Re: Real-Time Preemption, -RT-2.6.10-rc2-mm3-V0.7.31-7

[Ingo Molnar <mingo@elte.hu>]

* Mark_H_Johnson@raytheon.com <Mark_H_Johnson@raytheon.com> wrote:

> Unless I am mistaken, my "PK" config is the closest to 2.4 lowlat+preempt.

indeed, you are right.

> For the relevant differences in .config:
>   PK                                  RT
>   CONFIG_PREEMPT_DESKTOP=y            CONFIG_PREEMPT_DESKTOP is not set
>   CONFIG_PREEMPT_RT is not set        CONFIG_PREEMPT_RT=y
>   CONFIG_PREEMPT=y                    CONFIG_PREEMPT=y
>   CONFIG_PREEMPT_SOFTIRQS is not set  CONFIG_PREEMPT_SOFTIRQS=y
>   CONFIG_PREEMPT_HARDIRQS is not set  CONFIG_PREEMPT_HARDIRQS=y
> (though the system still creates ksoftirqd/0 and /1 on both...)
>   CONFIG_SPINLOCK_BKL is not set      [not present]
>   CONFIG_PREEMPT_BKL=y                CONFIG_PREEMPT_BKL=y
>   CONFIG_ASM_SEMAPHORES=y             [not present]
>   CONFIG_RWSEM_XCHGADD_ALGORITHM=y    [not present]
>   ...
>   [not present]                       CONFIG_RT_DEADLOCK_DETECT=y
>   ...

the only thing i'd suggest to change is to also generate an RT (and 
perhaps PK) result with all debugging options disabled - i.e. both 
RT_DEADLOCK_DETECT and all LATENCY_TRACING/timing related options 
disabled.

but your tests did trigger asserts not so long ago so it might not be
wise to run without debugging. But it's definitely a thing to try in the
future.

> Unless you are saying that I should back off to one of the other
> preempt settings (to replicate the 2.4 config on 2.6).

no, i think the PK kernel is supposed to be quite close to what
2.4+lowlat offers.

	Ingo