Re: ksoftirqd? (Was: Re: [PATCH] Preemption Latency Measurement Tool)

[Andrea Arcangeli]

On Sat, Sep 22, 2001 at 02:56:58PM +0200, Roger Larsson wrote:
> Hi,
> 
> We have a new kid on the block since we started thinking of a preemptive 
> kernel.
> 
> ksoftirqd...
> 
> Running with nice 19 (shouldn't it really be -19?)
> Or have a RT setting? (maybe not since one of the reasons for
> softirqd would be lost - would be scheduled in immediately)
> Can't a high prio or RT process be starved due to missing
> service (bh) after an interrupt?

It cannot be starved, if ksoftirqd is never scheduled the do_softirq()
will be run by the next timer irq or apic_timer irq.

> This will not show up in latency profiling patches since
> the kernel does what is requested...
> 
> Previously it was run directly after interrupt,
> before returning to the interrupted process...

It is still the case, that's also the common case actually.

Andrea