Mike Kravetz wrote:
> On Thu, Sep 06, 2001 at 01:40:35PM -0700, george anzinger wrote:
> > You might want to take a look at:
> > http://sourceforge.net/projects/rtsched/
> Thanks George,
> Your main page states,
> "Warning: The Real Time scheduler does not honor the "allowed_cpus" member of
> the task_struct, thus it will not honor any attempt to define cpu affinity.
> At this time the only known use of this is in TUX, where the dishonoring of
> this option may cause a performance hit."
> As you may already know, the new softirq code makes use of cpus_allowed
> field to 'bind' daemons to CPUs. Not sure what happens if the daemons
> are allowed to 'float'.
No, I was not aware of this. When did this get put in?
I have hopes of honoring the cpus allowed field, but planned to do it by
using something close to the MQ stuff. The issue I ran into was wanting
to KNOW on entry if the run queue had to go thru a recalculate. It
would not do to answer no to this question and then come up with no one
to run because of cpu affinity.
For you use, I was trying to point out the separation of the rt run list
and what it does to the goodness calculation.