Eric Schrock wrote On 11/18/05 17:42,:
> The '-L' flag requires manipulating the agent LWP in the target process.
> If you mix this with the '-F' flag, you're dancing with the devil.
> You'll basically have two processes fighting over the register sets of
> the agent LWP, and it's no surprise that the target process died
> unexpectedly.  I'm not exactly sure why the pmap(1) process itself died,
> that shouldn't be the case.  In any case, don't use '-F' unless you're
> really sure what you're doing.  For example, a safe use case is if the
> target is stopped via MDB, but you want to examine detailed state.

It hadn't clicked with me that the -L was done with the agent
LWP so I guess that makes sense.

I think it would be good to mention in the revised pmap man page
that it is bad to cross the streams in such a way. People get
used to forcing procbin tools and having a target of pmap
dump out unexpectedly could be shocking to say the least.

Let me know who wants the cores.


Jon.


> - Eric
> 
> On Fri, Nov 18, 2005 at 04:43:32PM +0000, Jonathan Haslam wrote:
> 
>>Hi,
>>
>>I've been using the new MPO observability tools on several
>>projects (as a few of you on this list know). They are proving
>>to be an *extremely* useful addition to the toolkit.
>>
>>Today I've just hit an issue that I wondered if you'd seen
>>before. A colleague and I both issued a 'pmap -L' on an
>>Oracle shadow process at roughly the same time (actually I
>>forced mine). Unfortunately, not only did both pmap invocations
>>core but the Oracle shadow process did as well (along with it's
>>28GB of SGA...).
>>
>>The stack from pmap looks like:
>>
>># pstack core
>>core 'core' of 1107:    pmap -FL 900
>> 0000000000404970 get_meminfo () + 70
>> 0000000000405057 look_lgmap () + 197
>> 0000000000402267 iter_map () + 47
>> 00000000004047f1 main () + e91
>> 0000000000401fbc _start () + 6c
>>
>>
>>Have you seen this before? I can get you the cores when
>>the machine is available again. Unfortunately, it's feeling
>>poorly at the minute though and needs to be massaged to
>>resume normal service. This is snv_27, ptools-bin-0.1.2 on
>>an amd based system.
>>
>>
>>Cheers.
>>
>>Jon.
>>
>>_______________________________________________
>>observability-discuss mailing list
>>observability-discuss at opensolaris.org
> 
> 
> --
> Eric Schrock, Solaris Kernel Development       http://blogs.sun.com/eschrock


Reply via email to