Dear Scott,
In message 20090406191242.GA4439@ld0162-tx32.am.freescale.net you wrote:
Is there any particular reason not to change the powerpc get_timer implementation to use the timebase (scaled down to ms)?
Side note: we already use the timebase for the implementation of udelay() and friends which are available as soon as we have a (limited) C runtime environment, i. e. long before relocation.
I think so. There are some boards where we actually measure the system clock frequency against a known reference clock. IIRC this requires the TB to be independent, or something like that. [It's a long time since that code was written.]
The existing get_timer uses the decrementer, which is not independent of the timebase.
Yes, you are right. I should have looked up the actual implementation. Sorry, volatile memory with too few/short refresh cycles ;-)
Best regards,
Wolfgang Denk