diff mbox series

mm/damon/reclaim: fix 'enabled' is incorrectly set because 'system_wq' is not initialized

Message ID 20220715061609.48463-1-songyuanzheng@huawei.com (mailing list archive)
State New
Headers show
Series mm/damon/reclaim: fix 'enabled' is incorrectly set because 'system_wq' is not initialized | expand

Commit Message

Yuanzheng Song July 15, 2022, 6:16 a.m. UTC
The 'enabled' will be incorrectly set because the 'system_wq'
might not initialized yet. This results in 'enabled=true',
but the 'damon_reclaim_timer' is inactive. So fix it by moving
the judgment logic of the 'damon_reclaim_initialized' to the
start position of the enable_store().

Fixes: 294928293813 ("mm/damon/reclaim: schedule 'damon_reclaim_timer' only after 'system_wq' is initialized")
Signed-off-by: Yuanzheng Song <songyuanzheng@huawei.com>
---
 mm/damon/reclaim.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

Comments

SeongJae Park July 15, 2022, 5:07 p.m. UTC | #1
Hello Yuanzheng,


On Fri, 15 Jul 2022 06:16:09 +0000 Yuanzheng Song <songyuanzheng@huawei.com> wrote:

> The 'enabled' will be incorrectly set because the 'system_wq'
> might not initialized yet. This results in 'enabled=true',
> but the 'damon_reclaim_timer' is inactive. So fix it by moving
> the judgment logic of the 'damon_reclaim_initialized' to the
> start position of the enable_store().

Thank you for this patch!

In the case, however, 'damon_reclaim_init()' will activate
'damon_reclaim_timer', which will check the 'enabled' and start DAMON later.
So 'enabled' will inconsistently set while DAMON_RECLAIM is not really enabled
for a moment, but those will eventually be consistent.  This patch could reduce
the duration of the inconsistent state.

However, this would break boot-time DAMON_RECLAIM enabling, which adds
'damon_reclaim.enabled=true' to the kernel parameter, as this change will set
'enabled' as 'false' in the early 'damon_reclaim_enabled_store()', so that the
later 'damon_reclaim_initialized()' activated 'damon_reclaim_timer' shows
'enabled' as 'false' and therefore doesn't start DAMON.

If there is anything I'm missing, please let me know.


Thanks,
SJ

> 
> Fixes: 294928293813 ("mm/damon/reclaim: schedule 'damon_reclaim_timer' only after 'system_wq' is initialized")
> Signed-off-by: Yuanzheng Song <songyuanzheng@huawei.com>
> ---
>  mm/damon/reclaim.c | 9 +++++----
>  1 file changed, 5 insertions(+), 4 deletions(-)
> 
> diff --git a/mm/damon/reclaim.c b/mm/damon/reclaim.c
> index e69b807fefe4..b13d5a02bf2e 100644
> --- a/mm/damon/reclaim.c
> +++ b/mm/damon/reclaim.c
> @@ -374,13 +374,14 @@ static bool damon_reclaim_initialized;
>  static int damon_reclaim_enabled_store(const char *val,
>  		const struct kernel_param *kp)
>  {
> -	int rc = param_set_bool(val, kp);
> -
> -	if (rc < 0)
> -		return rc;
> +	int rc;
>  
>  	/* system_wq might not initialized yet */
>  	if (!damon_reclaim_initialized)
> +		return -EINVAL;
> +
> +	rc = param_set_bool(val, kp);
> +	if (rc < 0)
>  		return rc;
>  
>  	schedule_delayed_work(&damon_reclaim_timer, 0);
> -- 
> 2.25.1
Yuanzheng Song July 18, 2022, 4:48 a.m. UTC | #2
Hello SJ,

Thank you very much for your reply, through your reply let me understand 
this principle, so please ignore this patch.

Thanks,

Yuanzheng

On 2022/7/16 1:07, SeongJae Park wrote:
> Hello Yuanzheng,
>
>
> On Fri, 15 Jul 2022 06:16:09 +0000 Yuanzheng Song <songyuanzheng@huawei.com> wrote:
>
>> The 'enabled' will be incorrectly set because the 'system_wq'
>> might not initialized yet. This results in 'enabled=true',
>> but the 'damon_reclaim_timer' is inactive. So fix it by moving
>> the judgment logic of the 'damon_reclaim_initialized' to the
>> start position of the enable_store().
> Thank you for this patch!
>
> In the case, however, 'damon_reclaim_init()' will activate
> 'damon_reclaim_timer', which will check the 'enabled' and start DAMON later.
> So 'enabled' will inconsistently set while DAMON_RECLAIM is not really enabled
> for a moment, but those will eventually be consistent.  This patch could reduce
> the duration of the inconsistent state.
>
> However, this would break boot-time DAMON_RECLAIM enabling, which adds
> 'damon_reclaim.enabled=true' to the kernel parameter, as this change will set
> 'enabled' as 'false' in the early 'damon_reclaim_enabled_store()', so that the
> later 'damon_reclaim_initialized()' activated 'damon_reclaim_timer' shows
> 'enabled' as 'false' and therefore doesn't start DAMON.
>
> If there is anything I'm missing, please let me know.
>
>
> Thanks,
> SJ
>
>> Fixes: 294928293813 ("mm/damon/reclaim: schedule 'damon_reclaim_timer' only after 'system_wq' is initialized")
>> Signed-off-by: Yuanzheng Song <songyuanzheng@huawei.com>
>> ---
>>   mm/damon/reclaim.c | 9 +++++----
>>   1 file changed, 5 insertions(+), 4 deletions(-)
>>
>> diff --git a/mm/damon/reclaim.c b/mm/damon/reclaim.c
>> index e69b807fefe4..b13d5a02bf2e 100644
>> --- a/mm/damon/reclaim.c
>> +++ b/mm/damon/reclaim.c
>> @@ -374,13 +374,14 @@ static bool damon_reclaim_initialized;
>>   static int damon_reclaim_enabled_store(const char *val,
>>   		const struct kernel_param *kp)
>>   {
>> -	int rc = param_set_bool(val, kp);
>> -
>> -	if (rc < 0)
>> -		return rc;
>> +	int rc;
>>   
>>   	/* system_wq might not initialized yet */
>>   	if (!damon_reclaim_initialized)
>> +		return -EINVAL;
>> +
>> +	rc = param_set_bool(val, kp);
>> +	if (rc < 0)
>>   		return rc;
>>   
>>   	schedule_delayed_work(&damon_reclaim_timer, 0);
>> -- 
>> 2.25.1
> .
diff mbox series

Patch

diff --git a/mm/damon/reclaim.c b/mm/damon/reclaim.c
index e69b807fefe4..b13d5a02bf2e 100644
--- a/mm/damon/reclaim.c
+++ b/mm/damon/reclaim.c
@@ -374,13 +374,14 @@  static bool damon_reclaim_initialized;
 static int damon_reclaim_enabled_store(const char *val,
 		const struct kernel_param *kp)
 {
-	int rc = param_set_bool(val, kp);
-
-	if (rc < 0)
-		return rc;
+	int rc;
 
 	/* system_wq might not initialized yet */
 	if (!damon_reclaim_initialized)
+		return -EINVAL;
+
+	rc = param_set_bool(val, kp);
+	if (rc < 0)
 		return rc;
 
 	schedule_delayed_work(&damon_reclaim_timer, 0);