linux-kernel - Re: About dmesg time and uptime

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <alpine.DEB.2.21.1803211107560.3754@nanos.tec.linutronix.de>
Date:   Wed, 21 Mar 2018 11:14:01 +0100 (CET)
From:   Thomas Gleixner <tglx@...utronix.de>
To:     Qixuan Wu <qixuan.wu@...ux.alibaba.com>
cc:     John Stultz <john.stultz@...aro.org>,
        Stephen Boyd <sboyd@...nel.org>, linux-kernel@...r.kernel.org
Subject: Re: About dmesg time and uptime

On Wed, 21 Mar 2018, Qixuan Wu wrote:
>    I have a doubt in kernel 4.9 about dmesg time and uptime. Would you
> please to help to clarify whether it's bug or not.
>    Below is our log. The dmesg time second is large than uptime second.
> 
> [25049978.919097] bond0.700: port 9(veth4e77160) entered forwarding state
> [25049978.926455] bond0.700: port 9(veth4e77160) entered forwarding state
> [chenggang.qcg@...g10383 ~]$ cat /proc/uptime
> 25023128.55 1416627078.34
> 
>    I found dmesg second is just directly get from tsc (x86 64 platform).
> While the uptime is get by a complicated computation.

Well it's not a complicated computation. It simply uses the timekeeping
infrastructure.

The dmesg timestamp comes from sched_clock() which is TSC based and it is
using a coarse calibration value obtained at boot time.

The timekeeping infrastructure is using a fine grained calibration which
can be further adjusted by NTP/PTP.

So yes, this can drift apart. The error from your numbers above is about
0.1% which is not an unusual value.

There is a patch is mainline which feed the fine grained calibration value
into sched_clock to make this slighty better, but it wont be perfect. See:

     aa7b630ea023 ("x86/tsc: Feed refined TSC calibration into sched_clock()")

There is also an effort to base dmesg time stamps on monotonic time, which
will fix this problem completely, but that's still work in progress.

Thanks,

	tglx