On Llu, 2003-06-02 at 21:44, McKown, John wrote:
Just an observation that may well be very stupid. Since Linux/390 can use NTP to sync time, then it must be that Linux/390 uses a "software" clock instead of the hardware clock (TOD clock). Is this true? If so, what happens if somebody down the line in the kernel maintainers decide that Linux should use the TOD clock instead of a "software" clock? Wouldn't that cause a problem?
It wouldn't neccessarily make sense to use the TOD clock like that. S/390 differs from the PC I guess in that the TOD clock is probably vaguely accurate.
Even then you'd want to query the TOD clock and use it as an NTP source to lock the software clock.
NTP uses the adjtime() system call. Even *if* the s390 architecture code issued SCK to reset the `real' clock, I would expect this to be intercepted if running under sie (ie vm) and a virtual clock would be updated. Likewise if running natively in an lpar. However, looking at the current code (2.4.19 + patches up thru #5) it only looks like the adjtime code adjusts the offset that is applied to current tod clock (as returned by stck) so the `real' clock is not affected at all.
Greg Smith
