Re: [patch] Real-Time Preemption, -RT-2.6.10-rc2-mm1-V0.7.27-3

[Ingo Molnar <mingo@elte.hu>]

* Florian Schmidt <mista.tapas@gmx.net> wrote:

> hmm, the output doens't look so much different (or am i just blind?).
> maybe i need to do make clean before building with this patch applied?

the trace is fine, note the extra 0x08000000:

>     5 88010003 0.000ms (+0.000ms): try_to_wake_up (wake_up_process)
>     5 88010003 0.000ms (+0.000ms): (0) ((1))
>     5 88010002 0.000ms (+0.000ms): preempt_schedule (try_to_wake_up)
>     5 88010002 0.000ms (+0.000ms): wake_up_process (redirect_hardirq)
         ^--- this one
>     5 88010001 0.000ms (+0.000ms): preempt_schedule (__do_IRQ)
>     5 88010001 0.000ms (+0.000ms): irq_exit (do_IRQ)
>     5 88000002 0.000ms (+0.000ms): do_softirq (irq_exit)
>     5 88000002 0.001ms (+0.990ms): __do_softirq (do_softirq)
>     5 08000000 0.991ms (+0.000ms): preempt_schedule (_mmx_memcpy)
>     5 98000000 0.991ms (+0.000ms): __schedule (preempt_schedule)

it was zero before - indeed hard to notice optically :-|

found the reason for this latency meanwhile: it's kernel_fpu_begin() and
kernel_fpu_end() disabling/enabling preemption. (these are used by the
mmx memcpy's)

i've uploaded the -11 patch with a preliminary fix:

    http://redhat.com/~mingo/realtime-preempt/

which turns off the FPU-based ops if PREEMPT_RT is specified. The speed
difference should be small but the latency difference is large ...

could you try -11, do you still see these large latencies?

	Ingo