[rtl] a bug in fpu-state saving/restoring?

[Abel Alberto Cuadrado Vega <cuadrado@pamela.etsiig.uniovi.es>]

Hello to all the people in rtlinux mail list,

by the moment only a quick question (now I'm very busy. I expect to give
more details the next week).

Why does this realtime task work all right?:

************************************************************************

unsigned char fpstat[108];

void functtr(int t)
{
    while(1) {
	
	/* Sorry about this if it is not completely right. This is the 
           first time I use the "asm" directive. But it works.      */
        
	__asm__("   fsave %0"   /* Save FPU state. */
        :"=m" (*fpstat)
        :
        :"memory");

        /*  A lot of mathematical operations (floating point) and
 	    functions of libm.a  here  ...*/

	/* ... and then data are sent to a Linux task. */
        

	__asm__("   frstor %0"   /* Restore FPU state.  */
        :
        :"m" (*fpstat)
        );

        rt_task_wait();
    }
}

*********************************************************************

and doesn't this one if it's the way must be done?: 

********************************************************************

void functtr(int n)
{
   
    rt_use_fp(1); /* This is its place in the examples of RT-Linux */ 
    while(1) {
	/* rt_use_fp(1);  I tried also to put it here  */

	/* A lot of mathematical operations (floating point) and functions  
           of libm.a here .... */

	/* ... and then data are sent to a Linux task.  */
	
	
	rt_task_wait();
  }
}

************************************************************************

The first one works OK, but the second one makes the Linux task (not real
time) to obtain sometimes wrong results **only** in floating point
operations. Even the "top"  command displays wrong data like this while
the second real time task is running:

CPU states: -35791394.-8% user, etc...

or something similar.

However with the first real time task running, my Linux task and the "top"
command both display correct values.

I'm using rtlinux-0.6-2.0.33 on a Pentium 120 with  Debian Linux 1.3.1.
(and also on a Pentium 200).

Does the real-time scheduler have a bug or maybe am I doing something
wrong?. The great problem with real-time tasks is that system's code and
data are not protected (although this case could be an error in my task's
code and data which can't be found by me). 

I expect somebody to have some idea with only this information. Soon I'll
give more details about the application I'm developing. 
Thanks a lot.



Alberto.

Abel Alberto Cuadrado Vega.
Area de Ingenieria de Sistemas y Automatica.
Universidad de Oviedo.
Spain.


--- [rtl] ---
For more information on Real-Time Linux see:
http://www.rtlinux.org/