Problems booting the 2nd processor

[Laksen]

Hey

I'm trying to get the second processor in my system to boot but I can't really make it do anything besides hanging.

So far I've managed to parse the SMP table seemingly correctly. The only thing I didn't get was the Apic ID specified for each processor entry. When I try to signal the Apic specified by that ID nothing happens, but if I instead signal by the logical processor number then at least it hangs

Next I've asserted and deasserted init ipi's. Then if I try to send the startup ipi with the bootpage loaded(0x1000) it just hangs.

The contents of the bootpage:

Code: Select all

[bits 16]

[org 0x1000]			;Will be loaded at address 0x1000
BOOTAPENTRY:
      lgdt [GDTDESC]

      mov eax, cr0
      or al, 1
      mov cr0, eax

      jmp GDT_DATA:GO_PMODE

[BITS 32]

GO_PMODE:
      mov ax, GDT_DATA
      mov ds, ax
      mov es, ax
      mov ss, ax

      JMP GDT_END
	

GDTDESC:
      dw GDT_END - GDT - 1
      dd GDT
GDT:
		times 8 db 0
GDT_CODE  equ  $-GDT
      dw 0xFFFF
      dw 0
      db 0
      db 0x9A
      db 0xCF
      db 0
GDT_DATA  equ  $-GDT
      dw 0xFFFF
      dw 0
      db 0
      db 0x92
      db 0xCF
      db 0
GDT_END:

Can anyone give a clue where I might be going wrong here? Another thing: Is it possible to get some debug info from QEmu while it's running?

[Combuster]

You should use Bochs - there is a prebuilt binary which supports multiple CPUs. You can compile your own that also includes a debugger which is very useful to have.

The code snippet you provided doesn't show us how the processors actually get started - which can be the actual cause of the bug.

[Laksen]

I've finally got the second processor started and am just having a few(probably solvable) problems with the gdt, but so far so good.

Seems Bochs(I haven't tested on a real machine) needs to have the physical bootpage number multiplied by 10 for it to work. If I load the code at address 0x8000, i must supply 80 in the startup ipi vector for it to hit the right address

[Brendan]

Hi,

Seems Bochs(I haven't tested on a real machine) needs to have the physical bootpage number multiplied by 10 for it to work. If I load the code at address 0x8000, i must supply 80 in the startup ipi vector for it to hit the right address

On both Bochs and real (Pentium or later) machines, the vector in the startup IPI is shifted left 8 times and loaded into CS, and IP is cleared.

For e.g. with 0x80 in the startup IPI vector, the (real mode) starting address (CS:IP) will be 0x8000:0x0000 (which is actually 0x00080000 as a flat 32-bit physical address).

I'm guessing you're getting real mode addressing mixed up with protected mode addressing...


Cheers,

Brendan

[Laksen]

Ah thanks

Yes indeed. I just skipped easily over the bootloader part by using grub so I haven't really tried that much of this real mode trickery