Re: readprofile: 0 total nan

Hi I found something intersting, and I wanted to share with all.

I copied the /proc/profile file of host (X86) to a file "my_profile" on the 
target system (XScale) and I execute the following command at the target, the
System.map file for the command is the one generated by cross compiling kernel
for XScale (target)  itself.

./readprofile -p ./my_profile | sort -nr +2 | head -2

Here is the result ....

./readprofile: profile address out of range. Wrong map file?
  5638 sock_wait_for_wmem                        22.3730
4186376 _stext                                    21.7461
   637 generic_osync_inode                        4.0833
   533 sock_kmalloc                               3.8071
   478 add_breakpoint_arm                         2.4896
  2473 icmp_send                                  2.4631
   835 sys_wait4                                  0.9237
    73 probe_irq_off                              0.7019
    57 it_real_fn                                 0.6786
    21 complete_and_exit                          0.5250
   393 do_exit                                    0.4962
    58 sys_getitimer                              0.4833
   309 sys_capset                                 0.4739
    37 tvtojiffies                                0.4625
    52 cpu_idle                                   0.4194
    40 free_uid                                   0.3571
   257 get_branch_address                         0.3418
    88 prepare_binprm                             0.3385
    21 iq80310_write_timer                        0.3281
    81 probe_irq_on                               0.3068
-----------------------------------------------------------------

Any comments? using which we could fix this readprofile problem for
XScale/ARM? I think there is a problem in the way, kernel for XScale/ARM is 
generating /proc/profile file. 

-Thanking you
Shesha

On Tue, 8 Apr 2003, Andy Pfiffer wrote:

> On Tue, 2003-04-08 at 15:41, [e-mail] wrote:
> > I am running 2.4.18 kernel for ARM. I have one of the boot parameters
> > "profile=2". The size of the /proc/profile file is shown as 16MB. But when I
> > execute "readprofile" the output is ...  
> > 0 total                                         nan
> > 
> > If I cat the file it just give me a ".". Can anyone suggest what i am doing
> > wrong?
> 
> [ I swear I was just talking about this problem with someone else... ]
> 
> 1. /proc/profile is a binary file.  cat won't show you anything
> meaningful.
> 
> 2. the 0 output by readprofile is a problem with the automatic
> byte-order detection heuristic built into the code.  Try invoking
> readprofile with the "-n" option, and see if your problem goes away.
> 
> FYI: For those that might also run into this, the essence of the problem
> is this piece of code in readprofile.c (fragmented for clarity):
> 
> "optNative" is 0.
> "buf" is an unsigned int *.
> 
> 
> if (!optNative) {
>         int entries = len/sizeof(*buf);
>         int big = 0,small = 0,i;
>         unsigned *p;
> 
>         for (p = buf+1; p < buf+entries; p++)
>                 if (*p) {
>                         if (*p >= 1 << (sizeof(*buf)/2)) big++;
>                         else small++;
>                 }       
>         if (big > small) {
>                 fprintf(stderr,"Assuming reversed byte order. "
>                    "Use -n to force native byte order.\n");
>                 <snipped>
>                 .
>                 .
>                 .
>         }       
> }
> 
> Based on my read of the code, "big" will be incremented if *p >= 4,
> otherwise "small" will be incremented.  I can't see how this would ever
> detect the byte order...
> 
> Werner proposed this as a solution, but it could still fail to correctly
> detect the byteorder (although with much less frequency):
> 
> 
>