Comment 58 for bug 1276705

I've tested with kernel from comment #56.

The kernel generated too much logs for IPMI serial console (which generated too much garbage), so I switched to a real serial console (and at 115kbauds).

I've attached a archive with 3 runs (the last run it the most interesting I think):

First run with serial console and a bigger kernel log buffer (log_buf_len=8M).
The hope with larger log buffer was to catch full kernel message with dmesg once server is running.
Sadly, after about 30 minutes, server was still printing stacks.
Serial console capture attached under name "01-serial-capture-large-buffer.txt"

Second run, still with serial console but with default kernel log (no log_buf_len)
This time, kernel booted fine (with exception to disk beeing discovered after rootdelay, but a ctrl+d resumed the boot process).
Note: this boot generated WAY less message and booted. The only change is the log_buf_len=8M present or not.
Serial console capture attached under name "02-serial-capture-default-buffer.txt"

Third run, this time with serial console disabled and very large kernel log buffer (log_buf_len=32M).
Probably the most interesting one, dmesg was complete (include very first messages). It is 12 MB large !
Server booted without any issue (disk detected before the end of rootdelay).