Re: + stupid-hack-to-make-mainline-build.patch added to -mm tree

[Dan Hecht <dhecht@vmware.com>]

On 03/06/2007 05:18 PM, Thomas Gleixner wrote:
> On Tue, 2007-03-06 at 16:53 -0800, Dan Hecht wrote:
>>> Ooops. I completely forgot, that you get the absolute expiry time
>>> already in ktime_t format (nanoseconds) when dev->set_next_event() is
>>> called.
>>>
>>> 	dev->next_event = expires;
>>>
>>> is done right before the call. 
>>>
>>> So it's already there for free.
>>>
>>>
>> Okay.  I noticed that but didn't think it was okay to use since it 
>> didn't seem like it was set up for the clock_event_device code's use, so 
>> seemed like a conceptual interface violation to go digging around in 
>> there.
> 
> Yes it is. 
> 
> I just wanted to point out that you can use it until I'm awake enough to
> implement it proper.
> 

Well, we'll probably just live with using the relative expiry for the 
first pass, and then revisit this later once that is working, rather 
than resort to hacking it out by reading ->next_event.

>> Also, wasn't one of the points of clockevents to prevent the device code 
>> from doing conversions between nanoseconds and clicks themselves?  Don't 
>> we really want the clockevents generic layer to do this conversion 
>> between monotonic nanonseconds to absolute device clicks and then give 
>> the device code that value, so the device layer doesn't perform any 
>> conversions?
> 
> Right. But this applies only to deltas, as the conversion of absolute
> time values gets ugly, i.e. 128bit math
> 

Yeah, hopefully we can come up with a clean way to do this.  But, like I 
said early, until we do, we'll stick with the relative expiry.

> IMO the paravirt interfaces should use nanoseconds anyway for both
> readout and next event programming. That way the conversion is done in
> the hypervisor once and the clocksources and clockevents are simple and
> unified (except for the underlying hypervisor calls).
> 

I disagree.  The clocksource/clockevents layer are always going to have 
to convert nanoseconds to/from hardware units, so why not use it?  And, 
some guests (say, a future version of linux that does trace-based 
process accounting) may want higher resolution than nanoseconds for 
certain uses.  In any case, this is beside the point; I'd prefer to 
stick to using the clockevents interface in the way it was intended 
rather than reaching into ->next_event.

thanks,
Dan