Messages in this thread | | | Subject | Re: [RFC][PATCH] linux-2.5.64_monotonic-clock_A1 | From | john stultz <> | Date | 11 Mar 2003 14:59:13 -0800 |
| |
On Tue, 2003-03-11 at 14:39, george anzinger wrote:
> I must have confused you. I am woking on a get time of day sort of > thing. In time.c, the gettimeofday code calls get_offset() and then > adds in lost ticks (ticks clocked by the PIT interrupt but not yet > rolled into the wall clock (xtime). I was thinking that get_offset > might be defined to add this its result.
I'm still not quite following that. But as long as we're both pointing at the same code and grunting in agreement I think I'll just let it slide ;)
> But, back to the problem I am trying to solve. The posixtimers code > is in the common kernel and needs the result returned by get_offset > OR, we could define a new function, get_monotonictimeofday(), which > returns the jiffies since boot + get_offset() + pending ticks (i.e. it > would be the same as gettimeofday except it would use jiffies_64 > instead of xtime to get its result. The format would be a timespec, > i.e. the same as xtime.
Actually, what is the difference between the call you're trying to implement and monotonic_clock() (outside of the timespec return)? Could you point me to the specific code you are describing? It sounds like we're working on basically the same solution from two different angles.
> This translates directly into a system call and is also used in the > timers code to convert from wall clock time to jiffies time for timers. > > Either way, we have a bit of a mess due to the arch dependency. I > don't really care which way it goes, but I do think it should be > resolved in 2.5.
Well, if the generic interfaces aren't providing what you need, then a new interface needs to be considered. This is precisely what the hangcheck-timer code ran into, and is why we're working on this monotonic_clock() code (which is intended be arch independent in the future).
thanks -john [unhandled content-type:application/pgp-signature] | |