Re: possible deadlock in console_unlock

[Petr Mladek]

On Fri 2018-06-15 17:38:04, Sergey Senozhatsky wrote:
> On (06/08/18 10:18), Petr Mladek wrote:
> [..]
> > > Could be.
> > > The good thing about printk_safe is that printk_safe sections can nest.
> > > I suspect there might be locks/printk_safe sections nesting at some
> > > point. In any case, switching to a new flavor of printk_safe will be
> > > pretty easy - just replace printk_safe_enter() with printk_foo_enter()
> > > and the same for printk_save_exit().
> > 
> > We could allow nesting. It is just a matter of how many bits we
> > reserve for it in printk_context variable.
> [..]
> > In each case, I would like to keep the printk_safe context usage
> > at minimum. It has its own problems caused by limited per-cpu buffers
> > and the need to flush them.
> 
> May be. Every new printk_safe flavour comes with increasing memory
> usage.

This must be a misunderstanding. My intention was to introduce
printk_deferred() context. Where any printk() called in this
context would behave like printk_deferred(). It does not need
any extra buffers.

IMHO, this problem is similar to the problems that we solve
in scheduler and timer code. The cure might be the same.
I just suggest to introduce a context to make our life easier.


> > It is basically needed only to prevent deadlocks related to logbuf_lock.
> 
> I wouldn't say that we need printk_safe for logbuf_lock only.
> printk_safe helps us to avoid deadlocks on:
> 
> - logbuf_lock spin_lock

logbuf_lock is already guarded by printk_safe context everywhere.


> - console_sem ->lock spin_lock
> - console_owner spin_lock
> - scheduler ->pi_lock spin_lock
> - and probably something else.

printk_deferred should be enough for others. Or do I miss anything?

Best Regards,
Petr