[PATCH -next] blk-iocost: fix sleeping in atomic warning for wbt_enable_default()

[PATCH -next] blk-iocost: fix sleeping in atomic warning for wbt_enable_default()

[Yu Kuai]

From: Yu Kuai <yukuai3@huawei.com>

There are following smatch warning:

block/blk-wbt.c:843 wbt_init() warn: sleeping in atomic context
ioc_qos_write() <- disables preempt
-> wbt_enable_default()
   -> wbt_init()

wbt_init() will be called from wbt_enable_default() if wbt is not
initialized, currently wbt is initialized in blk_register_queue(), hence
wbt_init() will never be called from iocost and this warning is false
positive.

However, we might support rq_qos destruction dynamically in the future,
and it's better to prevent that, hence move wbt_enable_default() outside
'ioc->lock'. This is safe because queue is still freezed.

Reported-by: Dan Carpenter <error27@gmail.com>
Link: https://lore.kernel.org/lkml/Y+Ja5SRs886CEz7a@kadam/
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
---
 block/blk-iocost.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/block/blk-iocost.c b/block/blk-iocost.c
index 7a2dc9dc8e3b..03bfe1dda07c 100644
--- a/block/blk-iocost.c
+++ b/block/blk-iocost.c
@@ -3279,11 +3279,9 @@ static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
 		blk_stat_enable_accounting(disk->queue);
 		blk_queue_flag_set(QUEUE_FLAG_RQ_ALLOC_TIME, disk->queue);
 		ioc->enabled = true;
-		wbt_disable_default(disk);
 	} else {
 		blk_queue_flag_clear(QUEUE_FLAG_RQ_ALLOC_TIME, disk->queue);
 		ioc->enabled = false;
-		wbt_enable_default(disk);
 	}
 
 	if (user) {
@@ -3296,6 +3294,10 @@ static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
 	ioc_refresh_params(ioc, true);
 	spin_unlock_irq(&ioc->lock);
 
+	if (enable)
+		wbt_disable_default(disk);
+	else
+		wbt_enable_default(disk);
 	blk_mq_unquiesce_queue(disk->queue);
 	blk_mq_unfreeze_queue(disk->queue);
 
-- 
2.31.1

Re: [PATCH -next] blk-iocost: fix sleeping in atomic warning for wbt_enable_default()

[Tejun Heo]

On Wed, Feb 08, 2023 at 11:48:03AM +0800, Yu Kuai wrote:
> diff --git a/block/blk-iocost.c b/block/blk-iocost.c
> index 7a2dc9dc8e3b..03bfe1dda07c 100644
> --- a/block/blk-iocost.c
> +++ b/block/blk-iocost.c
> @@ -3279,11 +3279,9 @@ static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
>  		blk_stat_enable_accounting(disk->queue);
>  		blk_queue_flag_set(QUEUE_FLAG_RQ_ALLOC_TIME, disk->queue);
>  		ioc->enabled = true;
> -		wbt_disable_default(disk);
>  	} else {
>  		blk_queue_flag_clear(QUEUE_FLAG_RQ_ALLOC_TIME, disk->queue);
>  		ioc->enabled = false;
> -		wbt_enable_default(disk);
>  	}
>  
>  	if (user) {
> @@ -3296,6 +3294,10 @@ static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
>  	ioc_refresh_params(ioc, true);
>  	spin_unlock_irq(&ioc->lock);
>  
> +	if (enable)
> +		wbt_disable_default(disk);
> +	else
> +		wbt_enable_default(disk);

Wouldn't this allow two competiting config attempts to race each other and
leave wbt in an unexpected state?

	task1				task2

	ioc_qos_write()			ioc_qos_write()
	lock()
	enable
	unlock()
					lock()
					disable
					unlock()
					wbt_enable_default()
	wbt_disable_default()

Thanks.

-- 
tejun