Re: [Xenomai-help] interrupt service routine issue

[stefano.franzoni@domain.hid]

>if it is Linux's pthread_cond_signal, this is normal, if this is
>Xenomai's pthread_cond_signal, then we have a problem.

How can I verify this point?



                                                                           
             Gilles                                                        
             Chanteperdrix                                                 
             <gilles.chanteper                                          To 
             drix@domain.hid>         stefano.franzoni@domain.hid        
                                                                        cc 
             29/01/2009 16:39          "xenomai@xenomai.org"              
                                       <xenomai@xenomai.org>              
                                                                   Subject 
                                       Re: [Xenomai-help] interrupt        
                                       service routine issue               
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           




stefano.franzoni@domain.hid wrote:
> Thanks Gilles,
>
> with your suggestion, setting the special mode bit T_WARNSW for the
> incriminated task, we found two critical functions that caused switches
to
> secondary mode.
>
> I report them for documentation:
>
> 1. We used mmap function used to map a file into memory. The access to
the
> memory caused sometimes the switch to secondary mode.

damn, this should not happen, we use mlockall and some additional tricks
in the I-pipe patch to avoid that.

> 2. A pthread_cond_signal call used to unlock a waiting task always
> generates the switch to secondary mode.

if it is Linux's pthread_cond_signal, this is normal, if this is
Xenomai's pthread_cond_signal, then we have a problem.

>
> We have implemented the two functions in a lower priority thread and the
> issue was solved.
> Thanks to your magic silver bullet!!
>
> To stress my application I have found a DoHell script by Ingo Molnar
> (http://marc.info/?l=linux-kernel&m=111947618802722&w=2) but it is
> uncompleted. Can you send your DoHell script?

I am afraid it is much the same. I have ltp installed in /ltp and do:

#! /bin/sh

mount | grep -q sda1 || mount /dev/sda1 /mnt
modprobe rtc-cmos

dortc() {
    chrt -f 98 -p `pidof 'IRQ 8'`
    /bin/rtc_wakeup -f 8192 -t 100000
}

run40() {
    while true; do ./runalltests.sh -x 40; done
}

while true; do dd if=/dev/zero of=/mnt/bigfile bs=1024000 count=100; done &
while true; do killall hackbench; sleep 5; done &
while true; do /ltp/testcases/bin/hackbench 20; done &
( cd /ltp; run40; ) &
ping -l 100000 -q -s 10 -f 192.168.0.5 &
while true; do ls -lR / > /dev/null 2>&1; done &
dortc &

--
                                                 Gilles.