David Laight wrote:

>On Thu, Nov 29, 2007 at 02:23:23PM +0100, theo borm wrote:
>  
>
>>My guess is that soekris NET4501 kernels will also be broken.
>>    
>>
>
>I think they are known to be 'broken'.
>
>The tsc on (at least some of) the soekris board doesn't count at all
>during 'halt'.
>
>Probably the TSC calibration code ought to try the calibration spinning
>and sleeping....
>
>	David
>
>  
>
They are already known to be broken, there is dmesg output telling me 
so. This knowledge does, however, not stop the timecounters framework 
from using the TSC (and neither does the kernel compilation option that 
I thought would do the trick).

In 3.0 there were no timecounters, and things "just worked", so from my 
(naive user) perspective something got broken somewhere between 3.0 and 
4.0_rc4, which took (me at least) some time to figure out (thanks all).

I don't know how difficult it would be to disable or significantly lower 
the priority (for timecounters) of the TSC when a broken TSC is 
detected, so that the i8254 is used "automagically" instead


For now
brouter -> gdb --write /netbsd
(gdb) set tsc_is_broken=1
(gdb) quit
brouter -> reboot

will do the trick for me, and I guess that goes for soekris boards too.


Thanks all for your patience.

with kind regards,

Theo