[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <44D26769.4070505@sgi.com>
Date: Thu, 03 Aug 2006 14:15:21 -0700
From: Jay Lan <jlan@....com>
To: Andrew Morton <akpm@...l.org>
Cc: Jay Lan <jlan@...r.sgi.com>, linux-kernel@...r.kernel.org,
nagar@...son.ibm.com, balbir@...ibm.com, jes@....com,
csturtiv@....com, tee@....com, guillaume.thouvenin@...l.net
Subject: Re: [patch 3/3] convert CONFIG tag for extended accounting routines
Andrew Morton wrote:
> On Wed, 02 Aug 2006 21:23:35 -0700
> Jay Lan <jlan@...r.sgi.com> wrote:
>
>
>>+/**
>>+ * acct_update_integrals - update mm integral fields in task_struct
>>+ * @tsk: task_struct for accounting
>>+ */
>>+void acct_update_integrals(struct task_struct *tsk)
>>+{
>>+ if (likely(tsk->mm)) {
>>+ long delta =
>>+ cputime_to_jiffies(tsk->stime) - tsk->acct_stimexpd;
>
>
> If a 32 architecture chooses to implement a 64-bit cputime_t, this
> expression might go wrong for very long-running tasks and high HZ.
>
> Perhaps we should do all this in terms of cputime_t and export everything
> to userspace as u64?
Andrew,
We export to userspace the acct_rss_mem1 and acct_vm_mem1, both as u64.
The above logic is to calculate stime delta since last update. Note that
acc_update_integrals() is invoked at do_execve, do_exit, _AND_ at
account_system_time, which is called every jiffy by timer interrupt
handler.
The tsk->acct_stimexpd is used to save the tsk->stime of last update.
It should be changed to cputime_t as well. I will include the fix in
my next fix patch.
>
>
>>+ if (delta == 0)
>>+ return;
>>+ tsk->acct_stimexpd = tsk->stime;
>>+ tsk->acct_rss_mem1 += delta * get_mm_rss(tsk->mm);
>>+ tsk->acct_vm_mem1 += delta * tsk->mm->total_vm;
>
>
> It's a bit weird to be multiplying RSS by time. What unit is a "byte
> second"?
>
> If this is not a bug then I guess this is an intermediate term for
> additional downstream processing. There is information loss here and I'd
> have thought that it would be better to simply send `delta' and the rss
> straight to userspace, let userspace work out what math it wants to perform
> on it. If that makes sense?
>
> I see that the code has been like this for a long time, so treat this as a
> "please educate me about BSD accounting" email ;)
This is not a BSD accounting thing. It came from UNICOS and IRIX.
I am pinging the person who knows how the real world users use these
two fields...
Regards,
- jay
>
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists