Subject: Re: [PATCH] sched: Buggy comparison in check_preempt_tick
From: Mike Galbraith <efault@gmx.de>
To: Venkatesh Pallipadi <venki@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>, Ingo Molnar <mingo@elte.hu>,
        linux-kernel@vger.kernel.org, Ranjit Manomohan <ranjitm@google.com>
In-Reply-To: <1293236813-31550-1-git-send-email-venki@google.com>
References: <1293236813-31550-1-git-send-email-venki@google.com>
Content-Type: text/plain; charset="UTF-8"
Date: Sat, 25 Dec 2010 08:50:18 +0100
Message-ID: <1293263418.6896.54.camel@marge.simson.net>
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1268
Lines: 31

On Fri, 2010-12-24 at 16:26 -0800, Venkatesh Pallipadi wrote:
> A preempt comparison line in check_preempt_tick has two bugs.
> * It compares signed and unsigned quantities, which breaks when signed
>   quantity happens to be negative

Oops, that's a bug.

> * It compares runtime and vruntime, which breaks when there are niced tasks

This imho isn't.

vruntimes advance at different rates for differently weighted tasks, so
they're already weighted, as is ideal_runtime.

For wakeup preemption we use wakeup_gran() as the weighted ruler to
limit spread.  This test just uses a different weighted ruler for the
same purpose at tick time, one already computed.

If you're a heavily niced task, you were very likely booted before you
got this far.  If you haven't run for considerable wall time, you won't
get further, you keep on running.  You only get booted if giving you a
full wall slice will spread vruntimes too much, for which you'll pay if
you keep running, and leftmost will pay now if we don't boot you.

	-Mike

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/