calcru-triggered panic?

[Scott Long]

John Birrell wrote:
> On Wed, Nov 29, 2006 at 12:04:03PM -0500, John Baldwin wrote:
>> On Monday 20 November 2006 16:38, Rene Ladan wrote:
>>> Hi,
>>>
>>> I got the attached panic on CURRENT 2006-11-17.
>>>
>>> It was preceded by a lot of 'calcru' messages.
>> The calcru messages you can ignore.
>>
>> This is the real problem:
>>
>> panic: blockable sleep lock (sleep mutex) tty 
>> @ /usr/src-current/sys/kern/kern_event.c:1670
>>
>> This panic is due to the recent changes to printf to try to buffer the printf 
>> output.  It now runs cnputc() under a critical section and syscons(4) tries 
>> to do a ttwakeup() which invokes KNOTE() and tries to lock a regular mutex 
>> while inside the critical section.  I'm not sure how best to fix this.
> 
> I would like to go back to using a printf buffer on the stack
> (like I had in my original implementation). The downside of
> doing that is that it places more stress on the stack and for
> things like cam, that is an issue, so I am told.
> 
> --
> John Birrell

It's probably less of an issue now that it used to be, since I/O is
decoupled through GEOM threads.  In 4.x, you could have a stack that
went from the syscall, through VFS, UFS, the block layer, CAM, and 
finally the device driver.  When I was working on RAIDFrame, adding
just a couple hundred bytes of stack usage would cause it to blow out.
But as I said, it might not be as much of an issue now.

Scott