On Fri, 2005-10-14 at 03:17, Keith Owens wrote:

> I am not keen on this patch.  Corrupt data can cause commands to loop,
> generating non-stop output.  Calling touch_nmi_watchdog() in the common
> output code loses the ability to catch the looping commands.  KDB
> already calls touch_nmi_watchdog() in commands that produce a lot of of
> output and I prefer to put those calls in the long running commands
> rather directly at the output point.  That way we get protection
> against looping output, while still protecting the commands that are
> expected to generate a lot of output.
> 
> Which KDB commands are tripping the nmi watchdog?

Hi Keith,

I reproduced this problem using kdb with serial console baud rate
set to 9600, setting LINES=1000 and then doing a dmesg command. 
I expect that ps would also produce the failure.

I guess I should have mentioned that this results in a panic/hang
rather than re-entering kdb (atleast on linux-2.6.13 on x86_64).

We have seen this a few times doing real debug with the serial console
at a more reasonable 38.4K baud.

Jim Houston - Concurrent Computer Corp.


---------------------------
Use http://oss.sgi.com/ecartis to modify your settings or to unsubscribe.

Reply via email to