Re: [RFC][PATCH v2 5/7] taskstats: Improve cumulative CPU time accounting

[Michael Holzheu <holzheu@linux.vnet.ibm.com>]

Hello Oleg,

On Sat, 2010-11-13 at 19:38 +0100, Oleg Nesterov wrote:
> I already asked you to split these changes, perhaps you can do this?
> Say, bacct_add_tsk() looks overcomplicated, the change in copy_process()
> shouldn't introduce the new CLONE_THREAD check, not sure I understand
> why release_task() was chosen for reparenting, other nits...

I want to establish the new hierarchy when a new process is forked and
not for new threads, therefore the check for CLONE_THREAD in
copy_process(). I do the reparenting with reparent_acct() when a process
dies, therefore the check for "group_dead" in exit_signal().

> But it is not easy to discuss these completely different things
> looking at the single patch.
> 
> Imho, it would be much better if you make a separate patch which
> adds acct_parent/etc and implements the parallel hierarchy. This
> also makes sense because it touches the core kernel.
> 
> Personally I think that even "struct cdata" + __account_ctime() helper
> needs a separate patch, and perhaps this change makes sense by itself
> as cleanup. And this way the "trivial" changes (like the changes in
> k_getrusage) won't distract from the functional changes.
>
> The final change should introduce cdata_acct and actually implement
> the complete cumulative accounting.

So you want to have the following three patches:
* Introduce "struct cdata" + __account_ctime() (no functional change)
* Add cdata_acct accounting + parallel accounting hierarchy
* Add taskstats interface to export the data to userspace

Correct?

Michael