Re: [PATCH v3 00/16] mod_devicetable.h: Split into per subsystem headers

["Uwe Kleine-König (The Capable Hub)" <u.kleine-koenig@baylibre.com>]

[-- Attachment #1: Type: text/plain, Size: 3447 bytes --]

Hello Takashi,

On Tue, Jun 30, 2026 at 08:58:04AM +0900, Takashi Sakamoto wrote:
> On Sun, Jun 28, 2026 at 11:58:35PM +0200, Uwe Kleine-König (The Capable Hub) wrote:
> >  ...
> >  drivers/firewire/core-device.c                |    1 -
> >  drivers/firewire/net.c                        |    1 -
> >  drivers/firewire/sbp2.c                       |    1 -
> >  ...
> >  drivers/media/firewire/firedtv-fw.c           |    1 -
> >  ...
> >  include/linux/firewire.h                      |    3 +-
> >  ...
> >  sound/firewire/isight.c                       |    1 -
> >  ...
> >  1649 files changed, 1575 insertions(+), 2544 deletions(-)
> >  ...
> >  create mode 100644 include/linux/device-id/ieee1394.h
> 
> I have no objection to the above changes relevant to firewire subsystem.
> 
> Acked-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>

Thanks! Added to my tree.

> Consequently, unit drivers just include include/linux/firewire.h to refer 
> to 'struct ieee1394_device_id'. After merging the changes, I'll post more
> patches to modify the following unit drivers, since they are left from the
> patchset.
> 
> * drivers/media/firewire/firedtv.h
> * sound/firewire/bebob/bebob.h
> * sound/firewire/dice/dice.h
> * sound/firewire/digi00x/digi00x.h
> * sound/firewire/fireface/ff.h
> * sound/firewire/fireworks/fireworks.h
> * sound/firewire/motu/motu.h
> * sound/firewire/oxfw/oxfw.h
> * sound/firewire/tascam/tascam.h
> 
> By the way, the changes touch so many files. If it takes more time to
> apply, I think it better to make steps to modify across several kernel
> generation.

Touching so many files is the motivation to do all the high-impact
changes in a single go. As of v7.2-rc1 <linux/mod_devicetable.h> is
included (transitively) in ~18000 of 21500 .o files below drivers/ for a
x86_64 allmodconfig[1]. So touching that file requires a near complete
rebuild. If we spread the changes over several trees/kernel releases the
need to recompile the whole tree repeats more often than necessary. Note
this bites not only during the merge window when building the
development tree, but also during bisection.

> For example, getting rid of direct reference of
> mod_devicetable.h from drivers at first (in the case of firewire
> subsystem, for 'Replace <linux/mod_devicetable.h> by more specific
> <linux/device-id/*.h> (headers))', then adding these per-subsystem
> headers and applying relevant changes.

The impact of these firewire changes is limited to the firewire
subsystem. So doing these later is a smaller issue. I plan to tackle
these when this series is in. (But if you want to care about the
firewire bits, that's fine, too.)
 
> Of course, I don't mind to apply the v3 patchset as is in upstream.

There are a few changes that I collected (Added SPDX markers, a few less
#includes added as my script evolved, a new patch to fix a hppa
fallout), so I will send a v4 later today.

Best regards
Uwe

[1] I did in my build tree (with the series applied)

	$ find drivers -name \*.o.cmd -not -name \*.mod.o.cmd | xargs grep -l /device-id/ | wc -l
	19221
	$ find drivers -name \*.o.cmd -not -name \*.mod.o.cmd | wc -l
	21597

That's a bit wrong as my series changes the numbers, but the ballpark
should be right. In
https://lore.kernel.org/all/20260622210733.1743063-2-u.kleine-koenig@baylibre.com/
I found similar numbers for 7.1 (17038/21330).

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]