Re: How to get a crash dump with recursive panic?

On 14/06/2014 12:32 AM, Darren Reed wrote:
> On 13/06/2014 5:35 AM, matthew green wrote:
>> Darren Reed writes:
>>> On 11/06/2014 4:33 AM, Martin Husemann wrote:
>>>> On Wed, Jun 11, 2014 at 04:19:20AM +1000, matthew green wrote:
>>>>> i haven't looked closely, but i'm pretty sure there is a bug in
>>>>> the sparc64 crash handling where it sets doing_shutdown and that
>>>>> triggers the "Skipping crash dump on recursive panic" message,
>>>>> when infact it hasn't gotten that far yet.
>>>> We should avoid leaking it, but I haven't looked closely either.
>>> So if I call panic() explicitly, that will write a crash dump?
>> i would try reverting the change i mentioned a few days ago,
>> or simply comment the inc/dec of "doing_shutdown" in the
>> sparc64 specific code and see what happens.
> Commenting out the doing_shutdown in db_interface.c allows the
> system to panic and there are no issues with lockmgr.

However forcing a panic does not lead to a happy place..

login: root
<hang, BREAK>
Stopped in pid 476.1 (getty) at netbsd:cpu_Debugger+0x4:        nop
db{1}> call panic
panic: Reader / writer lock error: rw_vector_enter: assertion failed:
!cpu_intr_p()

lock address : 0x00000000018c2fb8 type     :     sleep/adaptive
initialized  : 0x0000000001348aa0
shared holds :                  0 exclusive:                  0
shares wanted:                  0 exclusive:                  0
current cpu  :                  1 last held:              65535
current lwp  : 0x00000001092a0d00 last held: 000000000000000000
last locked  : 0x000000000133f6f8 unlocked*: 0x000000000133c754
owner/count  : 000000000000000000 flags    : 000000000000000000

Turnstile chain at 0x185daf0.
=> No active turnstile for this lock.

Skipping crash dump on recursive panic
panic: LOCKDEBUG: Reader / writer lock error: rw_vector_enter: assertion
failed: !cpu_intr_p()

<power-off required>

Darren