this gets overwritten

[Agares]

Hi!
In my kernel(x86, 32bit) I have some casual code to fill an array with 0's:

Code: Select all

	for (u32 i = 0; i < 1024; i++) {
		pageDirectory[i] = 0;
	}

This is what "pageDirectory" is(this code is in constructor, u32 is typdefed to unsigned int, pageDirectory is declared as private member: u32 *pageDirectory):

Code: Select all

pageDirectory = (u32 *)((((u32)&kernel_end) & 0xfffff000) + 0x1000) ;

kernel_end is declared as extern "C" u32 kernel_end and defined in my linker script(look below). My problem is, that right after the aforementioned loop, this is set to a strange value(at least that's what gdb[connected to qemu 1.4.0] says): 0x7f9dc00.
I am sure that pageDirectory is greater than &kernel_end(I've printed it to the screen, and printing'em out in gdb confirms that). What the hell am I doing wrong?
I compile the code with clang: clang -g -c -m32 -ffreestanding -fno-builtin -nostdlib -nostdinc -nostdinc++ -std=c++11 -fno-rtti paging.cpp -o paging.cpp.o
And link it with ld(from binutils compiled by myself): ~/opt/cross/bin/i586-elf-ld -L ~/opt/cross/lib/gcc/i586-elf/4.8.1/ -lgcc -T link.ld -o agos paging.cpp.o (and some more *.o files here of course)
this is my linker script:

Code: Select all

ENTRY(_start)

SECTIONS
{
	. = 1M;
	kernel_start = .;
	
	.text : ALIGN(4K) {
		*(.multiboot)
		*(.text)
	}

	.rodata : ALIGN(4K) {
		*(.rodata)
	}
	
	.data : ALIGN(4K) {
		*(.data)
	}

	.bss : ALIGN(4K) {
		*(COMMON)
		*(.bss)
		*(.bootstrapStack)
	}

	kernel_end = .;
}

[Combuster]

The appropriate type for kernel_end would technically be void because it doesn't actually store anything, it only exists to take it's reference - the language won't like it, so you'd at least declare it as something that's going to cause the least amount of errors, like declaring it as const uint8_t.

Another interesting consequence of this is that the value of &kernel_end should end up being hardcoded into the binary because that address is a constant. The compiler knows it and inserts symbol references, and then the linker fills it in for you - so if it actually changes you're doing something else wrong, like maybe actually writing to the variable instead or doing something significantly more scary.



Also, there's a minor bug in your aligning code that makes it sometimes skip unnecessary memory.

[Shaun]

is pageDirectory a class private variable? Is it global or static?

if pageDirectory is class private variable, and the class is built in runtime, and then you should know where the problem is.

[Agares]

I've changed type to u8, but it dint't change anything. The address ISN'T changing on runtime. And pageDirectory is a private variable, but it's only a pointer, so what's the difference?

[Agares]

I've checked esp after that loop and it's much less than pageDirectory address. Seriously, I don't have any more ideas. What am I missing? :C

[Shaun]

you can try declare pageDirectory as a global or static variable.

if it does not work ,paste your class related codes, please.

[Kevin]

The appropriate type for kernel_end would technically be void because it doesn't actually store anything, it only exists to take it's reference - the language won't like it, so you'd at least declare it as something that's going to cause the least amount of errors, like declaring it as const uint8_t.

This is perfectly valid C and what I usually use:

Code: Select all

extern const void kernel_end;

[Agares]

I declared pageDirectory as global and changed kernel_end type to void. Unfortunately, it didn't change anything.
Here is my class declaration(loop is in initializeDirectory, called from install):

Code: Select all

			
			class PagingInstaller {
			public:
				static const int pageSize = 0x1000;
				static const int addressMask = 0xFFFFF000;
				
				void install(Log *log, int availableMemory);
				
				void mapPage(u32 physicalAddress, u32 logicalAddress);
				void mapBlock(u32 physicalAddress, u32 logicalAddress, u32 size);
				void reloadTable();
			private:
				Log *log;
				u32 pagesToMap;
				
				void initializeDirectory();
				void mapMemory();
				void updateDirectory();
				void loadDirectory();
				void enablePaging();
			};

[Agares]

I've checked and even moving those loops to kernel_main messes up local variables. I really have no idea what is the reason. My GDT appears to be ok:

Code: Select all

Global Descriptor Table (base=0x0000000000117344, limit=24):
GDT[0x00]=??? descriptor hi=0x00000000, lo=0x00000000
GDT[0x01]=Code segment, base=0x00000000, limit=0xffffffff, Execute/Read, Non-Conforming, Accessed, 32-bit
GDT[0x02]=Data segment, base=0x00000000, limit=0xffffffff, Read/Write, Accessed

Edit:
I also tested and it appears that reordering declarations of variables(or removing some) helps. It seems to be a stack overflow, but I am quite sure that my stack is okay, this is my entry file(nasm -felf -ostart.asm.o start.asm, i tried changing from times 4096 db 0 to times 5*1024*1024 db 0, but it didn't help):

Code: Select all

MBALIGN 	equ 1<<0
MEMINFO 	equ 1<<1
FLAGS		equ MBALIGN | MEMINFO
MAGIC		equ 0x1BADB002
CHECKSUM	equ -(MAGIC+FLAGS)

section .multiboot
align 4
	dd MAGIC
	dd FLAGS
	dd CHECKSUM

section .bootstrapStack
align 4

global stackBottom
global stackTop

stackBottom:
	times 4096 db 0
stackTop:

section .text
global _start
_start:
	mov esp, stackTop

	push ebx
	push eax
	
	extern kernel_main
	call kernel_main

	cli
hang:
	hlt
	jmp hang