[Crash-utility] [crash] Analysing vmcore for a target with different endian-ness

Dave Anderson anderson at redhat.com
Tue Oct 11 13:07:30 UTC 2011



----- Original Message -----
> Hi Dave,
>
> I am working on enabling Kdump/crash support for PPC440x based chipsets.
>
> PPC44x is a family of embedded powerpc processors(BookE).
>
> Now to analyse the vmcore generated on those boards, one would want to
> run crash on his or her development machine, which is usually an x86 box.
>
> I started with this, but ended up in a problem with the endian-ness of the
> host and target. As of now this is not supported in crash.

Well, in that case, it's not so much an endian-ness issue as it is a
matching architecture issue.  There is/was a sourceforge project that 
attempted to handle that:
 
 http://sourceforge.net/projects/crosscrash/files

But it looks kind of dead...

> I did start looking at the code, trying to understand how we read the data.
> But before I jump in further I thought I would check if this is something
> that has been tried in the past ? Or something that is not feasible ?
> (We need to know the type of the data being read to convert the data)
>
> Please let me know your thoughts / inputs.

The best that the crash utility can do is this:

 - run a 32-bit x86 crash binary on an x86_64
 - run a cross-compiled 32-bit ARM crash binary on an x86
 - run a cross-compiled 32-bit ARM crash binary on an x86_64
 
And the only reason that ARM-on-x86[_64] works is because they are
both little-endian.  

Dave




More information about the Crash-utility mailing list