Re: [PATCH RESEND 1/3] printk: convert byte-buffer tovariable-length record buffer

[Peter Zijlstra]

On Thu, 2012-05-03 at 12:56 -0700, Linus Torvalds wrote:
> On Thu, May 3, 2012 at 12:48 PM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > On Thu, 2012-05-03 at 02:29 +0200, Kay Sievers wrote:
> >> + * This is printk(). It can be called from any context. We want it to work.
> >
> > This is not true..
> 
> Well, it is very very nearly true, and people should basically expect it to be.
> 
> There are exceptions, but they are ridiculously specific
>  - no NMI context

With the new record design you could actually make the thing lockless
and allow writing to it from NMI context. IIRC vsnprintf() doesn't do
any allocations/locks.

>  - we've had trouble with the percpu runqueue lock (due to the stupid wakeup)

Yep :-(

>  - for similar reasons, we've had trouble with printk called from
> within some terminal locks.
> 
> But basically, it damn well should be considered a hard rule that you
> can printk from pretty much anything. The few cases where that is not
> true are big problems, and should be considered to be real problems,
> not "oh, you can't printk from here".

Thing is, with bonkers stuff like usb-console and kms/drm that's a _lot_
of code running under the logbuf/console locks.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/