Re: [ANNOUNCE] Support 'kdump' command from KDB in 2.6.27-rc2-*-2.bz2 patchset

jidong xiao wrote:
> On Sat, Aug 16, 2008 at 3:49 AM, Jay Lan <[email protected]> wrote:
>> Hi,
>>
>> The kdb 2.6.27-rc2-*-2.bz2 patchset contains implementation
>> of 'kdump' command. It was based on the original patch posted
>> by Dan Aloni last year, then modified to provide i386 support
>> by Jason Xiao. I added IA64 support. I also added hooks to
>> intercept and drop to KDB from oops.
>>
>> It looks quite different from your patch, Jason, especially
>> in kdb/kdbmain.c to a style i like better. Sorry about that.
>>
>> This implementation would catch die, panic, MCA, NMI conditions
>> and drop into KDB. After analyze the oops situation and data,
>> you can issue 'kdump' command and a kdump vmcore will be
>> created.
>>
>> I do not intercept 'echo c > /proc/sysrq-trigger' since i see
>> no need to create extra works if users already decide to create
>> a vmcore from user space. Besides, you can use KDB key sequence
>> to break into KDB and do a 'kdump' command to take a dump as well.
>>
>> Doing a 'go' after panic is undefined, and it also depends on
>> the value of CONFIG_KDB_CONTINUE_CATASTROPHIC. So, do 'kdump'
>> after panic if you want take a vmcore.
>>
>> I have tested on IA64 and X86_64 to see a kdump kernel booted
>> up and /proc/vmcore created. Due to bugs of makedumpfile and
>> crash against the latest kernels, i did not run crash to
>> check validity of the vmcore though.
>>
>> Please report any bugs to me. Thanks!
>>
>> Regards,
>>  - jay
>>
>>
> Hi,Jay,
> 
> arch/ia64/kdb/kdba_support.c,
> 
> void
> kdba_kdump_prepare(struct pt_regs *fixed_regs)
> {
>         int i;
> 
>         /* Set on KEXEC bit on all onlinr cpus */
>         for (i = 1; i < NR_CPUS; ++i) {
>                 if (!cpu_online(i))
>                         continue;
> 
>                 KDB_STATE_SET_CPU(KEXEC, i);
>         }
> 
>         /* delaying for 5 seconds ... */
>         udelay(5*1000000);
>         machine_crash_shutdown(fixed_regs);
> }
>
> I wonder why do we need this 5-seconds-delay. Thanks.

I stole the code from ia64_init_handler() of arch/ia64/kernel/mca.c:

        /*
         * Wait for a bit.  On some machines (e.g., HP's zx2000 and
zx6000, INIT can be
         * generated via the BMC's command-line interface, but since the
console is on the
         * same serial line, the user will need some time to switch out
of the BMC before
         * the dump begins.
         */
        mprintk("Delaying for 5 seconds...\n");
        udelay(5*1000000);
        ia64_wait_for_slaves(cpu, "INIT");

Since i can not test on those machines mentioned above, i can not tell
if the delay is really necessary. But it is IA64 specific, i guess.

Have you tested the code on x86_32, Jason? I do not have an x86_32
mchine set up for kdump testing...

Regards,
 - jay

> 
> Regards
> Jason

---------------------------
Use http://oss.sgi.com/ecartis to modify your settings or to unsubscribe.