[Ocfs2-devel] [PATCH 00/34] OCFS2: Add trace event and replace mlog(0).

[Tao Ma]

On 12/31/2010 08:52 PM, Joel Becker wrote:
> On Thu, Dec 23, 2010 at 03:19:03PM +0800, Tao Ma wrote:
>> This patch set is my attempt to replace our old debug(mlog) with new
>> trace events.
>>
>> Wengang Wang has did some work in this field about one year ago.
>> http://oss.oracle.com/pipermail/ocfs2-devel/2009-September/005230.html
>> But at that time we don't have ocfs2 1.6 introduced, so we delay
>> this work until now after ocfs2 1.6 and UEK(Unbreakable Enterprise
>> Kernel) are released.
>>
>> All patches except one(see below) are only mlog related, so they
>> shouldn't affect the normal operations in ocfs2 and they should be
>> easy for review(patch 1-5 were already sent to the mail list around
>> one month ago).
>>
>> The only one that changes the codes somehow are patch 15, "ocfs2:
>> Little refactoring against ocfs2_iget", but it looks
>> straightforward.
>
> Tao,
> 	Overall this seems pretty straightforward.  There wasn't a lot
> of editing of masklog entries; we still have a million tracepoints.  I
> wonder how much memory that will use.  Have you checked the space usage
> of all the sysfs files for all of our tracepoints?
Sorry, I don't know how to check the space usage of these files. any tips?
> 	Aside from my comment about marketing terms in patch comments
> (hehe), I'm OK with going forward with these patches.  I would love the
> ability to turn specific traces on and off.
OK, I will regenerate the patches with advices from Christoph and your.
> 	Well, one generic comment.  It seems a lot of our log messages
> have gone from "ocfs2_foo(): Change blah 0x%p to new_blah 0x%p" to
> "ocfs2_foo(): %p %p".  I wonder if that is going to be hard to read.
Yeah, I used to think of adding all these descriptive strings to trace 
buffers, but finally I gave up. The reason is that no matter how we 
print the message in the trace buffer, we have to go to read the codes 
and with the codes, everything should be clear enough. ;)

Regards,
Tao