Jeff Webb wrote:
> Unfortunately, I get the same results with this patch applied. I
> haven't been able to isolate the problem yet. I did discover that
> putting in a clock_nanosleep call at certain points in my simulation
> loop allowed the loop to execute to completion without any FPU errors.
> Maybe this is a clue... I do still get an FPU error message at some
> point after the main simulation loop is done, though.
Okay... I have found a simple example that exhibits the problem on my
linux-2.4 system (see attached). I am running xenomai 2.2.2 with the two RTAI
FIFO patches (and the latest FPU-related patch). I get two unresolved
math-related symbols when I try to insert this module on my FC5 2.6 machine,
so I can't comment on the linux-2.6 behavior at this time. But that's a
puzzle for another day...
When I insert the attached module, I get the following kernel log output:
Sep 15 17:30:08 kernel: fptest_init
Sep 15 17:30:08 kernel: rtf_create returned: 1
Sep 15 17:30:08 kernel: invalid use of FPU in Xenomai context at [f1f410a4]
Sep 15 17:30:08 kernel: Xenomai: suspending kernel thread efe22d30
('efe22d30') at 0xf1f410a4 after exception #7
Sep 15 17:30:08 kernel: spawned task
Removing the module yields:
Sep 15 17:30:13 kernel: fptest_cleanup
Sep 15 17:30:13 kernel: rtf_destroy returned: 0
Sep 15 17:30:13 kernel: f = 1
If I change the rtf_put command to output 500 bytes instead of 5000, the
program works as expected:
Sep 15 17:29:54 simhost kernel: fptest_init
Sep 15 17:29:54 simhost kernel: rtf_create returned: 1
Sep 15 17:29:54 simhost kernel: loop 0, f=1
Sep 15 17:29:54 simhost kernel: spawned task
Sep 15 17:29:55 simhost kernel: loop 1, f=2
Sep 15 17:29:56 simhost kernel: loop 2, f=3
Sep 15 17:29:57 simhost kernel: loop 3, f=4
Sep 15 17:29:58 simhost kernel: loop 4, f=5
Sep 15 17:29:59 simhost kernel: loop 5, f=6
Sep 15 17:30:00 simhost kernel: fptest_cleanup
Sep 15 17:30:00 simhost kernel: rtf_destroy returned: 0
Sep 15 17:30:00 simhost kernel: f = 6
Does this mean there is some FPU-related problem in the RTAI FIFO code?
Thanks,
Jeff
--
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.
/* Linux kernel includes */
#include <linux/module.h>
/* Xenomai includes */
#include <pthread.h>
#include <time.h>
#include <rtai/fifo.h>
/* RTAI FIFO numbers /dev/rtf? */
#define FIFO1 3
/* Global variables */
pthread_t fptest_task = NULL;
double f = 0.0;
/* Real-time task */
void * fptest_routine (void *cookie)
{
struct timespec dt_ts;
int i;
char c[6000];
for (i=0; i<101; i++)
{
/* Do a floating point calclation */
f = f * 1.0 + 1.1;
/* Stuff some (garbage) data in a FIFO */
rtf_put(FIFO1, c, 5000);
/* Print message to the kernel log */
if (i % 1 == 0)
{
printk("loop %d, f=%d\n", i, (int)f);
}
/* Sleep for 1 second */
dt_ts.tv_sec = 1;
dt_ts.tv_nsec = 0;
clock_nanosleep(CLOCK_REALTIME, 0, &dt_ts, NULL);
};
return NULL;
}
static int fptest_init(void)
{
int err;
pthread_attr_t attr;
printk("fptest_init\n");
/* Create an FIFO */
err = rtf_create(FIFO1, 1024*1024*1);
printk("rtf_create returned: %d\n", err);
if (err < 0)
{
printk("could not create FIFO\n");
return -1;
}
/* Create a real-time task */
pthread_attr_init(&attr);
pthread_attr_setfp_np(&attr, 1);
err = pthread_create(&fptest_task, &attr, &fptest_routine, NULL);
if (err)
{
printk("could not create thread (error code: %d)\n", err);
err = rtf_destroy(FIFO1);
printk("rtf_destroy returned: %d\n", err);
return -1;
}
else
{
printk("spawned task\n");
};
return 0;
}
static void fptest_cleanup(void)
{
int err;
printk("fptest_cleanup\n");
/* Shut down the real-time thread */
pthread_cancel (fptest_task);
pthread_join (fptest_task, NULL);
/* Destroy the fifo */
err = rtf_destroy(FIFO1);
printk("rtf_destroy returned: %d\n", err);
/* Print the floating point result */
printk("f = %d\n", (int)f);
}
module_init(fptest_init);
module_exit(fptest_cleanup);
_______________________________________________
Xenomai-help mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-help
