problems about multitasking in diffrent rings

[one737]

I was trying to multitasking some process in ring3 and ring1. If there's only ring3 processes or ring1 processes, there's nothing at all.
But when I tried to make there's both ring1 and ring3 process, it'll cause a GP error, and the eflags register has turned into 0x96.
The lines causes error is main.c, line 36-37. (I'm pretty sure about that)

P.S.: I used a 64-bit elf toolchain under windows to compile
P.P.S: I tried to put the keyboard driver in ring3 too, but when I press the CapsLock/NumLock/ScrollLock, it gives me another GP

[devc1]

Where is the irq handler of task scheduling (in whicj file from your project) ? process_start seems fine, it's probably something with your scheduler that is causing a GPF when running a usermode process.

[one737]

Where is the irq handler of task scheduling (in whicj file from your project) ? process_start seems fine, it's probably something with your scheduler that is causing a GPF when running a usermode process.

task scheduling's irq handler is the hwint00() in kernel.asm, it turns out to be a wrapper of irq_table[0], and that's the clock_handler() in clock.c (set by init_clock() ). at the end of this function, it calls schedule() to do the actual task schedule.

[devc1]

Code: Select all

%macro hwint_master 1
    call save
    in al, INT_M_CTLMASK
    or al, (1 << %1)
    out INT_M_CTLMASK, al
    mov al, EOI
    out INT_M_CTL, al
    sti
    push %1
    call [irq_table + 4 * %1]
    pop ecx
    cli
    in al, INT_M_CTLMASK
    and al, ~(1 << %1)
    out INT_M_CTLMASK, al
    ret
%endmacro

This is what gets called ?
You should return from an interrupt using the iret instruction to make a context switch and change the code segment, just imagine with me : An interrupt occurs and the current task is in user mode, the CPU switches to kernel mode and executes the ISR, it saves the previous EIP and the code segment along with other segments in the stack frame.

But an interrupt doesn't just push EIP : Instead it pushes EIP, CS, EFLAGS, ESP, SS.
You should use iret to context switch to user mode and set CS with the right segment.

I wrote an example in the forum of the "WinExperiment" guy who asked about a syscall problem.
You can find it here : viewtopic.php?f=1&t=56515&start=15

Note that this is a task scheduler ISR and you are not required to use the same mechanism with other ISRs.

[Octocontrabass]

The lines causes error is main.c, line 36-37. (I'm pretty sure about that)

It's kernel.asm line 166. Looks like something is corrupting your stack.

Do you have one ring 0 stack per thread? If you do, your interrupt handling code is much more complicated than it needs to be.

[one737]

This is what gets called ?
You should return from an interrupt using the iret instruction to make a context switch and change the code segment, just imagine with me : An interrupt occurs and the current task is in user mode, the CPU switches to kernel mode and executes the ISR, it saves the previous EIP and the code segment along with other segments in the stack frame.

In the save() function, I saved [esp] as restart (if interrupt reenter occured, then restart_reenter), and it finally calls iretd, then it'll goes like what you said.

[one737]

Do you have one ring 0 stack per thread? If you do, your interrupt handling code is much more complicated than it needs to be.

If there's only ring1 or ring3 processes, everything's ok, so I don't think is a stack problem.

[Octocontrabass]

I don't think is a stack problem.

The instruction in kernel.asm at line 166 is "pop fs". It's a stack problem.

[one737]

So why there's totally nothing happened while only ring1/ring3?

[Octocontrabass]

You'll have to use a debugger to figure that out. Your code to manipulate the stack doesn't make sense to me.

[devc1]

In the save() function, I saved [esp] as restart (if interrupt reenter occured, then restart_reenter), and it finally calls iretd, then it'll goes like what you said.

But do you and how you save previous task's CS, EFLAGS, ESP and SS ?

[one737]

But do you and how you save previous task's CS, EFLAGS, ESP and SS ?

They're in the STACK_FRAME structure in the PCB (struct PROCESS).

[devc1]

Good.

Now how do you build the stack frame before returning with iretd ?


You're doing it inside the function called with call instruction right ?

Intel manual : The call instruction sets the Nested Task flag(NT). Iret will generate a #GPF if the NT Flag is set. You must iret outside a function, it is not recommended to fix it by clearing the NT Flag manually though,

[one737]

How do you build the stack frame before returning with iretd ?

After fill in the PCBs, I call the restart(), it left the stack exactly with the eip, cs, eflags, esp and ss, then I use the iretd to enter into the process.

[one737]

You'll have to use a debugger to figure that out. Your code to manipulate the stack doesn't make sense to me.

I checked my code, and found out that I do allocated an unique stack to every process.