How to write following asm for Linux rhel4

Hi,
Thanks for your reply. I want to target both x86 and x64 bit platform.
will this work on x64 platform ? Like on Win64bit i have to use some __cpuid  C++ instrinsic so the same applies for Linux64bit also ?

I've never tried it on an x86-64 platform, but it's also x86, so the instruction should be supported. Just give it a try ;)

Ok ,i will it a try and let you know.

Hi,

I tried the above solution, but it does not return any value, on Linux rhel4.tpp.com 2.6.9-5.EL i686 i686 i386 GNU/Linux machine.  where as i found a link, (http://fresh.t-systems-sfr.com/unix/privat/cpuid-20060917.src.tar.gz:a/cpuid-20060917/cpuid.c) and the same program return the desired output. But i am not able to undestand why the above program could not work, it does not return any value at all.

Could you please have a look at it, and suggest the require changes.
Thanks,

>> I tried the above solution, but it does not return any value

What do you mean by "not any value" ? On most systems, it will return 0. Is that what you get ?

>> and the same program return the desired output.

What is that desired output ?

Hi, Please have a look the below routine which i am trying to build to get the desired value, means different APIC ID's. I want to do the statatical sampling of APIC ID's.

When i tried to run the following routine, its give en error with gcc version 3.4.3 20041212 (Red Hat 3.4.3-9.EL4).
./neel: Exec format error. Wrong Architecture.

Please suggest.

#define INITIAL_APIC_ID_BITS 0xFF000000 /* EBX[31:24] unique APIC ID */
#include <stdio.h>

int get_APICID() {
    unsigned int reg_eax = 1;
    unsigned int reg_ebx = 0;
    __asm__ (
              "cpuid\n\t"
              : "=b" (reg_ebx)
              : "a" (reg_eax)
            );
    return ((reg_ebx & INITIAL_APIC_ID_BITS) >> 24);
}

int main()
{
  int n;
  for(int i=0; i<= 40; i++)
  {
       n = get_APICID();
       printf("value %d", n);
   }
}

>> ./neel: Exec format error. Wrong Architecture.

You have to compile on the same machine as you run it ... Did you do that ?

Can you also answer my questions from the post with id 20742593 ?

Hi,

The following code is working for x86 i686 architecture. Now, when i tried to run on x64_64 arch. its not giving me the same output ? Could anyone suggest what need to be taken care while porting the following asm to x86_64 bit architecture ?

#define WORD_EAX  0
#define WORD_EBX  1
#define WORD_ECX  2
#define WORD_EDX  3

#define WORD_NUM  4
unsinged int reg=1;
unsigned int  words[WORD_NUM];

     __asm__("cpuid"
         : "=a" (words[WORD_EAX]),      /* output */
           "=b" (words[WORD_EBX]),
           "=c" (words[WORD_ECX]),
           "=d" (words[WORD_EDX])
         : "a" (reg));                  /* input */
return ((words[WORD_EBX] & INITIAL_APIC_ID_BITS) >> 24)

Thanks,

>> its not giving me the same output ?

Can you show both values ?

Can you also do a sizeof(unsigned int) on the x86_64 machine ?

Please read this carefully:

You are barking up the wrong tree.  

On windows, it's quite possible to access the interrupt hardware.  Not trivial, as the system tries to protect that stuff, but possible.

On Linux, by design, the system hides all the hardware-specific stuff.   You might THINK you're doing a CPUID instruction, but all the hardware-dependent instructions generate an invisible trap to the kernel, which is free to do anything it wants.  Now in several cases, the kernel kinda emulates the instruction if it's not a dangerous one.  For example, there's not much harm in revealing some of the CPU attributes with CPUID.   But then again, no  guarantees.   The kernel might decide to present the CPU as a generic one, with some default and unrealistic values for its features.

Even if the kernel gives you some ACPID information, you generally can't do anything useful with this information.  Linux will most certainly not let you poke or peek at the control registers.    All low-level device control must be done  by the kernel and associated drivers.  users are not allowed to change critical system values as that will likely destabilize or crash the whole system.

You need to do a major rethink of what your goal is and design a way to accomplish it on Linux.  You can't just blindly convert this kind of code from Windows to Linux.

Hi,

Ok, agreed, but we can use something like:

The sample C program:

#include <stdio.h>
#define WORD_EAX  0
#define WORD_EBX  1
#define WORD_ECX  2
#define WORD_EDX  3
#define WORD_NUM  4

#define INITIAL_APIC_ID_BITS 0xFF000000

int get_APICID(){

        unsigned int words[WORD_NUM];
        unsigned int reg = 1;
        __asm__ ("pushl %%ebx      \n\t"        /* save %ebx */
                       "cpuid            \n\t"
                       "movl %%ebx, %1   \n\t"      /* save what cpuid just put in %ebx */
                       "popl %%ebx       \n\t"          /* restore the old %ebx */
                       : "=a" (words[WORD_EAX]),      /* output */
                         "=b" (words[WORD_EBX]),
                         "=c" (words[WORD_ECX]),
                        "=d" (words[WORD_EDX])
                       : "a" (reg));                               /* input */

                return ((words[WORD_EBX] & INITIAL_APIC_ID_BITS) >> 24);
}

found at : http://sam.zoy.org/blog/2007-04-13-shlib-with-non-pic-code-have-inline-assembly-and-pic-mix-well

but still i am getting an error: error: can't find a register in class `BREG' while reloading `asm'

Could you suggest where did i go wrong ? even though the site says its correct program?

Thanks in advance for your help.


 

 

I just ran Infinity08's code on my dual processor, single core system and it returned a value of 1. Nice job