Re: [PATCH] sched: cpuacct: Use bigger percpu counter batch values for stats counters

[Andrew Morton <akpm@linux-foundation.org>]

On Tue, 26 Jan 2010 11:47:15 +0530 Balbir Singh <balbir@linux.vnet.ibm.com> wrote:

> On Tuesday 26 January 2010 04:44 AM, Andrew Morton wrote:
> > On Mon, 18 Jan 2010 15:41:42 +1100
> > Anton Blanchard <anton@samba.org> wrote:
> > 
> >> When CONFIG_VIRT_CPU_ACCOUNTING and CONFIG_CGROUP_CPUACCT are enabled we can
> >> call cpuacct_update_stats with values much larger than percpu_counter_batch.
> >> This means the call to percpu_counter_add will always add to the global count
> >> which is protected by a spinlock and we end up with a global spinlock in
> >> the scheduler.
> > 
> > When one looks at the end result:
> > 
> > : static void cpuacct_update_stats(struct task_struct *tsk,
> > : 		enum cpuacct_stat_index idx, cputime_t val)
> > : {
> > : 	struct cpuacct *ca;
> > : 	int batch;
> > : 
> > : 	if (unlikely(!cpuacct_subsys.active))
> > : 		return;
> > : 
> > : 	rcu_read_lock();
> > : 	ca = task_ca(tsk);
> > : 
> > : 	batch = min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX);
> > : 	do {
> > : 		__percpu_counter_add(&ca->cpustat[idx], val, batch);
> > : 		ca = ca->parent;
> > : 	} while (ca);
> > : 	rcu_read_unlock();
> > : }
> > 
> > the code (which used to be quite obvious) becomes pretty unobvious.  In
> > fact it looks quite wrong.
> > 
> > Shouldn't there be a comment there explaining wtf is going on?
> 
> Andrew,
> 
> I guess a lot of the changelog and comments are in the email history,

Not a very useful location for it!

> Why does it look quite wrong to you?

Because it computes the correct value and then if it's larger than
INT_MAX, it inexplicably assigns INT_MAX to it, giving a wrong result!


Does that code actually work, btw?  percpu_counter_batch has type `int'
and cputime_one_jiffy has type `int' so their product has type `int'. 
So by the time min_t performs its comparison, the upper 32 bits of the
product are already lost.