Re: [PATCH v2] io_uring: reduced function parameter ctx if possible

[Jens Axboe <axboe@kernel.dk>]

On 11/8/19 7:33 AM, Jackie Liu wrote:
> 
> 
> 在 2019/11/8 22:16, Jens Axboe 写道:
>> On 11/8/19 7:11 AM, Jens Axboe wrote:
>>> On 11/8/19 12:29 AM, Jackie Liu wrote:
>>>> Many times, the core of the function is req, and req has already set
>>>> req->ctx at initialization time, so there is no need to pass in from
>>>> outside.
>>>>
>>>> Cleanup, no function change.
>>>
>>> I was curious if this patch netted us any improvements as well, but it
>>> actually blows up the text segment a lot on my laptop. Before the
>>> patch:
>>>
>>>       text	   data	    bss	    dec	    hex	filename
>>>      87504	  17588	    256	 105348	  19b84	fs/io_uring.o
>>>
>>> and after:
>>>
>>>       text	   data	    bss	    dec	    hex	filename
>>>      99098	  17876	    256	 117230	  1c9ee	fs/io_uring.o
>>>
>>> which seems really odd. I double checked just to be sure!
>>>
>>> axboe@x1:~ $ gcc --version
>>> gcc (Ubuntu 9.2.1-17ubuntu1~18.04.1) 9.2.1 20191102
>>
>> I took a look at the generated code, and it looks like it's just
>> enabling gcc to inline a lot more aggressively.
>>
> 
> Hum... I don't have a lot of attention to the size of the generated
> code, just to make the code look easier to understand. Will this affect
> performance?

Seems to be an extreme case on the laptop since I have various debug
options enabled for testing in qemu, but it's an increase on my
test box with gcc 7.2 as well. I don't think it's anything to worry
about, it's just inlining what it should inline. I did a quick perf
run and we're not down, so I'm going to apply this one as it cleans
up the code and that's always the most important.

-- 
Jens Axboe