Re: [PATCH v4 2/2] io_uring: Add support for napi_busy_poll

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

On 3/2/22 13:12, Olivier Langlois wrote:
> On Wed, 2022-03-02 at 02:31 +0800, Hao Xu wrote:
>>> +static void io_blocking_napi_busy_loop(struct list_head
>>> *napi_list,
>>> +                                      struct io_wait_queue *iowq)
>>> +{
>>> +       unsigned long start_time =
>>> +               list_is_singular(napi_list) ? 0 :
>>> +               busy_loop_current_time();
>>> +
>>> +       do {
>>> +               if (list_is_singular(napi_list)) {
>>> +                       struct napi_entry *ne =
>>> +                               list_first_entry(napi_list,
>>> +                                                struct napi_entry,
>>> list);
>>> +
>>> +                       napi_busy_loop(ne->napi_id,
>>> io_busy_loop_end, iowq,
>>> +                                      true, BUSY_POLL_BUDGET);
>>> +                       io_check_napi_entry_timeout(ne);
>>> +                       break;
>>> +               }
>>> +       } while (io_napi_busy_loop(napi_list) &&
>>> +                !io_busy_loop_end(iowq, start_time));
>>> +}
>>> +
>> How about:
>>
>> if (list is singular) {
>>
>>       do something;
>>
>>       return;
>>
>> }
>>
>> while (!io_busy_loop_end() && io_napi_busy_loop())
>>
>>       ;
>>
>>
>> Btw, start_time seems not used in singular branch.
> Hao,
>
> it takes me few readings before being able to figure out the idea
> behind your suggestions. Sorry about that!
>
> So, if I get it correctly, you are proposing extract out the singular
> block out of the while loop...
>
> IMHO, this is not a good idea because you could start iterating the
> do/while loop with a multiple entries list that ends up becoming a
> singular list after one or few iterations.
My bad, I get your point now.
>
> Check what io_napi_busy_loop() is doing...
>
> It does not look like that but a lot thoughts have been put into
> writing io_blocking_napi_busy_loop()...
True.