Multitasking

[spix]

Hi there,

I have a bug that's driving me crazy. I think I have narrowed it down to something to do with my task switching. The symptoms seem to be that on a task switch it's going to an incorrect eip and page faulting. I think my stacks are getting messed up or something.

I've been at this one bug for days, if anyone has any ideas, please tell me

here is my irq asm stubs:

Code: Select all

syscall_irq:
   push byte 0
   push byte 48

   jmp irq_common_stub


extern irq_handler


irq_common_stub:
   cld
   pushad
   push ds
   push es
   push fs
   push gs

   mov eax, [current]
   mov [eax], esp

   mov ax, 0x10
   mov ds, ax
   mov es, ax
   mov fs, ax
   mov gs, ax

   push esp

   call irq_handler

   add esp, 4

   jmp common_end

global switch_to

switch_to:
   mov eax, [current]   
   mov esp, [eax]   

   mov ebx, [eax+8];put content of the k-stack field into ebx.
   mov [morts_tss+4],ebx ;update system tss. (esp)

   mov ebx, [eax+20]
   mov cr3, ebx

common_end:

   mov eax, [esp]
   cmp eax, 40
   jl master_reg

   mov al, 0x20
   out 0xA0, al

master_reg:
   mov al, 0x20
   out 0x20, al

   pop gs    
   pop fs
   pop es
   pop ds
   popad

   add esp, 8

   iret

my task switches occur at a timer interrupt, basically it just changes the "current" task structure to a new one and calls "switch_to"

the task structure is:

Code: Select all

struct task_struct {
   uint32_t esp;
   uint32_t ss;
   uint32_t kstack;
   uint32_t kstack_mem;
   uint32_t ustack;
   uint32_t cr3;
   long state;
   long counter;
   long priority;
   uint8_t signal[NSIG];
   fn_ptr sig_restore;
   fn_ptr sig_fns[NSIG];
   int exit_code;
   unsigned long brk,start_stack;
   long pid,parent,pgrp,session,leader;
   unsigned short uid,euid,suid;
   unsigned short gid,egid,sgid;
   long alarm;
   long utime,stime,cutime,cstime,start_time;
   unsigned short used_math;
   int tty;
   unsigned short umask;
   struct ext2fs_inode_table *pwd; // TODO: should be a vfs node
   struct ext2fs_fs_data *pwd_fsd;
   struct ext2fs_inode_table *root; // TODO: should be a vfs node
   struct ext2fs_fs_data *root_fsd;
   unsigned long close_on_exec;
   struct file *filp[NR_OPEN];

   struct task_struct *next;
   struct task_struct *prev;
};

can anyone see any stupid mistakes in that?

The error seems to occur when I use my disk buffer code. To start with it only happened on real PCs and Qemu, but not Bochs or Vmware, so I rewrote my buffer code, and now it crashes on everything, so I switched back and studied it deeper, and it seems to do the wrong things but get away with it in bochs & vmware.

I have learnt thanks to Qemu & Gdb (yay OS Faq2!) that the offending eip is in my malloc function, (i am using a malloc function about the same as solar's PDClib malloc, at first I had my own malloc, but that faulted in free) but I have discovered that by adding breakpoints, that it gets to an eip in the malloc code without calling the malloc function, which made me think it's my stacks?

My stacks are 8k each, and are allocated using the malloc command. Also the error appears to happen in ring 3 and ring 0.

If anyone has any hints. please help me..

Thanks
Andrew

[Rob]

If you get inside your malloc() while you didn't call it, it sounds like you have forgotten a ret or iret somewhere. Or your call stack is indeed messing up and you end up (by chance) in your malloc() code.

That's the best I can offer I'm afraid

[Pype.Clicker]

you may want to run your code in bochs with the "trace-on" option to see from where it actually ended there ...

[spix]

you may want to run your code in bochs with the "trace-on" option to see from where it actually ended there ...

I think you told me to do that last time I had a problem, and I forgot :-/ it's good advice and led me to the source of my problem, again.

Seems it was my ext2fs code which was messed up..

Anyway, thanks again I'm writing a sticky note as we speak to use trace-on in bochs..

Andrew