[PATCH 1/2] procfs: /proc/schedstat fails on very very large machines.

[PATCH 1/2] procfs: /proc/schedstat fails on very very large machines.

[Nathan Zimmer]

On systems with 4096 cores doing a cat /proc/schedstat fails.
We are trying to push all the data into a single kmalloc buffer.
The issue is on these very large machines all the data will not fit in 4mb.
The failure can occur on smaller systems when memory fragmentation is high.

A better solution is to not us the single_open mechanism but to provide
our own seq_operations.

The output should be identical to previous version and thus not need the
version number.


Signed-off-by: Nathan Zimmer <nzimmer@sgi.com>
CC: Ingo Molnar <mingo@redhat.com>
CC: Peter Zijlstra <peterz@infradead.org>
CC: linux-kernel@vger.kernel.org
Reported-by: Dave Jones <davej@redhat.com>
---
 kernel/sched/stats.c |   73 ++++++++++++++++++++++++++++++++++++-------------
 1 files changed, 53 insertions(+), 20 deletions(-)

diff --git a/kernel/sched/stats.c b/kernel/sched/stats.c
index 903ffa9..33a85c9 100644
--- a/kernel/sched/stats.c
+++ b/kernel/sched/stats.c
@@ -21,9 +21,13 @@ static int show_schedstat(struct seq_file *seq, void *v)
 	if (mask_str == NULL)
 		return -ENOMEM;
 
-	seq_printf(seq, "version %d\n", SCHEDSTAT_VERSION);
-	seq_printf(seq, "timestamp %lu\n", jiffies);
-	for_each_online_cpu(cpu) {
+	if (v == (void *)1) {
+		seq_printf(seq, "version %d\n", SCHEDSTAT_VERSION);
+		seq_printf(seq, "timestamp %lu\n", jiffies);
+	} else {
+
+		cpu = (unsigned long)(v - 2);
+
 		struct rq *rq = cpu_rq(cpu);
 #ifdef CONFIG_SMP
 		struct sched_domain *sd;
@@ -72,35 +76,64 @@ static int show_schedstat(struct seq_file *seq, void *v)
 		}
 		rcu_read_unlock();
 #endif
+		kfree(mask_str);
 	}
-	kfree(mask_str);
 	return 0;
 }
 
-static int schedstat_open(struct inode *inode, struct file *file)
+static void *schedstat_start(struct seq_file *file, loff_t *offset)
 {
-	unsigned int size = PAGE_SIZE * (1 + num_online_cpus() / 32);
-	char *buf = kmalloc(size, GFP_KERNEL);
-	struct seq_file *m;
-	int res;
+	unsigned long n = *offset;
 
-	if (!buf)
-		return -ENOMEM;
-	res = single_open(file, show_schedstat, NULL);
-	if (!res) {
-		m = file->private_data;
-		m->buf = buf;
-		m->size = size;
-	} else
-		kfree(buf);
-	return res;
+	if (n == 0)
+		return (void *) 1;
+
+	n--;
+
+	if (n > 0)
+		n = cpumask_next(n - 1, cpu_online_mask);
+	else
+		n = cpumask_first(cpu_online_mask);
+
+	*offset = n + 1;
+
+	if (n < nr_cpu_ids)
+		return (void *)(unsigned long)(n + 2);
+	return NULL;
+}
+
+static void *schedstat_next(struct seq_file *file, void *data, loff_t *offset)
+{
+	(*offset)++;
+	return schedstat_start(file, offset);
+}
+
+static void schedstat_stop(struct seq_file *file, void *data)
+{
+}
+
+static const struct seq_operations schedstat_sops = {
+	.start = schedstat_start,
+	.next  = schedstat_next,
+	.stop  = schedstat_stop,
+	.show  = show_schedstat,
+};
+
+static int schedstat_open(struct inode *inode, struct file *file)
+{
+	return seq_open(file, &schedstat_sops);
 }
 
+static int schedstat_release(struct inode *inode, struct file *file)
+{
+	return 0;
+};
+
 static const struct file_operations proc_schedstat_operations = {
 	.open    = schedstat_open,
 	.read    = seq_read,
 	.llseek  = seq_lseek,
-	.release = single_release,
+	.release = schedstat_release,
 };
 
 static int __init proc_schedstat_init(void)
-- 
1.6.0.2

[PATCH 2/2] procfs: /proc/sched_debug fails on very very large machines.

[Nathan Zimmer]

On systems with 4096 cores attemping to read /proc/sched_debug fails.
We are trying to push all the data into a single kmalloc buffer.
The issue is on these very large machines all the data will not fit in 4mb.
The failure can occur on smaller systems when memory fragmentation is high.

A better solution is to not us the single_open mechanism but to provide
our own seq_operations and treat each cpu as an individual record.
Also sysrq_sched_debug_show must be updated to to not use the new 
sched_debug_show. 

