[En-Nut-Discussion] Re: Re: Real time clock on 1.3G not accurate anymore.

[Dusan Ferbas]

Hi,

we discovered, that formerly timer int was done and async. callbacks were 
fired from that context. Currently call requirement is only marked and done 
during earliest context switch. This reduces interrupt time and guarantees 
stack availability for timer callbacks. But I think accurateness can be 
gone, unless relative time is used for measurement.

Note: I my work I modified an OS to run such timer callbacks from a 
separate thread. This thread has extremely high priority and was put in run 
queue on a place for next thread to run. All other threads were run in 
round robin scheme. This change had extremely high impact on the system - 
it brought stability and defined stack for timer callbacks.

>Hello Lars,
>
>Right, there had been a significant change in timer handling. See
>ChangeLog entry 2005-06-12. If you need something more accurate,
>consider native interrupts. Nut/OS uses timer 0 only, all others
>are available. Timer 0 is configured to use the 32kHz crystal, if
>NUT_CPU_FREQ is not defined.
>
>I have no idea about the deviation right now and must confess,
>that I haven't tested it for some time.
>
>Harald

Dusan