[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <alpine.DEB.2.11.1601110905180.3575@nanos>
Date: Mon, 11 Jan 2016 09:15:25 +0100 (CET)
From: Thomas Gleixner <tglx@...utronix.de>
To: Nicholas Mc Guire <hofrat@...dl.org>
cc: John Stultz <john.stultz@...aro.org>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH RFC] timer: drop the unnecessary while loop in msleep
On Sat, 9 Jan 2016, Nicholas Mc Guire wrote:
> The while loop in msleep does not seem necessary as
> timeout is unsigned long and no larger than MAX_JIFFY_OFFSET (which is
> LONG_MAX/2 - 1) so the while-loop condition is always true at the beginning
> (msecs_to_jiffies will return >=0 always and with the +1 timeout is >= 1 so
> the while condition is always true at the start) and
> schedule_timeout_uninterruptible always returns 0, so the while loop always
> terminates after the first loop.
Err, no. schedule_timeout_uninterruptible() can return > 0 when there was a
non timer wakeup. Thinks spurious wakeups. So we need that loop.
> Q: what is the purpose of the + 1 offset to the jiffies here ?
>
> msleep was introduced in 2.6.7 but without the + 1, so with:
> unsigned long timeout = msecs_to_jiffies(msecs);
> in 2.6.10-rc2 the msecs_to_jiffies(msecs) + 1; is introduced.
> Nishanth Aravamudan <nacc@...ibm.com> (https://lkml.org/lkml/2004/11/19/294)
> seems to be the origin while converting msleep to a macro, but no reason
> for the + 1 is given there.
Not really. The +1 was introduced with the following commit:
https://git.kernel.org/cgit/linux/kernel/git/tglx/history.git/commit/kernel/timer.c?id=c259ef842622a5e64418d9dab3b62ee051867edf
Thanks,
tglx
Powered by blists - more mailing lists