Re: [PATCH] hwmon/sensors: fix SENSORS_LM75 dependencies

[Guenter Roeck <linux@roeck-us.net>]

On Tue, Jan 07, 2014 at 12:33:28PM -0400, Eduardo Valentin wrote:
> On 07-01-2014 10:21, Jean Delvare wrote:
> > Hi Eduardo,
> > 
> > On Tue, 7 Jan 2014 08:23:43 -0400, Eduardo Valentin wrote:
> >> On 07-01-2014 08:04, Jean Delvare wrote:
> >>> On Mon, 06 Jan 2014 18:26:34 -0800, Guenter Roeck wrote:
> >>>> This needs to be addressed
> >>>> in the thermal code, for example with dummy declarations if THERMAL=m but
> >>>> SENSORS_LM75=y. The functions are already declared as dummies if THERMAL_OF=n.
> >>>
> >>> This won't fly I'm afraid, the number of hwmon drivers affected will
> >>> grow in the future and you certainly don't want to have to change the
> >>> generic thermal code every time a new driver is added/converted.
> >>
> >> Agreed
> >>
> >>>
> >>> I've looked at the problem this morning and I will admit I do not even
> >>> understand what the problem is. In Randy's config, CONFIG_THERMAL_OF=y
> >>> so both thermal_zone_of_sensor_unregister and
> >>> thermal_zone_of_sensor_register are built-in. SENSORS_LM75=y so the
> >>> calls to these functions are built-in too. I just can't see how this
> >>> can be a problem at link time. Can anyone enlighten me?
> >>
> >> I believe the problem is more in the fact that THERMAL_OF is a bool, but
> >> the way it is in thermal Kconfig, it will link to the thermal module, in
> >> case CONFIG_THERMAL=m.
> > 
> > Doh. I've been looking at this for an hour and managed to miss
> > that :( Thanks for explaining.
> > 
> >> Thus I am proposing the following,
> >> which limits the user to have THERMAL_OF only as builtin and whenever
> >> is selected, it will select THERMAL too. That is:
> >>
> >>
> >> From b643aa260ed4f3514d1ca51b1ecbe4be7652a8d0 Mon Sep 17 00:00:00 2001
> >> From: Eduardo Valentin <eduardo.valentin@ti.com>
> >> Date: Tue, 7 Jan 2014 08:04:02 -0400
> >> Subject: [PATCH 1/1] thermal: fix compilation issue on CONFIG_THERMAL_OF
> >>  dependencies
> >>
> >> Users of API provided by THERMAL_OF config may suffer when
> >> CONFIG_THERMAL=y, causing linking issues, such as:
> >>
> >> drivers/built-in.o: In function `lm75_remove':
> >> lm75.c:(.text+0x12bd8c): undefined reference to
> >> `thermal_zone_of_sensor_unregister'
> >> drivers/built-in.o: In function `lm75_probe':
> >> lm75.c:(.text+0x12c123): undefined reference to
> >> `thermal_zone_of_sensor_register'
> >>
> >> Therefore, this patch limits the compilation build to always
> >> have THERMAL=y, whenever THERMAL_OF=y. In this way, whenever
> >> the API user is built, if THERMAL_OF=y, the build shall have
> >> the full thermal support, otherwise, the thermal API will provide
> >> stubs.
> >>
> >> Cc: Zhang Rui <rui.zhang@intel.com>
> >> Cc: linux-pm@vger.kernel.org
> >> Cc: linux-kernel@vger.kernel.org
> >> Signed-off-by: Eduardo Valentin <eduardo.valentin@ti.com>
> >> ---
> >>  drivers/thermal/Kconfig | 29 ++++++++++++++++-------------
> >>  1 file changed, 16 insertions(+), 13 deletions(-)
> >>
> >> diff --git a/drivers/thermal/Kconfig b/drivers/thermal/Kconfig
> >> index 58f98bd..dc315e9 100644
> >> --- a/drivers/thermal/Kconfig
> >> +++ b/drivers/thermal/Kconfig
> >> @@ -29,19 +29,6 @@ config THERMAL_HWMON
> >>  	  Say 'Y' here if you want all thermal sensors to
> >>  	  have hwmon sysfs interface too.
> >>
> >> -config THERMAL_OF
> >> -	bool
> >> -	prompt "APIs to parse thermal data out of device tree"
> >> -	depends on OF
> >> -	default y
> >> -	help
