Re: [PATCH kdumpid 0/3] Prevent segfault on missing disassembler

Sorry, I forgot to add one thing!

Stephen Brennan <[email protected]> writes:
> Hi Petr,
>
> I didn't know quite where to send patches for kdumpid, so I elected to
> send direct to you, and CC the linux-debuggers list. Hope that's ok!
>
> I encountered a segmentation fault of kdumpid (master branch):
>
> #0  0x0000000000000000 in ?? ()
> #1  0x0000000000406aa9 in disas_at (dd=0x7fffffffde70, info=0x7fffffffdc80, pc=0) at ppc64.c:112
> #2  0x0000000000406d42 in looks_like_kcode_ppc64 (dd=0x7fffffffde70, addr=0) at ppc64.c:174
> #3  0x0000000000405616 in explore_kernel (dd=0x7fffffffde70, fn=0x405880 <explore_utsname>) at util.c:269
> #4  0x0000000000405d3f in explore_raw_data (dd=0x7fffffffde70) at util.c:465

As you can see by the fact that the code fell into the need_explore() /
explore_kernel() path... this vmcore was missing much of the important
metadata, since it was generated by qemu without a vmcoreinfo device.

I think I shared before, but I have a small piece of code [1] which can
locate the vmcoreinfo note reasonably quickly by iterating over pages
and simply testing the first bytes against "OSRELEASE=", since the
vmcoreinfo can only occur on a page boundary. If you'd like, I could
send a patch with some sort of "explore_vmcoreinfo()" function to do
this? I know it's not necessarily the most efficient, but I believe that
it would be valuable when all other heuristics fail.

Thanks,
Stephen

[1] https://github.com/brenns10/kernel_stuff/blob/master/vmcoreinfo/dumpphys.c