On Fri, Jan 11 2008, Ingo Molnar wrote: > > (David, could you try the patch further below - does it fix bkltrace > timestamps too?) > > * Jens Axboe <jens.axboe@xxxxxxxxxx> wrote: > > > On Fri, Jan 11 2008, Ingo Molnar wrote: > > > > > > * Jens Axboe <jens.axboe@xxxxxxxxxx> wrote: > > > > > > > > they are from the scheduler git tree (except the first debug patch), > > > > > but queued up for v2.6.25 at the moment. > > > > > > > > So this means that blktrace will be broken with CONFIG_NO_HZ for > > > > 2.6.24? That's clearly a regression. > > > > > > 64-bit CONFIG_NO_HZ is a new feature in v2.6.24. If it happens on > > > 32-bit too and it didnt happen in v2.6.23 32-bit then it's a > > > regression. > > > > If blktrace worked in 2.6.23 and it doesn't in 2.6.24 because of some > > option that isn't immediately apparent, then it's a regression. > > Period. > > not completely correct. CONFIG_NO_HZ is a default-disabled option that > became newly available on 64-bit x86. So if NO_HZ does not completely > work on 64-bit, and if 32-bit works fine - which we dont know yet (my > guess would be that it's similarly broken on the same box) then it's not > a regression. Ingo, it doesn't matter if the option is disabled by default or not! The fact is that functionality foo works in 2.6.23 and doesn't in 2.6.24 because of something unrelated. And that IS a regression, no matter what kind of word play you are doing here :-) > But even if it's not "technically" a regression, it's something we want > to fix in .24 if we can, so i'm all with you Jens :) That's good :) > ktime_get() should have been used instead, which is a proper GTOD > clocksource. The patch below implements this. Will give it a whirl, it looks promising indeed and gets rid of the ugly cpu sync stuff. What is the cost of ktime_get() compared to sched_clock()? -- Jens Axboe - To unsubscribe from this list: send the line "unsubscribe linux-btrace" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
