[PATCH]: acct_process() locking and exit1()

[John Baldwin]

On Thursday 14 June 2007 03:58:30 am Roman Divacky wrote:
> On Thu, Jun 14, 2007 at 09:54:39AM +0200, Roman Divacky wrote:
> > hi
> > 
> > currently in exit1() we call acct_process() with Giant held.
> > I looked at the code and I think this is because of tty need
> > for Giant locking. Because of this we have to release process limit 
> > in a separate PROC_LOCK()/UNLOCK() block.
> > 
> > my just-to-look-at patch does this
> > 
> > 1) moves the acct_process() in exit1() out of Giant locked block (upward)
> > 2) acquires Giant in the acct_process() around tty handling
> > 3) moves the p->p_limit to a different PROC_LOCK/UNLOCK block (upward)
> > 
> > the result is saving 2 proc mtx operations in exit1()
> > 
> > can you look at it and tell me if its correct or wrong or something like 
that?
> > 
> > if it's correct is it worth committing? saving two mtx operations is a 
nice thing :)
> 
> erm... two bugs ;) "saving two mtx operations..." in the typical case of 
accounting
> NOT being enabled.
> 
> and forgot to show the patch ;) it's here
> 
> www.vlakno.cz/~rdivacky/kern_acct.patch

1) It would be nicer if lim_free() wasn't so far away from the rest of the 
limit freeing code (including the comment).  Also, missing a blank line 
before the comment now.

2) acct_process() saves resource usage information, so the later it is in 
exit1() the more accurate it is going to be, so it's probably best to leave 
it as it is unless you observe a measurable change in a benchmark with this 
patch.

-- 
John Baldwin