Re: [PATCH] io_uring: update sq_thread_idle after ctx deleted

[Hao Xu <haoxu@linux.alibaba.com>]

在 2021/4/25 上午8:03, Pavel Begunkov 写道:
> On 4/24/21 10:26 AM, Hao Xu wrote:
>> we shall update sq_thread_idle anytime we do ctx deletion from ctx_list
> 
> looks good, a nit below
> 
>>
>> Fixes:734551df6f9b ("io_uring: fix shared sqpoll cancellation hangs")
>> Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
>> ---
>>   fs/io_uring.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index 40f38256499c..15f204274761 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -8867,6 +8867,7 @@ static void io_sqpoll_cancel_cb(struct callback_head *cb)
>>   	if (sqd->thread)
>>   		io_uring_cancel_sqpoll(sqd);
>>   	list_del_init(&work->ctx->sqd_list);
>> +	io_sqd_update_thread_idle(sqd);
>>   	complete(&work->completion);
>>   }
>>   
>> @@ -8877,7 +8878,6 @@ static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx)
>>   	struct task_struct *task;
>>   
>>   	io_sq_thread_park(sqd);
>> -	io_sqd_update_thread_idle(sqd);
>>   	task = sqd->thread;
>>   	if (task) {
>>   		init_completion(&work.completion);
>> @@ -8886,6 +8886,7 @@ static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx)
>>   		wake_up_process(task);
>>   	} else {
>>   		list_del_init(&ctx->sqd_list);
>> +		io_sqd_update_thread_idle(sqd);
> 
> Not actually needed, it's already dying.
> 
Yep, we could keep it since it's not in hot path.
>>   	}
>>   	io_sq_thread_unpark(sqd);
>>   
>>
>