[PATCH 1/1] TTY: tty_buffer, warn on leaks

[PATCH 1/1] TTY: tty_buffer, warn on leaks

[Jiri Slaby]

When we leak something, warn about that. For that we need to account
the memory used also in the free_all method. It is handled elsewhere
correctly.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
---
 drivers/tty/tty_buffer.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/tty/tty_buffer.c b/drivers/tty/tty_buffer.c
index 5f08c39e76ab..20cb9492dbf8 100644
--- a/drivers/tty/tty_buffer.c
+++ b/drivers/tty/tty_buffer.c
@@ -115,21 +115,27 @@ void tty_buffer_free_all(struct tty_port *port)
 	struct tty_bufhead *buf = &port->buf;
 	struct tty_buffer *p, *next;
 	struct llist_node *llist;
+	int zero = 0;
 
 	while ((p = buf->head) != NULL) {
 		buf->head = p->next;
+		atomic_sub(p->size, &buf->memory_used);
 		if (p->size > 0)
 			kfree(p);
 	}
 	llist = llist_del_all(&buf->free);
-	llist_for_each_entry_safe(p, next, llist, free)
+	llist_for_each_entry_safe(p, next, llist, free) {
+		atomic_sub(p->size, &buf->memory_used);
 		kfree(p);
+	}
+
 
 	tty_buffer_reset(&buf->sentinel, 0);
 	buf->head = &buf->sentinel;
 	buf->tail = &buf->sentinel;
 
-	atomic_set(&buf->memory_used, 0);
+	atomic_xchg(&buf->memory_used, zero);
+	WARN_ON(zero != 0);
 }
 
 /**
-- 
1.8.4.2

Re: [PATCH 1/1] TTY: tty_buffer, warn on leaks

[Peter Hurley]

On 11/11/2013 03:05 PM, Jiri Slaby wrote:
> When we leak something, warn about that. For that we need to account
> the memory used also in the free_all method. It is handled elsewhere
> correctly.

Hi Jiri,

Good idea.

> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> ---
>   drivers/tty/tty_buffer.c | 10 ++++++++--
>   1 file changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/tty/tty_buffer.c b/drivers/tty/tty_buffer.c
> index 5f08c39e76ab..20cb9492dbf8 100644
> --- a/drivers/tty/tty_buffer.c
> +++ b/drivers/tty/tty_buffer.c
> @@ -115,21 +115,27 @@ void tty_buffer_free_all(struct tty_port *port)
>   	struct tty_bufhead *buf = &port->buf;
>   	struct tty_buffer *p, *next;
>   	struct llist_node *llist;
> +	int zero = 0;
>
>   	while ((p = buf->head) != NULL) {
>   		buf->head = p->next;
> +		atomic_sub(p->size, &buf->memory_used);

buf->memory_used doesn't need to be treated atomically here,
and doing so implies that it's necessary.

Accumulating p->size in a non-atomic local is probably better.

>   		if (p->size > 0)
>   			kfree(p);
>   	}
>   	llist = llist_del_all(&buf->free);
> -	llist_for_each_entry_safe(p, next, llist, free)
> +	llist_for_each_entry_safe(p, next, llist, free) {
> +		atomic_sub(p->size, &buf->memory_used);

Same here.

>   		kfree(p);
> +	}
> +
>
>   	tty_buffer_reset(&buf->sentinel, 0);
>   	buf->head = &buf->sentinel;
>   	buf->tail = &buf->sentinel;
>
> -	atomic_set(&buf->memory_used, 0);
> +	atomic_xchg(&buf->memory_used, zero);
> +	WARN_ON(zero != 0);

Then,

WARN_ON(atomic_read(&buf->memory_used) != <local accum>);
atomic_set(&buf->memory_used, 0);

?

Regards,
Peter Hurley

Re: [PATCH 1/1] TTY: tty_buffer, warn on leaks

[Peter Hurley]

On 11/11/2013 06:13 PM, Peter Hurley wrote:
> On 11/11/2013 03:05 PM, Jiri Slaby wrote:
>> When we leak something, warn about that. For that we need to account
>> the memory used also in the free_all method. It is handled elsewhere
>> correctly.
>
> Hi Jiri,
>
> Good idea.
>
>> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
>> ---
>>   drivers/tty/tty_buffer.c | 10 ++++++++--
>>   1 file changed, 8 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/tty/tty_buffer.c b/drivers/tty/tty_buffer.c
>> index 5f08c39e76ab..20cb9492dbf8 100644
>> --- a/drivers/tty/tty_buffer.c
>> +++ b/drivers/tty/tty_buffer.c
>> @@ -115,21 +115,27 @@ void tty_buffer_free_all(struct tty_port *port)
>>       struct tty_bufhead *buf = &port->buf;
>>       struct tty_buffer *p, *next;
>>       struct llist_node *llist;
>> +    int zero = 0;
>>
>>       while ((p = buf->head) != NULL) {
>>           buf->head = p->next;
>> +        atomic_sub(p->size, &buf->memory_used);
>
> buf->memory_used doesn't need to be treated atomically here,
> and doing so implies that it's necessary.
>
> Accumulating p->size in a non-atomic local is probably better.
>
>>           if (p->size > 0)
>>               kfree(p);
>>       }
>>       llist = llist_del_all(&buf->free);
>> -    llist_for_each_entry_safe(p, next, llist, free)
>> +    llist_for_each_entry_safe(p, next, llist, free) {
>> +        atomic_sub(p->size, &buf->memory_used);
>
> Same here.

Sorry, I forgot:  the buffers on the free list have already been
accounted for in tty_buffer_free() so don't form part of the
buf->mem_used total.

The buffer memory accounting was like that in 3.7 so I left it in.

static void tty_buffer_free(struct tty_port *port, struct tty_buffer *b)
{
	struct tty_bufhead *buf = &port->buf;

	/* Dumb strategy for now - should keep some stats */
	WARN_ON(atomic_sub_return(b->size, &buf->mem_used) < 0);

	if (b->size > MIN_TTYB_SIZE)
		kfree(b);
	else if (b->size > 0)
		llist_add(&b->free, &buf->free);
}

Regards,
Peter Hurley