[Soekris] sc1100 TSC bug
Jim Cromie
jim.cromie at gmail.com
Fri Sep 29 07:33:50 UTC 2006
dzpost at dedekind.net wrote:
> I'm confused about the status of the sc1100 TSC timer bug and its
> workaround. I just installed linux on my new net4801, with a standard
> 2.6.18 kernel which I built for the purpose. Based on what I'd read,
> I configured scx200_hrt module so it would use the high resolution
> timer as clock source instead of the TSC.
>
> When I install the scx200_hrt module, the clock starts running really
> _fast_: about 10 times normal speed. Putting idle=poll in the kernel
> command line seems to fix it (in fact it then seems to run slightly
> slow, but I'm not sure about that).
>
>
ilde=poll has no effect on the hi-res-timer based clocksource.
But it does help the TSC (see below)
> When I don't install the scx200_hrt module, I see this in the log
> files:
>
> kernel: TSC appears to be running slowly. Marking it as unstable
> kernel: Time: pit clocksource has been installed.
>
>
that looks right. 18 has Generic Timekeeping System, which checks the
TSC for 'sanity'
by comparing it for a short time against the 'jiffies' clocksource.
This TSC fails that test.
So that clocksource is 'de-rated', and the GTS selects the less good,
but working PIT.
You can check the currently selected clocksource :
soekris:/sys# more devices/system/clocksource/clocksource0/*
::::::::::::::
devices/system/clocksource/clocksource0/available_clocksource
::::::::::::::
jiffies tsc pit
::::::::::::::
devices/system/clocksource/clocksource0/current_clocksource
::::::::::::::
tsc
This listing shows that on my box, the TSC is not detected as insane.
This is the 1st time Ive seen that, it may be that some changes to GTS
in late -rcX's
reduced the sanity test time, and therefore its sensitivity.
Without the derating, modprobing scx200_hrt will not convince the GTS to
select it,
since the TSC has the highest rating (highest resolution, lowest cost of
reading)
soekris:/sys# modprobe scx200_hrt
[ 3123.253044] enabling scx200 high-res timer (27 MHz +0 ppm)
soekris:/sys#
soekris:/sys# more devices/system/clocksource/clocksource0/*
::::::::::::::
devices/system/clocksource/clocksource0/available_clocksource
::::::::::::::
scx200_hrt jiffies tsc pit
::::::::::::::
devices/system/clocksource/clocksource0/current_clocksource
::::::::::::::
tsc
Again, youre seeing the 'running slowly' message, so modprobing scx200_hrt
will replace the pit as the current_clocksource.
> I haven't checked it for more than a few minutes, but the clock seems
> be about as steady as with scx200_hrt and idle=poll.
>
with idle=poll, the halt instruction isnt used, which avoids the errata
(or the worst of it anyway)
http://soekris.com/Issue0003.htm
1) TSC stop during auto halt. The GX1 core has a bug (or feature ?)
where the TSC stop during auto halt. That mean that the TSC is somehow
useless and should not be used.
The cpu will run hotter, I dunno by how much.
soekris:/sys# date; ntpq -pcrv
Fri Sep 29 01:16:23 MDT 2006
remote refid st t when poll reach delay offset
jitter
==============================================================================
harpo .INIT. 16 u 23 64 0 0.000 0.000
0.000
-time.dlitz.net 39.217.151.255 3 u 13 64 377 68.677 33.494
48.481
*dsl081-199-165. .GPS. 1 u 18 64 377 97.208 -18.243
22.906
+wsip-68-226-79- 80.198.124.92 3 u 24 64 377 97.499 10.027
18.262
+dsl081-253-134. 192.5.41.40 2 u 19 64 377 75.940 9.848
18.037
LOCAL(0) .LOCL. 13 l 11 64 377 0.000 0.000
0.004
assID=0 status=0644 leap_none, sync_ntp, 4 events, event_peer/strat_chg,
version="ntpd 4.2.2 at 1.1532-o Fri Sep 15 09:33:32 UTC 2006 (1)",
processor="i586", system="Linux/2.6.18-sk", leap=00, stratum=2,
precision=-18, rootdelay=97.208, rootdispersion=49.444, peer=62615,
refid=64.81.199.165,
reftime=c8c746b6.320a9d1a Fri, Sep 29 2006 1:16:06.195, poll=7,
clock=c8c746c8.3baf4aeb Fri, Sep 29 2006 1:16:24.233, state=4,
offset=-8.301, frequency=32.007, jitter=28.369, noise=37.219,
stability=5.513, tai=0
you can force the selection of the scx200_hrt:
soekris:/sys# echo scx200_hrt >
devices/system/clocksource/clocksource0/current
_clocksource
[ 4598.986771] Time: scx200_hrt clocksource has been installed.
soekris:/sys#
soekris:/sys# more devices/system/clocksource/clocksource0/*
::::::::::::::
devices/system/clocksource/clocksource0/available_clocksource
::::::::::::::
scx200_hrt jiffies tsc pit
::::::::::::::
devices/system/clocksource/clocksource0/current_clocksource
::::::::::::::
scx200_hrt
with that done, jitter is reported as lower, but could partly be settling.
soekris:/sys# date; ntpq -pcrv
Fri Sep 29 01:23:09 MDT 2006
remote refid st t when poll reach delay offset
jitter
==============================================================================
harpo .INIT. 16 u 41 64 0 0.000 0.000
0.000
-time.dlitz.net 39.132.199.255 3 u 98 128 377 68.677 33.494
47.133
*dsl081-199-165. .GPS. 1 u 101 128 377 96.907 -16.173
7.870
+wsip-68-226-79- 80.198.124.92 3 u 45 128 377 97.361 -16.640
6.623
+dsl081-253-134. 192.5.41.40 2 u 40 128 377 76.261 -4.474
11.623
LOCAL(0) .LOCL. 13 l 29 64 377 0.000 0.000
0.004
assID=0 status=0644 leap_none, sync_ntp, 4 events, event_peer/strat_chg,
version="ntpd 4.2.2 at 1.1532-o Fri Sep 15 09:33:32 UTC 2006 (1)",
processor="i586", system="Linux/2.6.18-sk", leap=00, stratum=2,
precision=-18, rootdelay=96.907, rootdispersion=39.237, peer=62615,
refid=64.81.199.165,
reftime=c8c74777.31063eb4 Fri, Sep 29 2006 1:19:19.191, poll=7,
clock=c8c7485e.40e257b3 Fri, Sep 29 2006 1:23:10.253, state=4,
offset=-11.102, frequency=31.496, jitter=9.440, noise=34.830,
stability=5.160, tai=0
> What is the "pit"? Does this mean I don't need the scx200_hrt module
> at all? My understanding is that idle=poll makes the cpu consume more
> power and generate more heat, which I would rather avoid.
>
> Can someone set the record straight on the current situation? It
> would be nice to have this information updated on the Soekris web page
> as well :)
>
>
soekris:/sys# lsmod
Module Size Used by
scx200_hrt 2872 0 [permanent]
scx200_gpio 5444 0
scx200 5120 2 scx200_hrt,scx200_gpio
pc8736x_gpio 6220 0
nsc_gpio 3664 2 scx200_gpio,pc8736x_gpio
pc87360 21584 0
hwmon_vid 2896 1 pc87360
i2c_isa 6712 1 pc87360
i2c_core 24160 2 pc87360,i2c_isa
ndiswrapper 185032 0
usbcore 112884 1 ndiswrapper
Could you verify everything again ?
Id like to get to the bottom of the misbehaviors youre seeing.
> Thanks.
>
> -David
>
hth,
jimc
More information about the Soekris-tech
mailing list