> there really isn't any reason to cut off all interrupts (on dual CPU
> hardware) while waiting for an IRQ-masked critical section, especially if
> you can handle an interrupt during the time you were spinning rather than
> the time you were supposed to be running in the critical section.
>
> essentially you're saying that you want to be sure that the bad cases are
> visible so we can fix them. there would still be tools to measure how
> contended are specific spinlocks. i would prefer using such a tool since
> it gives me empirical data, rather than waiting to experience slight
> jerkiness in my GUI, for instance. in other words, why do we need to
> prevent a subjective improvement to preserve a subjective indicator of
> spinlock contention?
well, if i had to choose between 'look, our router benchmark goes 5%
faster' and 'look, although it makes no visible difference, it makes the
spinlock profile look better', i sure know which i prefer :)
(are you suggesting that increased performance as a development incentive
does not work? Ask Andrea! :) )
> (more time at the pub might be an acceptible answer :)
;)
-- mingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/