I've seen this crash twice recently:
Bad Address (null pointer deref?): Code=6 (Instruction TLB miss fault) at addr 00000000e5361d5f
Kernel panic - not syncing: Bad Address (null pointer deref?)
The message indicates parisc_terminate() was from do_page_fault(). It would seem either !mm or
!user_mode(regs). Might be an exception that wasn't handled by fixup_exception().
In parisc_terminate(), we have:
printk("\n");
pr_crit("%s: Code=%d (%s) at addr " RFMT "\n",
msg, code, trap_name(code), offset);
show_regs(regs);
spin_unlock(&terminate_lock);
/* put soft power button back under hardware control;
* if the user had pressed it once at any time, the
* system will shut down immediately right here. */
pdc_soft_power_button(0);
/* Call kernel panic() so reboot timeouts work properly
* FIXME: This function should be on the list of
* panic notifiers, and we should call panic
* directly from the location that we wish.
* e.g. We should not call panic from
* parisc_terminate, but rather the oter way around.
* This hack works, prints the panic message twice,
* and it enables reboot timers!
*/
panic(msg);
dave@mx3210:~$ cat /proc/sys/kernel/printk
7 4 1 7
Since the regs weren't displayed, this implies that user_mode(regs) is true. Thus I think we have a
user mode fault with no context:
tsk = current;
mm = tsk->mm;
if (!mm)
goto no_context;
It would be helpful level in show_regs if the printk log level was set to KERN_CRIT when this happens.
Dave
--
John David Anglin dave.anglin@xxxxxxxx
