Re: [RFC] Recursive printk

[Tejun Heo <tj@kernel.org>]

Hello,

Nick Andrew wrote:
>> Tejun had a thing a while ago which was kinda intended to solve the
>> same problem.  iirc his approach added a lot more code (bad), but
>> didn't go and add strange new semantics to printk.
> 
> I'm not trying to solve a huge problem (e.g. as Jason Baron's dynamic
> debug does), just a small problem. Why do tasks have to choose
> between double-buffering their messages and making multiple calls
> to printk?

Yeah, my patch happened to solve about the same problem (and mprintk
proper was ~370 lines, so it wasn't too bad).  Mine went the way of
assembling messages piece-by-piece and most of the complexities came
from buffer management and fallback (e.g. even if kmalloc fails to
allocate messages, messages still get printed out, albeit a bit
uglier).

>> IOW, for this to be halfway as useful as you expect, we need a
>> look-out-for-local-printk-hacks maintainer.
> 
> If my patch makes it in, it will be followed by fixes to all
> those local hacks. We might discover some common element so they
> can be made non-local. Once a better way to interface to printk
> becomes prevalent, it will probably be reused. And if it's not,
> then it affects only the one subsystem.

No matter which mechanism, I'd love to have something to solve this in
kernel.  I face this problem quite often and each and every time I
have to come up with some custom hack to work around it.

FWIW, this was the last take of the mprintk patchset.

  http://thread.gmane.org/gmane.linux.ide/28415

Thanks.

-- 
tejun