Skip to contentSkip to conversation

Linux-2.6.6 crash on Leon2-MMU (while init loads ELF busybox)

DISCUSSION Khalfallah Karim 1 participant 1 message Jul 7, 2005

lore
Share & tools
lore
#0 lore raw

In brief:
=========
While booting Snapgear-Linux-2.6.6 on Leon2-MMU, crash of IU due to an unresolved virtual address
of the 1st ELF binary launched by init process
(it is actually in kernel mode, not in user mode yet)
In detail:
==========
The situation:
When the init process is getting ready to load the Busybox (which is in ELF format), probably to do a mount, the system wants to clear its bss section. So there's a call made to padzero() (coded in file <fs/binfmt_elf.c>) with an argument 'elf_bss' which equals the virtual base address of .bss. Currently I am using ROMfs, for I still can't handle non-versatile memory on my board (I mean at a hardware level).
The trouble:
When padzero() attempts to clear the first word of the .bss section, the virtual address does not seem to have been previously "MMU-resolved" by the system. Therefore, it crashes the IU.
My questions:
Q1: Has anyone any idea why the MMU has not been settled yet at the time the system attempts to clear the .bss section ?
Q2: I should get an exception trap for accessing a 'bad' virtual address (I'm on Sparc) - why do the machine gets straight in IU error mode ?

I have not registered yet to <lkml.org>, so please CC'me if you intend to answer.

Thank you a lot,
Aurelyen.