On 22/11/2024 07:10, Matti Vaittinen wrote:
> Hi Javier,
>
> On 21/11/2024 15:54, Javier Carrasco wrote:
>> On 21/11/2024 14:04, Matti Vaittinen wrote:
>>> Use __cleanup.
>>>
>>> The series converts the rest of the ROHM sensors (maintained by me) to
>>> use guard(mutex). This simplifies the error paths.
>>>
>>> As a note, kx022a accelerometer driver is handled in another series,
>>> which also adds support for two new accelerometers. I did also patch the
>>> driver for the BU27008 and BU27010 - but when I was testing the changes
>>> I found that the BU27008 status is set to "obsolete". I'll try to dig
>>> some information about the BU27010 and decide if having the driver
>>> in-tree is still worth the effort, or if I should just send out patches
>>> to drop it all. Hence patch to rohm-bu27008.c is not included in the
>>> series. If someone is actually using the BU27008 or BU27010 and wants
>>> to patch it - feel free to pick
>>> 131315de97ff ("iio: bu27008: simplify using guard(mutex)")
>>> from
>>> https://github.com/M-Vaittinen/linux/tree/bu27008-cleanup
>>>
>>> ---
>>>
>>> Matti Vaittinen (2):
>>> iio: bu27034: simplify using guard(mutex)
>>> iio: bm1390: simplify using guard(mutex)
>>>
>>> drivers/iio/light/rohm-bu27034.c | 73 ++++++++++------------------
>>> drivers/iio/pressure/rohm-bm1390.c | 78 ++++++++++++------------------
>>> 2 files changed, 55 insertions(+), 96 deletions(-)
>>>
>>>
>>> base-commit: adc218676eef25575469234709c2d87185ca223a
>>
>> Hi Matti,
>>
>> Both patches look good to me, but I noticed that you kept a few
>> mutex_lock() + mutex_unlock() in both drivers, in particular in the
>> cases where a scoped_guard() could simplify the code. Did you leave
>> those cases untouched on purpose?
>
> Thanks for taking a look at the patches. Much appreciated :)
>
> I remember leaving couple of direct calls to mutex_lock() and
> mutex_unlock() - but I think I left them only to places where I saw no
> real improvement by the use of guard() or scoped_guard(). It is likely I
> considered the locking in these cases being trivial. (Probably only for
> a duration of one or couple of function calls, with no error handling
> when a lock is held). The direct mutex_lock()/mutex_unlock() has no real
> room for usual errors (like leaving the function while lock was taken)
> in such case.
>
> For me,
>
> mutex_lock();
> ret = foo();
> mutex_unlock();
>
> is as clear as it gets. I don't think scoped_guard() has benefits there.
> On the contrary, for me the scoped_guard() would be more complex and
> less obvious :)
>
> Yours,
> -- Matti
>
Yes, the cases I saw had very restricted scope. I just wanted to make
sure that you left them untouched on purpose. Often such refactoring of
mutex handling opts for removing all calls of mutex_lock/unlock to avoid
mixing both approaches in the same driver.
Personally, I like the scoped_guard() for short scopes too because it is
more robust if new code is added in that scope. But that is just a
preference :)
Best regards,
Javier Carrasco
