[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-rt
Subject:    Re: [Rtl] High interrupt latency
From:       Christian Hardt <hach0006 () hs-niederrhein ! de>
Date:       2003-11-23 15:33:35
[Download RAW message or body]

Der Herr Hofrat wrote

>>My configuration:
>>
>>- Kernel 2.4.20
>>- rtlinux-3.2-pre2
>>    
>>
>
>general comment:
>...everyone seems stuck at pre2 - work is only on pre3 tso
>getting feedback on pre3 would be more helpfull...
>
>  
>
I am using pre3 now.

>>I have modified the measurement method of Michael Barbanov (from his 
>>master thesis) to my needs. I am using two machines. Machine 1 with 
>>RT-Linux is the measuring object. Machine 2 sends interrupt requests 
>>over the parallel-port to Machine 1 and measures the respond time. The 
>>interrupt requests are served by a RT-ISR.
>>
>>Everything seems to work fine. I can compile and load the module. But 
>>during the measurement when I switch from KDE to a text console I get 
>>interrupt latencies up to 15ms!!! I did the same measurement under RTAI 
>>and the maximum latency was always under 30 us!
>>    
>>
>
>in examples/parallel you can find some tests to do that - could you check
>if they result in the same high delays ? 
>
>
>  
>
See below.

>>unsigned int MyISR_Handler(unsigned int irq, struct pt_regs *regs)
>>{
>>   outb(0x80, base);
>>   rtl_hard_enable_irq(irq);
>>   udelay(100);
>>    
>>
>
>hmm...30us response in an isr that has a udelay of 100 ? :)
>  
>
Yes, because I measure the time between sending the request and  the 
beginnig of  the  ISR.

>Aside from that the printk needs to be rtl_printf...looks like
>you copied from RTAI rt_printk -> printk...so you are calling an
>non-RT function here - not clear why this only pops up when switching
>consoles thoug - guess the VT code in the kernel is causing the delays you
>see.
>
>  
>

You were right. The latency was caused by the Linux printk. Using printk 
in RTAI results in the same bad behaviour.
Everything works fine now. Thank you!
I did the tests in examples/parallel, too. There was no strange behaviour.

C. Hardt


_______________________________________________
Rtl mailing list
Rtl@rtlinux.org
http://www2.fsmlabs.com/mailman/listinfo/rtl
[prev in list] [next in list] [prev in thread] [next in thread]