We want to be able to filter out the output on a threshold basis,
in this way we can get rid of a lot of noise and focus only on those
stacks which have an allegedly high counter.

We can control the threshold value by a new file called
'page_owner_threshold', which is 0 by default.

Signed-off-by: Oscar Salvador <osalvador@suse.de>
---
 include/linux/stackdepot.h |  3 +-
 lib/stackdepot.c           |  6 ++--
 mm/page_owner.c            | 61 +++++++++++++++++++++++++++++++++++++-
 3 files changed, 66 insertions(+), 4 deletions(-)

diff --git a/include/linux/stackdepot.h b/include/linux/stackdepot.h
index 20f62039f23a..ee66be40a152 100644
--- a/include/linux/stackdepot.h
+++ b/include/linux/stackdepot.h
@@ -26,7 +26,8 @@ depot_stack_handle_t __stack_depot_save(unsigned long *entries,
 					stack_action_t action);
 void stack_depot_dec_count(depot_stack_handle_t handle);
 int stack_depot_print_stacks_threshold(char *buf, size_t size, loff_t *pos,
-				       unsigned long *last_stack);
+				       unsigned long *last_stack,
+				       unsigned long threshold);
 
 /*
  * Every user of stack depot has to call stack_depot_init() during its own init
diff --git a/lib/stackdepot.c b/lib/stackdepot.c
index 3090ae0f3958..b4a04f09a7b7 100644
--- a/lib/stackdepot.c
+++ b/lib/stackdepot.c
@@ -528,7 +528,8 @@ depot_stack_handle_t stack_depot_save_action(unsigned long *entries,
 EXPORT_SYMBOL_GPL(stack_depot_save_action);
 
 int stack_depot_print_stacks_threshold(char *buf, size_t size, loff_t *pos,
-				       unsigned long *last_stack)
+				       unsigned long *last_stack,
+				       unsigned long threshold)
 {
 	struct stack_record *stack = NULL, *last;
 	struct stack_record **stacks;
@@ -547,7 +548,8 @@ int stack_depot_print_stacks_threshold(char *buf, size_t size, loff_t *pos,
 	for (; stack; stack = stack->next) {
 		if (!stack->size || stack->size < 0 ||
 		    stack->size > size || stack->handle.valid != 1 ||
-		    refcount_read(&stack->count) < 1)
+		    refcount_read(&stack->count) < 1 ||
+		    refcount_read(&stack->count) < threshold)
 			continue;
 
 		ret += stack_trace_snprint(buf, size, stack->entries, stack->size, 0);
diff --git a/mm/page_owner.c b/mm/page_owner.c
index 8c67c7eb2451..ef10cf44aaec 100644
--- a/mm/page_owner.c
+++ b/mm/page_owner.c
@@ -44,6 +44,7 @@ static depot_stack_handle_t early_handle;
 static void init_early_allocated_pages(void);
 
 static unsigned long last_stack = 0;
+static unsigned long threshold_count = 0;
 
 static int __init early_page_owner_param(char *buf)
 {
@@ -676,7 +677,8 @@ static ssize_t read_page_owner_stacks(struct file *file, char __user *buf,
 	if (!kbuf)
 		return ENOMEM;
 
-	ret += stack_depot_print_stacks_threshold(kbuf, count, pos, &last_stack);
+	ret += stack_depot_print_stacks_threshold(kbuf, count, pos, &last_stack,
+						  threshold_count);
 	if (copy_to_user(buf, kbuf, ret))
 		ret = -EFAULT;
 
@@ -687,6 +689,61 @@ static ssize_t read_page_owner_stacks(struct file *file, char __user *buf,
 	return ret;
 }
 
+static ssize_t read_page_owner_threshold(struct file *file, char __user *buf,
+					 size_t count, loff_t *pos)
+{
+	char *kbuf;
+	int ret = 0;
+
+	count = min_t(size_t, count, PAGE_SIZE);
+
+	if (*pos >= count)
+		return 0;
+
+	kbuf = kmalloc(count, GFP_KERNEL);
+	if (!kbuf)
+		return ENOMEM;
+
+	ret = scnprintf(kbuf, count, "%lu\n", threshold_count);
+	if (copy_to_user(buf, kbuf, ret))
+		ret = -EFAULT;
+
+	*pos += count;
+	kfree(kbuf);
+
+	return ret;
+}
+
+static ssize_t write_page_owner_threshold(struct file *file, const char __user *buf,
+					  size_t count, loff_t *pos)
+{
+	char *kbuf;
+	int ret = 0;
+
+	count = min_t(size_t, count, PAGE_SIZE);
+	kbuf = kmalloc(count, GFP_KERNEL);
+	if (!kbuf)
+		return -ENOMEM;
+
+	if (copy_from_user(kbuf, buf, count)) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	kbuf[count - 1] = '\0';
+
+	ret = kstrtoul(kbuf, 10, &threshold_count);
+
+out:
+	kfree(kbuf);
+	return ret ? ret : count;
+}
+
+static const struct file_operations proc_page_owner_threshold = {
+	.read = read_page_owner_threshold,
+	.write = write_page_owner_threshold,
+};
+
 static const struct file_operations proc_page_owner_stacks = {
 	.read = read_page_owner_stacks,
 };
@@ -706,6 +763,8 @@ static int __init pageowner_init(void)
 			    &proc_page_owner_operations);
 	debugfs_create_file("page_owner_stacks", 0400, NULL, NULL,
 			    &proc_page_owner_stacks);
+	debugfs_create_file("page_owner_threshold", 0600, NULL, NULL,
+			     &proc_page_owner_threshold);
 