The output should be identical to previous version.

Signed-off-by: Nathan Zimmer <nzimmer@sgi.com>
CC: Ingo Molnar <mingo@redhat.com>
CC: Peter Zijlstra <peterz@infradead.org>
CC: linux-kernel@vger.kernel.org
Reported-by: Dave Jones <davej@redhat.com>
---
 kernel/sched/debug.c |   84 +++++++++++++++++++++++++++++++++++++++++++------
 1 files changed, 73 insertions(+), 11 deletions(-)

diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c
index 6f79596..1ffdd42 100644
--- a/kernel/sched/debug.c
+++ b/kernel/sched/debug.c
@@ -253,11 +253,11 @@ static void print_cpu(struct seq_file *m, int cpu)
 	{
 		unsigned int freq = cpu_khz ? : 1;
 
-		SEQ_printf(m, "\ncpu#%d, %u.%03u MHz\n",
+		SEQ_printf(m, "cpu#%d, %u.%03u MHz\n",
 			   cpu, freq / 1000, (freq % 1000));
 	}
 #else
-	SEQ_printf(m, "\ncpu#%d\n", cpu);
+	SEQ_printf(m, "cpu#%d\n", cpu);
 #endif
 
 #define P(x)								\
@@ -314,6 +314,7 @@ do {									\
 	print_rq(m, rq, cpu);
 	rcu_read_unlock();
 	spin_unlock_irqrestore(&sched_debug_lock, flags);
+	SEQ_printf(m, "\n");
 }
 
 static const char *sched_tunable_scaling_names[] = {
@@ -322,11 +323,10 @@ static const char *sched_tunable_scaling_names[] = {
 	"linear"
 };
 
-static int sched_debug_show(struct seq_file *m, void *v)
+static void sched_debug_header(struct seq_file *m)
 {
 	u64 ktime, sched_clk, cpu_clk;
 	unsigned long flags;
-	int cpu;
 
 	local_irq_save(flags);
 	ktime = ktime_to_ns(ktime_get());
@@ -368,33 +368,95 @@ static int sched_debug_show(struct seq_file *m, void *v)
 #undef PN
 #undef P
 
-	SEQ_printf(m, "  .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
+	SEQ_printf(m, "  .%-40s: %d (%s)\n",
+		"sysctl_sched_tunable_scaling",
 		sysctl_sched_tunable_scaling,
 		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
+	SEQ_printf(m, "\n");
+}
 
-	for_each_online_cpu(cpu)
-		print_cpu(m, cpu);
+static int sched_debug_show(struct seq_file *m, void *v)
+{
+	int cpu = (unsigned long)(v - 2);
 
-	SEQ_printf(m, "\n");
+	if (cpu != -1)
+		print_cpu(m, cpu);
+	else
+		sched_debug_header(m);
 
 	return 0;
 }
 
 void sysrq_sched_debug_show(void)
 {
-	sched_debug_show(NULL, NULL);
+	int cpu;
+
+	sched_debug_header(NULL);
+	for_each_online_cpu(cpu)
+		print_cpu(NULL, cpu);
+
+}
+
+static void *sched_debug_start(struct seq_file *file, loff_t *offset)
+{
+	unsigned long n = *offset;
+
+	if (n == 0)
+		return (void *) 1;
+
+	n--;
+
+	if (n > 0)
+		n = cpumask_next(n - 1, cpu_online_mask);
+	else
+		n = cpumask_first(cpu_online_mask);
+
+	*offset = n + 1;
+
+	if (n < nr_cpu_ids)
+		return (void *)(unsigned long)(n + 2);
+	return NULL;
+}
+
+static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset)
+{
+	(*offset)++;
+	return sched_debug_start(file, offset);
+}
+
+static void sched_debug_stop(struct seq_file *file, void *data)
+{
+}
+
+
+static const struct seq_operations sched_debug_sops = {
+	.start = sched_debug_start,
+	.next = sched_debug_next,
+	.stop = sched_debug_stop,
+	.show = sched_debug_show,
+};
+
+static int sched_debug_release(struct inode *inode, struct file *file)
+{
+	seq_release(inode, file);
+
+	return 0;
 }
 
 static int sched_debug_open(struct inode *inode, struct file *filp)
 {
-	return single_open(filp, sched_debug_show, NULL);
+	int ret = 0;
+
+	ret = seq_open(filp, &sched_debug_sops);
+
+	return ret;
 }
 
 static const struct file_operations sched_debug_fops = {
 	.open		= sched_debug_open,
 	.read		= seq_read,
 	.llseek		= seq_lseek,
-	.release	= single_release,
+	.release	= sched_debug_release,
 };
 
 static int __init init_sched_debug_procfs(void)
-- 
1.6.0.2