintf(buf + len, "%-10s %-12s %-12s ", "Time(ms)",
+ "Old Freq(Hz)",
+ "New Freq(Hz)");
+ for_each_cpu(cpu, policy->cpus)
+ len += sprintf(buf + len, "%3s%d ", "CPU", cpu);
+ len += sprintf(buf + len, "\n");
+
+ i = stat->load_last_index;
+ do {
+ len += sprintf(buf + len, "%-10lld %-12d %-12d ",
+ load_table[i].time,
+ load_table[i].old,
+ load_table[i].new);
+
+ for_each_cpu(cpu, policy->cpus)
+ len += sprintf(buf + len, "%-4d ",
+ load_table[i].load[cpu]);
+ len += sprintf(buf + len, "\n");
+
+ if (++i == stat->load_max_index)
+ i = 0;
+ } while (i != stat->load_last_index);
+ spin_unlock(&cpufreq_stats_lock);
+
+ ret = simple_read_from_buffer(user_buf, count, ppos, buf, len);
+ kfree(buf);
+
+ return ret;
+}
+
+static const struct file_operations load_table_fops = {
+ .read = load_table_read,
+ .open = simple_open,
+ .llseek = no_llseek,
+};
+
+static int cpufreq_stats_reset_debugfs(struct cpufreq_policy *policy)
+{
+ struct cpufreq_stats *stat = per_cpu(cpufreq_stats_table, policy->cpu);
+ int size;
+
+ if (!stat)
+ return -EINVAL;
+
+ if (stat->load_table)
+ kfree(stat->load_table);
+ stat->load_last_index = 0;
+
+ size = sizeof(*stat->load_table) * stat->load_max_index;
+ stat->load_table = kzalloc(size, GFP_KERNEL);
+ if (!stat->load_table)
+ return -ENOMEM;
+
+ return 0;
+}
+
+static int cpufreq_stats_create_debugfs(struct cpufreq_policy *policy)
+{
+ struct cpufreq_stats *stat = per_cpu(cpufreq_stats_table, policy->cpu);
+ unsigned int idx, size;
+ int ret = 0;
+
+ if (!stat)
+ return -EINVAL;
+
+ if (!policy->cpu_debugfs)
+ return -EINVAL;
+
+ stat->load_last_index = 0;
+ stat->load_max_index = CONFIG_NR_CPU_LOAD_STORAGE;
+
+ /* Allocate memory for storage of CPUs load */
+ size = sizeof(*stat->load_table) * stat->load_max_index;
+ stat->load_table = kzalloc(size, GFP_KERNEL);
+ if (!stat->load_table)
+ return -ENOMEM;
+
+ /* Create debugfs directory and file for cpufreq */
+ idx = cpumask_weight(policy->cpus) > 1 ? policy->cpu : 0;
+ stat->debugfs_load_table = debugfs_create_file("load_table", S_IWUSR,
+ policy->cpu_debugfs[idx],
+ policy, &load_table_fops);
+ if (!stat->debugfs_load_table) {
+ ret = -ENOMEM;
+ goto err;
+ }
+
+ pr_debug("Creating debugfs file for CPU%d \n", policy->cpu);
+
+ return 0;
+err:
+ kfree(stat->load_table);
+ return ret;
+}
+
+/* should be called late in the CPU removal sequence so that the stats
+ * memory is still available in case someone tries to use it.
+ */
+static void cpufreq_stats_free_load_table(unsigned int cpu)
+{
+ struct cpufreq_stats *stat = per_cpu(cpufreq_stats_table, cpu);
+
+ if (stat) {
+ pr_debug("Free memory of load_table\n");
+ kfree(stat->load_table);
+ }
+}
+
+/* must be called early in the CPU removal sequence (before
+ * cpufreq_remove_dev) so that policy is still valid.
+ */
+static void cpufreq_stats_free_debugfs(unsigned int cpu)
+{
+ struct cpufreq_stats *stat = per_cpu(cpufreq_stats_table, cpu);
+
+ if (stat) {
+ pr_debug("Remove load_table debugfs file\n");
+ debugfs_remove(stat->debugfs_load_table);
+ }
+}
+
+static void cpufreq_stats_store_load_table(struct cpufreq_freqs *freq,
+ unsigned long val)
+{
+ struct cpufreq_stats *stat;
+ int cpu, last_idx;
+
+ stat = per_cpu(cpufreq_stats_table, freq->cpu);
+ if (!stat)
+ return;
+
+ spin_lock(&cpufreq_stats_lock);
+
+ switch (val) {
+ case CPUFREQ_POSTCHANGE:
+ if (!stat->load_last_index)
+ last_idx = stat->load_max_index;
+ else
+ last_idx = stat->load_last_index - 1;
+
+ stat->load_table[last_idx].new = freq->new;
+ break;
+ case CPUFREQ_LOADCHECK:
+ last_idx = stat->load_last_index;
+
+ stat->load_table[last_idx].time = freq->time;
+ stat->load_table[last_idx].old = freq->old;
+ stat->load_table[last_idx].new = freq->old;
+ for_each_present_cpu(cpu)
+ stat->load_table[last_idx].load[cpu] = freq->load[cpu];
+
+ if (++stat->load_last_index == stat->load_max_index)
+ stat->load_last_index = 0;
+ break;
+ }
+
+ spin_unlock(&cpufreq_stats_lock);
+}
+
static int freq_table_get_index(struct cpufreq_stats *stat, unsigned int freq)
{
int index;
@@ -204,7 +386,7 @@ static int cpufreq_stats_create_table(struct cpufreq_policy *policy,
unsigned int alloc_size;
unsigned int cpu = policy->cpu;
if (per_cpu(cpufreq_stats_table, cpu))
- return -EBUSY;
+ return 0;
stat = kzalloc(sizeof(struct cpufreq_stats), GFP_KERNEL);
if ((stat) == NULL)
return -ENOMEM;
@@ -257,6 +439,14 @@ static int cpufreq_stats_create_table(struct cpufreq_policy *policy,
spin_lock(&cpufreq_stats_lock);
stat->last_time = get_jiffies_64();
stat->last_index = freq_table_get_index(stat, policy->cur);
+
+ ret = cpufreq_stats_create_debugfs(data);
+ if (ret < 0) {
+ spin_unlock(&cpufreq_stats_lock);
+ ret = -EINVAL;
+ goto error_out;
+ }
+
spin_unlock(&cpufreq_stats_lock);
cpufreq_cpu_put(data);
return 0;
@@ -297,11 +487,18 @@ static int cpufreq_stat_notifier_policy(struct notifier_block *nb,
if (val != CPUFREQ_NOTIFY)
return 0;
table = cpufreq_frequency_get_table(cpu);
- if (!table)
- return 0;
- ret = cpufreq_stats_create_table(policy, table);
- if (ret)
+ if (table) {
+ ret = cpufreq_stats_create_table(policy, table);
+ if (ret)
+ return ret;
+ }
+
+ ret = cpufreq_stats_reset_debugfs(policy);
+ if (ret < 0) {
+ pr_debug("Failed to reset CPUs data of debugfs\n");
return ret;
+ }
+
return 0;
}

