null pointer dereference in GC with LDC 0.9.2 on FreeBSD

Hi sogho, this sounds bad.

Are you in a position to help with further debugging this? FreeBSD is somewhat sparse among us main developers. The LDC folks may also be able to help.

Replying to larsivi:

Hi sogho, this sounds bad. Are you in a position to help with further debugging this? FreeBSD is somewhat sparse among us main developers. The LDC folks may also be able to help.

Hi larsivi, thanks for your quick reply.

I will report this issue at the LDC's site for further debugging.

I think it's a Tango issue. The stack bottom is probably wrong. See http://dsource.org/projects/tango/browser/trunk/tango/core/rt/compiler/ldc/rt/memory.d

Afaik, it was working for me (speaking of Hello World). Well, maybe smth. was messed up. I will try to do some test to find out what is going wrong.

Btw.: Here is the link to the ldc ticket: http://dsource.org/projects/ldc/ticket/404

sysctl version

sysctlbyname version

I have created patches to make the "Hello World" program run. I attached three types of patches as a starting point. But I do not know whether these solutions are appropriate. I would appreciate if you would review the patch and refine the code.

Thank you.

Thanks for the contribution!

So far, I've made a test case. (FreeBSD 7.2 i386)

import tango.io.Stdout;

import tango.sys.freebsd.consts.sysctl; 
import tango.stdc.stdlib;

extern (C) int sysctl(int *, uint, void *, size_t *, void *, size_t); 
extern (C) int sysctlbyname(char *, void *, size_t *, void *, size_t); 

extern (C) extern void* __stack_end_address_from_main_func;

enum SysCtl
{
CTL_KERN = 1,
KERN_USRSTACK = 33
}

void* se1()
{
        int    mib[2];
        uint   userStack;   // vm_size_t
        size_t len;
        int    retval;

        mib[]  = [ SysCtl.CTL_KERN, SysCtl.KERN_USRSTACK ];
        len    = userStack.sizeof;
        retval = sysctl(cast(int*)&mib, 2, &userStack, &len, null, 0);
        if (retval < 0)
        assert(false, "cannot get the stack end address");

        return cast(void*)userStack;
}

void* se2()
{
        uint   userStack;   // vm_size_t 
        size_t len; 
        int    retval; 
 
        len    = userStack.sizeof; 
        retval = sysctlbyname("kern.usrstack", &userStack, &len, null, 0); 
        if (retval < 0) 
        assert(false, "cannot get the stack end address"); 
 
        return cast(void*)userStack;
}

void main()
{
        Stdout("se0: ")(cast(ulong) __stack_end_address_from_main_func).newline;
        Stdout("se1: ")(cast(ulong) se1()).newline; 
        Stdout("se2: ")(cast(ulong) se2()).newline;

        uint x;
        uint y;
        Stdout("x:   ")(cast(ulong) cast(void*) &x).newline;
        Stdout("y:   ")(cast(ulong) cast(void*) &y).newline;
        void* a = alloca(16);
        Stdout("a:   ")(cast(ulong) a).newline;
}

se0: 3217026232
se1: 3217031168
se2: 3217031168
x:   3217025972
y:   3217025968
a:   3217025896

thanks sohgo :)

Which patch is the appropriate one, mwarning?

stack-end-ldc-freebsd-sysctl.patch + more complete sysctl.d

stack-end-ldc-freebsd-sysctl.patch + more complete sysctl.d

thank you

(In [5464]) fixes #1907 :: null pointer dereference in GC with LDC 0.9.2 on FreeBSD

kudos to mwarning & sohgo :)

Thank you, mwarning & kris.