Joerg Schilling wrote:
> "max at bruningsystems.com" <max at bruningsystems.com> wrote:
>
>   
>> Hi All,
>> OK.  I figured out how to get my blog stuff working in English, so you
>> can also read about the modified mdb on my blog, 
>> http://mbruning.blogspot.com/
>>     
>
> Well, let me take your use your work on mdb as a cause for thoughts on kmdb 
> usability on Solaris x86.
>
> Problem:
>
>       If you do not own a x86 system that itself implements remote 
>       administration, you need to run Solaris x86 permanently under kmdb.
>
>       This is because the only way to abort a hung x86 system is to send
>       a <break> to ttya and to let kmdb handle abortion.
>
> Pitfall:
>
>       If such a system panics for some reason, it will not auto-reboot
>       but hang until the sysadmin types e.g. ":c" at the kmdb prompt.
>
> Idea for solution:
>
>       Allow to configure kmdb to fall through to dumping core in case
>       of a panic() call and allow to configure kmdb to wait some time
>       for possible adminstrative input from ttya.
>
>
> I asume that for a simple solution it would be sufficient to prepend a 
>
>       if (!panicstr) to the for(;;) loop in kmdb_main.c::kmdb_main()
>
> Is this correct?
>   
I assume you know about the "snooping" flag and deadman timer.  This can be
useful in getting a dump when the system is "hard" hung (i.e., you may 
or may not
have kmdb loaded, but you can't get in).

I think better is a flag that says that when the system panics, bypass 
kmdb and go immediately
to the dump code.  And I think this should be the default behaviour.  I 
mostly use kmdb
for analyzing hangs, not panics.  When the system panics, you're 
generally too late anyway, and I would
rather run mdb on the dump (so I have full access to the machine resources).
The only time I am interested in kmdb other than hangs is
during development.  In this case, the flag that says to bypass kmdb on 
panic could be turned off.

Just my 2 cents.

max


> My other question: how to let a stand alone program / co-routine like
> kmdb to check for let's say 60 seconds of unsuccessful wait for input
> from stdin?
>
> Any thoughts? Any help?
>
> J?rg
>
>   


Reply via email to