One-shot-oriented event timers management
mav at FreeBSD.org
Thu Sep 2 13:33:48 UTC 2010
b. f. wrote:
> you need to offset the declaration of 'cpu' in getnextevent() on line
> 256 of src/sys/kern/kern_clocksource.c by #ifdef SMP, because it is
> not used otherwise, and will break UP kernel builds with our default
> warnings and -Werror.
> Incidentally, do you intend to commit the tm6292_idle.patch along with
> the new timer code, after testing is satisfactory? Or is this not
> appropriate for general use? If it isn't suitable for all users,
> perhaps some of the periods of the events in that patch can be
> abstracted and made tunable, so that we can make it possible to
> conserve power, and also keep others happy?
No, I am not going to commit it in present form. It is mostly made for
technology preview. May be some parts of it could be committed as-is,
but most of others are dirty hacks. I hope they should be safe enough,
but they made with assumption of some tradeoffs and require proper
rework or at least reevaluating. It would be nice if people aware with
these areas (SCHED_ULE, TCP, CPU time resources, yarrow, console, GEOM)
could review and reimplement them properly.
More information about the freebsd-current