Re: [ckrm-tech] Re: [Patch 6/6] Delay accounting: Connector interface

[Matt Helsley <matthltc@us.ibm.com>]

On Wed, 2006-01-04 at 11:04 -0800, Jay Lan wrote:
> Shailabh Nagar wrote:
<snip>
> >Index: linux-2.6.15-rc7/kernel/exit.c
> >===================================================================
> >--- linux-2.6.15-rc7.orig/kernel/exit.c
> >+++ linux-2.6.15-rc7/kernel/exit.c
> >@@ -29,6 +29,7 @@
> > #include <linux/syscalls.h>
> > #include <linux/signal.h>
> > #include <linux/cn_proc.h>
> >+#include <linux/cn_stats.h>
> >
> > #include <asm/uaccess.h>
> > #include <asm/unistd.h>
> >@@ -865,6 +866,7 @@ fastcall NORET_TYPE void do_exit(long co
> >
> > 	tsk->exit_code = code;
> > 	proc_exit_connector(tsk);
> >+	cnstats_exit_connector(tsk);
> >  
> 
> We need to move both proc_exit_connector(tsk) and
> cnstats_exit_connector(tsk) up to before exit_mm(tsk) statement.
> There are task statistics collected in task->mm and those stats
> will be lost after exit_mm(tsk).
> 
> Thanks,
>  - jay
> 
> > 	exit_notify(tsk);
> > #ifdef CONFIG_NUMA
> > 	mpol_free(tsk->mempolicy);
> >-

	Good point. The assignment of the task exit code will also have to move
up before exit_mm(tsk) because the process event connector exit function
retrieves the exit code from the task struct.

	Moving these may also affect the job/pagg/task_notify/cpuset exit
notification if we're eventually going to remove *direct* calls to these
from kernel/exit.c.

Cheers,
	-Matt Helsley