[gPXE] gpxe boots only with "make DEBUG=undi"

[Stefan Hajnoczi]

I have pushed a crash and hang debugging tool called fnrec (function
recorder) to my git repo.  It is an experimental tool that collects a
trace of functions called before the hang.  It will allow us to
determine which function hangs.  I hope you will try it :).

This tool is based on an idea that Thomas Miletich (meteger) and I
were playing with about a year ago.  By using the gcc
-finstrument-functions compiler flag, we can trace each function call
and record it into a memory buffer.  The memory buffer has a magic
number at the start to identify it as valid and not just random bytes
in memory.  As gPXE runs, it records each function call into the
memory buffer.  When the crash occurs and the system is reset, we
assume that memory is not cleared or overwritten (this is true on many
machines but you may have to experiment with pressing Ctrl+Alt+Del
versus the reset switch versus other alternatives).

When the system loads gPXE again, it checks for the memory buffer and
if the magic number is still there, the contents of the memory buffer
are printed out.  This output is all the recorded function calls,
which allows developers to see what code paths were taken and what the
last function called was.

Here is the commit on my 'fnrec' branch:

http://git.etherboot.org/?p=people/stefanha/gpxe.git;a=commitdiff;h=87a60a77e8f9cccceaf0c4ed55d06bc64c7ebdac

How to try it:

$ git clone -b fnrec git://git.etherboot.org/scm/people/stefanha/gpxe.git
$ cd gpxe/src
$ make bin/sky2.usb

1. Boot gPXE and reach the point where it hangs.  Now reset your
system and boot into gPXE again.
2. This time gPXE should print out lots of numbers before the banner
and prompt.  Copy these numbers into a text file (serial console is
useful for capturing this output).  If you can't use serial or easily
copy the numbers, please take at least the last two lines of numbers.

$ cat >fnrec.dat
<paste the numbers from gPXE output here>
^D
$ util/fnrec.sh bin/sky2.hd.tmp fnrec.dat

Please send the output produced by the fnrec.sh utility.  It contains
the function calls recorded before the hang.

Hacked up two days ago and tested on three machines, QEMU, and
Virtualbox.  It works on all except one machine where gPXE fails to
start and the system resets immediately.  Hopefully it will work for
you :).

Stefan