On Tue, Feb 16, 2021 at 07:27:20PM +0000, Jonathan Cameron wrote:
> On Tue, 16 Feb 2021 11:16:03 +0900
> William Breathitt Gray <vilhelm.gray@gmail.com> wrote:
> 
> > On Sun, Feb 14, 2021 at 06:19:46PM +0000, Jonathan Cameron wrote:
> > > On Fri, 12 Feb 2021 21:13:45 +0900
> > > William Breathitt Gray <vilhelm.gray@gmail.com> wrote:
> > >   
> > > > This patch replaces the mutex I/O lock with a spinlock. This is in
> > > > preparation for a subsequent patch adding IRQ support for 104-QUAD-8
> > > > devices; we can't sleep in an interrupt context, so we'll need to use a
> > > > spinlock instead.
> > > > 
> > > > Cc: Syed Nayyar Waris <syednwaris@gmail.com>
> > > > Signed-off-by: William Breathitt Gray <vilhelm.gray@gmail.com>  
> > > 
> > > Why do these need to be raw_spin_locks?
> > > Normally only need to do that if in code related to interrupt chips etc,
> > > not their use.
> > > 
> > > Jonathan  
> > 
> > This lock can be taken in an interrupt context in a subsequent patch:
> > counter_push_event() called by quad8_irq_handler() can end up calling
> > the Counter component callbacks which take this lock. We can't use a
> > mutex nor a regular spinlock because those can sleep on RT setups [1]
> 
> But on RT setups the interrupts become threads and are preemptable etc,
> so that shouldn't matter.  There are a few corner cases that cause
> trouble, but in most normal drivers you should be fine with a
> spin_lock.
> 
> Jonathan

Thinking this over again I realize you're probably right. A simple
spin_lock should be fine for these drivers, so I'll switch over to that
instead of the raw_spin_lock for this driver and the Counter interface
code.

William Breathitt Gray

> > which would result in a deadlock due to the interrupt context here -- so
> > therefore we're left with using raw_spin_lock.
> > 
> > [1] https://lwn.net/Articles/367219/
> > 
> > William Breathitt Gray
>