missing RELOC()s in prom.c?

Anton Blanchard anton at samba.org
Sat Feb 7 20:40:35 EST 2004

> Cause on legacy iSeries there are no real interrupts.  There are only lp
> events put on a queue.  And we check the queue on each decrementer.  And
> there's an option somewhere or other (search for spread_lp_events) for
> whether we check the queue only with cpu0 or with all cpus..
> By the way, at one point there was a cool option to stagger the
> decrementers so that even with overclocking, multiple CPUs would check the
> lp event queues at staggered intervals within the decrementer intervals.

In fact we are currently staggering the decrementers in 2.6. It was done
to avoid all cpus hitting some global locks in the timer code. The
global lock has been removed and x86 has moved back to triggering all
timer irqs at the same time.

Considering that staggered interrupts should help iseries, we might
continue to do it on ppc64. Also I notice there is a spread_events boot
option, should we make this the default?

> As to whether HZ helps, It's too late at night, and Hugh made me drink too
> much beer to contemplate that.

Watch out for Hugh :) My theory is if our usual overclocking of the decr
in 2.4 ends up the same frequency as base 2.6 (1000Hz) then we can remove
the option.


** Sent via the linuxppc64-dev mail list. See http://lists.linuxppc.org/

More information about the Linuxppc64-dev mailing list