This patch-to-your-patch seems to fix it:

RCS file: /nfs/projects/cvs/crash/filesys.c,v
retrieving revision 1.45
diff -r1.45 filesys.c
2015c2015,2016
<       ulong files_struct_addr, fdtable_addr;
---
>       ulong files_struct_addr;
>       ulong fdtable_addr = 0;
2153c2154,2155
<       if (!fdtable_addr || !files_struct_addr || max_fdset == 0 || max_fds == 0) {
---
>       if ((VALID_MEMBER(files_struct_fdt) && !fdtable_addr) ||
>           !files_struct_addr || max_fdset == 0 || max_fds == 0) {

With fdtable_addr uninitialized, it only gets past the "if" statement
above on a pre-2.6.14 kernel if there were non-zero garbage left there
on the stack.  If there happened to be a zero there, the "if" path is
taken inadvertently, and it just prints "No open files".

But do you agree with my logic of the enclosed checking of the validity first
and then the fdtable_addr?  Seems right.

Dave

AFAICT, if we're running 2.6.14 with fdtables, it doesn't appear
that there's any possibility of fdtable_addr ever being zero?
It begins life in alloc_files() pointing to the embedded fdtable
in the files_struct, and potentially can be reassigned to a
replacement fdtable allocated from expand_fdtable().  But I don't
see anywhere that it could be cleared.

For that matter, my original check for a NULL files_struct_addr
also looks to be impossible, even in the case of kernel threads.
Maybe back in the 2.2 timeframe it could be NULL -- I just don't
remember...

In any case, I prefer to leave the code as it is above -- just to
be absolutely safe.

Dave