> Pardon the suggestion of a dumb hueristic, feel free to ignore me:
> would it work to run-first processes that have modified their iopl()
> level? i.e. "if you access hardware directly, we'll treat you specially
> in the scheduler"?
>
> An alternative is to encourage distros to set some sort of flag for
> processes like the X server, when it is run. This sounds suspiciously
> like the existing "renice X server" hack, but it could be something like
> changing the X server from SCHED_OTHER to SCHED_HW_ACCESS instead.
>
> Just throwing out some things, because I care about apps which access
> hardware from userspace :)
Well, close. But any low-latency access, even if somewhat buffered,
is likely to be user visible if unresponsive. Clearly that means video
memory like X and DRI, and sound, and mouse. If you define this generally
it would include serial as well, probably not a bad thing.
The proposal to backfeed priority from interractive processes to processes
waking them sounds useful, perhaps that might be limited a bit however.
Someone (Ingo?, Linus?) proposed limiting this to a fraction of the
priority of the interractive process, but it might be more useful to
simply limit how much could be added at one time, perhaps as a fraction of
the delta between max and current interractive bonus, which would have the
waker increase asymptomically toward max. Hysteresis is nice.
I do agree that it would be better to identify processes using physical
devices, far better than trying to identify some subset because it's
obvious.
I think this would include parallel port, although other than PL/IP I
don't think of a case where it matters much.
-- bill davidsen <davidsen@tmr.com> CTO, TMR Associates, Inc Doing interesting things with little computers since 1979.- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/