Using sysarch specific syscalls in assembly?

[Daan Vreeken [PA4DAN]]

On Wednesday 10 August 2005 15:09, alexander wrote:
> I tried to write a little C app that uses sysarch and i386_set_ioperm to
> gain access to certain ports and after a bit of testing I'm pretty sure
> that there is a bug or better a timing issue with the sysarch syscall or
> the
> i386_set_ioperm procedure. Please have a look at the following code:
>
> //CODE START
>
> #include <machine/sysarch.h>
>
> int main (void) {
>
> unsigned int port = 0x378;
> unsigned char val = 'A';
> int number = 4;
>
> static inline void outb (unsigned short int port, unsigned char val) {
>         __asm__ volatile ("outb %0,%1\n"::"a" (val), "d" (port) );
> }
>
> struct i386_ioperm_args {
>         unsigned int start;
>         unsigned int length;
>         int     enable;
> };
>
> struct i386_ioperm_args *args;
> struct i386_ioperm_args arg;
> args = &arg;
>
> args->start = 0x378;
> args->length = 1;
> args->enable = 1;
>
> if(sysarch(number,args) == 0) {
> /* int i;
>    for(i=0; i < 100; i++) {
>    printf("DELAY\n");
>    }
> */
>    outb(0x378,0xF);
>    exit(0);
> }
>
> else {
>    printf("Error during syscall");
>    exit(1);
> }
> }
>
> //eof
>
> //CODE END
>
> On my PC this code will cause a core dump (Bus error: 10). If I however add
> a delay (the code that's commented out) the app will end without any
> errors.
>
> It seems FBSD needs some time to set the I/O permissions for an app. Can
> somebody test this code on his computer? Maybe this is a bug in RELENG_6.
> I'm running:
>
> FreeBSD 6.0-BETA1 #0: Mon Jul 18 03:00:45 CEST 2005

I can confirm that. I have tested the program on 5.4-RELEASE here. Testing 
your program (I called it "p") 10 times gives the following output :

root at Racebeest# for a in 0 1 2 3 4 5 6 7 8 9;do echo "starting p"; ./p ;done
starting p
starting p
starting p
Bus error (core dumped)
starting p
Bus error (core dumped)
starting p
starting p
starting p
Bus error (core dumped)
starting p
Bus error (core dumped)
starting p
starting p
root at Racebeest# 

However, opening /dev/io to gain IO privileges instead of using sysarch always 
works. I tested that with the following program :

#include <fcntl.h>

static inline void outb (unsigned short int port, unsigned char val) {
        __asm__ volatile ("outb %0,%1\n"::"a" (val), "d" (port) );
}

int main (void) {

        if (open("/dev/io", O_RDONLY) == -1) {
                printf("EEK!\n");
                exit(1);
        }

        outb(0x378, 0xff);
}

--- EOF ---

grtz,
Daan