Re: [PATCH] iio: core: make iio_device_claim_direct_mode() __must_check

[David Lechner <dlechner@baylibre.com>]

On 10/4/24 12:57 PM, Dan Williams wrote:
> kernel test robot wrote:
>> Hi David,
>>
>> kernel test robot noticed the following build warnings:
>>
>> [auto build test WARNING on 431c39f6d3edbab14f48dbf37a58ccdc0ac3be1e]
>>
>> url:    https://github.com/intel-lab-lkp/linux/commits/David-Lechner/iio-core-make-iio_device_claim_direct_mode-__must_check/20241002-233644
>> base:   431c39f6d3edbab14f48dbf37a58ccdc0ac3be1e
>> patch link:    https://lore.kernel.org/r/20241002-iio-must-check-claim-direct-v1-1-ab94ce728731%40baylibre.com
>> patch subject: [PATCH] iio: core: make iio_device_claim_direct_mode() __must_check
> [..]
>>>> include/linux/iio/iio.h:669:50: warning: ignoring return value of function declared with 'warn_unused_result' attribute [-Wunused-result]
>>      669 | DEFINE_GUARD(iio_claim_direct, struct iio_dev *, iio_device_claim_direct_mode(_T),
> 
> So I think this points to the fact that iio_device_claim_direct_mode()
> should not be using DEFINE_GUARD() in the first instance. I think
> iio_claim_direct() really wants to be using DEFINE_CLASS() directly.
> Skip usage of DEFINE_GUARD() which I now see is unable to interoperate
> with a __must_check locking function.
> 
> Perhaps the new class can be something like:
> 
>     DEFINE_GUARD_EXCL_COND()
> 
> ...which creates a guard that is exclusively conditional and has no
> unconditional flavor. However, maybe that only lives in iio unless and
> until another user arrives.

Hmm... I see what you mean. All other conditional guards (e.g.
mutex_trylock) have an unconditional counterpart (mutex_lock)
that is used with DEFINE_GUARD.