 	return 0;
 }
-- 
2.35.3

On Thu 01-09-22 06:42:49, Oscar Salvador wrote:
[...]
> +static ssize_t read_page_owner_threshold(struct file *file, char __user *buf,
> +					 size_t count, loff_t *pos)
> +{
> +	char *kbuf;
> +	int ret = 0;
> +
> +	count = min_t(size_t, count, PAGE_SIZE);
> +
> +	if (*pos >= count)
> +		return 0;
> +
> +	kbuf = kmalloc(count, GFP_KERNEL);

No, you do not want to trigger user defined allocation like that. I
would use seq_file.
-- 
Michal Hocko
SUSE Labs

On Thu, Sep 01, 2022 at 10:40:32AM +0200, Michal Hocko wrote:
> On Thu 01-09-22 06:42:49, Oscar Salvador wrote:
> [...]
> > +static ssize_t read_page_owner_threshold(struct file *file, char __user *buf,
> > +					 size_t count, loff_t *pos)
> > +{
> > +	char *kbuf;
> > +	int ret = 0;
> > +
> > +	count = min_t(size_t, count, PAGE_SIZE);
> > +
> > +	if (*pos >= count)
> > +		return 0;
> > +
> > +	kbuf = kmalloc(count, GFP_KERNEL);
> 
> No, you do not want to trigger user defined allocation like that. I
> would use seq_file.

Sure, will use that.

Thanks!


-- 
Oscar Salvador
SUSE Labs

On 9/1/22 11:42 AM, Oscar Salvador wrote:> +static ssize_t read_page_owner_threshold(struct file *file, char __user *buf,
> +					 size_t count, loff_t *pos)
> +{
> +	char *kbuf;
> +	int ret = 0;
> +
> +	count = min_t(size_t, count, PAGE_SIZE);
> +
> +	if (*pos >= count)
> +		return 0;
> +
> +	kbuf = kmalloc(count, GFP_KERNEL);
> +	if (!kbuf)
> +		return ENOMEM;

Missing a negative sign, return -ENOMEM.

> +	ret = scnprintf(kbuf, count, "%lu\n", threshold_count);
> +	if (copy_to_user(buf, kbuf, ret))
> +		ret = -EFAULT;
> +
> +	*pos += count;
> +	kfree(kbuf);
> +
> +	return ret;
> +}
> +
> +static ssize_t write_page_owner_threshold(struct file *file, const char __user *buf,
> +					  size_t count, loff_t *pos)
> +{
> +	char *kbuf;
> +	int ret = 0;
> +
> +	count = min_t(size_t, count, PAGE_SIZE);
> +	kbuf = kmalloc(count, GFP_KERNEL);

This looks overestimating to me. For unsigned long, on a 64-bit system
has max val 18446744073709551615 (20 chars).

You can use stack a allocated local variable with length 21. No need
to use kmalloc(). The same way with the read() op.

> +	if (!kbuf)
> +		return -ENOMEM;
> +
> +	if (copy_from_user(kbuf, buf, count)) {
> +		ret = -EFAULT;
> +		goto out;
> +	}
> +
> +	kbuf[count - 1] = '\0';
> +
> +	ret = kstrtoul(kbuf, 10, &threshold_count);
> +
> +out:
> +	kfree(kbuf);
> +	return ret ? ret : count;
> +}
> +
> +static const struct file_operations proc_page_owner_threshold = {
> +	.read = read_page_owner_threshold,
> +	.write = write_page_owner_threshold,
> +};

-- 
Ammar Faizi

On Thu, Sep 01, 2022 at 03:31:51PM +0700, Ammar Faizi wrote:
> On 9/1/22 11:42 AM, Oscar Salvador wrote:> +static ssize_t read_page_owner_threshold(struct file *file, char __user *buf,
> > +	kbuf = kmalloc(count, GFP_KERNEL);
> > +	if (!kbuf)
> > +		return ENOMEM;
> 
> Missing a negative sign, return -ENOMEM.

Will fix.

> > +static ssize_t write_page_owner_threshold(struct file *file, const char __user *buf,
> > +					  size_t count, loff_t *pos)
> > +{
> > +	char *kbuf;
> > +	int ret = 0;
> > +
> > +	count = min_t(size_t, count, PAGE_SIZE);
> > +	kbuf = kmalloc(count, GFP_KERNEL);
> 
> This looks overestimating to me. For unsigned long, on a 64-bit system
> has max val 18446744073709551615 (20 chars).
> 
> You can use stack a allocated local variable with length 21. No need
> to use kmalloc(). The same way with the read() op.

Probably could do that, but I'll go with Michal's option and will use
seq_file.

Thanks!


-- 
Oscar Salvador
SUSE Labs