From: Hou Tao <houtao1@huawei.com>

fscache_create_volume_work() uses wake_up_bit() to wake up the processes
which are waiting for the completion of volume creation. According to
comments in wake_up_bit() and waitqueue_active(), an extra smp_mb() is
needed to guarantee the memory order between FSCACHE_VOLUME_CREATING
flag and waitqueue_active() before invoking wake_up_bit().

Considering clear_bit_unlock() before wake_up_bit() is an atomic
operation, use smp_mb__after_atomic() instead of smp_mb() to provide
such guarantee.

Signed-off-by: Hou Tao <houtao1@huawei.com>
---
 fs/fscache/volume.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/fs/fscache/volume.c b/fs/fscache/volume.c
index fc3dd3bc851d..734d17f404e7 100644
--- a/fs/fscache/volume.c
+++ b/fs/fscache/volume.c
@@ -281,6 +281,11 @@ static void fscache_create_volume_work(struct work_struct *work)
 				 fscache_access_acquire_volume_end);
 
 	clear_bit_unlock(FSCACHE_VOLUME_CREATING, &volume->flags);
+	/*
+	 * Paired with barrier in wait_on_bit(). Check wake_up_bit() and
+	 * waitqueue_active() for details.
+	 */
+	smp_mb__after_atomic();
 	wake_up_bit(&volume->flags, FSCACHE_VOLUME_CREATING);
 	fscache_put_volume(volume, fscache_volume_put_create_work);
 }
-- 
2.29.2

On 12/26/22 6:33 PM, Hou Tao wrote:
> From: Hou Tao <houtao1@huawei.com>
> 
> fscache_create_volume_work() uses wake_up_bit() to wake up the processes
> which are waiting for the completion of volume creation. According to
> comments in wake_up_bit() and waitqueue_active(), an extra smp_mb() is
> needed to guarantee the memory order between FSCACHE_VOLUME_CREATING
> flag and waitqueue_active() before invoking wake_up_bit().
> 
> Considering clear_bit_unlock() before wake_up_bit() is an atomic
> operation, use smp_mb__after_atomic() instead of smp_mb() to provide
> such guarantee.
> 
> Signed-off-by: Hou Tao <houtao1@huawei.com>
> ---
>  fs/fscache/volume.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/fs/fscache/volume.c b/fs/fscache/volume.c
> index fc3dd3bc851d..734d17f404e7 100644
> --- a/fs/fscache/volume.c
> +++ b/fs/fscache/volume.c
> @@ -281,6 +281,11 @@ static void fscache_create_volume_work(struct work_struct *work)
>  				 fscache_access_acquire_volume_end);
>  
>  	clear_bit_unlock(FSCACHE_VOLUME_CREATING, &volume->flags);
> +	/*
> +	 * Paired with barrier in wait_on_bit(). Check wake_up_bit() and
> +	 * waitqueue_active() for details.
> +	 */
> +	smp_mb__after_atomic();
>  	wake_up_bit(&volume->flags, FSCACHE_VOLUME_CREATING);
>  	fscache_put_volume(volume, fscache_volume_put_create_work);
>  }

LGTM.

Actually I'm thinking if clear_and_wake_up_bit() could be used here.
Ditto for patch 1.

-- 
Thanks,
Jingbo

Hi,

On 1/12/2023 12:34 PM, Jingbo Xu wrote:
>
> On 12/26/22 6:33 PM, Hou Tao wrote:
>> From: Hou Tao <houtao1@huawei.com>
>>
>> fscache_create_volume_work() uses wake_up_bit() to wake up the processes
>> which are waiting for the completion of volume creation. According to
>> comments in wake_up_bit() and waitqueue_active(), an extra smp_mb() is
>> needed to guarantee the memory order between FSCACHE_VOLUME_CREATING
>> flag and waitqueue_active() before invoking wake_up_bit().
>>
>> Considering clear_bit_unlock() before wake_up_bit() is an atomic
>> operation, use smp_mb__after_atomic() instead of smp_mb() to provide
>> such guarantee.
>>
>> Signed-off-by: Hou Tao <houtao1@huawei.com>
>> ---
>>  fs/fscache/volume.c | 5 +++++
>>  1 file changed, 5 insertions(+)
>>
>> diff --git a/fs/fscache/volume.c b/fs/fscache/volume.c
>> index fc3dd3bc851d..734d17f404e7 100644
>> --- a/fs/fscache/volume.c
>> +++ b/fs/fscache/volume.c
>> @@ -281,6 +281,11 @@ static void fscache_create_volume_work(struct work_struct *work)
>>  				 fscache_access_acquire_volume_end);
>>  
>>  	clear_bit_unlock(FSCACHE_VOLUME_CREATING, &volume->flags);
>> +	/*
>> +	 * Paired with barrier in wait_on_bit(). Check wake_up_bit() and
>> +	 * waitqueue_active() for details.
>> +	 */
>> +	smp_mb__after_atomic();
>>  	wake_up_bit(&volume->flags, FSCACHE_VOLUME_CREATING);
>>  	fscache_put_volume(volume, fscache_volume_put_create_work);
>>  }
> LGTM.
>
> Actually I'm thinking if clear_and_wake_up_bit() could be used here.
> Ditto for patch 1.
Good idea. Just find the presence of clear_and_wake_up_bit(). Will do in v3 for
both patch 1 & patch 2.
>

Hou Tao <houtao@huaweicloud.com> wrote:

> fscache_create_volume_work() uses wake_up_bit() to wake up the processes
> which are waiting for the completion of volume creation. According to
> comments in wake_up_bit() and waitqueue_active(), an extra smp_mb() is
> needed to guarantee the memory order between FSCACHE_VOLUME_CREATING
> flag and waitqueue_active() before invoking wake_up_bit().

What two values are you ordering?

If we're using this to create a critical section, then yes, we would need a
barrier to order the changes inside the critical section before changing the
memory location that forms the lock - but this is not a critical section.

David

Hi,

On 1/12/2023 12:09 AM, David Howells wrote:
> Hou Tao <houtao@huaweicloud.com> wrote:
>
>> fscache_create_volume_work() uses wake_up_bit() to wake up the processes
>> which are waiting for the completion of volume creation. According to
>> comments in wake_up_bit() and waitqueue_active(), an extra smp_mb() is
>> needed to guarantee the memory order between FSCACHE_VOLUME_CREATING
>> flag and waitqueue_active() before invoking wake_up_bit().
> What two values are you ordering?
>
> If we're using this to create a critical section, then yes, we would need a
> barrier to order the changes inside the critical section before changing the
> memory location that forms the lock - but this is not a critical section.
It is similar with patch #1. The smp_mb() is used for order between
volume->flags and wq->head.
> David
>

David Howells <dhowells@redhat.com> wrote:

> What two values are you ordering?
> 
> If we're using this to create a critical section, then yes, we would need a
> barrier to order the changes inside the critical section before changing the
> memory location that forms the lock - but this is not a critical section.

Actually, that said, the ordering is probably between the bit being cleared
and the task state.

David