How to setup the stack?

[pcmattman]

I've run into a problem in my multitasker... whenever I create a process, it never returns (and with the current design, doing so causes exceptions). The problem is, I would like others to write their processes as though they are just functions.

At the moment, a process must do this (note that Yield is just basically resetting the processes timeslice, my multitasker is preemptive):

Code: Select all

void MyProcess()
{
	printf( "Hello World!" );

	while( true ) { Yield(); };
}

I want to be able to do this:

Code: Select all

void MyProcess()
{
	printf( "Hello World!" );
}

The multitasker sets up the stack like so:

Code: Select all

*stackSpace-- = (unsigned int) &procret;

Unfortunately, whenever a process tries to return it crashes with an "Invalid Opcode" error. I'm assuming this means that the process stack isn't setup right.

[Brendan]

Hi,

Unfortunately, whenever a process tries to return it crashes with an "Invalid Opcode" error. I'm assuming this means that the process stack isn't setup right.

You can't "return" unless there's something to return to....

Normally for C there's a run time library, with something like this:

Code: Select all

void STARTUP_ENTRY_POINT(char *commandLineString) {
    int argc;
    char *argv[];
    int status;

    parseCommandLine(&argc, &argv);
    status = main(argc, argv);
    exit(status);
}

This means when "main()" returns it returns to the C library, and the C library calls some function to terminate the task (which never returns, as the task is terminated).

Of course the code above is just an example - someone who is more familiar with C would have a better idea of what is actually in the C library's startup code. I can imagine a collection of OS dependant things, like setting up STDIN, STDOUT and STDERR, initialising the heap, installing default signal handlers, etc.


Cheers,

Brendan

[XCHG]

I suspect that the Yield() is corrupting your stack. Can you put the code for this function/procedure?

I recommend reading the value of ESP at the beginning of each of the processes' code and print it to the screen and do the same BEFORE the procedure is about to exit.

[frank]

When I set up my stack for my tasks I push the value of a Exit_Task function onto the stack. When the task finally returns it returns the to Exit_Task function which kills the task. This way I don't have to worry about calling Exit_Task from the task and there is no way the task could return to bogus memory.

[pcmattman]

That's what I'm trying to do at the moment, I just don't know how to push that onto the stack

[carbonBased]

That's what I'm trying to do at the moment, I just don't know how to push that onto the stack

Uhm... well... you would have created the stack for the task, so just write to it!

I find its easier (and more scalable) to simply have a common entry for all tasks, which accepts, as parameters, the user task entry function, and a parameter for it, as well, perhaps (similar to Brendan's suggestion):

Code: Select all

void commonEntry(void (*userEntry)(void*), void *param)
{
  // do all your initialization here
  userEntry(param);
  // do all your task clean up here
}

--Jeff

[pcmattman]

That's a really good idea...

I did create the stack, I did try to write to it but it didn't work... If you had read my first post I'm sure I put the stack writing code there...

Either way, your idea sounds much better because I can (finally) send arguments to processes .

Edit: would it work to move the process stack address into ESP, then execute push instructions, then restore the old ESP value?