> >> -	  This options provides helpers to add the support to
> >> -	  read and parse thermal data definitions out of the
> >> -	  device tree blob.
> >> -
> >> -	  Say 'Y' here if you need to build thermal infrastructure
> >> -	  based on device tree.
> >> -
> >>  choice
> >>  	prompt "Default Thermal governor"
> >>  	default THERMAL_DEFAULT_GOV_STEP_WISE
> >> @@ -235,3 +222,19 @@ source "drivers/thermal/samsung/Kconfig"
> >>  endmenu
> >>
> >>  endif
> >> +
> >> +menuconfig THERMAL_OF
> >> +	bool
> >> +	prompt "APIs to parse thermal data out of device tree"
> >> +	depends on OF
> >> +	select THERMAL
> >> +	default y
> >> +	help
> >> +	  This options enables DT thermal API which adds support to
> >> +	  read and parse thermal data definitions out of the
> >> +	  device tree blob. This option is mostly used by embedded
> >> +	  thermal drivers.
> >> +
> >> +	  Say 'Y' here if you need to build thermal infrastructure
> >> +	  based on device tree.
> >> +
> > 
> > I suppose this works, however I believe there is value in allowing for
> > modular building of as much code as possible.
> > 
> > I have an alternative proposal, which lets thermal be built as module,
> > and hopefully also addresses the issue (I can't test...) The only
> > drawback is that the same dependency must be added for every other
> > hwmon driver which optionally uses THERMAL_OF.
> > 
> > From: Jean Delvare <khali@linux-fr.org>
> > 
> > Based on an earlier attempt by Randy Dunlap.
> > 
> > Fix SENSORS_LM75 dependencies to eliminate build errors:
> > 
> > drivers/built-in.o: In function `lm75_remove':
> > lm75.c:(.text+0x12bd8c): undefined reference to `thermal_zone_of_sensor_unregister'
> > drivers/built-in.o: In function `lm75_probe':
> > lm75.c:(.text+0x12c123): undefined reference to `thermal_zone_of_sensor_register'
> > 
> > Add depends on THERMAL since that is what provides the
> > register/unregister functions above, but only if THERMAL_OF was
> > selected as this is an optional feature of the driver.
> > 
> > Signed-off-by: Jean Delvare <khali@linux-fr.org>
> > Cc: Randy Dunlap <rdunlap@infradead.org>
> > Cc: Eduardo Valentin <eduardo.valentin@ti.com>
> 
> I agree with this approach, as it can properly serve the purpose of
> fixing this specific issue. The only drawback here is the fact that we
> need to add this change to every possible user of this API. If you want
> to push this fix via hwmon, you can add my
> Acked-by: Eduardo Valentin <eduardo.valentin@ti.com>
> 
> and I request Rui to drop the patch I sent earlier.
> 
> Now, other issue is the fact that the thermal framework config entry
> needs to be revisited, despite the link to this issue. As I mentioned in
> other email thread, there are other system requirements that may prevent
> proper usage of the thermal framework as a module. This is, I know,
> another topic, and it shall be discussed in a different thread.
> 
> > ---
> >  drivers/hwmon/Kconfig |    1 +
> >  1 file changed, 1 insertion(+)
> > 
> > --- linux-3.13-rc7.orig/drivers/hwmon/Kconfig	2014-01-07 09:01:24.812848091 +0100
> > +++ linux-3.13-rc7/drivers/hwmon/Kconfig	2014-01-07 15:19:11.039472329 +0100
> > @@ -650,6 +650,7 @@ config SENSORS_LM73
> >  config SENSORS_LM75
> >  	tristate "National Semiconductor LM75 and compatibles"
> >  	depends on I2C
> > +	depends on THERMAL || !THERMAL_OF
> >  	help
> >  	  If you say yes here you get support for one common type of
> >  	  temperature sensor chip, with models including:
> > 
Excellent idea. Looks good to me, only I think you'll need the same for TMP102.

Guenter