Re: [PATCH] fbmem: copy_from/to_user() with mutex held (v3)

[Harvey Harrison <harvey.harrison@gmail.com>]

On Tue, 2009-01-27 at 12:52 -0800, Andrew Morton wrote:
> On Tue, 20 Jan 2009 10:27:56 +0100
> Andrea Righi <righi.andrea@gmail.com> wrote:
> 
> > Avoid to call copy_from/to_user() with fb_info->lock mutex held in fbmem
> > ioctl().
> > 
> > fb_mmap() is called under mm->mmap_sem (A) held, that also acquires
> > fb_info->lock (B); fb_ioctl() takes fb_info->lock (B) and does
> > copy_from/to_user() that might acquire mm->mmap_sem (A), causing a
> > deadlock.
> > 
> > NOTE: it doesn't push down the fb_info->lock in each own driver's
> > fb_ioctl(), so there're still potential deadlocks somewhere.
> > 
> 
> Looks good to me.
> 
> > ...
> >
> >  static long fb_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
> > -__acquires(&info->lock)
> > -__releases(&info->lock)
> 
> Should the __acquires/__releases annotation be relocated to
> do_fb_ioctl()?  I've never actually got down and understood those
> things.
> 
> 

There's not really any point as the way lock_fb_info is done, it's essentially
impossible to annotate properly.  You could look at how spin_trylock is done,
but it will require a wrapping macro.  As I suggested the first time this
was posted.

Harvey