On Wed, Mar 06, 2013 at 11:00:00AM -0500, Dave Anderson wrote:
 
 
 ----- Original Message -----
 > 
 > 
 > ----- Original Message -----
 > > 
 > > It just occured to me that we compile makedumpfile for ARM with following
 > > flags:
 > > 
 > > 	-D_FILE_OFFSET_BITS=64
 > > 	-D_LARGEFILE_SOURCE
 > > 	-D_LARGEFILE64_SOURCE
 > > 
 > > and looks like crash doesn't have such flags set. Do you think setting
those
 > > might help here?
 > 
 > Crash does use -D_FILE_OFFSET_BITS=64 for all the 32-bit architectures.
 > 
 > And it should be noted that the problem does not exist with natively-compiled
 > ARM crash binaries.
 > 
 > So the question is, would -D_LARGEFILE_SOURCE and -D_LARGEFILE64_SOURCE affect
 > alignment when building on an x86/x86_64 host with "make target=ARM".
 
 And the answer is: they don't make a difference.
   
 I patched configure.c like this:
 
  $ diff configure.c.orig configure.c
  138c138
  < #define TARGET_CFLAGS_ARM_ON_X86_64  "TARGET_CFLAGS=-m32
-D_FILE_OFFSET_BITS=64"
  ---
  > #define TARGET_CFLAGS_ARM_ON_X86_64  "TARGET_CFLAGS=-m32
-D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE"
  $
 
 and built with "make target=ARM".  And it fails the same way:
 
  $ ./crash --osrelease /usr/dumps/ARM/vmdump.v3
  crash: compressed kdump: cannot lseek dump vmcoreinfo
  unknown
  $ ./crash --log /usr/dumps/ARM/vmdump.v3 
 
  crash: /usr/dumps/ARM/vmdump.v3: no VMCOREINFO section
 
  $  
OK, thanks. Yeah, I suspected that it really doesn't make difference and now
both you and Jan confirmed that.
 Anyway, I've got a patch that fixes things up in a much more
rational
 manner than the first kludge/patch that I put into crash-6.1.3 for
 header_version 3. 
Good that this problem got solved :-)