@@ -312,32 +509,40 @@ static int cpufreq_stat_notifier_trans(struct notifier_block *nb,
struct cpufreq_stats *stat;
int old_index, new_index;

- if (val != CPUFREQ_POSTCHANGE)
- return 0;
+ switch (val) {
+ case CPUFREQ_POSTCHANGE:
+ stat = per_cpu(cpufreq_stats_table, freq->cpu);
+ if (!stat)
+ return 0;

- stat = per_cpu(cpufreq_stats_table, freq->cpu);
- if (!stat)
- return 0;
+ old_index = stat->last_index;
+ new_index = freq_table_get_index(stat, freq->new);

- old_index = stat->last_index;
- new_index = freq_table_get_index(stat, freq->new);
+ /* We can't do stat->time_in_state[-1]= .. */
+ if (old_index == -1 || new_index == -1)
+ return 0;

- /* We can't do stat->time_in_state[-1]= .. */
- if (old_index == -1 || new_index == -1)
- return 0;
+ cpufreq_stats_update(freq->cpu);

- cpufreq_stats_update(freq->cpu);
+ if (old_index == new_index)
+ return 0;

- if (old_index == new_index)
- return 0;
-
- spin_lock(&cpufreq_stats_lock);
- stat->last_index = new_index;
+ spin_lock(&cpufreq_stats_lock);
+ stat->last_index = new_index;
#ifdef CONFIG_CPU_FREQ_STAT_DETAILS
- stat->trans_table[old_index * stat->max_state + new_index]++;
+ stat->trans_table[old_index * stat->max_state + new_index]++;
#endif
- stat->total_trans++;
- spin_unlock(&cpufreq_stats_lock);
+ stat->total_trans++;
+ spin_unlock(&cpufreq_stats_lock);
+
+ cpufreq_stats_store_load_table(freq, CPUFREQ_POSTCHANGE);
+
+ break;
+ case CPUFREQ_LOADCHECK:
+ cpufreq_stats_store_load_table(freq, CPUFREQ_LOADCHECK);
+ break;
+ }
+
return 0;
}

@@ -352,12 +557,16 @@ static int __cpuinit cpufreq_stat_cpu_callback(struct notifier_block *nfb,
cpufreq_update_policy(cpu);
break;
case CPU_DOWN_PREPARE:
+ cpufreq_stats_free_debugfs(cpu);
cpufreq_stats_free_sysfs(cpu);
break;
case CPU_DEAD:
+ cpufreq_stats_free_load_table(cpu);
cpufreq_stats_free_table(cpu);
break;
case CPU_UP_CANCELED_FROZEN:
+ cpufreq_stats_free_debugfs(cpu);
+ cpufreq_stats_free_load_table(cpu);
cpufreq_stats_free_sysfs(cpu);
cpufreq_stats_free_table(cpu);
break;
@@ -418,6 +627,7 @@ static void __exit cpufreq_stats_exit(void)
unregister_hotcpu_notifier(&cpufreq_stat_cpu_notifier);
for_each_online_cpu(cpu) {
cpufreq_stats_free_table(cpu);
+ cpufreq_stats_free_debugfs(cpu);
cpufreq_stats_free_sysfs(cpu);
}
}
diff --git a/include/linux/cpufreq.h b/include/linux/cpufreq.h
index 825f379..4d0a4fd 100644
--- a/include/linux/cpufreq.h
+++ b/include/linux/cpufreq.h
@@ -141,12 +141,18 @@ static inline bool policy_is_shared(struct cpufreq_policy *policy)
#define CPUFREQ_POSTCHANGE (1)
#define CPUFREQ_RESUMECHANGE (8)
#define CPUFREQ_SUSPENDCHANGE (9)
+#define CPUFREQ_LOADCHECK (10)

struct cpufreq_freqs {
unsigned int cpu; /* cpu nr */
unsigned int old;
unsigned int new;
u8 flags; /* flags of cpufreq_driver, see below. */
+
+#ifdef CONFIG_CPU_FREQ_STAT
+ int64_t time;
+ unsigned int load[NR_CPUS];
+#endif
};


--
1.8.0

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/