Welterlen Benoit (on Wed, 22 Mar 2006 17:57:42 +0100) wrote: >I made a first experimental patch to apply on Nan hai´s patch for 2.6.15.4. > >I registered an INIT notifier for monarch and slave cpus to call kexec >and restart on a kdump kernel using the INIT trigger. This only work on >a single processor kernel because hotplug is not reliable at this moment. > >So, I changed the panic and sysrq call to kexec to use IPI INIT instead >of just a maskable IPI. >Now, all cpus call ´crash_save_this_cpu´, but I don´t save regs from >INIT signal. How do you think you will save cpus datas ? In a specific >exported structure like in LKCD ? How will they be used by gdb ?
See the notes in Documentation/ia64/mca.txt, it applies to both MCA and INIT events. The bottom line is that you cannot trust the kernel stack when MCA/INIT is delivered so both events switch to a dedicated stack. If the previous state is satisfactory then the previous registers and saved on the old stack. mca.txt has topics "How do we get a backtrace on the tasks that were running when MCA/INIT was delivered?" and "How do we identify the tasks that were running when MCA/INIT was delivered?". _______________________________________________ fastboot mailing list [email protected] https://lists.osdl.org/mailman/listinfo/fastboot
