Re: [PATCH] Re: boot time, process start time, and NOW time

[George Anzinger <george@mvista.com>]

john stultz wrote:
> On Tue, 2004-08-31 at 12:27, George Anzinger wrote:
> 
>>Tim Schmielau wrote:
>>
>>>On Mon, 30 Aug 2004, john stultz wrote:
>>>
>>>>On Mon, 2004-08-30 at 16:00, Tim Schmielau wrote:
>>>>
>>>>
>>>>>George, please excuse my lack of understanding. What again where the
>>>>>precise reasons to have an ntp-corrected uptime?
>>>>
>>>>Ah, here's the thread with the first mention of it that I could find.
>>>>
>>>>http://www.uwsg.iu.edu/hypermail/linux/kernel/0306.1/1471.html
>>
>>As I recall the problem was that jiffies since boot was being converted to get 
>>uptime base on 1/HZ = 1 jiffie.  Since it is really not quite that, there was an 
>>error.  Using clock_monotonic seemed like the right answer as it eliminated the 
>>error AND made the result consistant with get_clock(CLOCK_MONOTONIC,..).
>>
>>The alternate answer is, of course, to directly convert the elapsed jiffies. 
>>The main problem with this is that this can be a BIG number and, therefor, the 
>>math needs to be carefully.  And, of course, it is inconsistant with 
>>get_clock(), but that is a new interface...
> 
> 
> Hmmm. Well, I may be starting to lean in Tim's direction of pulling the
> clock_monotonic based uptime and going back to the jiffies based uptime.
> Atleast until we can make all the /proc/ output consistent. 
> 
> I just worry that it actually fixed a problem for someone, and backing
> it out would just reopen that.
> 
> Thoughts?

Well, it was done in reaction to some complaint.  I don't, at this point, recall 
who or why.  If it is done with the correct values (i.e. NOT 1/HZ, but what the 
wall clock uses) I think it will not reopen that complaint.

-- 
George Anzinger   george@mvista.com
High-res-timers:  http://sourceforge.net/projects/high-res-timers/
Preemption patch: http://www.kernel.org/pub/linux/kernel/people/rml