Re: [PATCH v2] percpu: improve percpu_alloc_percpu event trace

[Vasily Averin]

On 5/11/22 05:33, Roman Gushchin wrote:
> On Fri, May 06, 2022 at 10:29:25PM +0300, Vasily Averin wrote:
>>  TRACE_EVENT(percpu_alloc_percpu,
>>  
>> -	TP_PROTO(bool reserved, bool is_atomic, size_t size,
>> -		 size_t align, void *base_addr, int off, void __percpu *ptr),
>> +	TP_PROTO(unsigned long call_site,
>> +		 bool reserved, bool is_atomic, size_t size,
>> +		 size_t align, void *base_addr, int off,
>> +		 void __percpu *ptr, size_t bytes_alloc, gfp_t gfp_flags),
> 
> Don't we want to preserve the order and add the call_site at the end?
> Trace events are not ABI, but if we don't have a strong reason to break it,
> I'd preserve the old order.

I checked recent trace patches and found that order changes is acceptable.

commit 8c39b8bc82aafcc8dd378bd79c76fac8e8a89c8d
Author: David Howells <dhowells@xxxxxxxxxx>
Date:   Fri Jan 14 11:44:54 2022 +0000

    cachefiles: Make some tracepoint adjustments

-           TP_printk("o=%08x i=%lx e=%d",
-                     __entry->obj, __entry->ino, __entry->error)
+           TP_printk("o=%08x dB=%lx B=%lx e=%d",
+                     __entry->obj, __entry->dino, __entry->ino, __entry->error)

On the other hand I'm agree to keep old order by default.
that's why I added bytes_alloc and gfp_flags to end of output.
However I think call_site is an exception. In all cases found, 
call_site is output first.
For me personally it simplified output parsing.

So I would like to know Steven's position on this question.

Thank you,
	Vasily Averin