linux-kernel - Re: [RFC] tick_sched_timer() is not properly aligned, fixed by chance

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

Message-ID: <20230406150254.ZrawA2Y-@linutronix.de>
Date:   Thu, 6 Apr 2023 17:02:54 +0200
From:   Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To:     Thomas Gleixner <tglx@...utronix.de>
Cc:     Peter Zijlstra <peterz@...radead.org>,
        linux-kernel@...r.kernel.org, Ingo Molnar <mingo@...hat.com>,
        Pavel Tatashin <pasha.tatashin@...cle.com>
Subject: Re: [RFC] tick_sched_timer() is not properly aligned, fixed by chance

On 2023-04-06 13:08:29 [+0200], Thomas Gleixner wrote:
> On Thu, Apr 06 2023 at 12:55, Peter Zijlstra wrote:
> > On Thu, Apr 06, 2023 at 11:57:35AM +0200, Sebastian Andrzej Siewior wrote:
> >
> >> The tick_sched_timer story is that it uses ktime_get() to set the
> >
> > But but but, ktime_get() does not use sched_clock(), it has it's own TSC
> > reader.
> 
> Correct and at that point during early boot the clocksource which feeds
> ktime_get() should be jiffies and not some other random clocksource.
> 
> Sebastian, can you please evaluate which clocksource is used for
> ktime_get() at the point where the first clock event device is set up?

The first one added is hpet. It points to "jiffies" as name and
jiffies_read() as ->read().

Before the change/ with PeterZ change I get from ktime_get() for
base/nsec something like:
|[    0.004000] tick_next_period Name: jiffies jiffies_read+0x0/0x10
|[    0.004000] tick_next_period base/nsec: -401771248 405771248
|[    0.004000] tick_next_period hpet 4000000

base/nsec is different on each boot but it always ends up with 4000000.
With the optimisation, ktime_get() is:
| [    1.179079] tick_next_period base/nsec: 647439581 518613145
| [    1.179646] tick_next_period hpet 1166052726

so something is using it during init. And this is
	read_persistent_wall_and_boot_offset()

> Thanks,
> 
>         tglx
> 
Sebastian