Messages in this thread | | | From | "Ulrich Windl" <> | Date | Mon, 06 Sep 2004 08:08:50 +0200 | Subject | Re: [RFC][PATCH] new timeofday core subsystem (v.A0) |
| |
On 3 Sep 2004 at 11:07, Albert Cahalan wrote:
> On Fri, 2004-09-03 at 05:08, Ulrich Windl wrote: > > On 3 Sep 2004 at 2:42, Albert Cahalan wrote: > > > > HZ not being HZ is the source of many foul problems. > > > > > > NTP should be able to correct for the error. For systems > > > not running NTP, provide a fake NTP to make corrections > > > based on the expected frequency error. > > > > > > Based on that, skip or double-up on the ticks to make > > > them be exactly HZ over long periods of time. > > > > I think nobody wants a sawtooth-like timing. Time should > > proceed as smoothly as possible. > > Of course, and all hardware should have ideal clocks. > Now, back to the real world...
Albert,
I am serious.
> > The kernel is broken if: > > a. HZ is not really HZ
HZ is an integer however.
> b. Timekeeping is via 2 unrelated clocks. (jiffies+offset)
What do you do if one clock lacks resolution, and the other clock lacks digits? The interrupt clock may suck, but the TSC sucks even more.
> c. HZ is not an integer
HZ is HZ, but the true interrupt frequency is something completely different.
> > So on box using only clock ticks, steer jiffies > toward HZ using NTP (or default frequency error value).
Are you saying you are happy with a lcok that less than 1 HZ off? That would be -- in a extreme case -- about 86000 seconds off per day.
> On a box with high-res time, use that instead, and make > jiffies follow it to satisfy various kernel-internal > uses of jiffies.
Make jiffies follow variable CPU clock? Are you serious?
> > Look, if HZ won't be HZ then you can just remove > the "HZ" define from the kernel. It's useless.
It's not useless, it's a historic standard.
> > I think we're all sick of the recent time-related > bugs. I could go for ripping out all the fancy and
Yes.
> broken stuff that was added recently, replacing it > with the simple Linux 2.4.xx or 2.2.xx code. Swiping
It depends on what you want. The kernel really needs a working framework for nanoseconds; at least regarding the variables' precision.
> code from DragonflyBSD would be worth investigating. > > >
Regards, Ulrich
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |