Re: [PATCH v4 03/11] iio: amplifiers: ad8366: remove unused include headers

[Jonathan Cameron]

On Wed, 11 Feb 2026 15:35:32 +0200
Andy Shevchenko <andriy.shevchenko@xxxxxxxxx> wrote:

> On Wed, Feb 11, 2026 at 12:55:30PM +0000, Rodrigo Alencar wrote:
> > On 26/02/10 09:57PM, Andy Shevchenko wrote:  
> > > On Tue, Feb 10, 2026 at 07:42:03PM +0000, Rodrigo Alencar via B4 Relay wrote:
> > >   
> > > > Apply IWYU principle, removing the following headers:
> > > > - linux/device.h: no usage of devm_add_action_or_reset, device_attr...
> > > > - linux/kernel.h: no usage of container_of, kasprintf, ...
> > > > - linux/slab.h: memory management handled by iio
> > > > - linux/sysfs.h: sysfs interaction is managed by iio
> > > > - linux/iio/sysfs.h: not using iio device attributes in this driver  
> > > 
> > > Yeah, but it also means to add (a lot of) missed headers...
> > > 
> > > array_size,h
> > > dev_printk.h
> > > mod_devicetable.h
> > > mutex.h
> > > stddef.h  
> > 
> > Are there proper guidelines for IWYU in the kernel?
> > Include headers end up including a bunch of others, so
> > the build finishes successfully anyways.  
> 
> This is global stuff, no need to repeat this in the kernel documentation.
> It's the same as asking documentation for KISS principle.
> 
> > I understand that the concern is build time, so we better
> > include all small parts that are needed rather than a generic
> > header that includes that and much more.  
> 
> > This is the output of the iwyu tool without this patch series:  
> 
> Have you applied the configuration Jonathan made for this tool in relation
> to the Linux kernel project? By default the tool has a lot of noise, indeed.

Worth noting that I'm still evolving my config and suspect there will still
be a bit of 'taste' applied to the output even once I conclude what the
best combinations are.  So to give my view on the following..

> 	#include <stddef.h>                     // for NULL
Not this one.

> 	#include "asm-generic/errno-base.h"     // for EINVAL, ENOMEM
Something to get to errnos is good, but not that one.

> 	#include "linux/array_size.h"           // for ARRAY_SIZE
Yes - this is part of the kernel.h split up work.

> 	#include "linux/compiler_attributes.h"  // for __aligned
Never that one, but I sometimes feel compiler.h is fine.

> 	#include "linux/dev_printk.h"           // for dev_err
Yes if device.h doesn't have to be there for other reasons.

> 	#include "linux/iio/types.h"            // for iio_chan_info_enum, iio_chan_...
No. We always need iio.h which will always include that.

> 	#include "linux/math.h"                 // for abs
yes
> 	#include "linux/minmax.h"               // for __cmp_op_max
yes

> 	#include "linux/mod_devicetable.h"      // for spi_device_id
yes
> 	#include "linux/mutex.h"                // for mutex_lock, mutex_unlock, mut...
yes
> 	#include "linux/mutex_types.h"          // for mutex
no as mutex.h will alays include that.

> 	#include "vdso/bits.h"                  // for BIT

Usually via bitops.h or similar. But if none of those are there anyway linux/bits.h

Jonathan

> 
> And note, tool != principle. The tool is just an implementation of the helper
> to enforce the principle in practice, but it may be not always suitable for
> the certain project "as is".
> 
> In the below output some are valid, but some are just noise as there are
> guarantees for the "proxying".
>