2.6.31-rc5-git2 crash on a idle system.

[Michael Ellerman]

On Thu, 2009-08-06 at 19:03 +0530, Sachin Sant wrote:
> Benjamin Herrenschmidt wrote:
> > Thanks. Since it's a memory corruption (or seems to be) however, it's
> > possible that the bisection will mislead you. IE. The culprit could be
> > somewhere else, and the commit you'll find via bisection just happens to
> > move things around in the kernel in such a way that the corruption hits
> > that code path instead of another rarely used one.
> >
> > I would suggest using printk to print out the content of memory where
> > the code appears to have been smashed at different stages during boot
> > (maybe even in the initcalls loop in init/main.c) to try to point out
> > what appears to be causing the corruption.
> >   
> By the time machine is up and running the particular memory location
> in question is already overwritten. So seems like the corruption occurs
> during the boot.
> 
> I added few printks in the initcall debug code patch. The o/p suggests
> that by the time first initicall debug message is printed the code is
> already corrupted. Further debug suggests, when start_kernel() is
> called the code at address(0xc000000000600000) is already corrupted.
> About 28 bytes of code starting from the above address is overwritten. 
> 
> I will try to add few more debug statements to find the place where
> this corruption might me happening.

Is it always the exact same pattern at the exact same address? Or does
it change and if so how?

cheers
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
URL: <http://lists.ozlabs.org/pipermail/linuxppc-dev/attachments/20090806/396f70d3/attachment.pgp>