[Crash-utility] Why module's global symbol cannot be displayed in crash? [ARM]

[Dave Anderson]

----- Original Message -----
> On Wed, Mar 20, 2013 at 10:56 PM, Dave Anderson <anderson at redhat.com>
> wrote:
> >
> >
> > ----- Original Message -----
> >
> >> >
> >> > On that note, I only have a handful of sample ARM vmcore
> >> > dumpfiles,
> >> > and none of their kernels have any modules.  My supposition was
> >> > that
> >> > it is simply common usage that ARM kernels are built without
> >> > modules
> >> > because of their typically embedded use-cases.
> >> >
> >> > In any case, if somebody can make a sample ARM vmcore available
> >> > to
> >> > me that does have modules, I'd appreciate having it on hand for
> >> > testing.
> >> >
> >>
> >> Here is one such dump file: http://sdrv.ms/ZTjmUn
> >> For I don't want anyone to access it, I don't post towards the
> >> maillist.
> >>
> >> Thanks,
> >> Lei
> >>
> >
> > Got them -- thanks!
> >
> > I note that when running with an x86_64 binary that I built
> > using "make target=ARM", that I don't see data or bss symbols
> > either before or after loading the modules.  I did this:
> 
> So what it means?...
> Kernel build method is wrong?
> 
> I don't understand why module would be showed in the "before" file,
> shouldn't it only existed in "after" file for module being loaded?
> 
> Should I do anything else to check?...
> 
> Thanks,
> Lei

With respect to "mod -[sS]", it seems to be a problem that can be partially
addressed by doing this:

--- crash/symbols.c	2013-03-21 10:00:10.173201102 -0400
+++ ARM_crash-6.1.4/symbols.c	2013-03-21 10:06:01.531186455 -0400
@@ -10560,7 +10560,7 @@
                                         break;
                                 }
                         }
-                        if (i < lm->mod_sections) {
+                        if (i <= lm->mod_sections) {
                                 if (CRASHDEBUG(2))
                                     fprintf(fp, "%08lx (%c) [%s] %s\n",  
                                         (ulong)syminfo.value, 


And then it would also require filtering out the symbols types, because doing
the above also captures (U) and (a) types.  

But it's ARM-specific, though, because the patch above causes problems with
x86_64.

Dave