Doing a kernel in C++

(This was taken from a message on the OS-Dev message board)

Using C++ is quite easy for the most part just a few snags that have to be delt with namely the builtin functions that g++ inserts into your code use the following make line to get rid of most of them.

g++ -nostdlib -nostdinc -fno-builtin -fno-exceptions

of course it is easier to just define those options as some macro the only builtin function you will have to deal with after this will be __builtin-delete for deleting this __builtin_new for creating new objs and __builtin_vec_new for arrays of objs simply.

Simply don't use NEW and you are safe from the last two builtins and just declare this in a header file somewhere

void inline __builtin_delete(void* arg){}; 

then things should compile without errors if you want to be able to use new and new[] then you have to define the builtins correctly just like regular functions where new and new[] take a size number and an unsigned int and return a void*

i use flat binary mode for the kernel initialization stuff. Compile everything as modules using (nolib is the crap i described earlier)

g++ $(NOLIB) file.cpp -r 
the link them in the end using (this is 1 line, but split over two for display)
ld $(NOLIB) -o kernel.bin -e main -oformat binary
-Ttext 0x00100000 file1.o file2.o file3.o ...

The entry point where execution will start is int main(); i find that if you define main() after you define any other functiosn the linker doesn't produce code properly for binary output so just make main the first function in your file and make sure that it is the first file in the list of files linked by ld in the line i described earlier.

Finally -Ttext 0x00100000 tells the linker where the file will be in memory when it gets executed is at 1MB mark.


Aiyah! Whats RTTI? (Run Time Type Info)

(This info take from a message on the OS-Dev message board)

This is the way i understand RTTI...
RTTI is for Run Time Type Information. Because you are using a virtual function and inherritance, it would be possible to pass two completely different objects to the same function. Because the function is virtual the compiler cannot create two versions at link time and sort it out ahead of time, instead it has to be figured out just before it is executed.

RTTI is a means of (for lack of a better term) enumerating types while the machine runs and then providing a means to get that type information for any object (type number). I think you did the rignt thing by putting -no-rtti in the compile string but if you ever dynamically change an object from one type to another during run time you will get into trouble.

This eg should clarify for all what i think would happen (in a C equivalent manner).

print(char* string) 
...prints a string which is given as a 32bit pointer on the stack... 

print(int number) 
...prints a 32bit integer that is passed as an argument on the stack 

//now for the types 
int num1=12345; 
char name[]="ABCDEFG; 
int_or_string variabletype = "I am a string for now" 

//now call them 
variabletype = 98765; 
variabletype = "a string again"; 

The compiler can figure out a compile time that print(num1); is a call to the integer version of print() and print(name) is a call to the string version.

But because our variabletype can change from one type (class) to another it is not possible at compile time to determine which function to call until we are actually calling it.


How do I disable RTTI in GCC?

You can disable RTTI in gcc by adding the switch "-fno-rtti"


Can I use NEW and DELETE in my kernel?

(This message taken from a message on the OS-Dev message board)

The new a delete will be available if you define them as functions and it is possible to (from real 16bit mode setup a gdt and idt in a kind of crude fashion then turn on the pmode bit and long jump into the code using the right selector and offset which will start right into c++ code with objects and everything you want.