newlines in message macros

newlines in message macros

[Bob Pearson]

Li,

I see that you removed the built-in newlines in the debug macros in rxe.h which is ok by me. But,
for some reason the rxe_xxx() macros all still have built-in newlines. Why shouldn't we be consistent
and make them all the same. (Maybe they don't get used much or at all.)

Bob

Re: newlines in message macros

[Bob Pearson]

On 9/13/23 15:50, Bob Pearson wrote:
> Li,
> 
> I see that you removed the built-in newlines in the debug macros in rxe.h which is ok by me. But,
> for some reason the rxe_xxx() macros all still have built-in newlines. Why shouldn't we be consistent
> and make them all the same. (Maybe they don't get used much or at all.)
> 
> Bob

On closer examination about half the debug messages that should have newlines added don't. About equally
easy to fix this mess by reverting the change and/or removing or adding newlines as needed.

Which way should we go.

Bob

Re: newlines in message macros

[Zhu Yanjun]

在 2023/9/14 4:50, Bob Pearson 写道:
> Li,
> 
> I see that you removed the built-in newlines in the debug macros in rxe.h which is ok by me. But,

I made tests for many times about adding newline speeding up flush 
messages. With or without new line, I can not find out the difference on 
flushing messages. Not sure if Li Zhijian found this difference in a 
specific scenario or not.
And even without new line, after output the line, the message still goes 
to a new line. I suspect if a newline is appended in the PRINTK subsystem.

Zhu Yanjun

> for some reason the rxe_xxx() macros all still have built-in newlines. Why shouldn't we be consistent
> and make them all the same. (Maybe they don't get used much or at all.)
> 
> Bob

Re: newlines in message macros

[Jason Gunthorpe]

On Thu, Sep 14, 2023 at 07:12:04AM +0800, Zhu Yanjun wrote:
> 在 2023/9/14 4:50, Bob Pearson 写道:
> > Li,
> > 
> > I see that you removed the built-in newlines in the debug macros in rxe.h which is ok by me. But,
> 
> I made tests for many times about adding newline speeding up flush messages.
> With or without new line, I can not find out the difference on flushing
> messages. Not sure if Li Zhijian found this difference in a specific
> scenario or not.
> And even without new line, after output the line, the message still goes to
> a new line. I suspect if a newline is appended in the PRINTK
> subsystem.

Kernel standard is to have newlines in the format strings.

Jason

Re: newlines in message macros

[Zhijian Li (Fujitsu)]

Yanjun,

On 14/09/2023 07:12, Zhu Yanjun wrote:
> 在 2023/9/14 4:50, Bob Pearson 写道:
>> Li,
>>
>> I see that you removed the built-in newlines in the debug macros in rxe.h which is ok by me. But,
> 
> I made tests for many times about adding newline speeding up flush messages. With or without new line, I can not find out the difference on flushing messages. Not sure if Li Zhijian found this difference in a specific scenario or not.
> And even without new line, after output the line, the message still goes to a new line. I suspect if a newline is appended in the PRINTK subsystem.


When i'm using something like: `dmesg --follow` monitor the dmesg, I can notice that delay clearly.
you will see that the timestamp is correct, but the messages don't appear until a next newline.

Thanks
Zhijian



> 
> Zhu Yanjun
> 
>> for some reason the rxe_xxx() macros all still have built-in newlines. Why shouldn't we be consistent
>> and make them all the same. (Maybe they don't get used much or at all.)
>>
>> Bob
> 

Re: newlines in message macros

[Zhu Yanjun]

在 2023/9/15 9:38, Zhijian Li (Fujitsu) 写道:
> Yanjun,
>
> On 14/09/2023 07:12, Zhu Yanjun wrote:
>> 在 2023/9/14 4:50, Bob Pearson 写道:
>>> Li,
>>>
>>> I see that you removed the built-in newlines in the debug macros in rxe.h which is ok by me. But,
>> I made tests for many times about adding newline speeding up flush messages. With or without new line, I can not find out the difference on flushing messages. Not sure if Li Zhijian found this difference in a specific scenario or not.
>> And even without new line, after output the line, the message still goes to a new line. I suspect if a newline is appended in the PRINTK subsystem.
>
> When i'm using something like: `dmesg --follow` monitor the dmesg, I can notice that delay clearly.
> you will see that the timestamp is correct, but the messages don't appear until a next newline.

Thanks. To verify what you said, I made a simple tests. In my test, I 
output logs with several printk lines without newlines.

 From the test result, except the last printk line, the other printk 
lines can output logs correctly in time.

Perhaps kernel standards add newlines in the format strings. And the 
last printk without newlines can not output logs in time.

To fix this problem, I add a newline in the last printk line. Then all 
the printk logs can output correctly in time.

I think your commit and Bob's commit have added newlines in RXE.

Thanks.

Zhu Yanjun

>
> Thanks
> Zhijian
>
>
>
>> Zhu Yanjun
>>
>>> for some reason the rxe_xxx() macros all still have built-in newlines. Why shouldn't we be consistent
>>> and make them all the same. (Maybe they don't get used much or at all.)
>>>
>>> Bob