Re: Embedded Linux Vs. Real time Linux

Michael Schnell wrote:

So what you are saying is that the Linux kernel WILL disable global
interrupts from time to time.

AFAIK, this is correct (maybe depending on the CPU architecture). You
need to inspect the Kernel sources to be sure.

But this is the approach of the RT Preempt patch: To reduce the need of
disabling the interrupt and replace this code by simple spin locks or
mutexes. This means the word "preemption". Ensure every piece of code can
be preempted if something with a higher priority is ready to work. So also
the interrupt service routines are kernel threads now, and with a priority.

On a DualCore-Pentium with RT Preempt we need more than 10000 hackbench jobs
and a two ping floods to see latency violations on this box....

JB
.

Relevant Pages

  • Re: [this_cpu_xx V5 19/19] SLUB: Experimental new fastpath w/o interrupt disable
    ... A disadvantage is that we have to disable preempt. ... void **object; ... interrupt waiting on the interrupt line to ensure scheduler timings. ... goto another_slab; ...
    (Linux-Kernel)
  • Re: Is a non-prempt kernel faster and better then a prempt one?
    ... again (complete the task that caused them to enter the kernel). ... Like for example a par port thing that needs to do IO within a specific time? ... I think what you are saying is interrupt service. ... service has the highest prority that can preempt every process. ...
    (comp.os.linux.development.apps)
  • Re: scheduler (sched_4bsd) questions
    ... The other place is supposed to preempt, and it should be ok to ... > Thread A holds a mutex x contested by Thread B and has priority pri. ... > An interrupt occurs and the interrupt handler ... it is before the mtx_unlock_spinof the turnstile chain lock. ...
    (freebsd-arch)
  • Re: Begin/EndCriticalRegion
    ... Preempt is the term you want to use, not interrupt. ... It means the scheduler will not preempt the thread to run any lower priority ... Chris Tacke, Embedded MVP ...
    (microsoft.public.dotnet.framework.compactframework)
  • Re: [patch] Real-Time Preemption, -RT-2.6.10-rc2-mm2-V0.7.29-0, and 30-9
    ... no interrupt would preempt another. ... > irqs registered later. ... send the line "unsubscribe linux-kernel" in ...
    (Linux-Kernel)