How to Measure Time to execute a function in Kernel space

barunparichha · 12-19-2007, 06:51 AM

Hi,
I have written one kernel level program and
I want to measure the execution time of my kernel function inside this program. I am using linux kernel 2.6.

Can you suggest me some function to measure the start time and end time so that I can find the difference between both.

With thanks in advance,

tronayne · 12-20-2007, 03:34 PM

You probably want to use the clock() function -- be sure to read the manual page and determine that the value of CLOCKS_PER_SEC for your system is appropriate for your use.

Here is an example program that shows usage:

Code:

#include <stdio.h>
#include <stdlib.h>
#include <time.h>

void    foo     (void);

void    main    (void)
{
        clock_t t0, t1;

        t0 = clock ();
        foo ();
        t1 = clock ();
        (void) fprintf (stdout, "foo took %f secs.\n",
            (float) (t1 - t0) / CLOCKS_PER_SEC);
        exit (EXIT_SUCCESS);
}

void    foo     (void)
{
        long    i;

        for (i = 0; i < 10e6L; i++)
                ;
        return;
}

barunparichha · 12-24-2007, 03:31 AM

I need to measure time in kernel space not in user space.

ANd more over to measure acurate time in user space, you should always use "struct timeval". Because clock() can't give accracy in msec.
The code in user space is as below.

#include<stdio.h>
#include<stdlib.h>
#include<time.h>

void abc();

int main()
{
struct timeval t0,t1;
int ret;
double diff;
ret=gettimeofday(&t0,NULL);
abc();
ret=gettimeofday(&t1,NULL);
diff=(t1.tv_sec - t0.tv_sec)*1000 + (t1.tv_usec - t0.tv_usec)/1000.0;
printf("%10.5f msec",diff);

}

void abc()
{
long i=0;
while(i<100000)
i++;
}

barunparichha · 12-24-2007, 06:14 AM

For kernel space following code will work ....

#include <linux/time.h>
int init_module (void) /* Loads a module in the kernel */
{
struct timeval t0,t1;
long i=0;
float diff;
double temp;
do_gettimeofday(&t0);

//Your code goes here
while(i<1000)
{
//printk("%ld",i++);
}

do_gettimeofday(&t1);

temp = (double) (t1.tv_sec - t0.tv_sec)*1000 + (t1.tv_usec - t0.tv_usec)/1000;
return 0;
}

mat_the_w · 01-07-2008, 11:48 AM

I'm pretty sure that you aren't supposed to use floating point calculations in kernel code, as the kernel doesn't keep track of the FPU's state when the kernel is running code.

When I just tried to use the code

temp = (double) (t1.tv_sec - t0.tv_sec)*1000 + (t1.tv_usec - t0.tv_usec)/1000;

in a device driver, it complied just fine, but the linker(?) complained that:

*** Warning: "__adddf3" [drivers/matthew/test1/test1.ko] undefined!
*** Warning: "__muldf3" [drivers/matthew/test1/test1.ko] undefined!
*** Warning: "__floatsidf" [drivers/matthew/test1/test1.ko] undefined!

--mat_the_w