Re: [PATCH 1/3] printk: Release console_sem after logbuf_lock

[Peter Zijlstra <peterz@infradead.org>]

On Fri, 2011-06-10 at 11:40 +0200, Peter Zijlstra wrote:
> On Fri, 2011-06-10 at 11:33 +0200, Ingo Molnar wrote:
> > * Andrew Morton <akpm@linux-foundation.org> wrote:
> > 
> > > > Some notes:
> > > 
> > > Stupid thing doesn't explain the magical locking though :( I'm 
> > > 99.9% sure that putting an up() inside a spinlock_irq()ed region 
> > > was deliberate.
> > 
> > My guess would be it's done so that pending irqs that have queued up 
> > during our current printk-ing activities do not hit us with the 
> > console still locked.
> 
> Ah, so we already flushed the buffer, but have console_sem locked, so
> any interrupt that comes in and prints something will place it in the
> buffer but find console_sem is taken, so not flush it.
> 
> Then when we're back to doing up() the buffer is filled and nobody will
> flush it.
> 
> I guess, we can write it like:
> 
> 	spin_unlock(&logbuf_lock);
> 	up(&console_sem);
> 	local_irq_restore(flags);
> 
> which would keep interrupt disabled over up(), but have the logbuf_lock
> dropped.

Ah, not so, another cpu could start printing stuff.

So what we need is another test of log_start - log_end after up().