[Soekris] net4826/TSC Bug

[Chris Boot]

Aaron Isotton wrote:
> Chris Boot wrote:
>> On 13 Jul 2006, at 22:30, Aaron Isotton wrote:
>>
>>> Hi,
>>>
>>> I'm working with some net4826 boxes and I suspect that they are affected
>>> by the TSC bug
>>> (http://lists.soekris.com/pipermail/soekris-tech/2005-May/008432.html)
>>> since they are based on the same processor as the 4801. Is that true?
>> They are indeed.
> 
> Ok, thanks.
> 
>> 2.6.18 will have the new scx200_hrt Hi-Res timer code that works really
>> rather nicely along with NTP. I didn't have too much trouble with my TSC
>> while I was using that, however. If you want to use the Hi-Res timer
>> just download 2.6.18-rc1. 
> 
> I don't so much care about NTP. What I do is analyzing TCP traffic and I
> need high-resolution timestamps for that (I need resolutions of 0.1 ms
> or less). Until now I was using the TSC for that because jiffies don't
> have the resolution I need.
> 
> Does this hi-res timer "magically fix" the TSC or do I have to use a
> separate time source to use this scx200 timer?

This is a separate time source, but it is built into the SC1200 CPU that the 
net48xx series use. However, as these aren't always that accurate I do suggest 
you run ntpd which, with the new timesource code, will smooth out most 
inaccuracies in time. I have no idea how accurate the timing is when passed back 
to userspace mind you. The scx200_hrt ticks at either 1MHz or 27MHz.

>> If you get the latest git snapshot you'll also get my error LED code! :-D
> 
> Now THAT'S what I call good news ;-)

:-)

Chris


-- 
Chris Boot
bootc at bootc.net
http://www.bootc.net/