From: Keith Owens <kaos@sg...> - 2006-06-02 01:14:12
Tom Morano (on Thu, 01 Jun 2006 09:47:42 -0700) wrote:
>Attached is a patch derived from a sugestion from Keith Owens. It puts
>in place a check to ensure that runaway unwinds (of task backtraces)
>don't happen for ia64 dumps. The patch imposes a hard coded limit of 200
>stack frames. Unless there is a likelihood that a task will exceed this
>limit (in kernel mode), having a hard line should be sufficient. If this
>turns out not to be the case, then a new command line option to lcrash.
>Please let me know if you see any problems with this solution.
Years of KDB usage has shown that the maximum sensible kernel backtrace
has ~65 entries, I have never seen one that even reached 100 entries.
200 leaves plenty of room without adding yet another knob to be
BTW Tom, I strongly recommend doing the check on all architectures, not
just ia64. With i386 and x86_64 moving to dwarf unwinding, they will
have the same potential problems as ia64. Corrupt or missing unwind
info leads to loops.
Get latest updates about Open Source Projects, Conferences and News.