Re: [lkcd-devel] xlcrash cross-architecture dump header access

[Andreas H. <AHE...@de...>]

"Jan Beulich" <JBe...@no...> wrote:

> Hmm, interesting. This is at least something. Nothing exists for the
> architecture specific headers, though,

Ok, now I've got it.  This did not attract my attention yet, because
up to now I did not really debug little endian dumps with big endian
lcrash.  (s390-dumps do pretty well with i386-lcrash because there is
no arch specific dump header at all ;-)

Of course, this is a bug.

>  and having to explicitly call
> a second function after reading the header is not very straitforward

Well, I could say it is just a matter of taste to first read the whole
dump header from file and afterwards swap each element of the header
or read and byte swap each header element individually.

But as a matter of fact the dump accessing routines are not suitable
to access the dump header.
Neither you need address translation (kl_vread_* routines) nor you
need decompression (kl_read_* routines) to access the dump header.
That is the reason why the dump header is read from file
and afterwards the elements are byte swapped.

> (or is there something you could do with an unswapped dump header?),
> so I'd still re-raise the question in the form: Along with adding
> similar functionality for the architecture sepcific headers,
> shouldn't the swapping be integrated into the reading routines?

OK, architecture specific headers have to be byte swapped.  I think
it is better not to implement another bunch of reading routines solely
for dump header access which incorporate byte swapping.

IMHO you have two alternatives:

- Read the arch specific dump header as a whole and swap bytes for all
elements subsequently

- Try to extent kl_readmem() for dump header access and read/byte-swap
all dump header elements individually using the existing dump access
routines.

> Thanks again, Jan


Regards,

Andreas