Re: [PATCH v5 2/2] io_uring: avoid unnecessary io_wq_work copy for fast poll feature

[Pavel Begunkov <asml.silence@gmail.com>]

On 02/06/2020 04:16, Xiaoguang Wang wrote:
> hi Jens, Pavel,
> 
> Will you have a look at this V5 version? Or we hold on this patchset, and
> do the refactoring work related io_wq_work firstly.

It's entirely up to Jens, but frankly, I think it'll bring more bugs than
merits in the current state of things.

> 
> 
>> Basically IORING_OP_POLL_ADD command and async armed poll handlers
>> for regular commands don't touch io_wq_work, so only REQ_F_WORK_INITIALIZED
>> is set, can we do io_wq_work copy and restore.
>>
>> Signed-off-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
>> ---
>> V3:
>>    drop the REQ_F_WORK_NEED_RESTORE flag introduced in V2 patch, just
>>    use REQ_F_WORK_INITIALIZED to control whether to do io_wq_work copy
>>    and restore.
>> ---
>>   fs/io_uring.c | 13 +++++++++----
>>   1 file changed, 9 insertions(+), 4 deletions(-)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index 8e022d0f0c86..b761ef7366f9 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -4527,7 +4527,8 @@ static void io_async_task_func(struct callback_head *cb)
>>       spin_unlock_irq(&ctx->completion_lock);
>>         /* restore ->work in case we need to retry again */
>> -    memcpy(&req->work, &apoll->work, sizeof(req->work));
>> +    if (req->flags & REQ_F_WORK_INITIALIZED)
>> +        memcpy(&req->work, &apoll->work, sizeof(req->work));
>>       kfree(apoll);
>>         if (!canceled) {
>> @@ -4624,7 +4625,8 @@ static bool io_arm_poll_handler(struct io_kiocb *req)
>>           return false;
>>         req->flags |= REQ_F_POLLED;
>> -    memcpy(&apoll->work, &req->work, sizeof(req->work));
>> +    if (req->flags & REQ_F_WORK_INITIALIZED)
>> +        memcpy(&apoll->work, &req->work, sizeof(req->work));
>>       had_io = req->io != NULL;
>>         get_task_struct(current);
>> @@ -4649,7 +4651,8 @@ static bool io_arm_poll_handler(struct io_kiocb *req)
>>           if (!had_io)
>>               io_poll_remove_double(req);
>>           spin_unlock_irq(&ctx->completion_lock);
>> -        memcpy(&req->work, &apoll->work, sizeof(req->work));
>> +        if (req->flags & REQ_F_WORK_INITIALIZED)
>> +            memcpy(&req->work, &apoll->work, sizeof(req->work));
>>           kfree(apoll);
>>           return false;
>>       }
>> @@ -4694,7 +4697,9 @@ static bool io_poll_remove_one(struct io_kiocb *req)
>>                * io_req_work_drop_env below when dropping the
>>                * final reference.
>>                */
>> -            memcpy(&req->work, &apoll->work, sizeof(req->work));
>> +            if (req->flags & REQ_F_WORK_INITIALIZED)
>> +                memcpy(&req->work, &apoll->work,
>> +                       sizeof(req->work));
>>               kfree(apoll);
>>           }
>>       }
>>

-- 
Pavel Begunkov