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

[David Lechner <dlechner@baylibre.com>]

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@gmail.com> wrote:
>>
>>> On Thu, Dec 4, 2025 at 4:35 PM Nuno Sá <noname.nuno@gmail.com> 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.