On Thu, 9 Nov 2006 09:20:12 -0800 , Trevor Hamm <Trevor_Hamm@pmc-sierra.com>
wrote:
> > Hmm, could you try init=/bin/sh? If the shell invoked successfully it
> > might be COW issue.
>
> Yes, /bin/sh works. I've also run different /sbin/init programs
> successfully (sysvinit, busybox; I think I mentioned sysvinit worked
> in my original post). It's just with simpleinit from util-linux
> 2.12r that we've seen this issue.
Oh, I had missed that point. And I realize you said the fault was
happend in /sbin/init itself. So it should not be COW issue.
> > In this case, could you try deleting
> > __HAVE_ARCH_COPY_USER_HIGHPAGE in include/asm-mips/page.h?
>
> Okay, I did this, but /sbin/init still hangs in the same place. I
> also had to delete the copy_user_highpage function in
> arch/mips/mm/init.c to get the kernel to build. It's now using the
> copy_user_highpage from include/linux/highmem.h. I assume this was
> your intention?
Yes, then copy_user_highpage would not be guilty. Your trial
confirmed this. Thanks anyway.
So now I doubt flush_dcache_page/update_mmu_cache change caused your
problem, which was happen during 2.6.17.8 and 2.6.17.9. This is a bit
inconsistent from your analysis (2.6.17.10 was OK), but in general
cache troubles are very sensitive anyway...
Could you confirm that removing whole "if (mapping ..." block from
__flush_dcache_page can hide your problem?
Or if you changed a line in __update_cache():
int exec = (vma->vm_flags & VM_EXEC) && !cpu_has_ic_fills_f_dc;
to
int exec = 1;
then your problem still happen?
---
Atsushi Nemoto
|