[Bug threads/28014] gdb coredumps when remote+kgdbing a system that OOMs too hard

["simark at simark dot ca" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=28014

--- Comment #8 from Simon Marchi <simark at simark dot ca> ---
I investigated a little bit why GDB fails to get the exec displacement for your
program, and found that program header validation here fails:

https://gitlab.com/gnutools/gdb/-/blob/f1fa7a3d88561cef54dd5cf9422c29a802af6ce3/gdb/solib-svr4.c#L2625-2628

The program headers read from the core is all zeroes.  The address of the PHDR
in your core is:

    $ eu-readelf -n qemu_gdb_20210625-192829_18450.core | grep PHDR
        PHDR: 0x4000000040

The corresponding LOAD in the core, for that address, is:

    $ eu-readelf -l qemu_gdb_20210625-192829_18450.core
      Type           Offset   VirtAddr           PhysAddr           FileSiz 
MemSiz   Flg Align
      LOAD           0x005000 0x0000004000000000 0x0000000000000000 0x000000
0xa04000 R E 0x1000

FileSiz == 0 hints that this section of memory was not dumped in the core.

I then found that bit 4 of /proc/$pid/coredump_filter controls that, as
documented in the core(5) man page.  I tried flipping that bit off, generated a
core, and also got a core with FileSiz == 0.  Is it possible that your system
does not include that information in the code dumps?

Simon

-- 
You are receiving this mail because:
You are on the CC list for the bug.