On Tue, 9 Mar 2004 11:41:44 -0800 (PST), Yi Sun <[EMAIL PROTECTED]> wrote: >Hi There, >KDB prints following messages, can anyone tell me that in which case LKCD will hang? > >>This kernel supports KDB but LKCD was invoked directly, not via KDB. >>Falling back to the old and broken LKCD method of getting data from all >>cpus, >>do not be surprised if LKCD hangs.
When lkcd is invoked directly, it uses its own code to try to silence and restart the system. That code is unreliable, especially on ia64. It has been known to deadlock, to spin forever waiting for other cpus to respond and even to oops when leaving lkcd (which is one of the reasons that lkcd does not reboot reliably). kdb has much more reliable code for stopping and restarting the other processors, it knows about all the nasty corner cases on ia64. That reliability comes at a cost, including a 600+ line patch to the ia64 MCA handler. Instead of duplicating that code in lkcd, I put a hook in to call lkcd from kdb so lkcd can take advantage of all the work that kdb has done. Since you have both lkcd and kdb, do not use sysrq-c to take a dump. Drop into kdb first then type 'sr c'. That invokes the same code as sysrq-c but in a clean state. --------------------------- Use http://oss.sgi.com/ecartis to modify your settings or to unsubscribe.
