Still seeing "calcru: runtime went backwards" messages

[Poul-Henning Kamp]

In message <200603011724.08214.jhb at freebsd.org>, John Baldwin writes:

>Maybe we could make the dynamic flag for the tsc controllable via tunable?

Ideally ACPI would tell us the max TSC frequency and the TSC wouldn't need
to be marked variable. Havn't had time to look at this yet though.

>If I have a server machine without any fancy pentium-m cpufreq type stuff,
>then I would rather just use the tsc frequency snapshot taken at boot and
>just stick with that as the static frequency then have a bunch of warnings
>on the console if the max freq changes sometime later.

Right, I would actually prefer to try to make the calibration more precise
first, rather than add yet another knob for roots to remember to tweak.

>Also, note that
>I am still getting a bunch of the backwards messages on my DS20 Alpha which
>is _not_ using the TSC. :)

That's interesting...  Alpha should be using the timecounter and that
is not marked "variable".

My alpha hasn't been able to boot in a long time now :-(

> (I should probably fix Alpha to use its own
>version of the TSC at some point, but for now I want to get it to stop
>spitting out the warnings when using the timecounters as that shouldn't
>be causing any warnings).

I'd be tempted to just supress the warning if the negative delta is less
than 3usec (for some value of 3).

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.