Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Thu, 18 Jul 2019 16:55:49 +0200
From:   Oleg Nesterov <oleg@redhat.com>
To:     Ingo Molnar <mingo@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Thomas Gleixner <tglx@linutronix.de>
Cc:     Andrew Fox <afox@redhat.com>,
        Stephen Johnston <sjohnsto@redhat.com>,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] sched/cputime: make scale_stime() more precise
Message-ID: <20190718145549.GA22902@redhat.com>
References: <20190718131834.GA22211@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190718131834.GA22211@redhat.com>
User-Agent: Mutt/1.5.24 (2015-08-30)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 07/18, Oleg Nesterov wrote:
>
> + * NOTE! currently the only user is cputime_adjust() and thus
> + *
> + *	stime < total && rtime > total
> + *
> + * this means that the end result is always precise and the additional
> + * div64_u64_rem() inside the main loop is called at most once.

Ah, I just noticed that the comment is not 100% correct... in theory we
can drop the precision and even do div64_u64_rem() more than once, but this
can only happen if stime or total = stime + utime is "really" huge, I don't
think this can happen in practice...

We can probably just do

	static u64 scale_stime(u64 stime, u64 rtime, u64 total)
	{
		u64 res = 0, div, rem;

		if (ilog2(stime) + ilog2(rtime) > 62) {
			div = div64_u64_rem(rtime, total, &rem);
			res += div * stime;
			rtime = rem;

			int shift = ilog2(stime) + ilog2(rtime) - 62;
			if (shift > 0) {
				rtime >>= shift;
				total >>= shitt;
				if (!total)
					return res;
			}
		}

		return res + div64_u64(stime * rtime, total);
	}

but this way the code looks less generic.

Oleg.