Re: Deadlock between fbcon and fb_defio?

["Bruno Prémont" <bonbons@linux-vserver.org>]

Hi Jaya,

On Mon, 10 May 2010 08:00:51 Jaya Kumar wrote:
> On Mon, May 10, 2010 at 12:49 AM, Bruno Prémont
> <bonbons@linux-vserver.org> wrote:
> >
> > Fix crash if we are the first framebuffer loaded as in that case
> > fbcon wants to flush framebuffer at the end of fb registration,
> > before we have setup fb_defio.
> 
> Bruno,
> 
> Please help me understand, how does this scenario occur? I'm
> interpreting what you've written above to mean that fbcon is accessing
> the framebuffer before you've called defio_init()? Is that correct?

That was the original state as I called defio_init after
register_framebuffer() and defio_cleanup() before
unregister_framebuffer(), the opposite to the typical sequence you
detail below. Fixed by the patch.

My deadlock issue, after applying the patch, is during set_par() when I
replace the framebuffer and wish defio to start using the new page(s)
instead of the old one(s).

I want to make sure that I won't be accessing the old framebuffer after
freeing it and also that defio monitors the new framebuffer.

For that reason I defio_cleanup(), replace framebuffer, defio_init().
All of this happens while do_fb_ioctl() holds lock on fb_info and
console_sem.

Thanks,
Bruno

> The typical defio use sequence is: defio_init(),
> register_framebuffer() and the typical remove sequence is in the
> reverse order unregister_framebuffer(), defio_cleanup(). So, I don't
> see how fbcon is accessing the framebuffer either before
> register_framebuffer() completes (at which point defio init is already
> done) or after unregister_framebuffer() completes.
> 
> Thanks,
> jaya