On 8/9/21 8:08 AM, Hao Xu wrote:
> 在 2021/8/9 下午10:01, Jens Axboe 写道:
>> On 8/8/21 7:54 AM, Hao Xu wrote:
>>> The former patch to add check between nr_workers and max_workers has a
>>> bug, which will cause unconditionally creating io-workers. That's
>>> because the result of the check doesn't affect the call of
>>> create_io_worker(), fix it by bringing in a boolean value for it.
>>>
>>> Fixes: 21698274da5b ("io-wq: fix lack of acct->nr_workers < acct->max_workers judgement")
>>> Signed-off-by: Hao Xu <haoxu@xxxxxxxxxxxxxxxxx>
>>> ---
>>> fs/io-wq.c | 19 ++++++++++++++-----
>>> 1 file changed, 14 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/fs/io-wq.c b/fs/io-wq.c
>>> index 12fc19353bb0..5536b2a008d1 100644
>>> --- a/fs/io-wq.c
>>> +++ b/fs/io-wq.c
>>> @@ -252,14 +252,15 @@ static void io_wqe_wake_worker(struct io_wqe *wqe, struct io_wqe_acct *acct)
>>>
>>> raw_spin_lock_irq(&wqe->lock);
>>> if (acct->nr_workers < acct->max_workers) {
>>> - atomic_inc(&acct->nr_running);
>>> - atomic_inc(&wqe->wq->worker_refs);
>>> acct->nr_workers++;
>>> do_create = true;
>>> }
>>> raw_spin_unlock_irq(&wqe->lock);
>>> - if (do_create)
>>> + if (do_create) {
>>> + atomic_inc(&acct->nr_running);
>>> + atomic_inc(&wqe->wq->worker_refs);
>>> create_io_worker(wqe->wq, wqe, acct->index);
>>> + }
>>> }
>>
>> I don't get this hunk - we already know we're creating a worker, what's the
>> point in moving the incs?
>>
> Actually not much difference, I think we don't need to protect
> nr_running and worker_refs by wqe->lock, so narrow the range of
> raw_spin_lock_irq - raw_spin_unlock_irq
Agree, we don't need it, but it's not a fix as such. I'd rather defer that
one to a separate cleanup for the next release.
--
Jens Axboe
