Re: Strange deadlock

BERTRAND Joël a écrit :
> Michael a écrit :
>> Hello,
>>
>> On Mon, 21 Jul 2014 21:30:47 +0200
>> BERTRAND Joël <[email protected]> wrote:
>>
>>>     Hello,
>>>
>>>     Some weeks ago, I have indicated here I have a stability issue
>>> with my
>>> Blade2000 running NetBSD :
>>>
>>> trap type 0x68: cpu 1, pc=410657c0 npc=410657ac
>>> pstate=0xffffffff99820092<,PEF,IE>
>>> Skipping crash dump on recursive panic
>>> panic: +fast data access MMU miss
>>> cpu1: Begin traceback...
>>> cpu1: End traceback...
>>> cpu0: shutting down
>>> cpu1: rebooting
>>>
>>>     That was with 6.99.40 kernel. I have changed a faulty memory module,
>>> rebuilt all packages I use from sources (perl, gtk and all dependencies)
>>> and rebuilt a new kernel :
>>>
>>> NetBSD legendre.systella.fr 6.99.47 NetBSD 6.99.47 (CUSTOM) #1: Sun Jul
>>> 20 15:21:37 CEST 2014
>>> [email protected]:/usr/src/sys/arch/sparc64/compile/CUSTOM
>>> sparc64
>>>
>>>     CUSTOM stands for GENERIC.MP with some minor modifications :
>>>
>>> sd0     at scsibus4 target 0 lun 0
>>> sd1     at scsibus4 target 1 lun 0
>>> cd0     at scsibus0 target 6 lun 0
>>> ...
>>> # Wedge support
>>> options         DKWEDGE_AUTODISCOVER
>>> options         DKWEDGE_METHOD_GPT
>>>
>>>     With 6.99.40 kernel, this workstation was stable enough to build
>>> packages from sources (load average : 6). With 6.99.47, it has hanged
>>> just after a few hours.
>>>
>>>     At 12:48 (CEST), this workstation has crashed under heavy I/O (gcc48
>>> compilation). Something very similar to a deadlock and I have rebooted
>>> it with power button. Console didn't respond, network subsystem didn't
>>> answer to ping... There was no information on console or serial line. No
>>> panic and no dump file.
>>>
>>>     I don't think if it is sparc specific, I don't have any other
>>> information. Has anyone ever seen the same problem?
>>
>> I've seen my sb2500 spontaneously reboot under heavy load with .47,
>> it's been rock stable with .44 ( and many, many older revisions )
>
>      In my case, it's not a reboot, only a complete deadlock and I
> haven't found any sparc64 specific modification between .40 that panics
> and reboots and .47 that locks...
>
>      Regards,
>
>      JKB

	I think I have found why my blade2000 was not stable. One of CPU was 
dying. With a new one, system is seems to be stable with a load average 
greater than 6 during several days (kernel 6.99.49).

	I only have to buy new CPU's :-(

	Regards,

	JKB