2023-12-23 03:39:34

by Li Nan

Subject: [PATCH v3 2/2] md: don't account sync_io if iostats of the disk is disabled

From: Li Nan <[email protected]>

If iostats is disabled, disk_stats will not be updated and
part_stat_read_accum() only returns a constant value. In this case,
continuing to count sync_io and to check is_mddev_idle() is no longer
meaningful.

Signed-off-by: Li Nan <[email protected]>
---
drivers/md/md.h | 3 ++-
drivers/md/md.c | 4 ++++
2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/md/md.h b/drivers/md/md.h
index 1a4f976951c1..e2d03a7a858c 100644
--- a/drivers/md/md.h
+++ b/drivers/md/md.h
@@ -584,7 +584,8 @@ extern void mddev_unlock(struct mddev *mddev);

static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
{
- atomic64_add(nr_sectors, &bdev->bd_disk->sync_io);
+ if (blk_queue_io_stat(bdev->bd_disk->queue))
+ atomic64_add(nr_sectors, &bdev->bd_disk->sync_io);
}

static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
diff --git a/drivers/md/md.c b/drivers/md/md.c
index a6829ea5b560..b56614eae8dc 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -8502,6 +8502,10 @@ static int is_mddev_idle(struct mddev *mddev, int init)
rcu_read_lock();
rdev_for_each_rcu(rdev, mddev) {
struct gendisk *disk = rdev->bdev->bd_disk;
+
+ if (!blk_queue_io_stat(disk->queue))
+ continue;
+
curr_events =
(long long)part_stat_read_accum(disk->part0, sectors) -
atomic64_read(&disk->sync_io);
--
2.39.2

2024-01-03 03:16:33

by Yu Kuai

Subject: Re: [PATCH v3 2/2] md: don't account sync_io if iostats of the disk is disabled

?? 2023/12/23 11:37, [email protected] д??:
> From: Li Nan <[email protected]>
>
> If iostats is disabled, disk_stats will not be updated and
> part_stat_read_accum() only returns a constant value. In this case,
> continuing to count sync_io and to check is_mddev_idle() is no longer
> meaningful.
>
> Signed-off-by: Li Nan <[email protected]>
> ---
> drivers/md/md.h | 3 ++-
> drivers/md/md.c | 4 ++++
> 2 files changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/md/md.h b/drivers/md/md.h
> index 1a4f976951c1..e2d03a7a858c 100644
> --- a/drivers/md/md.h
> +++ b/drivers/md/md.h
> @@ -584,7 +584,8 @@ extern void mddev_unlock(struct mddev *mddev);
>
> static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
> {
> - atomic64_add(nr_sectors, &bdev->bd_disk->sync_io);
> + if (blk_queue_io_stat(bdev->bd_disk->queue))
> + atomic64_add(nr_sectors, &bdev->bd_disk->sync_io);
> }
>
> static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
> diff --git a/drivers/md/md.c b/drivers/md/md.c
> index a6829ea5b560..b56614eae8dc 100644
> --- a/drivers/md/md.c
> +++ b/drivers/md/md.c
> @@ -8502,6 +8502,10 @@ static int is_mddev_idle(struct mddev *mddev, int init)
> rcu_read_lock();
> rdev_for_each_rcu(rdev, mddev) {
> struct gendisk *disk = rdev->bdev->bd_disk;
> +
> + if (!blk_queue_io_stat(disk->queue))
> + continue;

Consider that the queue flag can be set/cleared through sysfs, let's
keep set rdev->last_events in the case 'init'. To prevent a false
positive(althrough highly unlikely) if iostat is enabled during
md_do_sync().

Thanks,
Kuai

> +
> curr_events =
> (long long)part_stat_read_accum(disk->part0, sectors) -
> atomic64_read(&disk->sync_io);
>

2024-01-03 06:36:01

by Li Nan

Subject: Re: [PATCH v3 2/2] md: don't account sync_io if iostats of the disk is disabled

在 2024/1/3 11:16, Yu Kuai 写道:
>
>
> 在 2023/12/23 11:37, [email protected] 写道:
>> From: Li Nan <[email protected]>
>>
>> If iostats is disabled, disk_stats will not be updated and
>> part_stat_read_accum() only returns a constant value. In this case,
>> continuing to count sync_io and to check is_mddev_idle() is no longer
>> meaningful.
>>
>> Signed-off-by: Li Nan <[email protected]>
>> ---
>>   drivers/md/md.h | 3 ++-
>>   drivers/md/md.c | 4 ++++
>>   2 files changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/md/md.h b/drivers/md/md.h
>> index 1a4f976951c1..e2d03a7a858c 100644
>> --- a/drivers/md/md.h
>> +++ b/drivers/md/md.h
>> @@ -584,7 +584,8 @@ extern void mddev_unlock(struct mddev *mddev);
>>   static inline void md_sync_acct(struct block_device *bdev, unsigned
>> long nr_sectors)
>>   {
>> -    atomic64_add(nr_sectors, &bdev->bd_disk->sync_io);
>> +    if (blk_queue_io_stat(bdev->bd_disk->queue))
>> +        atomic64_add(nr_sectors, &bdev->bd_disk->sync_io);
>>   }
>>   static inline void md_sync_acct_bio(struct bio *bio, unsigned long
>> nr_sectors)
>> diff --git a/drivers/md/md.c b/drivers/md/md.c
>> index a6829ea5b560..b56614eae8dc 100644
>> --- a/drivers/md/md.c
>> +++ b/drivers/md/md.c
>> @@ -8502,6 +8502,10 @@ static int is_mddev_idle(struct mddev *mddev,
>> int init)
>>       rcu_read_lock();
>>       rdev_for_each_rcu(rdev, mddev) {
>>           struct gendisk *disk = rdev->bdev->bd_disk;
>> +
>> +        if (!blk_queue_io_stat(disk->queue))
>> +            continue;
>
> Consider that the queue flag can be set/cleared through sysfs, let's
> keep set rdev->last_events in the case 'init'. To prevent a false
> positive(althrough highly unlikely) if iostat is enabled during
> md_do_sync().
>

Thanks for your review, I will add checks of 'init' in next version.

> Thanks,
> Kuai
>
>> +
>>           curr_events =
>>               (long long)part_stat_read_accum(disk->part0, sectors) -
>>               atomic64_read(&disk->sync_io);
>>

--
Thanks,
Nan