Re: [PATCH RFC 0/6] iio: core: Introduce cleanup.h support for mode locks

[Nuno Sá]

On Tue, 2025-12-09 at 11:05 -0600, David Lechner wrote:
> On 12/9/25 4:34 AM, Nuno Sá wrote:
> > On Sat, 2025-12-06 at 18:46 +0000, Jonathan Cameron wrote:
> > > On Thu, 4 Dec 2025 17:07:28 +0200
> > > Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote:
> > > 
> > > > On Thu, Dec 4, 2025 at 4:35 PM Nuno Sá <noname.nuno@xxxxxxxxx> wrote:
> > > > > On Wed, 2025-12-03 at 14:18 -0500, Kurt Borja wrote:  
> > > > > > 
> > > > > > In a recent driver review discussion [1], Andy Shevchenko suggested we
> > > > > > add cleanup.h support for the lock API:
> > > > > > 
> > > > > >       iio_device_claim_{direct,buffer_mode}().  
> > > > > 
> > > > > We already went this patch and then reverted it. I guess before we did not had
> > > > > ACQUIRE() and ACQUIRE_ERR() but I'm not sure that makes it much better. Looking at the
> > > > > last two patches on how we are handling the buffer mode stuff, I'm really not convinced...
> > > > > 
> > > > > Also, I have doubts sparse can keep up with the __cleanup stuff so I'm not sure the
> > > > > annotations much make sense if we go down this path. Unless we want to use both
> > > > > approaches which is also questionable.  
> > > > 
> > > > This, indeed, needs a (broader) discussion and I appreciate that Kurt
> > > > sent this RFC. Jonathan, what's your thoughts?
> > > 
> > > I was pretty heavily involved in discussions around ACQUIRE() and it's use
> > > in CXL and runtime PM (though that's still evolving with Rafael trying
> > > to improve the syntax a little).  As you might guess I did have this use
> > > in mind during those discussions.
> > > 
> > > As far as I know by avoiding the for loop complexity of the previous
> > > try we made and looking (under the hood) like guard() it should be much
> > > easier and safer to use.  Looking at this was on my list, so I'm very happy
> > > to see this series from Kurt exploring how it would be done.
> > > 
> > > Sparse wise there is no support for now for any of the cleanup.h magic
> > > other than ignoring it.  That doesn't bother me that much though as these
> > > macros create more or less hidden local variables that are hard to mess
> > > with in incorrect ways.
> > > 
> > > So in general I'm very much in favour of this for same reasons I jumped
> > > in last time (which turned out to be premature!)
> > > 
> > > This will be particularly useful in avoiding the need for helper functions
> > > in otherwise simple code flows.
> > > 
> > 
> > Ok, it seems we are going down the path to introduce this again. I do agree the new ACQUIRE()
> > macros make things better (btw, I would be in favor of something similar to pm runtime). Though
> > I'm still a bit worried about the device lock helper (the iio_device_claim one). We went through
> > some significant work in order to make mlock private (given historical abuse of it) and this
> > is basically making it public again. So I would like to either think a bit harder to see if we
> > can avoid it or just keep the code in patches 5 and 6 as is (even though the dance in there is
> > really not pretty).
> > 
> > At the very least I would like to see a big, fat comment stating that lock is not to be randomly
> > used by drivers to protect their own internal data structures and state.
> > 
> > - Nuno Sá
> 
> Due to the way that conditional guards only extend regular guards, I don't
> think there is a way to not expose the basic mlock wrapper. So "don't use this
> unless you really know what you are doing" docs seem like the best option.

Right! I figured my first option would be very unlikely... But for the comment I hope we can
elaborate a bit more. Like "don't use this lock to protect your own driver state/data ... you might
need this together iio_buffer_enabled() and if for some reason you cannot use the claim helpers).

- Nuno Sá