Hello,
(2015/02/09 12:29), Baoquan He wrote:> On 02/09/15 at 12:22pm, HATAYAMA Daisuke wrote:
>> From: Baoquan He <bhe at redhat.com>
>>> On 02/05/15 at 05:59pm, HATAYAMA Daisuke wrote:
>>>> diff --git a/kernel/kexec.c b/kernel/kexec.c
>>>> index 9a8a01a..0ecf252 100644
>>>> --- a/kernel/kexec.c
>>>> +++ b/kernel/kexec.c
>>>> @@ -84,6 +84,8 @@ struct resource crashk_low_res = {
>>>>
>>>> int kexec_should_crash(struct task_struct *p)
>>>> {
>>>> + if (crash_kexec_post_notifiers)
>>>> + return 0;
>>>> if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
>>>> return 1;
>>>
>>> What if these two conditions !p->pid || is_global_init(p) are satisfied?
>>> Seems the behavious is changed.
>>>
>>
>> Please further follow do_exit() path. For each condition, there are
>> the corresponding panic() calls. In summary:
>>
>> oops_end
>> 1) panic() for in_interrupt()
>> 2) panic() for panic_on_oops
>> do_exit
>> 3) panic() for !p->pid (idle task)
>> exit_notify
>> forget_original_parent
>> find_child_reaper
>> 4) panic() for p->pid == 1 (init task)
>
> Yes, all conditions have been covered.
>
> So this patch is necessary, ACK it. Thanks
>
> Acked-by: Baoquan He <bhe at redhat.com>
Thanks for the patch!
I tested it in following cases on x86_64 and it worked well;
my panic notifier was called, then 2nd kernel booted.
- Null pointer dereference in each context of
- hard IRQ
- pid == 0
- pid == 1
- others with panic_on_oops=1
- Zero-divide in the context of normal process
(panic_on_oops=1)
Tested-by: Hidehiro Kawai <hidehiro.kawai.ez at hitachi.com>
--
Hidehiro Kawai
Hitachi, Yokohama Research Laboratory
