sys_process_exec crashes on qemu-2.11.1

[jmazanec15]

I am trying to create a window manager user program in which a list of programs are passed to the window manager and the window manager forks and execs each of them in their own windows. However, there seems to be a bug in the exec call.

Initially, running this fork and exec program fails:

#include "library/syscalls.h"
#include "library/string.h"

int main()
{
	int pid = process_fork();

	if (pid == 0) {
		printf("hello world, I am the child %d.\n", process_self());
		const char *args[] = { "snake.exe" };
		process_exec("snake.exe", args, 1);
	} else {
		printf("hello world, I am the parent %d.\n", process_self());
		struct process_info info;
		process_wait(&info, -1);
		process_reap(info.pid);
	}

	return 0;
}

It produces the following exception:

After doing some debugging, I have found that something goes wrong in the process_switch function. At the end of the sys_process_exec, it calls process_yield, which calls process_switch(PROCESS_STATE_READY). Here is the code for process_switch:

static void process_switch(int newstate)
{
	interrupt_block();

	if(current) {
		if(current->state != PROCESS_STATE_CRADLE) {
			asm("pushl %ebp");
			asm("pushl %edi");
			asm("pushl %esi");
			asm("pushl %edx");
			asm("pushl %ecx");
			asm("pushl %ebx");
			asm("pushl %eax");
		      asm("movl %%esp, %0":"=r"(current->kstack_ptr));
		}

		interrupt_stack_pointer = (void *) INTERRUPT_STACK_TOP;
		current->state = newstate;

		if(newstate == PROCESS_STATE_READY) {
			list_push_tail(&ready_list, &current->node);
		}
		if(newstate == PROCESS_STATE_GRAVE) {
			list_push_tail(&grave_list, &current->node);
		}
	}

	current = 0;

	while(1) {
		current = (struct process *) list_pop_head(&ready_list);
		if(current)
			break;

		interrupt_unblock();
		interrupt_wait();
		interrupt_block();
	}
	current->state = PROCESS_STATE_RUNNING;
	interrupt_stack_pointer = current->kstack_top;

	asm("movl %0, %%cr3"::"r"(current->pagetable));
	asm("movl %0, %%esp"::"r"(current->kstack_ptr));

	asm("popl %eax");
	asm("popl %ebx");
	asm("popl %ecx");
	asm("popl %edx");
	asm("popl %esi");
	asm("popl %edi");
	asm("popl %ebp");

	interrupt_unblock();
}

I have found that the program works by adding a dumby for loop before the while loop as a delay:

	for (int i = 0; i < 100000; ++i)
	{
		continue;
	}

	while(1) {

While this obviously is not a solution, I think it might indicate that there is some kind of race condition going on.

[jmazanec15]

Just adding something as simple as int t = 0; instead of adding the for loop gets rid of the exception as well.

[ethanmw]

Sounds like it's allocating the integer in both cases that fixes it, so maybe it's a stack thing, not a time thing?

[jmazanec15]

Yes I think you are right.

[dthain]

Hmm, I am not able to reproduce the problem on using the argv-repair branch:

make
qemu-system-i386 -cdrom basekernel.iso
mount 2 cdrom
cd bin
run exctst.exe

[dthain]

[dthain]

@jmazanec15 Can you double check whether you have the same problem with a fresh checkout of the argv-repair branch, and also the master branch?

[jmazanec15]

I ran the following commands on the terminal in a new repository:
git clone [email protected]:dthain/basekernel.git
git checkout origin/argv-repair

I then changed the Makefile.config to look like this so it would work on my Mac:

KERNEL_CCFLAGS=-Wall -c -ffreestanding -fno-pie -g
ISOGEN=mkisofs

# These settings select the native compiler,
# which is likely to work on native linux-x86.
#
CC=gcc -m32
LD=ld -melf_i386
AR=ar
OBJCOPY=objcopy

# If you are compiling from another platform,
# then use the script build-cross-compiler.sh
# add cross/bin to your path, and uncomment these lines:
CC=i686-elf-gcc
LD=i686-elf-ld
AR=i686-elf-ar
OBJCOPY=i686-elf-objcopy

And then ran in the terminal:
make
qemu-system-i386 -cdrom basekernel.iso

And then ran in base kernel:
mount 2 cdrom
chdir bin
run exctst.exe

And got the same error:

(I also removed the \f in the interrupt.c file so that the exception would not flush the entire screen.)

[dthain]

Urk, sounds like we have a subtle different between mac and linux.
I'm hard-pressed to understand why a debug interrupt is raised.

Some more ideas to try:

• Try running it with gdb attached, see if you can determine what is the instruction at eip 0x06fe0000
• Try changing the interrupt handler to simply return in the case of a debugging trap.

[dthain]

Er wait -- do you get the same problem when running from the master branch?

[jmazanec15]

I got the same problem running from the master branch. I also just ran it from Ubuntu and got the same problem. I will try using gdb and determine the instruction. I know Ethan setup GDB so I'll see if he can help me set that up.

[dthain]

FWIW, I'm using qemu-2.0.0, what do you have?

[jmazanec15]

It looks I am using 2.11.1 for Ubuntu and 3.0.0 for my Mac.

[dthain]

Is this one actually fixed?
I believe I observed it a few weeks back on my mac, but didn't have time to work on it then...

[jmazanec15]

My mistake. I was interpreting the output of exectest wrong. It is broken for qemu 3.0.0 as well.