On Thu, 06 Feb 2003 17:22:25 -0600, "Foris, Jim (MED)" <[EMAIL PROTECTED]> wrote: >Keith Owens wrote: >> Use the nmi watchdog, boot with nmi_watchdog=1. > >Did that... made no difference. > >We think it is because CPU0 handles all NMIs.... so if it dies, >so does the NMI watchdog.
NMI should be broadcast. Before the system hangs, cat /proc/interrupts. The NMI count on each cpu must be non-zero and increasing continuously if the nmi watchdog is working. >[1]kdb> bt >ESP EIP Function (args) >0xc46cde28 0xc01ed880 <unknown>+0xc01ed880 > kernel <unknown> 0x0 0x0 0x0 >0xc46cde2c 0xc46cdde0 <unknown>+0xc46cdde0 > kernel <unknown> 0x0 0x0 0x0 That is wierd. grep kallsyms System.map and insmod -nm some_module_name.
