Thread: Get CPU info

Hello , I have read your post on forum and it was helpful
however i cant understand some lines of volatile asm.
I googled and read http://www.ibm.com/developerworks/library/l-ia.html
However still no bulb in my head =)

Code:

mov %5, %%eax 
mov %6, %%ecx 
" mov %7, %%ebx"

What are %5 , %6, %7 ? Cant resolve it.
But i can sense that

Code:

                        " mov %%eax, %1;"
                        " mov %%ebx, %2;"
                        " mov %%ecx, %3;"
                        " mov %%edx, %4;"

%1 is *a , %2 is *b and so on so forth.

Plus ,
Why did u define a "keeper" named variable and used it in output operands if youre not gonna send it out of the "void cpuid" function ?

Originally Posted by lastsurvivor

But i can sense that

Code:

                        " mov %%eax, %1;"
                        " mov %%ebx, %2;"
                        " mov %%ecx, %3;"
                        " mov %%edx, %4;"

%1 is *a , %2 is *b and so on so forth.

The inline assembly for the GCC can't work with variables directly, so you have to use %0 for the first variable you will be using, %1 for the second and so on.
I guess that's what you already found out.
If you have Registers you're writing the value of a value to, those are input operations. Writing the register value to a variable is a output operation.
Since you're going to name the output variables first and the input ones after, input variable numbering will start at %0. Output variable numbering starts at the max input variable +1, so if we've got 5 input variables (max %4 since 0 counts), input variables will be named %5 and up.
If you've got no output variables at all, input variables will start at %0 of course

Originally Posted by lastsurvivor

Plus ,
Why did u define a "keeper" named variable and used it in output operands if youre not gonna send it out of the "void cpuid" function ?

That "keeper" variable keeps the value of register EBX before executing the CPUID instruction and is used to write that value back after the EAX, EBX, ECX and EDX register values have been saved to *a,*b,*c and *d.
If you modify the EBX register (or "clobber" it as it's called in assembly) and compile code with the -fpic option of GCC you'll get an error, because the EBX register is supposed to hold some pretty important data. (I think it's about memory adresses)
Details on PIC (Position Independent Code) can be found at http://en.wikipedia.org/wiki/Position_independent_code .
It's particularly interesting for shared libraries or making code invulnerable against some exploits because the code's memory adresses are "hidden".
So I added this in case some guys running hardened gentoo (which uses -fpic AFAIK) try to compile this with their default CFLAGS. Dunno if it's the perfect workaround, but at least I know it works.


If you're interested in the CPUID instruction, I can send you a small tool for Linux I'm developing that detects the CPU, the microarchitecture, codename, core stepping and has more sophisticated cache recognition than /proc/cpuinfo for multicore CPUs.

Peter

Originally Posted by pgzh

That "keeper" variable keeps the value of register EBX before executing the CPUID instruction and is used to write that value back after the EAX, EBX, ECX and EDX register values have been saved to *a,*b,*c and *d.
If you modify the EBX register (or "clobber" it as it's called in assembly) and compile code with the -fpic option of GCC you'll get an error, because the EBX register is supposed to hold some pretty important data. (I think it's about memory adresses)

This is all unnecessary. GCC will automatically save and restore any registers it needs to. Just make sure that EBX is listed in the clobber list or the output list.

Code:

         /* Output */ : "=a" (*a), "=r" (*b), "=c" (*c), "=d" (*d)
         /* Input  */ : "a" (*a), "c" (*c) );

I'm not -too- familiar with assembly code or what you're doing here, but it looks like you use r for the first time here, instead of b. All of the other variables seem to correspond to the pointer of the same name. I highlighted it in red above.

In case I wasn't clear, I'm fully aware I'm possibly being an idiot here. Just wanted to make sure you knew this was there and that it was supposed to be that way.

Originally Posted by JDGATX

Code:

         /* Output */ : "=a" (*a), "=r" (*b), "=c" (*c), "=d" (*d)
         /* Input  */ : "a" (*a), "c" (*c) );

I'm not -too- familiar with assembly code or what you're doing here, but it looks like you use r for the first time here, instead of b. All of the other variables seem to correspond to the pointer of the same name. I highlighted it in red above.

In case I wasn't clear, I'm fully aware I'm possibly being an idiot here. Just wanted to make sure you knew this was there and that it was supposed to be that way.

It's basically telling the compiler, "use the register indicated by asm param %1", in this case, the param of line " mov %%ebx, %1;" In other words, load %ebx into any available register.

Ah, alright! Makes sense.